Rob ae5738ea2f spec: §5.9.8 reactions + §5.9.9 reply-to (app extensions) + §5.5 raw-wire-bytes rule

Resolves #8. Documents the convergent fields[16] / fields[0x30]+[0x31]
shapes shared by reticulum-mobile-app, MeshChatX, and Columba.

§5.5 gains the normative line that msgpack_payload MUST be hashed from
raw wire bytes — without it, reactions/replies miss the relay rewrite
cache for any message whose fields don't round-trip byte-identically.

§5.9.8 (tap-back reactions on fields[16]) and §5.9.9 (reply-to on
fields[0x30] + optional fields[0x31]) carry > UNVERIFIED markers since
these keys are outside the upstream LXMF allocation range. Columba's
legacy fields[16] = {reply_to} shape is recorded as an inbound-only
tolerance branch with a finite lifetime — torlando-tech/columba#926
shipped the dual-key shape in v2.0.0-beta on 2026-05-23.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

2026-05-24 08:36:39 -04:00

274 KiB

Raw Blame History

Reticulum Wire Specifications

A byte-level reference for implementing Reticulum-compatible clients. This document focuses on what implementations need to interop with the canonical Python implementation (markqvist/Reticulum and markqvist/LXMF) plus the existing client ecosystem (Sideband, Nomadnet, MeshChat, the various firmware projects).

Last verified against: RNS 1.2.4 / LXMF 0.9.7 / RNode_Firmware (master at the spec's last revision date). Each section's source citations were re-checked against these versions; runtime verifiers in tools/ lock the wire-format claims in against actually-running upstream code. When you upgrade past these, re-run every tools/verify_*.py and look for FAILs.

Source citations refer to the standard pip install rns lxmf install layout (RNS/, LXMF/).

Contents — click to expand the per-section table of contents (regenerate with python tools/_gen_toc.py)

1. Identity and destination hashes
2. Packet header
3. Token cryptography (modified Fernet)
4. Announce wire format
5. LXMF wire format
6. Reticulum Link protocol
7. Transport behavior — the parts that bite
8. Transport framing
9. Implementation gotchas
10. Resource fragmentation protocol
11. REQUEST/RESPONSE protocol (NomadNet pages, propagation /get, custom RPC)
12. Transport-relay behaviour
13. Threading and concurrency model
14. Failure modes — symptom → root cause
15. Time and clock requirements
16. Bounded-state inventory (memory limits at a glance)
17. Implementation taxonomy: who needs which sections
18. Test vectors
19. Source map

1. Identity and destination hashes

1.1 Identity composition

A Reticulum identity is two keypairs concatenated:

public_key  = X25519_pub(32) || Ed25519_pub(32)        // 64 bytes
private_key = X25519_priv(32) || Ed25519_priv(32)      // 64 bytes

X25519 for ECDH (encryption / shared-secret derivation), Ed25519 for signatures.

identity_hash = SHA256(public_key)[:16]                // 16 bytes truncated

The 16-byte truncation is consistent across all hashes Reticulum stores on the wire (destinations, link IDs, packet hashes, etc.). The full SHA-256 is used internally for signing inputs but never appears in headers.

1.2 Destination hash

The 16-byte destination hash that appears in packet headers and announces is:

name_hash = SHA256(full_app_name_string)[:10]
dest_hash = SHA256(name_hash || identity_hash)[:16]

Where full_app_name_string is e.g. "lxmf.delivery", "nomadnetwork.node", "rnstransport.path.request". The hex-encoded identity hash is NOT part of the input — only the plain ASCII app-name string. This is the identity=None branch of upstream's expand_name() function (RNS/Destination.py). The identity hex appears only in the human-readable Destination.name debug string.

Common pre-computed name_hash values:

10-byte hex	App name
`6ec60bc318e2c0f0d908`	`lxmf.delivery`
`e03a09b77ac21b22258e`	`lxmf.propagation`
`213e6311bcec54ab4fde`	`nomadnetwork.node`
`0ad8bff9ff75737c058e`	`nomadnetwork.gossip`
`9efb9c771eeb5ae90ea6`	`rnstransport.broadcasts`
`4848a053c16415bed6c8`	`rnstransport.remote.management`
`7926bbe7dd7f9aba88b0`	`rnstransport.path.request` (resulting `dest_hash` with `identity=None`: `6b9f66014d9853faab220fba47d02761`)

1.3 Private key on-disk format

RNS.Identity.to_file(path) writes the raw 64-byte private-key blob with no header, no version byte, no checksum, no encryption. The byte order is the same as the public_key concatenation in §1.1 — verified by tools/verify_destination_hash.py's existing Identity.from_bytes round-trip:

prv_bytes_blob  =  X25519_priv(32) || Ed25519_priv(32)         // 64 bytes total

Identity.get_private_key() at RNS/Identity.py:723-728 returns this exact concatenation:

def get_private_key(self):
    return self.prv_bytes + self.sig_prv_bytes
    #      ^^^^^^^^^^^^^   X25519 priv (set at line 679 from X25519PrivateKey.generate())
    #                       ^^^^^^^^^^^^^^^   Ed25519 priv (set at line 682)

Identity.load_private_key(prv_bytes) at line 706-717 slices it back the same way:

self.prv_bytes     = prv_bytes[:32]   # X25519
self.sig_prv_bytes = prv_bytes[32:]   # Ed25519

to_file is a thin wrapper that writes get_private_key() to the path; from_file reads back with no extra parsing.

File-system facts

Size: exactly 64 bytes. No magic, no length prefix.
Encryption: none. Anyone with read access can fully impersonate the identity.
Permissions: upstream doesn't chmod the file; clients are expected to put it in a directory protected by OS permissions (~/.reticulum/storage on Linux/macOS, %APPDATA%/Reticulum/storage on Windows by default).
Filename: caller-controlled. RNS itself uses transport_identity for the transport node and lets app-level callers choose for delivery destinations (LXMF puts these in LXMF.LXMRouter.storagepath).

Constructing from raw bytes — `from_bytes` HAZARD

Identity.from_bytes(prv_bytes) at line 611-623 takes the same 64-byte concat and reconstitutes an Identity. The upstream docstring explicitly warns:

HAZARD! Never use this to generate a new key by feeding random data in prv_bytes.

The reason: X25519PrivateKey.from_private_bytes and Ed25519PrivateKey.from_private_bytes both accept arbitrary 32-byte values without scalar clamping or rejection — a clean-room implementation that feeds raw random data into from_bytes skips the keypair-generation invariants enforced by the upstream cryptography library's .generate() methods (e.g. X25519 scalar clamping per RFC 7748 §5). Always generate fresh keys via the cryptography (or equivalent) library's keypair generator, then concatenate; never invent your own bytes.

Cross-implementation portability

The format is portable across implementations because there's nothing in it but the raw bytes. A 64-byte file written by Python RNS is byte-identical to one written by any clean-room implementation that follows this section, and both produce the same identity_hash and lxmf.delivery destination_hash when fed back through §1.1 and §1.2 — test vectors at test-vectors/identities.json demonstrate the round-trip against RNS 1.2.4.

⚠️ Spec correction: Earlier revisions of this section described the on-disk order as Ed25519 first, X25519 second ("opposite of the public_key concatenation"). That was wrong — verified by re-running Identity.to_file and reading back the bytes against the test vector at test-vectors/identities.json, the actual order is X25519 first, Ed25519 second, identical to the public_key order. Implementations following the prior spec wording would have corrupted identity files when interoperating with upstream Python RNS.

1.4 GROUP destinations (symmetric-key alternative to SINGLE)

Most Reticulum traffic — including all LXMF — uses SINGLE destinations with the X25519 ECDH + Ed25519 signing scheme described above. There is also a GROUP destination type (RNS.Destination.GROUP = 0x01, in the dest_type field of the packet header per §2.1) that uses a pre-shared symmetric key instead, intended for closed channels where every participant should be able to decrypt every message without per-recipient ECDH.

1.4.1 Key generation

Destination.create_keys() for GROUP calls Token.generate_key() (RNS/Cryptography/Token.py:53-56):

@staticmethod
def generate_key(mode=AES_256_CBC):
    if   mode == AES_128_CBC: return os.urandom(32)    # 16B signing + 16B encryption
    elif mode == AES_256_CBC: return os.urandom(64)    # 32B signing + 32B encryption

The default is AES-256-CBC, so the symmetric key is 64 random bytes, split into:

signing_key     = key[ 0..32]      // HMAC-SHA256 input
encryption_key  = key[32..64]      // AES-256-CBC key

A clean-room implementation that needs to interop with a GROUP destination must use AES-256-CBC by default and derive the same split. AES-128-CBC mode (32-byte key, 16/16 split) is supported by the Token class but no upstream caller currently selects it for GROUPs.

1.4.2 Wire format

GROUP destinations encrypt and decrypt via Token.encrypt / Token.decrypt — the same Token format used by Link-derived encryption (§3.1, the no-ephemeral-pub form):

wire_body  =  iv(16) || aes_ciphertext || hmac_sha256(32)

There is no ephemeral_pub prefix because there is no ECDH — every participant already shares the same (signing_key, encryption_key) pair. The format is identical to a Link DATA payload after the link is established (§6.4). Reticulum's Token class is shared across both code paths; see RNS/Destination.py:601-609 and :645-653 for GROUP encrypt/decrypt, and RNS/Cryptography/Token.py:87-114 for the underlying primitive.

1.4.3 Destination hash for GROUP

GROUP destinations use the same dest_hash recipe as SINGLE (§1.2) — SHA256(name_hash || identity_hash)[:16] — with two wrinkles:

The constructor accepts an identity argument optionally. If provided, identity_hash = SHA256(identity.public_key)[:16] per §1.1; the resulting dest_hash is keyed to that identity's public key as well as the group name. Different identities → different group destinations even with the same name.
If identity is not provided (None), dest_hash = SHA256(name_hash)[:16] (same recipe as PLAIN destinations — see the path-request example in §1.2).

The identity (if any) does NOT participate in encryption — it's purely a way to disambiguate group destinations sharing a name across owners. The actual encryption uses the symmetric key from §1.4.1.

1.4.4 On-disk format

Destination.get_private_key() for GROUP returns self.prv_bytes — the 64 (or 32) raw key bytes. Destination.load_private_key(key) accepts the same. There is no canonical file path or filename — the application chooses where to store the symmetric key, and is responsible for distributing it to every group member out of band.

Like the SINGLE identity file (§1.3), the GROUP key file has no header, no encryption-at-rest, no checksum. Anyone with read access can fully impersonate the group.

1.4.5 Why most clients don't bother

GROUP destinations are rarely seen in the wild because:

LXMF doesn't use them (every chat is one-to-one between SINGLE destinations, even in multi-party rooms — those are application-layer constructs).
NomadNet pages use SINGLE destinations.
The forward-secrecy properties of SINGLE (per-message ephemeral X25519 + ratchet rotation per §7.3) are absent for GROUP — once the symmetric key is leaked, every past and future message decryptable by that key is compromised.
Group key distribution is an unsolved problem at the protocol level — Reticulum doesn't help with this.

A clean-room client that targets LXMF interop only can ignore GROUP destinations entirely. Implementations of Destination.encrypt/decrypt should still recognize the GROUP (0x01) type byte in the packet header to gracefully reject (rather than crash on) inbound packets to a GROUP whose key the receiver doesn't hold.

2. Packet header

2.1 Flag byte layout

Every Reticulum packet starts with a 1-byte flag field:

bit 7   : ifac_flag        (0 = open / no IFAC, 1 = IFAC field present)
bit 6   : header_type      (0 = HEADER_1, 1 = HEADER_2)
bit 5   : context_flag     (1 = announce includes a ratchet pubkey)
bit 4   : transport_type   (0 = BROADCAST, 1 = TRANSPORT)        — the official manual calls this "propagation_type"
bit 3-2 : destination_type (0 = SINGLE, 1 = GROUP, 2 = PLAIN, 3 = LINK)
bit 1-0 : packet_type      (0 = DATA, 1 = ANNOUNCE, 2 = LINKREQUEST, 3 = PROOF)

Each subfield is 1 bit (or 2 for destination_type / packet_type). Upstream's parser extracts them with these masks (RNS/Packet.py:246-250 in RNS 1.2.4):

self.header_type      = (self.flags & 0b01000000) >> 6     # bit 6 only
self.context_flag     = (self.flags & 0b00100000) >> 5
self.transport_type   = (self.flags & 0b00010000) >> 4
self.destination_type = (self.flags & 0b00001100) >> 2
self.packet_type      = (self.flags & 0b00000011)

Bit 7 (ifac_flag) is set by Transport.transmit immediately before transmission when the egress interface has an IFAC identity attached (RNS/Transport.py:994-1024). The setter is unambiguous:

# Set IFAC flag
new_header = bytes([raw[0] | 0x80, raw[1]])    # 0x80 = bit 7
# Assemble new payload with IFAC
new_raw    = new_header + ifac + raw[2:]       # IFAC field is inserted between header and addresses

When ifac_flag = 1, an ifac_size-byte IFAC field appears immediately after byte 2 of the header — i.e. between the hops byte and the start of ADDRESSES. ifac_size is interface-configured and ranges from IFAC_MIN_SIZE = 1 byte (RNS/Reticulum.py:148-152) up to 64 bytes (full Ed25519 signature). The receiving side strips the IFAC after verification and rebuilds the un-IFACed packet for upstream processing (RNS/Transport.py:1343-1369).

⚠️ Spec correction. Earlier revisions of this section (through commit 8c4d550) treated header_type as a 2-bit field occupying bits 7-6, with bit 7 reserved. That was wrong: bit 7 has always been the IFAC flag (Transport.transmit line 1003), and header_type is 1 bit at position 6. The official manual §4.6.3 documents the correct layout. Implementations that followed the prior wording will mis-parse IFAC-protected packets as header_type = 2 or 3 and reject them. Surfaced by the reporter on issue #4 item #1.

2.2 Two header forms

HEADER_1: flags(1) hops(1) dest_hash(16) context(1) data(...)        // min 19 bytes
HEADER_2: flags(1) hops(1) transport_id(16) dest_hash(16) context(1) data(...)   // min 35 bytes

HEADER_2 carries a transport_id (the next-hop transport node's identity hash) before the final destination hash. A relay converts a HEADER_1 packet to HEADER_2 by setting bit 6 of flags, inserting its own identity at offset 2, and re-transmitting.

2.3 Originator HEADER_1 → HEADER_2 conversion

This is non-obvious and matters: when an originator (not a relay) sends a packet to a destination known to be more than 1 hop away, the originator MUST also do the HEADER_2 conversion. From RNS/Transport.py::outbound (lines 1077-1108 in RNS 1.2.4; verified by tools/verify_packet_header.py):

if path_entry[IDX_PT_HOPS] > 1:
    if packet.header_type == RNS.Packet.HEADER_1:
        new_flags = (RNS.Packet.HEADER_2) << 6 | (Transport.TRANSPORT) << 4 | (packet.flags & 0b00001111)
        new_raw  = struct.pack("!B", new_flags)
        new_raw += packet.raw[1:2]                       # hops byte unchanged
        new_raw += path_entry[IDX_PT_NEXT_HOP]           # 16B transport_id at offset 2
        new_raw += packet.raw[2:]                        # original dest_hash + context + payload

For destinations 0 or 1 hops away, the originator may stay HEADER_1 — the receiving rnsd auto-fills the transport_id when the destination matches a local client (for_local_client branch at RNS/Transport.py:1454 in RNS 1.2.4). Implementations that always emit HEADER_1 will silently fail to deliver to multi-hop destinations even with a known path.

2.4 Hop count

Byte 1 is hops, an 8-bit counter that each transit relay increments by 1. 0 for a packet still on the originator. 255 would in theory wrap, but no Reticulum mesh in practice has paths anywhere near that long.

2.5 Context byte

Single byte after the destination hash (offset 18 for HEADER_1, offset 34 for HEADER_2). Common values:

Full context inventory from RNS/Packet.py:74-92 (RNS 1.2.4):

Hex	Name	Used for
`0x00`	NONE	Generic / opportunistic DATA packet
`0x01`	RESOURCE	One part (chunk) of a Resource transfer (§10)
`0x02`	RESOURCE_ADV	Resource advertisement
`0x03`	RESOURCE_REQ	Resource part request (from receiver to sender)
`0x04`	RESOURCE_HMU	Resource hashmap update (next-segment hashmap)
`0x05`	RESOURCE_PRF	Resource proof (a PROOF-type packet using this context)
`0x06`	RESOURCE_ICL	Resource cancel from the initiator
`0x07`	RESOURCE_RCL	Resource cancel from the receiver / reject of an advertisement
`0x08`	CACHE_REQUEST	Cache lookup over a Link
`0x09`	REQUEST	Link REQUEST (NomadNet page fetch, propagation `/get`)
`0x0A`	RESPONSE	Link RESPONSE matching a REQUEST
`0x0B`	PATH_RESPONSE	An ANNOUNCE emitted in response to a `path?` request — distinguishes it from a periodic re-announce. Receivers handle the two paths differently (see §7.2 and §4.5)
`0x0C`	COMMAND	Channel-style remote-execution command
`0x0D`	COMMAND_STATUS	Status reply for a COMMAND
`0x0E`	CHANNEL	Link channel multiplexed payload
`0xFA`	KEEPALIVE	Link keepalive (sent periodically while a Link is idle)
`0xFB`	LINKIDENTIFY	Backchannel-identify proof on an established Link (§5 backchannel)
`0xFC`	LINKCLOSE	Link teardown notification
`0xFD`	LINKPROOF	Defined but not actually emitted by upstream RNS 1.2.4 in this revision. Both `Identity.prove` and `Link.prove_packet` build their proof packets with `context = NONE (0x00)` — the proof-ness is conveyed by `packet_type = PROOF (3)`, not by this context byte. Reserved for a future revision; see §6.5
`0xFE`	LRRTT	Link round-trip-time reply — initiator's link-activation packet (§6.4.2)
`0xFF`	LRPROOF	Link request proof (§6.2)

2.6 Source

RNS/Packet.py for the constants and _pack / _unpack methods. RNS/Transport.py for the routing-side HEADER_1↔HEADER_2 transitions.

3. Token cryptography (modified Fernet)

Reticulum's "Token" construction is a modified Fernet used for opportunistic destination encryption (single packet), as well as for derived-key channels on established Links.

3.1 Wire format

ephemeral_pub(32) || iv(16) || aes_ciphertext(...) || hmac_sha256(32)

For Link-derived-key encryption (after the Link handshake has produced a session key), the ephemeral_pub prefix is omitted and the wire form is just iv || ciphertext || hmac.

3.2 Encrypt steps (opportunistic)

Generate ephemeral X25519 keypair (eph_priv, eph_pub).
ECDH: shared = X25519(eph_priv, recipient_X25519_pub). The recipient's X25519 pub is either their long-term encPub (first 32 bytes of public_key) or their currently-announced ratchet_pub if present.
HKDF-SHA256: derived = HKDF(shared, salt = recipient_identity_hash, info = "", L = 64). The salt is the recipient's 16-byte identity hash — not their destination hash, not the ratchet hash.
Split: signing_key = derived[0..32], encryption_key = derived[32..64].
Random 16-byte IV.
AES-256-CBC encrypt plaintext with encryption_key and iv. Do NOT manually pad — the platform AES-CBC API (AES/CBC/PKCS5Padding on JCA, Web Crypto's default) auto-pads PKCS#7. Manual padding on top causes 16 garbage bytes of double-padding.
hmac = HMAC-SHA256(signing_key, iv || ciphertext).
Concatenate as the wire format above.

3.3 Decrypt steps

Reverse of encrypt. Critically:

Verify HMAC BEFORE attempting decryption (encrypt-then-MAC; prevents AES padding-oracle attacks).
A receiver that has multiple candidate X25519 private keys (typically the current ratchet privkey + the long-term identity privkey) should try each in order until one produces a matching HMAC. Senders that haven't seen the receiver's latest ratchet announce will encrypt to the long-term key as a fallback.

3.4 Source

RNS/Cryptography/Token.py (and the equivalents in vendor crypto modules). The webclient's reference/js-reference/crypto.js is a faithful port.

4. Announce wire format

4.1 Packet body

The Reticulum packet header (HEADER_1, packet_type=ANNOUNCE, dest_type=SINGLE, transport_type=BROADCAST) is followed by an announce body:

public_key(64) || name_hash(10) || random_hash(10) || [ratchet_pub(32) if context_flag] || signature(64) || app_data(...)

The 64-byte public_key is the X25519 || Ed25519 concat described in section 1.1.

random_hash is NOT 10 random bytes — only the first 5 bytes are random; the trailing 5 bytes carry the emission timestamp as a big-endian unsigned 40-bit Unix-seconds integer (RNS/Destination.py:282):

random_hash = RNS.Identity.get_random_hash()[0:5] + int(time.time()).to_bytes(5, "big")

Transit relays read the timestamp portion via Transport.timebase_from_random_blob(random_blob) = int.from_bytes(random_blob[5:10], "big") (RNS/Transport.py:3106-3107) to make ordering decisions when an inbound announce carries a higher hop count than the cached path: only newer-emitted announces can refresh the path table (see §4.5). 5 bytes of seconds covers ~34,000 years, so wraparound is not a near-term concern. Implementations MUST emit this exact format, including a clock value that's monotonically non-decreasing across announces from the same destination — clockless sender devices (per §9.6) may end up locked out of long-range path table updates.

⚠️ UNVERIFIED — Known deviation: attermann/microReticulum/src/Destination.cpp:270-272 (and therefore every project that uses microReticulum unmodified, including thatSFguy/reticulum-lora-repeater and the Faketec sibling project) currently emits 10 fully-random bytes for random_hash — the timestamp half is a TODO that never landed:
//p random_hash = Identity::get_random_hash()[0:5] << int(time.time()).to_bytes(5, "big")
// CBA TODO add in time to random hash
Bytes random_hash = Cryptography::random(Type::Identity::RANDOM_HASH_LENGTH/8);
Python RNS receivers interpret random_hash[5:10] as a big-endian uint40 unix_seconds. A uniformly-random uint40 has median value ~5.5×10¹¹ ≈ year 19403 AD, so a microReticulum announce will (with overwhelming probability) appear "far-future" to a Python receiver. Effect: once one such announce populates path_table[dest][IDX_PT_RANDBLOBS], the equal-or-greater-hop branch at RNS/Transport.py:1723-1755 will reject any real-timestamped announce as "stale" until the path TTL expires. First-contact path-table population is unaffected; the bug only surfaces on path replacement under §4.5 step 6.3. The microReticulum receive side does NOT consult the timestamp half so microReticulum-to-microReticulum traffic is unaffected. The repeater repo's pre_build.py patches several microReticulum protocol bugs but not this one (as of thatSFguy/reticulum-lora-repeater@95823ad-vintage upstream). Verifying by capture-and-decode against an actual mixed-vendor mesh is the work that would let this callout be removed.

The optional 32-byte ratchet_pub (an X25519 public key) is present iff the packet header's context_flag bit is 1. Indexing through this layout accordingly is mandatory; see RNS/Identity.py::validate_announce for the canonical parser.

4.2 Signed data

signed_data = dest_hash(16) || public_key(64) || name_hash(10) || random_hash(10) || [ratchet_pub(32)] || app_data
signature   = Ed25519_sign(signed_data, identity.Ed25519_priv)

Note that dest_hash is INCLUDED in the signed data even though it's not in the wire-format announce body (the receiver gets it from the packet header). The signing key is the Ed25519 half (last 32 bytes) of the identity's private_key.

4.3 `app_data` format for LXMF delivery destinations

Upstream LXMF/LXMRouter.py::get_announce_app_data produces a 2-element msgpack array (verified against LXMF 0.9.7 by tools/verify_announce_app_data.py):

# LXMF/LXMRouter.py:985-1002 in LXMF 0.9.7
peer_data = [display_name, stamp_cost]   # stamp_cost = None unless 1 ≤ N ≤ 254
return msgpack.packb(peer_data)

Wire bytes for display_name = "Reticulum5", stamp_cost = None:

92         # fixarray, 2 elements
c4 0a      # bin8, length 10
52 65 74 69 63 75 6c 75 6d 35    # "Reticulum5"
c0         # nil (stamp_cost)

Encoding the display name as msgpack bin (0xc4 NN) is required for upstream interop — see section 9.3 below. The stamp_cost field can be int 0 (0x00) or nil (0xc0); upstream's stamp_cost_from_app_data doesn't strict-type-check.

A third optional [capability_flags] element (e.g. [SF_COMPRESSION], the only flag currently defined at LXMF/LXMF.py:108) is read by the parser (compression_support_from_app_data at LXMF/LXMF.py:154-167) but is not emitted by the LXMF 0.9.7 producer — LXMRouter.py:998 computes supported_functionality = [SF_COMPRESSION] but never appends it to peer_data. Implementations should accept the 3-element form on inbound (a future LXMF version may re-enable it; older deployments may emit it) but should not rely on receiving it.

The parser also tolerates a 1-element msgpack array (just the name) and a raw UTF-8 string ("original announce format" branch at LXMF/LXMF.py:138-139) — see LXMF/LXMF.py::display_name_from_app_data for all four accepted shapes.

4.4 Announce filtering by `name_hash`

When ingesting an announce, clients should distinguish by name_hash:

lxmf.delivery (6ec60bc318e2c0f0d908) — messagable peers, surface in contacts UI
lxmf.propagation (e03a09b77ac21b22258e) — propagation node, surface separately
nomadnetwork.node (213e6311bcec54ab4fde) — page-serving NomadNet host
rnstransport.broadcasts / rnstransport.remote.management — transport-internal, ignore for user UI
Any other name_hash — non-LXMF custom destination (telemetry beacons, application-specific)

Treating every announce as a contact (the naive default) populates the UI with hundreds of irrelevant rows.

4.5 Announce validation rules (receive side)

These are the MUST rules a receiver applies to every inbound announce before considering the announced destination "known". The canonical implementation is RNS/Identity.py::validate_announce (line 509-612 in RNS 1.2.4); the dispatch site that calls it is RNS/Transport.py::inbound line 1631-1655.

1. Body parse — branch on `context_flag`

The context_flag bit (bit 5 of the packet's 1-byte flag field, §2.1) selects between two body layouts. Slice offsets, with keysize = 64, name_hash_len = 10, random_hash_len = 10, ratchet_size = 32, sig_len = 64:

context_flag == 1 (ratchet present):
   public_key   = data[ 0                                     :  64]
   name_hash    = data[ 64                                    :  74]
   random_hash  = data[ 74                                    :  84]
   ratchet_pub  = data[ 84                                    : 116]
   signature    = data[116                                    : 180]
   app_data     = data[180                                    :    ]   # may be empty

context_flag == 0 (no ratchet):
   public_key   = data[ 0                                     :  64]
   name_hash    = data[ 64                                    :  74]
   random_hash  = data[ 74                                    :  84]
   signature    = data[ 84                                    : 148]
   app_data     = data[148                                    :    ]   # may be empty

A client that uses a fixed offset for signature regardless of the flag (a real bug from the SF webclient's first cut) silently rejects every ratchet-bearing announce as having a bad signature.

2. Signature verification

Reconstruct the signed_data exactly per §4.2:

signed_data = destination_hash || public_key || name_hash || random_hash || ratchet || app_data

Where ratchet is b"" (empty, not absent) when context_flag == 0, and app_data is b"" when not present in the packet. destination_hash comes from the outer packet header, NOT from the announce body — re-using the body bytes as the dest_hash would let a sender forge announces for arbitrary destinations.

Verify the 64-byte signature with the announced public_key's Ed25519 half (last 32 bytes). Reject on failure.

3. `destination_hash` recomputation

Recompute the dest_hash from the announced inputs:

identity_hash    = SHA256(public_key)[:16]
expected_hash    = SHA256(name_hash || identity_hash)[:16]

Reject the announce iff expected_hash != packet.destination_hash (the value from the outer header). This catches both random hash collisions and active spoofing attempts that pair a valid signature with an unrelated dest_hash. (RNS/Identity.py:562-565).

4. Public-key collision rejection

If the receiver already has a different public_key cached for this destination_hash (from a prior announce), the new announce MUST be rejected with a critical-severity log even if the signature is otherwise valid. Per the upstream comment: "In reality, this should never occur, but in the odd case that someone manages a hash collision, we reject the announce" (RNS/Identity.py:569-575).

This rule means: first-announcer-wins for any given destination_hash within a receiver's lifetime. A peer who loses their identity material and regenerates with the same display name + app_name will produce a different identity_hash → different destination_hash → no collision. A peer who tries to replace their announced public key under the same destination_hash, however, gets rejected — the real defense against this class of attack.

5. Blackhole list check

Before everything else, check RNS.Transport.blackholed_identities. An identity_hash on the blackhole list is dropped silently regardless of signature validity (RNS/Identity.py:551-554). This is operator-controlled state, not a wire feature.

6. Caching the announce contents

On a fully validated announce, the receiver MUST update its caches in this order:

known_destinations[destination_hash] ← [recv_time, packet_hash, public_key, app_data, last_used] — populates the table that RNS.Identity.recall(dest_hash) reads when constructing outbound destinations (RNS/Identity.py::remember, line 101-113). Without this, every subsequent outbound message to this peer fails because no public key is available for Token encryption.
known_ratchets[destination_hash] ← ratchet_pub (only if context_flag == 1 and ratchet_pub != b"") — Identity._remember_ratchet, line 395-428. The ratchet is also persisted to disk under {storagepath}/ratchets/{hexhash} for use across restarts.
path_table entry update or insertion (see §4.6 — TBD when the relay rebroadcast spec lands), gated by:
- random_blob (= random_hash) not in the cached random_blobs history for this destination — cheap replay defence (RNS/Transport.py:1710, 1735, 1748).
- Hop count comparison against any existing entry: equal-or-fewer hops always win; more hops win only if the cached path has expired or the new announce's emission timestamp (from random_hash[5:10]) is more recent than every cached blob's timestamp (RNS/Transport.py:1700-1755).

7. `PATH_RESPONSE` distinction

An announce whose outer packet context == PATH_RESPONSE (0x0B) is the responder's reply to a recent path? request, not a periodic re-announce. Validation is identical (rules 1-6 above), but listener dispatch differs:

The default behavior of Transport.announce_handlers registered via RNS.Transport.register_announce_handler is to skip path-response announces unless the handler sets receive_path_responses = True on itself (RNS/Transport.py:1991-1995).
The path table population path is the same either way — both regular and path-response announces refresh the path entry — so a leaf client that ignores PATH_RESPONSE entirely at the application layer still benefits from the path-table side effect.

8. Implementation-private behavior (SHOULD)

These are not wire-spec MUST rules but most working clients implement them; without them the implementation will misbehave in busy meshes:

Per-interface ingress rate limiting. When the inbound announce rate on an interface exceeds IC_BURST_FREQ_NEW = 6 Hz (interfaces less than 2 hours old) or IC_BURST_FREQ = 35 Hz (older), and the announced destination is not in path_table and not in path_requests, the announce is held in the interface's held_announces dict for later release rather than processed immediately. Released later in lowest-hop-count-first order. (RNS/Interfaces/Interface.py:60-200.) Without this, a flood of unknown-destination announces can drown out everything else.
random_blob history cap. The cached random_blobs list per destination is bounded by Transport.MAX_RANDOM_BLOBS (= 64 in RNS 1.2.4 at RNS/Transport.py:97) to keep the path table from growing without bound under a long-lived destination's announce stream (RNS/Transport.py:1823).
Self-announce filter. §9.5 — drop announces where destination_hash matches one of the receiver's own destinations to avoid populating its own contact list with itself.

9. Source map for §4.5

File	What it pins down
`RNS/Identity.py:509-612`	`validate_announce` — body parse, signed_data, sig verify, dest_hash recompute, collision check
`RNS/Identity.py:101-113`	`Identity.remember` — `known_destinations` update
`RNS/Identity.py:408-441`	`_remember_ratchet` — ratchet persistence
`RNS/Transport.py:1631-2030`	inbound dispatch for `packet_type == ANNOUNCE`: quick sig check, ingress limiting, path table population, handler dispatch
`RNS/Transport.py:3106-3122`	`timebase_from_random_blob`, `announce_emitted`
`RNS/Interfaces/Interface.py:60-200`	ingress-limit constants, `should_ingress_limit`, `hold_announce`, `process_held_announces`
`RNS/Packet.py:83`	`PATH_RESPONSE = 0x0B` context constant

4.6 `rrc.hub` announce app_data (Reticulum Relay Chat)

Reticulum Relay Chat hubs announce a destination on the rrc.hub aspect — name_hash = SHA256("rrc.hub")[:10] = ac9fd3a81e4036f86e1d. Unlike §4.3 (LXMF delivery), the app_data is not a msgpack [name, cost] array, and the two hub implementations disagree on its shape:

rrcd — the Python reference hub. app_data is a CBOR (RFC 8949) map: {"proto": "rrc", "v": 1, "hub": <hub_name>} — CBOR, not msgpack, because RRC's wire codec (rrcd/codec.py, Python cbor2) is CBOR throughout. The human hub name is the "hub" key's value (a text string); "proto" is always "rrc" and "v" is the app_data schema version (1). Source: rrcd service.py — app_data = encode({"proto": "rrc", "v": 1, "hub": self.config.hub_name}), where encode is the CBOR encoder.
reticulum-relay-chat — the Go hub. app_data is the hub name as plain UTF-8 bytes, unwrapped. Source: internal/service/service.go — BuildAnnounce(id, "rrc.hub", []byte(s.cfg.Hub.Name), ...).

⚠️ CBOR-vs-msgpack gotcha. A CBOR 3-entry map begins with byte 0xa3. In msgpack 0xa3 is fixstr of length 3 — so a client that blindly msgpack-decodes the rrcd app_data reads the next three bytes (0x65 0x70 0x72, the CBOR text-string header of "proto" plus its first two characters) as the 3-character string "epr". Decode rrc.hub app_data with a CBOR decoder, keyed on the rrc.hub name_hash — do not feed it to the LXMF (msgpack) app_data parser.

A client listing RRC hubs should resolve the name as: the "hub" value when app_data CBOR-decodes to a map; else a bare UTF-8 string (the Go hub's shape); else a generic "RRC hub" label. The same name is also delivered authoritatively in the RRC WELCOME body key B_WELCOME_HUB once a session is established.

5. LXMF wire format

LXMF has two delivery methods with different plaintext layouts.

5.1 Opportunistic delivery (single Reticulum DATA packet)

Plaintext (after Token decryption):

source_hash(16) || signature(64) || msgpack_payload(...)

The recipient's destination_hash is stripped (the outer Reticulum packet's dest_hash already conveys it; including it would waste bytes).

5.2 Direct delivery (over an established Reticulum Link)

destination_hash(16) || source_hash(16) || signature(64) || msgpack_payload(...)

Full layout. The Link's session key encrypts the whole blob.

5.3 `msgpack_payload`

A msgpack array of 4 elements (5th optional):

[timestamp_seconds_double, title_bytes, content_bytes, fields_dict]
# optional 5th element: stamp (varies)

Times are seconds-since-Unix-epoch as a double-precision float. Title and content are msgpack bin (Python bytes). Fields is a msgpack map; usually {} for plain text, but used for attachments, stickers, etc.

5.4 Source/destination semantics

source_hash is the SENDER's destination hash (SHA256(name_hash || identity_hash)[:16]), NOT the raw identity hash. A common implementation bug is to write the identity_hash here; the recipient then can't look the sender up in their contacts (which are keyed by destination_hash).

5.5 Signed data

hashed_part  = destination_hash(16) || source_hash(16) || msgpack_payload
message_hash = SHA256(hashed_part)
signed_data  = hashed_part || message_hash
signature    = Ed25519_sign(signed_data, sender_identity.Ed25519_priv)

For opportunistic delivery, destination_hash is the recipient's destination hash (from the outer packet header, not from the LXMF body).

The msgpack_payload MUST be the exact bytes received on the wire — not a decode-then-re-encode — when computing message_hash. The same value is reused as the LXMF message_id (workblock input for stamps in §5.7, target identifier for reactions in §5.9.8 and replies in §5.9.9). A receiver that re-encodes will compute a divergent identifier for any message whose fields map does not round-trip byte-identically — most commonly reply messages — which breaks both the §5.6 path-1 raw-signature check and any relay rewrite-cache keyed on the value. For a stamped message (§5.7.1) the raw payload on the wire is a 5-element array; the hash is over the first 4 elements, which requires a byte-canonical re-pack per §5.6.1.

5.6 Signature verification — msgpack variant tolerance

Different msgpack encoders produce subtly different byte sequences for the same logical value (e.g. integer encoding choice, string vs bin selection). The signer signed over THEIR encoder's output. A receiver should try verifying against:

The raw msgpack bytes from the wire as-received (msgpack_payload exactly).
A stripped re-encoded version (decode then re-encode the first 4 elements, omitting the optional stamp field).

If either matches, the signature is valid. Strict raw-only verification fails interop with anything that's been through a msgpack re-encode somewhere in the chain.

5.6.1 Canonical encoder for senders

RNS bundles umsgpack (RNS/vendor/umsgpack.py) and uses it for every signing input on the upstream Python side. Senders SHOULD produce signing-input bytes that match umsgpack's output for the LXMF payload types so receivers' path-1 (raw) verification succeeds and the path-2 (decode + re-encode) fallback stays defensive rather than load-bearing:

Logical type	umsgpack canonical form
Python `str` / UTF-8 text	`str` family — `fixstr` / `str8` / `str16` / `str32`, smallest that fits
Python `bytes`	`bin` family — `bin8` / `bin16` / `bin32`, smallest that fits
Integer	smallest signed integer envelope that fits — positive fixint, `uint8`/`16`/`32`/`64`, negative fixint, `int8`/`16`/`32`/`64` (per umsgpack's `_pack_integer`)
Float	`float64` always (9 bytes including the type byte) — never `float32`, even for integer-valued doubles
Map	sorted-by-insertion-order — umsgpack preserves input order, does NOT lex-sort keys

Mismatches most often originate from integer width (a timestamp encoded as uint32 by one library and float64 by another round-trips the same logical value but produces different bytes) and from JS encoders that prefer str for byte strings or float32 for non-integer numbers. Implementing umsgpack's "minimum width that fits" rule for ints and "always float64" rule for floats is sufficient for byte-identical signature inputs against upstream Python LXMF.

5.7 LXMF stamps and tickets (anti-spam)

LXMF.LXMessage.payload[4] (the optional 5th element of the msgpack body — see §5.3) is a stamp: a proof-of-work value that lets a recipient gate inbound messages against unsolicited senders. Modern Sideband installs (≥ 1.x) treat unstamped messages as low-trust and may drop them at the application layer.

5.7.1 Stamp wire format

LXMessage.STAMP_SIZE = HASHLENGTH//8 = 32 bytes (LXMF/LXStamper.py). The stamp is appended to the payload msgpack array as the 5th element only if the receiver requires one or the sender has an outbound ticket. Wire form is just 32 raw bytes inside a bin8/bin16 msgpack envelope.

When stripping the stamp during signature verification (§5.6), the receiver removes element [4] from the unpacked array and re-encodes the first 4 elements as packed_payload for hash computation. This is what lets a sender add or remove a stamp without invalidating the Ed25519 signature.

5.7.2 Stamp generation (proof-of-work)

LXMF/LXStamper.py::generate_stamp(message_id, target_cost) and ::stamp_valid(stamp, target_cost, workblock). The algorithm:

Workblock construction — expensive HKDF-driven memory inflation:

def stamp_workblock(material, expand_rounds=3000):       # WORKBLOCK_EXPAND_ROUNDS
    workblock = b""
    for n in range(expand_rounds):
        workblock += RNS.Cryptography.hkdf(
            length=256,
            derive_from=material,
            salt=RNS.Identity.full_hash(material + msgpack.packb(n)),
            context=None)
    return workblock                                     # 768 KiB total

material is the 32-byte message_id (= SHA256(dest_hash || src_hash || msgpack_payload)). 3000 rounds of 256-byte HKDF produces a 768 KiB workblock — designed to be cache-unfriendly enough that GPU/ASIC speedup is limited.

Stamp search — find a 32-byte value such that SHA256(workblock || stamp) has at least target_cost leading zero bits:
```
def stamp_valid(stamp, target_cost, workblock):
    target = 1 << (256 - target_cost)
    return int.from_bytes(SHA256(workblock + stamp), "big") <= target
```
The target_cost is the Hamming-distance-from-2^256 in bit-leading-zeros — target_cost = 8 means the result must be ≤ 2^248, i.e. start with at least 8 zero bits.
Stamp value — for received valid stamps, stamp_value(workblock, stamp) returns the actual leading-zero count, which can exceed the recipient's required cost. Exceeded cost = "extra effort spent" and is exposed to the application for prioritization.

The default WORKBLOCK_EXPAND_ROUNDS = 3000 (regular stamps), WORKBLOCK_EXPAND_ROUNDS_PN = 1000 (propagation-node stamps — cheaper because store-and-forward already throttles), WORKBLOCK_EXPAND_ROUNDS_PEERING = 25 (peering keys between propagation nodes — even cheaper).

5.7.3 Tickets — pre-shared shortcut around proof-of-work

A ticket is a 16-byte (TICKET_LENGTH = TRUNCATED_HASHLENGTH//8) shared secret a recipient hands to a known sender, letting them skip the PoW step. With a ticket, the "stamp" becomes:

stamp = SHA256(ticket || message_id)[:32]      # truncated to STAMP_SIZE

(LXMessage.py::get_stamp line 297). The recipient validates by trying every ticket they've issued the sender against the inbound stamp:

# LXMessage.py::validate_stamp line 270-280
for ticket in tickets:
    if self.stamp == RNS.Identity.truncated_hash(ticket + self.message_id):
        self.stamp_value = LXMessage.COST_TICKET
        return True

COST_TICKET is a sentinel value (not a real cost) that just marks "valid by ticket".

Tickets are exchanged via the FIELD_TICKET = 0x0C key in the fields dict of an inbound message:

# LXMRouter.lxmf_delivery, line 1741-1752
if message.signature_validated and FIELD_TICKET in message.fields:
    ticket_entry = message.fields[FIELD_TICKET]   # [expires_unix_seconds, ticket_bytes]
    if type(ticket_entry) == list and len(ticket_entry) > 1:
        if time.time() < ticket_entry[0]:
            self.remember_ticket(message.source_hash, ticket_entry)

Format: fields[FIELD_TICKET] = [expires_unix_seconds(int), ticket(bytes, 16)]. Stored under the sender's source_hash in the receiver's persistent ticket cache. Subsequent outbound messages from the receiver to the same sender automatically use this ticket via LXMessage.outbound_ticket. Tickets expire at expires_unix_seconds; expired tickets are evicted and the next outbound message falls back to PoW.

5.7.4 The full stamp-cost field inventory

LXMF announces (§4.3) carry a stamp_cost integer in the app_data msgpack array's element [1]. A receiver tells potential senders "you must do this much PoW to message me" by setting their delivery destination's stamp_cost and re-announcing. Senders who get this announce store the cost in RNS.Identity.known_destinations[dest_hash][3].app_data and apply it to outbound messages via LXMRouter.outbound_stamp_costs.

When a receiver gets a message:

If delivery_destination.stamp_cost == None: no stamp required; messages without one are accepted.
If delivery_destination.stamp_cost != None AND the inbound message has no valid stamp AND _enforce_stamps == True: the message is dropped (LXMRouter.py:1768-1770).
If _enforce_stamps == False (default): the message is accepted regardless, and the application is told via message.stamp_valid whether the stamp checked out.

A clean-room implementation that doesn't implement stamps at all will:

Successfully send to peers with stamp_cost = None.
Be silently rejected by peers with stamp_cost != None AND _enforce_stamps.
Be flagged as "untrusted" / "spam" in receiver UIs that promote stamp validation to a UX signal even without enforcement.

For interop coverage today, "implement PoW for outbound; tolerate-but-don't-validate inbound" is the minimum. Full ticket support is a Tier-3 nice-to-have.

5.7.5 Source map

File	What
`LXMF/LXStamper.py:18-46`	`stamp_workblock`, `stamp_value`, `stamp_valid`
`LXMF/LXMessage.py:41`	`TICKET_LENGTH = 16`
`LXMF/LXMessage.py:270-291`	`validate_stamp` (ticket-then-PoW dispatch)
`LXMF/LXMessage.py:293-324`	`get_stamp` (ticket-or-PoW emission)
`LXMF/LXMRouter.py:1741-1774`	inbound dispatch — ticket cache + stamp validation + drop logic
`LXMF/LXMF.py:19`	`FIELD_TICKET = 0x0C` constant

5.8 Propagation node protocol (offline message store-and-forward)

A propagation node is an LXMF node configured to accept and store messages on behalf of recipients who are temporarily offline, then deliver them when the recipient comes back online and asks. Without propagation nodes, every message requires both peers online simultaneously — a fatal assumption for mobile / mesh-edge deployments. Propagation nodes form a peer mesh that syncs messages between themselves so a recipient can retrieve mail from any one of them.

The PROPAGATED LXMF method (LXMessage.py:423-441, mentioned in flows/send-link-lxmf.md step 3) submits a message to a propagation node rather than directly to the recipient. The propagation node stores it and offers it to peers via §5.8.2 sync, and to the recipient via §5.8.3 retrieval.

5.8.1 The `lxmf.propagation` destination

Every propagation node registers a SINGLE destination of name lxmf.propagation (LXMRouter.py:173):

self.propagation_destination = RNS.Destination(
    self.identity, IN, SINGLE, APP_NAME, "propagation",
)

Per §1.2, the well-known name_hash is e03a09b77ac21b22258e (SHA256("lxmf.propagation")[:10]). The propagation node's identity is its own — different propagation nodes have different identity hashes and therefore different destination hashes. Receivers of lxmf.propagation announces filter by name_hash to surface "propagation node available" UI separately from "messageable peer available" UI per §4.4.

The destination registers four request handlers via register_request_handler (LXMRouter.py:651-655):

Path	Constant	Allow	Purpose
`/offer`	`LXMPeer.OFFER_REQUEST_PATH`	`ALLOW_ALL`	Peer-to-peer message-set offer (§5.8.2)
`/get`	`LXMPeer.MESSAGE_GET_PATH`	`ALLOW_ALL`	Client message retrieval (§5.8.3)
`/stats`	`LXMRouter.STATS_GET_PATH`	implementation-defined	Operator stats query
`/sync`	`LXMRouter.SYNC_REQUEST_PATH`	`ALLOW_LIST`	Operator-triggered sync push

All four are reached over an active Reticulum Link via the §11 REQUEST/RESPONSE protocol. The link must be identify()-d before /offer and /get requests are honored — that's how the propagation node knows which client / peer is making the request.

5.8.2 Peer-to-peer sync via `/offer`

Two propagation nodes that have peered with each other periodically sync. The initiator sends an /offer REQUEST whose data is:

data = [peering_key(32), [transient_id_1, transient_id_2, ...]]

Where:

peering_key is a 32-byte proof-of-work key per §5.8.4.
transient_id_N is the 16-byte hash of an LXM (= SHA256(lxmf_data)[:16] — the truncated hash of the full encrypted LXMF body) that the offering node has and thinks the receiving node might want.

The receiving node validates the peering_key, then for each transient_id:

If it already has the message in propagation_entries: skip.
Otherwise: add to wanted_ids.

Then returns one of three response shapes (LXMRouter.py:2185-2187):

Response	Meaning
`False` (boolean)	Peer already has every offered message; no transfer needed.
`True` (boolean)	Peer wants every offered message.
`[wanted_id_1, ...]` (list)	Peer wants the listed subset only.

If the response indicates the peer wants any messages, the offering node packs them into a Resource (§10) and sends:

resource_data = msgpack.packb([time.time(), [lxmf_data_1, lxmf_data_2, ...]])
RNS.Resource(resource_data, link, callback=...)

The Resource contains the full encrypted LXMF bodies — the bytes that were signed and encrypted by the original sender; the propagation nodes never decrypt them. The receiving node writes each one to its propagation store under its transient_id key.

Error responses (LXMPeer.py:14-50):

Constant	Hex	Meaning
`ERROR_NO_IDENTITY`	`0xf0`	Link wasn't `identify()`-d before the offer arrived. Initiator should retry with `link.identify()`.
`ERROR_NO_ACCESS`	`0xf1`	Peer rejected (e.g. `from_static_only=True` on the receiver).
`ERROR_THROTTLED`	`0xf2`	Peer is rate-limiting; postpone for `PN_STAMP_THROTTLE` (default 30 minutes).
`ERROR_INVALID_KEY`	`0xf3`	Peering key failed proof-of-work validation.
`ERROR_INVALID_DATA`	`0xf4`	Offer payload didn't match the expected `[key, [ids]]` shape.
`ERROR_NOT_FOUND`	`0xf5`	(Used by `/sync` and stats-query paths)

5.8.3 Client retrieval via `/get`

A regular LXMF client (Sideband, NomadNet client, custom) retrieves stored messages with an /get REQUEST whose data is:

data = [wanted_ids, have_ids, optional_transfer_limit_kb]

Where:

wanted_ids = None AND have_ids = None triggers a listing query: the propagation node returns [transient_id_1, transient_id_2, ...] of every message it holds for the requesting identity, sorted by size ascending.
wanted_ids is a list of transient_ids the client wants delivered. Propagation node responds with a Resource (or single packet if small enough) carrying msgpack.packb([time.time(), [lxmf_data_1, ...]]).
have_ids is a list of transient_ids the client confirms it has stored locally. Propagation node deletes those from its store. (Equivalent to "ack and purge".)
optional_transfer_limit_kb lets the client cap the transfer size — propagation node skips messages that would exceed the cap.

Common usage: client first sends /get with [None, None] to get the list, picks which ones it wants based on size, then sends /get with [wanted_subset, prior_subset_to_purge] to fetch the new ones and acknowledge previously-fetched ones.

The propagation node only returns messages whose propagation_entries[tid][0] == requester's destination_hash (LXMRouter.py:1440, 1455) — each message is keyed to its intended recipient and the propagation node is structurally unable to deliver it to the wrong address. The LXMF body is still encrypted to the recipient's public key as a defence-in-depth.

5.8.4 Peering keys (PoW for peer-to-peer auth)

Two propagation nodes that want to peer must each compute a peering key for the relationship (LXStamper.py::validate_peering_key and stamp_workblock with WORKBLOCK_EXPAND_ROUNDS_PEERING = 25):

peering_id = self.identity.hash + remote_identity.hash    # 32 bytes (16 + 16)
workblock  = stamp_workblock(peering_id, expand_rounds=25)
peering_key = (find any 32B value such that
               SHA256(workblock || peering_key)
               has at least target_cost leading zero bits)

target_cost is the receiving node's peering_cost (announced in element [5][2] of the propagation announce app_data, see §5.8.5). With only 25 rounds of HKDF expansion (vs 3000 for regular message stamps in §5.7), the workblock is ~6 KiB and peering keys can be computed in milliseconds. Peering keys are amortized: computed once between two propagation nodes and reused for every subsequent /offer for the lifetime of the peering.

5.8.5 Propagation node announce app_data

Distinct from §4.3 (which is for lxmf.delivery). For lxmf.propagation announces, LXMRouter.get_propagation_node_app_data (line 307-319) emits a 7-element msgpack array:

announce_data = [
    False,                                  # [0] legacy-LXMF-PN-support flag (always False now)
    int(time.time()),                       # [1] node timebase (unix seconds, big-int)
    node_state,                             # [2] bool — accepting messages right now?
    propagation_per_transfer_limit,         # [3] int — per-transfer cap in KB
    propagation_per_sync_limit,             # [4] int — per-sync incoming cap in KB
    [stamp_cost, stamp_cost_flexibility,
     peering_cost],                         # [5] list of three ints
    metadata,                               # [6] dict — operator-supplied node metadata
]
return msgpack.packb(announce_data)

Element [5] sub-fields:

Index	Name	Meaning
`[5][0]`	`stamp_cost`	PoW cost (leading zero bits) for client `/get` retrieval stamps
`[5][1]`	`stamp_cost_flexibility`	Tolerance — client stamps within this many bits below `stamp_cost` are still accepted
`[5][2]`	`peering_cost`	PoW cost for peering keys per §5.8.4

Receivers parse this via pn_announce_data_is_valid (LXMF/LXMF.py:191-206), which insists on exactly 7 elements with type-correct positions. A client that misparses element [5] as a single integer (rather than a 3-element list) silently fails to compute the right peering / retrieval stamp and is rejected — this is the most common interop break in custom propagation-node implementations.

5.8.6 Source map

File	What
`LXMF/LXMRouter.py:173`	propagation_destination construction
`LXMF/LXMRouter.py:306-322`	propagation announce app_data shape
`LXMF/LXMRouter.py:650-651`	`/offer` and `/get` handler registration
`LXMF/LXMRouter.py:1426-1500`	`message_get_request` handler (client `/get`)
`LXMF/LXMRouter.py:2145-2200`	`offer_request` handler (peer `/offer`)
`LXMF/LXMPeer.py:14-50`	path constants and error-response constants
`LXMF/LXMPeer.py:370-486`	initiator-side `/offer` flow
`LXMF/LXStamper.py::validate_peering_key`	peering-key PoW validation
`LXMF/LXMF.py:191-206`	`pn_announce_data_is_valid` parser

5.9 LXMF field constants and helper specifiers

The fields dict inside an LXMF message (the 4th element of the msgpack array described in §5.3) is keyed by 1-byte integers. Upstream LXMF/LXMF.py (verified against LXMF 0.9.7 by tools/verify_lxmf_fields.py) defines the following allocations.

5.9.1 Top-level `fields` dict keys

Sender and receiver agree on these keys; each value's structure is field-specific (described below where it matters at byte level).

Key	Constant	Purpose
`0x01`	`FIELD_EMBEDDED_LXMS`	A list of further LXMF messages embedded inside this one (used for forwarding / bundling).
`0x02`	`FIELD_TELEMETRY`	A single telemetry snapshot (Sideband telemetry — see Sideband for the inner format; LXMF is opaque to the contents).
`0x03`	`FIELD_TELEMETRY_STREAM`	A list of telemetry snapshots (history flush).
`0x04`	`FIELD_ICON_APPEARANCE`	Sender-supplied avatar / appearance hint.
`0x05`	`FIELD_FILE_ATTACHMENTS`	A list of attached files (multiple attachments per message). See §5.9.7 for the wire shape.
`0x06`	`FIELD_IMAGE`	Single embedded image — `[extension_string, image_bytes]`. See §5.9.2 for the wire shape.
`0x07`	`FIELD_AUDIO`	Single embedded audio clip — `[mode_byte, audio_bytes]`. Mode byte chooses the codec; see §5.9.3.
`0x08`	`FIELD_THREAD`	Conversation thread ID (links related messages).
`0x09`	`FIELD_COMMANDS`	List of commands the sender is requesting the receiver execute (Sideband node/command protocol).
`0x0A`	`FIELD_RESULTS`	List of results for commands previously requested via `FIELD_COMMANDS`.
`0x0B`	`FIELD_GROUP`	Group / channel association metadata.
`0x0C`	`FIELD_TICKET`	Stamp ticket grant — `[expires_unix_seconds(int), ticket_bytes(16)]`. See §5.7 for the anti-spam protocol.
`0x0D`	`FIELD_EVENT`	Event-style payload (alert, state change).
`0x0E`	`FIELD_RNR_REFS`	Reticulum Node Registry references.
`0x0F`	`FIELD_RENDERER`	Renderer hint for the message `content` body — see §5.9.4 for accepted values.
`0xFB`	`FIELD_CUSTOM_TYPE`	App-defined type identifier accompanying `FIELD_CUSTOM_DATA`.
`0xFC`	`FIELD_CUSTOM_DATA`	App-defined opaque data — meaning given by `FIELD_CUSTOM_TYPE`.
`0xFD`	`FIELD_CUSTOM_META`	App-defined metadata alongside `FIELD_CUSTOM_DATA`.
`0xFE`	`FIELD_NON_SPECIFIC`	Development / unstructured payload — not for production.
`0xFF`	`FIELD_DEBUG`	Debug payload — not for production.

⚠️ UNVERIFIED: the byte-level shape of FIELD_EMBEDDED_LXMS, FIELD_TELEMETRY*, FIELD_COMMANDS, FIELD_RESULTS, FIELD_GROUP, FIELD_EVENT, and FIELD_RNR_REFS is not described here because no test vectors have been captured against upstream Sideband emissions for these. The constants are verified (see tools/verify_lxmf_fields.py) but the value structures are application-defined and not pinned by LXMF itself. Future PRs should add per-field byte layouts as test vectors arrive. (FIELD_FILE_ATTACHMENTS was on this list until 2026-05-18 — its shape is now documented in §5.9.7 from upstream Sideband source.)

5.9.2 `FIELD_IMAGE` (`0x06`) value shape

fields[0x06] = [extension_string(bytes-or-str), image_bytes(bytes)]

The extension_string is the lowercase file extension WITHOUT a leading dot ("jpg", "png", "webp"). The image_bytes is the raw image file content. Receivers must tolerate the extension arriving as either msgpack str (0xa0..0xbf / 0xd9..0xdb) or msgpack bin (0xc4..0xc6) — different encoders pick differently. See §9.3 for the str-vs-bin distinction and §10 for how images larger than a single Reticulum DATA packet are delivered via Resource over a Link.

5.9.3 `FIELD_AUDIO` (`0x07`) value shape

fields[0x07] = [mode_byte(int), audio_bytes(bytes)]

mode_byte is one of the AM_* constants defined in LXMF/LXMF.py (verified by tools/verify_lxmf_fields.py):

Byte	Constant	Codec
`0x01`	`AM_CODEC2_450PWB`	Codec2 450 bps pseudo-wideband
`0x02`	`AM_CODEC2_450`	Codec2 450 bps
`0x03`	`AM_CODEC2_700C`	Codec2 700C
`0x04`	`AM_CODEC2_1200`	Codec2 1200 bps
`0x05`	`AM_CODEC2_1300`	Codec2 1300 bps
`0x06`	`AM_CODEC2_1400`	Codec2 1400 bps
`0x07`	`AM_CODEC2_1600`	Codec2 1600 bps
`0x08`	`AM_CODEC2_2400`	Codec2 2400 bps
`0x09`	`AM_CODEC2_3200`	Codec2 3200 bps
`0x10`	`AM_OPUS_OGG`	Opus in OGG container
`0x11`	`AM_OPUS_LBW`	Opus low-bandwidth
`0x12`	`AM_OPUS_MBW`	Opus medium-bandwidth
`0x13`	`AM_OPUS_PTT`	Opus push-to-talk profile
`0x14`	`AM_OPUS_RT_HDX`	Opus realtime half-duplex
`0x15`	`AM_OPUS_RT_FDX`	Opus realtime full-duplex
`0x16`	`AM_OPUS_STANDARD`	Opus standard
`0x17`	`AM_OPUS_HQ`	Opus high-quality
`0x18`	`AM_OPUS_BROADCAST`	Opus broadcast
`0x19`	`AM_OPUS_LOSSLESS`	Opus lossless
`0xFF`	`AM_CUSTOM`	Client-detected — inspect `audio_bytes` to determine the codec

5.9.4 `FIELD_RENDERER` (`0x0F`) value shape

fields[0x0F] = renderer_byte(int)

One of the RENDERER_* constants:

Byte	Constant	Rendering
`0x00`	`RENDERER_PLAIN`	Plain text — no formatting
`0x01`	`RENDERER_MICRON`	NomadNet Micron markup (see NomadNet docs)
`0x02`	`RENDERER_MARKDOWN`	CommonMark / GitHub-flavored Markdown
`0x03`	`RENDERER_BBCODE`	BBCode-style tags

Implementations should fall back to RENDERER_PLAIN for any unknown renderer byte rather than rejecting the message.

5.9.5 Propagation-node metadata keys

Distinct from the top-level fields dict, these PN_META_* keys are used inside the fields[0x02] element of a propagation-node announce (§5.8.5 element [2]) or in /get-flow metadata responses. Allocations may change before LXMF 1.0.0 — code defensively.

Byte	Constant	Purpose
`0x00`	`PN_META_VERSION`	Propagation protocol version
`0x01`	`PN_META_NAME`	Operator-supplied node name
`0x02`	`PN_META_SYNC_STRATUM`	Sync tier in the propagation mesh
`0x03`	`PN_META_SYNC_THROTTLE`	Operator-imposed sync throttle
`0x04`	`PN_META_AUTH_BAND`	Auth requirement (open / restricted / private)
`0x05`	`PN_META_UTIL_PRESSURE`	Utilization back-pressure hint
`0xFF`	`PN_META_CUSTOM`	Operator-defined extensions

⚠️ UNVERIFIED: the value type for each PN_META_* key is not yet pinned by this spec — upstream still treats them as a soft contract. Implementations should preserve unknown keys round-trip rather than dropping them.

5.9.6 Functionality signalling keys

For announce-level capability negotiation:

Byte	Constant	Meaning
`0x00`	`SF_COMPRESSION`	Sender supports compressed message bodies (see §10.12)

5.9.7 `FIELD_FILE_ATTACHMENTS` (`0x05`) value shape

fields[0x05] = [ [filename(str-or-bytes), file_bytes(bytes)], ... ]

A list of attachments — one LXMF message may carry more than one file. Each attachment is itself a 2-element list: element [0] is the file name, element [1] is the raw file content. As with FIELD_IMAGE (§5.9.2) the file name may arrive as msgpack str or bin depending on the encoder — receivers must tolerate both (see §9.3).

The file name is sender-controlled and untrusted. Upstream Sideband strips ../ from it on receive (sbapp/ui/messages.py: filename = str(attachment[0]).replace("../", "")); receivers MUST sanitise it — reject or strip path separators, .. segments and control characters — before display or save, and never let it influence a write path. The extension is likewise untrusted: do not auto-open or auto-execute an attachment based on its claimed type.

Files larger than a single Reticulum DATA packet are delivered as a Resource over a Link, identically to large FIELD_IMAGE payloads (§10).

Source: markqvist/Sideband sbapp/sideband/core.py (fields[LXMF.FIELD_FILE_ATTACHMENTS] = [attachment], where each attachment = [filename, filedata]) and sbapp/ui/messages.py (receive side indexes attachment[0] filename / attachment[1] bytes). Confirmed from upstream source 2026-05-18; a captured wire test vector would further pin the msgpack str-vs-bin choice.

5.9.8 `fields[16]` (`0x10`): tap-back reactions (app extension)

⚠️ UNVERIFIED: this key is outside the upstream LXMF allocation range (§5.9.1) — there is no FIELD_REACTION constant in LXMF/LXMF.py. The shape below is a convergent app-extension across three shipping FOSS clients (reticulum-mobile-app, Quad4-Software/MeshChatX, torlando-tech/columba release/v0.10.x and later). Documented here because the convergence is stable enough for new implementations to interop. Replace with the upstream allocation if/when markqvist/LXMF blesses one.

fields[16] = {
  "reaction_to": <target LXMF message_id, 64-char hex string>,
  "emoji":       <unicode>,
  "sender":      <reactor's identity hash, 32-char hex string>,
}

The carrying LXMF has empty content and empty title; fields[16] IS the entire payload.
reaction_to is the canonical LXMF message_id (= message_hash of §5.5; the same value used as the workblock input for stamps in §5.7). It MUST be derived from the recipient-side raw wire payload bytes per the §5.5 normative rule — a decode-re-encode diverges for any non-trivial fields map (including reply-carrying messages, §5.9.9) and the reaction misses the relay rewrite-cache for that target.
sender is the reactor's identity hash (SHA256(identity_public_key)[:16]), NOT the lxmf.delivery source_hash. Both are 16-byte SHA-256 truncations and trivially easy to conflate, but receivers key reaction aggregation by (emoji, sender), so emitting the destination hash mis-buckets against every (identity-hash-emitting) peer. All three convergent clients emit the identity hash (python/reticulum_wrapper.py::send_reaction in Columba release/v0.10.x and v2.0.0-beta+; meshchat.py::send_reaction in MeshChatX; ReticulumEngine.kt::sendReaction in reticulum-mobile-app).
Receivers MUST aggregate by (emoji, sender) with dedup and SHOULD NOT render the reaction-carrying LXMF as a separate bubble.
String-vs-bytes tolerance: msgpack values may arrive as str or bin depending on encoder. Receivers MUST accept both — same precedent as §5.6 for the LXMF signature.
Inner-map decode tolerance. The fields[16] value is a msgpack map. Strongly-typed decoders may surface it as either a string-keyed map or an Any-keyed map (Map<String, Any> / Map<Any?, Any?>, map[string]any / map[any]any, etc.), depending on the runtime msgpack library's choice given the key types seen on the wire. Receivers MUST tolerate both via a runtime cast that does not depend on the outer map's static key type. A silent type-assertion failure on the "wrong" variant produces a no-log no-error drop indistinguishable on the receive side from "the message never arrived."
Relay routing. If the target message arrived over a Link whose validated §6.7.6 LINKIDENTIFY peer differs from the LXMF body's source_hash, the reaction-carrying LXMF MUST egress to the link peer's destination hash, not source_hash; otherwise the reaction bypasses the relay's fanout and is delivered direct to the original sender instead of the relay group.

5.9.9 `fields[0x30]` + optional `fields[0x31]`: reply-to threading (app extension)

⚠️ UNVERIFIED: as with §5.9.8, these keys are outside the upstream LXMF allocation range. The shape below is the convergent app-extension across the same three FOSS clients. Columba's earlier release/v0.10.x builds used a legacy single-key shape (fields[16] = {"reply_to": <hex>}); Columba v2.0.0-beta (released 2026-05-23) and later emit the dual-key shape per torlando-tech/columba#926. The legacy-tolerance branch in the inbound rule below has a finite lifetime — drop it once < v2.0.0-beta Columba is no longer in the field.

fields[0x30] = <raw 32-byte LXMF message_id (NOT hex-encoded)>
fields[0x31] = <UTF-8 quoted content>     # optional

fields[0x30] IS the canonical LXMF message_id from §5.7 — the raw 32 bytes on the wire, not a hex-encoded string. The message_id MUST be computed over raw wire payload bytes per §5.5; a decode-re-encode diverges for reply messages (their fields map is non-trivial) and breaks any relay rewrite-cache keyed on the value.
fields[0x31] (optional) carries the quoted text for offline-receiver fallback so the recipient can render a quote preview even when the original message hasn't arrived locally. Useful on intermittent links; SHOULD be omitted when the sender doesn't want to pay the airtime.
Legacy-tolerance (inbound only). Receivers SHOULD also accept Columba's earlier fields[16] = {"reply_to": <64-char hex>} shape for interop with already-deployed release/v0.10.x builds. Senders MUST NOT emit it — emit only [0x30]/[0x31]. The same inner-map dual-shape tolerance described in §5.9.8 applies to this legacy parse branch (the value is a msgpack map).
Bandwidth rationale. The dual-key shape is roughly 2× smaller than the legacy single-key shape for the same information — 32 raw bytes + 1-byte key vs. a msgpack-encoded dict carrying a hex-string hash. This is the Zen of Reticulum alignment cited by the MeshChatX maintainer in Quad4-Software/MeshChatX#14 and accepted by the Columba maintainer in torlando-tech/columba#926.
The raw-bytes branches (fields[0x30], fields[0x31]) are not maps and do not have the §5.9.8 inner-map risk, but receivers MUST still tolerate both bytes and str carriers per §5.6 in case an encoder ships the hash hex-encoded by mistake.
Relay routing. Same as §5.9.8 — if the target message arrived over a Link whose §6.7.6 LINKIDENTIFY peer differs from source_hash, the reply-carrying LXMF MUST egress to the link peer's destination hash.

Reference implementations: Quad4-Software/MeshChatX (meshchatx/src/backend/lxmf_utils.py; meshchat.py::send_reaction); torlando-tech/columba v2.0.0-beta and later (python/reticulum_wrapper.py); thatSFguy/reticulum-mobile-app (shared/.../engine/ReticulumEngine.kt::extractField16, sendReaction, tryDeliverOverLink::replyFields).

5.10 Source

LXMF/LXMessage.py for pack/unpack; LXMF/LXMF.py for the app_data extraction helpers and the field/audio/renderer constants enumerated in §5.9; LXMF/LXStamper.py for stamps; LXMF/LXMRouter.py for receive-side stamp/ticket dispatch and propagation handlers; LXMF/LXMPeer.py for the propagation peer-to-peer state machine.

6. Reticulum Link protocol

A Link is an ephemeral encrypted channel between two destinations, established via a 2-packet handshake (LINKREQUEST → LRPROOF) and used afterward for full-duplex DATA.

6.1 LINKREQUEST (initiator → responder)

A regular packet with packet_type = LINKREQUEST (2), dest_type = SINGLE, addressed to the responder's destination hash. Body:

initiator_X25519_pub(32) || initiator_Ed25519_pub(32) || [signalling(3)]

Both initiator-side keys are fresh ephemeral keys (not the initiator's long-term identity). The optional 3-byte signalling field encodes a packed 21-bit MTU and 3-bit link mode — see §6.6 for the bit layout and the negotiation rules. Receivers detect its presence by body length: len(data) == 64 means no signalling, len(data) == 67 means signalling present.

6.2 LRPROOF (responder → initiator)

A packet_type = PROOF (3) with context = 0xff, addressed to the link itself — i.e. dest_hash in the packet header is the 16-byte link_id (RNS/Packet.py:182-184: when context is LRPROOF, header += destination.link_id and the body is appended unencrypted).

Body (proof_data at RNS/Link.py:371):

signature(64) || responder_X25519_pub(32) || [signalling(3)]

Only the responder's X25519 is fresh-ephemeral; the responder signs with its long-term Ed25519 private key (asymmetric with the initiator). The responder's long-term Ed25519 public key is not sent on the wire — both sides already know it from the responder's prior announce, and it is included implicitly in the signature input. Signature input (RNS/Link.py:373-374 for the signer, :417 for the validator):

signed_data = link_id || responder_X25519_pub || responder_long_term_Ed25519_pub || [signalling]

The full wire packet is therefore: flags(1) || hops(1) || link_id(16) || context=0xff(1) || signature(64) || responder_X25519_pub(32) || [signalling(3)].

The signalling slot, when present, carries the responder's confirmed_mtu and the link mode in a 24-bit packed integer — see §6.6. Receivers detect its presence by length: body 96 vs 99 bytes. Critical for interop: the signalling bytes (when present) MUST be included in signed_data exactly where shown above; an implementation that signs without them on a peer that emits them — or vice versa — fails signature validation and the link never establishes. This is the most common cause of link-handshake failures with mixed-version peers.

6.3 link_id derivation

link_id = SHA256(hashable_part_of_LINKREQUEST_packet)[:16]

hashable_part is built by Packet.get_hashable_part (RNS/Packet.py:354-361):

hashable_part = byte(flags & 0x0F) || raw[N:]
   where N = 2  for HEADER_1   (strip flags + hops)
         N = 18 for HEADER_2   (strip flags + hops + transport_id)

The "hashable part" deliberately strips header_type, context_flag, transport_type (top 4 bits of flags — modifiable by transit relays), the hops byte (modified by every relay), and (for HEADER_2) the transport_id (added by the originator and re-written by each relay). What remains in both cases is the low nibble of flags + dest_hash + context + body, so the resulting link_id is the same whether the LINKREQUEST is hashed at the initiator (HEADER_1) or at the responder after one or more transport relays (HEADER_2). Both sides agree on the 16-byte ID.

For LINKREQUEST packets specifically, the trailing signalling bytes (if present, indicated by len(packet.data) > Link.ECPUBSIZE in link_id_from_lr_packet at RNS/Link.py:341-348) are stripped from the END of hashable_part before hashing, so the link_id is invariant under MTU-discovery signalling.

6.4 Session keys and link activation

After the LINKREQUEST/LRPROOF exchange completes, both peers must (a) derive matching link session keys, (b) drive the link state machine to ACTIVE, and (c) settle on the wire conventions for all subsequent traffic on the link. The three subsections below cover each step. Skipping any of them silently breaks interop in distinct ways: §6.4.1 wrong → no peer can decrypt anything; §6.4.2 omitted → the responder never reaches ACTIVE and silently drops all link DATA; §6.4.3 wrong on multi-hop → packets are dropped at the destination's transport filter.

6.4.1 Session key derivation

Both sides compute:

shared       = X25519(my_ephemeral_priv, peer_ephemeral_pub)
session_key  = HKDF(shared, salt = link_id, info = "", L = 64)
signing_key  = session_key[0..32]
encrypt_key  = session_key[32..64]

Subsequent DATA packets on the link use the Link-derived-key Token format (section 3.1, no ephemeral_pub prefix).

6.4.2 LRRTT — initiator's link-activation packet

After validating the responder's LRPROOF and deriving session keys, the initiator MUST send a Link Round-Trip Time packet to the responder before transmitting any application DATA. The wire form is:

Field	Value
`header_type`	`HEADER_1`
`packet_type`	`DATA (0x00)`
`destination_type`	`LINK (0x03)`
`dest_hash`	`link_id`
`context`	`LRRTT (0xFE)`
body (plaintext)	`umsgpack.packb(rtt_seconds)` — a single msgpack float64 (9 bytes) carrying the initiator's measured RTT in seconds (LRREQ-send to LRPROOF-receive)
body (wire)	the plaintext above, encrypted with the link's session keys per §3.1 (link form Token, no `eph_pub` prefix)

Source: RNS/Link.py:440-442 constructs and sends the packet immediately after LRPROOF validation:

rtt_data   = umsgpack.packb(self.rtt)
rtt_packet = RNS.Packet(self, rtt_data, context=RNS.Packet.LRRTT)
rtt_packet.send()

The responder uses receipt of LRRTT as the trigger to transition its link state from HANDSHAKE to ACTIVE (RNS/Link.py:534-553). The initiator transitions independently upon LRPROOF validation (Link.py:430-432); the responder MUST NOT transition before LRRTT arrives. The responder routes context LRRTT to Link.rtt_packet() from its main receive() dispatch at RNS/Link.py:1056-1058.

Link.rtt_packet() is also the only path on the responder side that fires the link_established callback (Link.py:550-551). Without that callback, application layers cannot install link-state policies that depend on ACTIVE — most importantly, LXMF's LXMRouter.delivery_link_established (LXMF/LXMRouter.py:1852-1859) only calls link.set_resource_strategy(ACCEPT_APP) from this callback. Until that strategy is installed, the responder's Link.receive() hits the silent-drop branch elif self.resource_strategy == Link.ACCEPT_NONE: pass (RNS/Link.py:1087) on every inbound RESOURCE_ADV, and any oversize LXMF delivered as a Resource is discarded with no log line at default levels. This is silent, end-to-end, default-config message loss for an initiator that completes LRPROOF and immediately sends RESOURCE_ADV without first sending LRRTT.

The exact RTT value reported is non-load-bearing: the responder takes max(its_own_measurement, initiator_reported) (Link.py:540). Implementations that don't have an accurate RTT measurement at this point may report a coarse estimate or zero — the responder's measurement carries forward when the initiator reports a smaller value. The value is, however, included in the encrypted body and so is integrity-bound to the link session keys; a peer that fails to encrypt this body with the correct link keys will fail decrypt and rtt_packet returns without transitioning to ACTIVE.

6.4.3 Header type for post-handshake DATA and Resource

All packets sent on an active Link — link DATA (packet_type=DATA, context=NONE), Resource control packets (context ∈ {RESOURCE_ADV (0x02), RESOURCE_REQ (0x03), RESOURCE_HMU (0x04), RESOURCE_ICL (0x06), RESOURCE_RCL (0x07)}), Resource part packets (context=RESOURCE (0x01)), and link control packets (KEEPALIVE, LRRTT, LINKCLOSE, LINKIDENTIFY, REQUEST, RESPONSE, CHANNEL) — MUST be emitted with header_type=HEADER_1 and no transport_id, regardless of whether the responder is reachable directly or through one or more transit relays.

This is asymmetric to the LINKREQUEST that established the link. LINKREQUEST is destination-hash-routed via path_table and therefore eligible for HEADER_2 with transport_id set to the next-hop relay; the relay's path_table-forwarding branch strips transport_id (HEADER_2 → HEADER_1) at the last hop (RNS/Transport.py:1500-1517):

if remaining_hops > 1:
    # Just increase hop count and transmit (HEADER_2 preserved)
elif remaining_hops == 1:
    # Strip transport headers and transmit
    new_flags = (RNS.Packet.HEADER_1) << 6 | (Transport.BROADCAST) << 4 | (packet.flags & 0b00001111)
    new_raw  = struct.pack("!B", new_flags)
    new_raw += struct.pack("!B", packet.hops)
    new_raw += packet.raw[(RNS.Identity.TRUNCATED_HASHLENGTH//8)+2:]

Once the link exists, post-handshake traffic addressed to link_id is routed via link_table, not path_table. The link_table forwarding branch (RNS/Transport.py:1587-1622) does NOT touch the header — it bumps hops and forwards the bytes verbatim:

new_raw = packet.raw[0:1]
new_raw += struct.pack("!B", packet.hops)
new_raw += packet.raw[2:]
Transport.transmit(outbound_interface, new_raw)

A HEADER_2 link DATA packet would therefore arrive at the destination with transport_id intact, where the receiver's Transport.packet_filter (RNS/Transport.py:1283-1285) drops it as "for another transport instance" because the embedded transport_id is the relay's identity, not the receiver's:

if packet.transport_id != None and packet.packet_type != RNS.Packet.ANNOUNCE:
    if packet.transport_id != Transport.identity.hash:
        RNS.log("Ignored packet ... in transport for other transport instance", RNS.LOG_EXTREME)
        return False

The drop fires only at LOG_EXTREME (level 7) and is invisible at the default LOG_NOTICE (level 3); see §9.9.

Upstream RNS does not trip this in practice because its own constructors use Packet's defaults (HEADER_1, transport_id=None) regardless of multi-hop. The relevant ones, all in RNS/Resource.py and RNS/Link.py, take the form RNS.Packet(self.link, body, context=...) (e.g. Resource.py:521 for RESOURCE_ADV); see Packet.__init__ at RNS/Packet.py:122-123 for the default values. Upstream-to-upstream interop therefore never exercises this path. A clean-room implementation that copies the LINKREQUEST multi-hop pattern verbatim — setting transport_id on every link-bound packet — silently fails on any link with one or more transit relays in the path. The unit tests for that implementation pass (both sides agree on the same wire mistake) but localhost-rnsd interop with a real upstream destination drops every Resource part.

The asymmetry summarized: the same Link is set up via a HEADER_2-eligible LINKREQUEST, but uses HEADER_1 for everything else once established.

6.5 Packet receipts (regular `PROOF` packets)

A PROOF-type packet (packet_type = 3, context = NONE (0x00)) is the receipt that closes the loop on every CTX_NONE DATA packet — both opportunistic DATA addressed to a SINGLE destination and DATA flowing on an active Link. Without it, the sender's PacketReceipt never resolves, its retransmit queue fires repeatedly, and on a Link the KEEPALIVE budget is exhausted and the link torn down.

This section specifies the regular PROOF body. Two related proof formats are documented elsewhere and are NOT compatible with this format:

LRPROOF (context = 0xFF) is the link-establishment proof (§6.2). Different body, different signature input.
RESOURCE_PRF (context = 0x05) is the proof for a completed Resource transfer (§10.8). Different body (resource_hash || full_proof), no signature.

6.5.1 Two body formats: explicit vs implicit

Regular PROOFs come in two wire forms (RNS/Packet.py:413-414):

EXPL_LENGTH = HASHLENGTH//8 + SIGLENGTH//8 = 32 + 64 = 96 bytes
IMPL_LENGTH = SIGLENGTH//8                 =      64 = 64 bytes

explicit body  =  packet_hash(32) || signature(64)
implicit body  =                     signature(64)

Where:

packet_hash = Identity.full_hash(original_packet.get_hashable_part()) — the full SHA-256 (32 bytes, not truncated to 16) of the prove-target packet's hashable part. get_hashable_part is the same recipe used for link_id derivation in §6.3, so the proof binds to the version of the packet that survived any HEADER_1↔HEADER_2 conversion in transit (the high nibble of flags, hops byte, and any HEADER_2 transport_id are stripped before hashing).
signature is the destination's (or link's) Ed25519 signature over packet_hash, NOT over the proof body itself. The signing key is the destination's long-term Ed25519 private key for an opportunistic DATA proof, or the link's Ed25519 signing key (Link.sig_prv, RNS/Link.py:279 / :286 in RNS 1.2.9) for a Link DATA proof — the owner identity's long-term key on the responder side, the link's ephemeral Ed25519 keypair on the initiator side. (The HKDF-derived signing_key from §6.4.1 is a separate symmetric HMAC key for the link Token form §3.1 — it cannot produce an Ed25519 signature and is not used here.)

The two forms are distinguished purely by length at the receiver. PacketReceipt.validate_proof (RNS/Packet.py:497-548) dispatches on len(proof) == 96 (explicit) vs len(proof) == 64 (implicit); lengths matching neither are rejected outright. There is no flag bit or context byte that signals which form is being used — wire length is the only signal.

6.5.2 Choosing which form to emit

Sender side, two distinct policies:

Opportunistic DATA addressed to a SINGLE destination — RNS.Identity.prove(packet, destination) at RNS/Identity.py:943-954:

def prove(self, packet, destination=None):
    signature = self.sign(packet.packet_hash)
    if RNS.Reticulum.should_use_implicit_proof():
        proof_data = signature                                  # 64 bytes
    else:
        proof_data = packet.packet_hash + signature             # 96 bytes
    proof = RNS.Packet(destination_or_proof_dest, proof_data,
                       RNS.Packet.PROOF, attached_interface=...)
    proof.send()

The default upstream value is Reticulum.__use_implicit_proof = True (RNS/Reticulum.py:256), so upstream emits the 64-byte implicit form by default. The 96-byte explicit form is only emitted when the operator's [reticulum] config sets use_implicit_proof = No. A clean-room implementation that hardcodes either single form will fail to interop with peers running the other one — receiver-side validators handle both, but a hardcoded sender writing the wrong length to the wire is not negotiable.

DATA on an active Link — RNS.Link.prove_packet(packet) at RNS/Link.py:383-394:

def prove_packet(self, packet):
    signature = self.sign(packet.packet_hash)
    proof_data = packet.packet_hash + signature                 # 96 bytes — always
    proof = RNS.Packet(self, proof_data, RNS.Packet.PROOF)
    proof.send()

with the upstream comment # TODO: Hardcoded as explicit proof for now. Link DATA proofs are always the 96-byte explicit form in RNS 1.2.4 regardless of the use_implicit_proof setting, and the matching validate_link_proof at RNS/Packet.py:449-494 has the implicit-form branch commented out with the same note. Today, Link DATA proofs are explicit-only on both ends; an implementation may match this behavior with a single hardcoded length on the link path, but should be ready to revisit if upstream re-enables the implicit branch (no fixed timeline).

6.5.3 Where the proof packet is addressed

The dest_hash position in the proof packet's outer header depends on which side of which transport the proven packet was on:

Opportunistic DATA proof: dest_hash = packet_hash[:16] (the 16-byte truncation of the full SHA-256 of the proved packet's hashable part, used as a synthetic ProofDestination — RNS/Packet.py:390-396). The proof rides through Transport.outbound and follows the reverse path home via the receiver's reverse_table.
Link DATA proof: dest_hash = link.link_id (the 16-byte link id, just like all other Link traffic; RNS/Packet.py:182-184 notes this position is filled by destination.link_id whenever the destination object is a Link). The proof rides on the link itself.

6.5.4 Wire summary

explicit form (96 bytes total body):
[ 1B flags ][ 1B hops ][ 16B dest_hash || link_id ][ 1B context=0x00 ]
[ 32B SHA256(get_hashable_part(original_packet)) ]
[ 64B Ed25519_signature(SHA256(get_hashable_part(original_packet))) ]

implicit form (64 bytes total body):
[ 1B flags ][ 1B hops ][ 16B dest_hash || link_id ][ 1B context=0x00 ]
[ 64B Ed25519_signature(SHA256(get_hashable_part(original_packet))) ]

Note context = 0x00 (NONE) in both cases — the proof-ness is conveyed by packet_type = PROOF (3) in the flag byte, not by a context. This is in contrast to LRPROOF (which uses context = 0xFF) and RESOURCE_PRF (which uses context = 0x05). The LINKPROOF (0xFD) context constant defined at RNS/Packet.py:90 is reserved but not actually used by either prove path in RNS 1.2.4.

6.5.5 Receiver tolerance

A new implementation's PROOF validator MUST accept both 64- and 96-byte bodies for opportunistic DATA proofs (per validate_proof's length-dispatch above) so it interops with peers running either policy. Hardcoding only one form at the validator silently fails on traffic from a peer with the opposite setting. Length-dispatch is also the only place the validator ever distinguishes the two — there is no "I want explicit" hint a sender can express.

A receiver that gets a PROOF whose length matches neither form treats it as malformed and returns False from validate_proof; no NACK is sent to the originator.

6.5.6 Why this matters for Link interop

After processing each NONE DATA packet on an active link, the receiver MUST emit the explicit-form PROOF described above. Without it, the sender's retransmit queue fires and the same packet arrives repeatedly, eventually exceeding the link's KEEPALIVE budget and tearing down the link. This is Packet.prove_packet upstream — non-optional for any client that wants to receive content over a Link without spamming the sender.

6.6 MTU and mode signalling (3-byte trailer on LINKREQUEST and LRPROOF)

The optional 3-byte signalling slot referenced in §6.1 and §6.2 carries two negotiated parameters in a single 24-bit big-endian packed integer: the link MTU (21 bits) and the link mode (3 bits, top of the high byte). When present, the signalling bytes are also included in the LRPROOF's signed_data, so the responder's signature commits to the negotiated values and a peer flipping a bit in transit invalidates the proof.

6.6.1 Wire layout

3 bytes total, big-endian. Byte 0 is split: top 3 bits are mode, low 5 bits are the most-significant 5 bits of the 21-bit mtu. Bytes 1 and 2 are the remaining 16 bits of mtu:

byte 0 :  M M M m m m m m       — top 3 bits = mode (0..7), low 5 bits = mtu[20..16]
byte 1 :  m m m m m m m m       — mtu[15..8]
byte 2 :  m m m m m m m m       — mtu[7..0]

Encoded by RNS/Link.py:148-152:

@staticmethod
def signalling_bytes(mtu, mode):
    if not mode in Link.ENABLED_MODES:
        raise TypeError(f"Requested link mode {Link.MODE_DESCRIPTIONS[mode]} not enabled")
    signalling_value = (mtu & Link.MTU_BYTEMASK) + (((mode << 5) & Link.MODE_BYTEMASK) << 16)
    return struct.pack(">I", signalling_value)[1:]    # big-endian uint32, drop top byte

with MTU_BYTEMASK = 0x1FFFFF (21 bits) and MODE_BYTEMASK = 0xE0 (top 3 bits of a byte).

Decoded mode and mtu (from mode_from_lr_packet line 171-176, mtu_from_lr_packet line 153-157):

mode = (signalling[0] & 0xE0) >> 5
mtu  = ((signalling[0] << 16) + (signalling[1] << 8) + signalling[2]) & 0x1FFFFF

The mtu decode trick: the full 24-bit value of all three bytes is masked with the 21-bit MTU_BYTEMASK, which strips the top 3 bits (i.e. the mode bits) without any explicit byte 0 masking step. Implementations that use (signalling[0] & 0x1F) << 16 | … instead get the same answer.

6.6.2 Mode field

3-bit value (0..7) at the top of byte 0. Defined values, with RNS/Link.py:126-145:

Mode	Name	Status in RNS 1.2.4	Derived key length
`0x00`	`MODE_AES128_CBC`	Defined, NOT enabled (sender-side will raise `TypeError`)	32 bytes
`0x01`	`MODE_AES256_CBC`	Default; the only enabled mode (`ENABLED_MODES = [0x01]`)	64 bytes
`0x02`	`MODE_AES256_GCM`	Reserved, not enabled	—
`0x03`	`MODE_OTP_RESERVED`	Reserved, not enabled	—
`0x04`–`0x07`	`MODE_PQ_RESERVED_*`	Reserved for the post-quantum migration; not enabled	—

The derived_key_length at RNS/Link.py:359-361 is what the HKDF in §6.4 produces, split as signing_key(32) || encrypt_key(32) for the AES-256 path or signing_key(16) || encrypt_key(16) for the AES-128 path.

A receiver MUST tolerate seeing any 3-bit value in the mode field on inbound traffic — mode_from_lr_packet returns the raw integer without validating it against ENABLED_MODES. The mode is enforced at handshake time (Link.handshake at line 353-368): unknown / disabled modes raise TypeError and the link transitions to CLOSED rather than ACTIVE. Senders MUST NOT emit any mode value not in ENABLED_MODES — signalling_bytes() raises if you try.

A clean-room implementation today can safely hardcode mode = 0x01 on emit. On receive, it should accept 0x01 and reject the rest as "mode not supported by this implementation" rather than silently treating them as the default — a future RNS version that flips the default to 0x04 (one of the PQ slots) would render a hardcoded-default decoder ambiguous about whether the wire bytes mean "AES_256_CBC" or "the new default".

6.6.3 MTU field

21-bit unsigned integer in the low 21 bits of the 24-bit signalling value. Max representable: 0x1FFFFF = 2,097,151 bytes. Real Reticulum HW_MTU values are radically smaller (LoRa: 508; TCP: typical HW MTU ~64 KiB; AutoInterface: matches its bearer). The 21-bit width is forward-looking: it leaves room for future high-bandwidth interfaces without a wire-format change.

When the initiator emits a LINKREQUEST with signalling, the encoded mtu is the next-hop interface's HW_MTU (RNS/Link.py:309-314):

nh_hw_mtu = RNS.Transport.next_hop_interface_hw_mtu(destination.hash)
if RNS.Reticulum.link_mtu_discovery() and nh_hw_mtu:
    signalling_bytes = Link.signalling_bytes(nh_hw_mtu, self.mode)
else:
    signalling_bytes = Link.signalling_bytes(RNS.Reticulum.MTU, self.mode)

When the responder emits an LRPROOF with signalling, the encoded mtu is the min of its own next-hop view and what arrived in the LINKREQUEST. In RNS 1.2.4 the clamp is performed by transit relays in the DATA forwarding path (RNS/Transport.py:1539-1556), which rewrite the LINKREQUEST's signalling bytes in place before forwarding so the responder's Link.validate_request (RNS/Link.py:186-200) sees the already-clamped value:

path_mtu = Link.mtu_from_lr_packet(packet) or Reticulum.MTU
nh_mtu   = receiving_interface.HW_MTU if AUTOCONFIGURE_MTU/FIXED_MTU else Reticulum.MTU
if nh_mtu < path_mtu:
    path_mtu = nh_mtu
    clamped_signalling = Link.signalling_bytes(path_mtu, mode)
    packet.data = packet.data[:-LINK_MTU_SIZE] + clamped_signalling

The clamp is rewritten into the LINKREQUEST packet's data buffer in place before that packet enters the responder's Destination.receive path, so the responder's eventual LRPROOF carries the clamped value, not the originally-requested one. The clamp also affects link_id derivation: link_id_from_lr_packet strips trailing signalling bytes before hashing (per §6.3), so this in-place rewrite doesn't change the link_id even though it does change the wire bytes.

The initiator reads confirmed_mtu back via mtu_from_lp_packet during LRPROOF validation (RNS/Link.py:404-408), accepts it as link.mtu, and the link's mdu (max data unit per packet for §3.1 link-derived Token traffic) is recomputed via update_mdu().

6.6.4 Presence detection — length only

Both directions detect the optional signalling slot purely by packet body length:

Packet	Body length without signalling	Body length with signalling
LINKREQUEST	`ECPUBSIZE = 64`	`ECPUBSIZE + LINK_MTU_SIZE = 67`
LRPROOF	`SIGLENGTH//8 + ECPUBSIZE//2 = 96`	`... + LINK_MTU_SIZE = 99`

Where ECPUBSIZE = 64 is the combined initiator ephemeral X25519 + Ed25519 public key (Link.py:70), and SIGLENGTH//8 = 64 is the responder's Ed25519 signature.

Receivers MUST handle both forms. validate_request at RNS/Link.py:186-190 checks len(data) == ECPUBSIZE OR len(data) == ECPUBSIZE+LINK_MTU_SIZE and rejects anything else. The same length-dispatch is in validate_proof for the LRPROOF side at RNS/Link.py:404-410. There is no flag bit signalling presence — wire length is the only signal.

6.6.5 Inclusion in LRPROOF signed_data

Per §6.2, the LRPROOF's signed_data when signalling is present is:

signed_data = link_id || responder_X25519_pub || responder_long_term_Ed25519_pub || signalling

A clean-room implementation that omits the signalling bytes when present (or includes them when absent) computes a different signed_data than the responder did, fails signature validation, and the link never establishes. This is the most common interop break in this area; cross-check against RNS/Link.py:373-374 (signer) and :417 (validator).

6.6.6 Disabling MTU discovery

The [reticulum] config option link_mtu_discovery = No makes Reticulum.link_mtu_discovery() return False, so the initiator skips signalling on outbound LINKREQUESTs (RNS/Link.py:311-314). In that case the link uses Reticulum.MTU (default 500 bytes) globally, no per-link MTU clamping happens, and all four lengths fall back to the no-signalling sizes in §6.6.4.

A receiver doesn't need its own copy of the disable switch — it just stops seeing trailing signalling bytes from peers that have it disabled. Its own MTU reporting on the LRPROOF return path runs unaffected for peers that send it.

6.7 KEEPALIVE and link teardown

A Link goes through five states (RNS/Link.py:110-114): PENDING → HANDSHAKE → ACTIVE → STALE → CLOSED. KEEPALIVE and LINKCLOSE are the two control-plane packet types that drive transitions out of ACTIVE.

6.7.1 KEEPALIVE (`context = 0xFA`)

Cadence (RNS/Link.py:845-847):

def __update_keepalive(self):
    self.keepalive = max(min(self.rtt * (KEEPALIVE_MAX / KEEPALIVE_MAX_RTT), KEEPALIVE_MAX), KEEPALIVE_MIN)
    self.stale_time = self.keepalive * STALE_FACTOR

with constants KEEPALIVE_MAX = 360s, KEEPALIVE_MIN = 5s, KEEPALIVE_MAX_RTT = 1.75s, STALE_FACTOR = 2. The interval is RTT × 205.7 clamped to [5, 360] seconds. Before the first RTT is measured (set in validate_proof), the link uses KEEPALIVE = KEEPALIVE_MAX = 360s.

The watchdog (Link.__watchdog_job, line 751-821) fires on every active link. When now >= last_inbound + keepalive AND the local node is the initiator, it emits a KEEPALIVE:

def send_keepalive(self):
    keepalive_packet = RNS.Packet(self, bytes([0xFF]), context=RNS.Packet.KEEPALIVE)
    keepalive_packet.send()

Body is a single byte 0xFF — the "ping" sentinel. KEEPALIVE is NOT Token-encrypted — RNS/Packet.py pack() puts KEEPALIVE in its not-encrypted branch alongside RESOURCE, RESOURCE_PRF, link PROOF, and CACHE_REQUEST (self.ciphertext = self.data, Packet.py:206-209 in RNS 1.2.9). The wire body is the single sentinel byte in the clear — no IV, no ciphertext expansion, no HMAC.

The responder receives this in Link.receive at RNS/Link.py:1149-1153 and answers with the "pong" sentinel (in 1.2.4 the body is bytes([0xFE])):

elif packet.context == RNS.Packet.KEEPALIVE:
    if not self.initiator and packet.data == bytes([0xFF]):
        keepalive_packet = RNS.Packet(self, bytes([0xFE]), context=RNS.Packet.KEEPALIVE)
        keepalive_packet.send()

So:

Ping = initiator → responder, body 0xFF.
Pong = responder → initiator, body 0xFE.
Only the initiator originates KEEPALIVE traffic. The responder never spontaneously pings.

Both sentinel bytes are arbitrary; what actually matters for keep-alive purposes is that any inbound traffic on the link refreshes last_inbound (the watchdog's anchor for staleness decisions). KEEPALIVE packets do not generate a PROOF receipt — the link-DATA proof path in RNS/Link.py receive() is gated on packet.context == RNS.Packet.NONE (Link.py:988 in RNS 1.2.9), and KEEPALIVE takes its own branch at line 1149-1153. So a successful ping/pong exchange resets the staleness clock on both sides via the two-packet exchange itself: ping → pong (no pong-proof).

A clean-room responder MUST emit the pong on inbound 0xFF; without it the initiator's watchdog will declare the link stale on the next cycle.

6.7.2 STALE → CLOSED transition

When now >= last_inbound + stale_time (= 2 × keepalive), the watchdog moves the link from ACTIVE to STALE (line 796-800), then on its next pass emits a teardown packet and transitions to CLOSED (line 805-810):

elif self.status == Link.STALE:
    sleep_time = 0.001
    self.__teardown_packet()                  # see §6.7.3
    self.status = Link.CLOSED
    self.teardown_reason = Link.TIMEOUT
    self.link_closed()

teardown_reason is set to Link.TIMEOUT (constant value 0x01) so the application's link_closed_callback can distinguish "the peer went dark" from "the peer cleanly closed".

There is also an explicit-cleanup path: after a STALE-induced teardown the watchdog adds a final grace period of RTT × KEEPALIVE_TIMEOUT_FACTOR + STALE_GRACE (= RTT × 4 + 5s) at line 797 to allow a delayed reply to bring the link back into ACTIVE before final teardown — but in upstream RNS 1.2.4 the STALE → CLOSED transition runs immediately on the next watchdog pass without consulting that grace period. The grace constant lives in case a future revision restores the soft-stale window.

6.7.3 LINKCLOSE (`context = 0xFC`)

Either side can cleanly tear down a link by calling Link.teardown() (line 699-708), which sends a single LINKCLOSE packet and transitions the local state to CLOSED:

def __teardown_packet(self):
    teardown_packet = RNS.Packet(self, self.link_id, context=RNS.Packet.LINKCLOSE)
    teardown_packet.send()

Wire form:

packet_type = DATA (0), context = 0xFC, dest_hash = link_id.
Body is the 16-byte link_id, Token-encrypted by the link's session key.

The peer's receiver path at RNS/Link.py:1061-1063 calls teardown_packet(packet) (line 710-728):

def teardown_packet(self, packet):
    plaintext = self.decrypt(packet.data)
    if plaintext == self.link_id:                # auth check
        self.status = Link.CLOSED
        if self.initiator:
            self.teardown_reason = Link.DESTINATION_CLOSED
        else:
            self.teardown_reason = Link.INITIATOR_CLOSED
        self.link_closed()

The body's plaintext MUST equal link_id for the close to take effect — this is the on-link auth check. A peer that doesn't share the session key can't decrypt the body, and even if it could, the link_id check rejects bodies with arbitrary content. Combined with the Token HMAC, this gives both "encrypted" and "authenticated" guarantees on the teardown signal.

After link_closed() (line 724-743) runs:

All incoming_resources and outgoing_resources are cancelled (cancels propagate into the §10 Resource state machine).
The Link's session keys (self.shared_key, self.derived_key) are zeroed by reassignment to None — the upstream comment at line 700-702 notes this is the forward-secrecy property: "encryption keys are purged. New keys will be used if a new link to the same destination is established."
The link_closed_callback registered via set_link_closed_callback fires.
The Link is removed from its destination's links list (responders only — initiators don't have a destination-list entry).

6.7.4 Teardown reason codes

Link.teardown_reason is set to one of (RNS/Link.py:116-118):

Constant	Hex	Meaning
`TIMEOUT`	`0x01`	Watchdog STALE → CLOSED transition. No LINKCLOSE was received.
`INITIATOR_CLOSED`	`0x02`	This side is the responder; the initiator sent a LINKCLOSE.
`DESTINATION_CLOSED`	`0x03`	This side is the initiator; the responder sent a LINKCLOSE.

These are local-state values, not on the wire — the LINKCLOSE packet itself doesn't carry a reason code. The recipient just infers whether the close came from the other side based on whether they're initiator or responder.

6.7.5 Receiver responsibilities (minimum)

For a clean-room implementation that wants links to survive idle periods longer than a few seconds:

Keep a per-link last_inbound timestamp updated on every inbound packet on the link (DATA, PROOF, KEEPALIVE — anything).
On the initiator side, run a watchdog that emits a 0xFF KEEPALIVE every link.keepalive seconds since last_inbound. Default link.keepalive = 360s is fine until you measure RTT.
On the responder side, reply to every 0xFF KEEPALIVE with a 0xFE KEEPALIVE. Don't originate.
On both sides, transition to CLOSED if last_inbound + 2*keepalive elapses with no traffic, AND emit a LINKCLOSE packet so the peer doesn't have to wait for its own watchdog to time out.
On every inbound LINKCLOSE, decrypt, verify body equals link_id, transition to CLOSED.
On CLOSED, zero the session keys and cancel any in-progress Resources.

6.7.6 LINKIDENTIFY (`context = 0xFB`)

A separate Link control DATA packet — not part of the keepalive / teardown cycle, but documented here alongside the other context-dispatched Link control packets. Used by the initiator to prove which long-term identity is making the request without re-running the Link handshake; §11.6 covers the calling context on the NomadNet REQUEST path.

Link.identify(identity) (RNS/Link.py:459-475 in RNS 1.2.9):

def identify(self, identity):
    if self.initiator and self.status == Link.ACTIVE:
        signed_data = self.link_id + identity.get_public_key()
        signature  = identity.sign(signed_data)
        proof_data = identity.get_public_key() + signature
        proof      = RNS.Packet(self, proof_data, RNS.Packet.DATA,
                                context = RNS.Packet.LINKIDENTIFY)
        proof.send()

Wire body (128 bytes):

public_key(64) || signature(64)

public_key is the initiator's full 64-byte Identity public key — the same get_public_key() that LINKREQUEST and announces use, the concatenation of the Ed25519 verification key and the X25519 encryption key per §3.
signature is identity.sign(link_id(16) || public_key(64)) — the signature is over the link_id concatenated with the public_key, NOT over link_id alone. The responder verifies with public_key over that same concatenation.

The packet is a DATA packet on the active Link, so it IS link-encrypted per §3.1 link-derived form like ordinary link DATA — context = 0xFB is NOT in RNS/Packet.py pack()'s not-encrypted set, unlike KEEPALIVE (§6.7.1) or link PROOF (§6.5).

The responder's receive() (RNS/Link.py:1010-1029) decrypts the body, splits off the 64-byte public_key prefix, reconstructs signed_data = link_id || public_key, verifies the signature, and on success sets self.remote_identity so subsequent REQUEST handlers can check the caller against per-page allowlists (§11.6).

A clean-room implementation MUST build the payload as public_key || signature (NOT just the signature) and sign the concatenation link_id || public_key (NOT link_id alone). Either mistake makes every ALLOW_LIST-protected page return DEFAULT_NOTALLOWED.

6.8 Channel mode (`CHANNEL = 0x0E`)

A Channel is a continuous, bi-directional, message-typed stream on top of an established Link. Distinct from §11 REQUEST/RESPONSE (single-shot, client-server) and §10 Resources (large unidirectional transfers): Channel messages are short, can flow in either direction at any time, and carry an application-defined type byte the receiver dispatches on. NomadNet uses it for its "channel" API (live chat over a Link), and any application can register custom message types via RNS.Channel.Channel.register_message_type.

6.8.1 Wire form

A Channel message rides as a single Link DATA packet with context = CHANNEL (0x0E). The body is 6-byte fixed-prefix header + variable-length payload (RNS/Channel.py:192-198):

msgtype(2)  ||  sequence(2)  ||  length(2)  ||  data(length bytes)

All three header fields are big-endian unsigned 16-bit integers (Python struct.pack(">HHH", msgtype, sequence, length)):

Field	Width	Meaning
`msgtype`	uint16 BE	Application-defined message type. Distinguishes the payload schema.
`sequence`	uint16 BE	Per-direction sequence number, starting at 0 and incrementing each emission. Wraps at 65536.
`length`	uint16 BE	Length in bytes of the payload that follows.

The whole 6-byte header + payload is the Link DATA packet's plaintext, which is then Token-encrypted by the link's session key (§3.1 link-derived form, no eph_pub prefix) before transmission.

6.8.2 Reserved system message types

RNS/Channel.py:45-46:

class SystemMessageTypes(enum.IntEnum):
    SMT_STREAM_DATA = 0xff00

0xff00 is reserved for upstream's stream-over-channel implementation. Application-defined message types should stay in the 0x0000..0xfeff range to avoid collisions with reserved system types. There's no centralized registry — each Link's Channel instance maintains its own message_factories dict mapping msgtype to a constructor.

6.8.3 MSGTYPE registration

Both endpoints of a Link must register matching message types via Channel.register_message_type(message_class) before they can send or receive that type. The constructor must implement:

class MyMessage(MessageBase):
    MSGTYPE = 0x1234           # uint16 in [0x0000, 0xfeff]
    def pack(self) -> bytes: ...
    def unpack(self, raw: bytes): ...

A receiver that gets a msgtype it didn't register raises ChannelException(ME_NOT_REGISTERED) and drops the message. A sender that tries to send a class without MSGTYPE defined raises ChannelException(ME_NO_MSG_TYPE).

6.8.4 Reliable delivery

Channel uses the standard §6.5 mandatory PROOF receipt mechanism for each message — every Channel DATA packet generates a PROOF, and the sender's Channel retries on timeout up to a per-packet limit. Reliability is the main reason to use Channel over plain Link DATA: the application doesn't need to implement its own retransmit logic.

The receiver-side Channel uses a sliding window with the same window-growth dynamics as §10 Resources (Channel.WINDOW = 2 initial, with rate-thresholded growth/shrink). Sequence numbers in the channel header let the receiver detect gaps and request retransmits; out-of-order arrivals are buffered until the gap fills.

6.8.5 When to use Channel vs the alternatives

Use case	Best mechanism
One-shot small request → one-shot small response	§11 REQUEST/RESPONSE
One-shot large transfer (file, page)	§10 Resource
Continuous bi-directional small messages (chat, telemetry stream, command/event flow)	§6.8 Channel
Continuous bi-directional large transfers	§10 Resources sequenced in time, OR Channel with `SMT_STREAM_DATA` chunks

A clean-room client that only implements opportunistic LXMF can ignore Channel entirely. NomadNet-aware clients need it for the channel API; custom-RPC applications may prefer it over §11 for its bi-directional nature.

6.8.6 Source map

File	What
`RNS/Channel.py:174-211`	`Envelope` — wire-form pack/unpack
`RNS/Channel.py:214-...`	`Channel` — windowed reliable delivery on a Link
`RNS/Channel.py:45-46`	`SMT_STREAM_DATA = 0xff00` reserved system type
`RNS/Channel.py:317-325`	`register_message_type` — per-Link MSGTYPE dispatch table
`RNS/Packet.py:86`	`CHANNEL = 0x0E` context constant

6.9 Source

RNS/Link.py, RNS/Packet.py::prove, RNS/Identity.py::prove, RNS/PacketReceipt.py::validate_proof, RNS/Channel.py. The webclient's reference/js-reference/link.js is a faithful port.

7. Transport behavior — the parts that bite

7.1 Path requests: peers send `path?` before opportunistic LXMF when no path is known

The path-request preamble in upstream LXMF is conditional, not unconditional (verified by tools/verify_path_request.py against LXMF 0.9.7):

# LXMF/LXMRouter.py::handle_outbound, ~line 1672
if not RNS.Transport.has_path(destination_hash) and lxmessage.method == LXMessage.OPPORTUNISTIC:
    RNS.log("Pre-emptively requesting unknown path for opportunistic ...", RNS.LOG_DEBUG)
    RNS.Transport.request_path(destination_hash)
    lxmessage.next_delivery_attempt = time.time() + LXMRouter.PATH_REQUEST_WAIT

In other words: a path? is sent before the LXM only when no entry exists in Transport.path_table for the target — has_path() is just a key-presence check via LXMRouter.handle_outbound. Existing-but-stale path entries are NOT replaced by this preamble; LXMF instead leans on the periodic Transport.jobs cycle to evict expired path entries (stale_paths accumulator at RNS/Transport.py:750+), after which the next outbound LXM rediscovers the unknown-path branch and triggers the request_path. A second request_path is issued from the retry path (LXMRouter.py:2568+) once lxmessage.delivery_attempts >= MAX_PATHLESS_TRIES, so on a flaky path peers can see multiple path? retransmits without intervening DATA — that matches BLE-trace observations.

A path? request itself is a regular DATA packet (verified by tools/verify_path_request.py):

dest_hash = SHA256(SHA256("rnstransport.path.request")[:10])[:16] = 6b9f66014d9853faab220fba47d02761
dest_type = PLAIN, transport_type = BROADCAST, header_type = HEADER_1, context = CTX_NONE
payload (RNS/Transport.py::request_path):
- leaf clients (transport disabled): target_dest_hash(16) || random_tag(16) — 32 bytes
- transport-enabled originators: target_dest_hash(16) || transport_id(16) || random_tag(16) — 48 bytes — so the responding announce can be routed back along the request's reverse path

7.2 Responding to path requests

Every node — including non-transport leaf clients — that knows the requested target MUST respond by re-announcing. This is the only way the requester learns a path back. If you implement only the "send a path request" half but not the "respond to incoming requests for our own destination" half, peers can never message you after the path expires (typically within minutes after your last announce).

7.2.1 Path-request packet parse rules

The path-request handler at RNS/Transport.py:2806-2850 parses inbound packets addressed to path_request_destination (the dest_hash in §7.1). The handler is registered as the destination's packet_callback at Transport.py:241, so any DATA packet to that dest_hash flows through it.

The path-request destination is a PLAIN destination with no identity attached, which is why its dest_hash derives only from the name: dest_hash = SHA256(SHA256("rnstransport.path.request")[:10])[:16] per the PLAIN/GROUP recipe in §1.4.3 (the identity == None branch of Destination.hash at RNS/Destination.py:121-130). The result is a constant — 6b9f66014d9853faab220fba47d02761 — that every node on the mesh resolves identically without needing to discover a per-peer identity first.

def path_request_handler(data, packet):
    if len(data) >= 16:
        destination_hash = data[:16]                                  # mandatory 16B target
        if len(data) > 32:
            requesting_transport_instance = data[16:32]               # optional 16B transport_id
        else:
            requesting_transport_instance = None

        # tag bytes — required, anything past the fixed prefix
        tag_bytes = data[32:] if len(data) > 32 else (data[16:] if len(data) > 16 else None)
        if tag_bytes is None:                                         # tagless requests are dropped
            return
        if len(tag_bytes) > 16:
            tag_bytes = tag_bytes[:16]                                # cap to 16B

Three observations that matter for interop:

Tagless requests are dropped. A path? packet with exactly 16 bytes payload (just target_dest_hash, no tag) is logged at DEBUG level and discarded. The tag is what makes the request unique enough to dedup — without it, a relay would loop forever on retransmits of the same packet. A clean-room implementation MUST emit at least one tag byte; the upstream emitter (RNS.Transport.request_path) uses 16 random bytes.
The transport_id field is optional and detected by length. If the payload is exactly 32 bytes the second 16B slot is the tag; if it's >32 bytes the second 16B is transport_id and the rest is the tag. This is consistent with the §7.1 description (leaf: 32B; transport: 48B) but the boundary case len == 32 lands in the leaf-client interpretation.
The tag is capped at 16 bytes. Any tail beyond that is silently truncated. Senders may emit longer tags but receivers normalize to 16B for dedup table keys.

7.2.2 Tag-based deduplication

The handler builds unique_tag = destination_hash || tag_bytes and consults Transport.discovery_pr_tags (Transport.py:2835-2845):

unique_tag = destination_hash + tag_bytes

with Transport.discovery_pr_tags_lock:
    if not unique_tag in Transport.discovery_pr_tags:
        Transport.discovery_pr_tags.append(unique_tag)
        Transport.path_request(destination_hash,
                               from_local_client(packet),
                               packet.receiving_interface,
                               requestor_transport_id=requesting_transport_instance,
                               tag=tag_bytes)
    else:
        # ignore duplicate path request

discovery_pr_tags is bounded at Transport.max_pr_tags = 32000 entries (Transport.py:127); older entries are aged out by the periodic Transport.jobs cycle. Every node — leaf or transport — that wants to respond to path requests MUST maintain this dedup table or it will respond to every retransmit, and a transport-enabled node will additionally re-forward to all other interfaces, generating a broadcast storm.

The unique_tag = dest_hash || tag format means the same tag bytes against different destination_hashes are distinct — so two different requesters racing for the same target with happenstance-identical random tags don't suppress each other. Senders MUST use a fresh random tag per fresh request (the upstream emitter calls Identity.get_random_hash()); reusing tags across requests for the same destination_hash makes the second request appear to be a duplicate.

7.2.3 The five-way dispatch in `Transport.path_request`

RNS/Transport.py:2852-2980. After dedup, the handler calls into path_request() which decides how to respond. Five mutually-exclusive branches in priority order:

destination_hash is local (i.e. it's one of our own registered destinations, line 2873-2875):
```
local_destination.announce(path_response=True, tag=tag,
                           attached_interface=attached_interface)
```
We answer by emitting a path-response announce (§7.2.4 below) on the interface the request arrived on. This is the only branch a leaf client must implement — the others are transport-mode behaviours.
Path is known via the path_table AND (transport_enabled OR is_from_local_client) (line 2877-2938): retrieve the cached announce packet from the path table, set its hops to the cached value, and queue it for retransmit. If the next hop happens to be the requestor itself (path-loop indicator), drop instead. This is the transport-mode path-resolver: a relay that already knows where the destination lives answers on its behalf, saving the requester from another hop of broadcast.
Request is from a local-client interface, no path known (line 2940-2947): forward the request to every OTHER interface so the broader mesh can answer. Generates a fresh random tag for the forwarded request to avoid loop-back through the same dedup table.
transport_enabled AND no path known AND interface allows discovery (line 2949-2963): record a discovery_path_requests entry (capped at PATH_REQUEST_TIMEOUT = 15s) and forward the request to every other interface, preserving the original tag to prevent loops. This is recursive transport-mode discovery — we don't know the destination but we'll go ask the rest of the mesh.
No path known and not transport-enabled (line 2972-2973): log "no path known" and drop. Leaf clients hit this branch when they receive a path? for someone else's destination.

Branch 1 is the only MUST for any node that wants to be reachable. Branches 2-4 are transport-node behaviours; a leaf client safely ignores them by never being in transport_enabled mode.

7.2.4 Path-response announce wire format

When branch 1 fires, Destination.announce(path_response=True, tag=tag, ...) runs. The wire bytes are identical to a regular announce (§4.1) except the outer Reticulum packet's context byte is set to PATH_RESPONSE = 0x0B instead of NONE = 0x00 (RNS/Destination.py:307-308):

if path_response: announce_context = RNS.Packet.PATH_RESPONSE
else:             announce_context = RNS.Packet.NONE

The body — public_key || name_hash || random_hash || [ratchet_pub] || signature || app_data — is built identically; the random_hash carries a fresh emission timestamp, the signature is computed over the same signed_data per §4.2. A receiver running the validation flow in §4.5 can't tell from the announce body that this is a response to a query rather than a periodic re-announce; only the context byte distinguishes them.

A tag argument hands a previously-built path-response announce body back unchanged when the same tag is requested twice within Destination.PR_TAG_WINDOW = 30s (RNS/Destination.py:260-278). This is what prevents a flood of identical path-response announces when several relays simultaneously forward the same path? request to a leaf — the leaf serves the cached body to all of them with the same wire bytes, lining up dedup decisions on every transit relay.

7.2.5 Timing: `PATH_REQUEST_GRACE` and roaming

When branch 2 fires (transit relay answering on behalf of a remote destination), the rebroadcast is delayed by PATH_REQUEST_GRACE = 0.4s (Transport.py:80) — extra grace to let directly-reachable peers respond first if they're in earshot. On MODE_ROAMING interfaces an additional PATH_REQUEST_RG = 1.5s is added on top (Transport.py:81) so well-connected fixed nodes get a chance to answer before mobile ones.

Branch 1 (local destination answers) fires immediately with no grace, since the leaf is the authoritative source for its own destination — there's no point waiting for someone else to potentially answer faster.

Local-client originators also bypass the grace period (Transport.py:2915-2920): a relay answering for a destination that lives on a local-client interface can send back the cached announce instantly because the answer doesn't need to compete with peer-mesh announces.

7.2.6 Minimum responsibility for a leaf

The minimum path-request response logic for a non-transport leaf, in protocol terms:

Receive a DATA packet with dest_hash == 6b9f66014d9853faab220fba47d02761.
Parse target_dest_hash = data[:16] and tag_bytes = data[16:32] (or data[32:48] if len(data) > 32).
Drop if len(tag_bytes) == 0 (tagless requests).
Drop if (target_dest_hash, tag_bytes) already in the dedup table. A leaf-appropriate cap is 128–256 entries with FIFO eviction; the upstream max_pr_tags = 32000 (§7.2.2) is sized for a transit node maintaining dedup across all destinations on the mesh, not a leaf that only sees requests for itself.
If target_dest_hash == our_destination_hash for any of our registered destinations: emit a path-response announce (§7.2.4) on the receiving interface, with the request's tag passed through.
Otherwise: do nothing — leaves can't fulfill path requests for destinations they don't OWN.

Steps 4 and 5 are both required. Skipping the dedup table makes the leaf storm the network with redundant announces; skipping the local-destination check means peers can never message you after the path expires.

Leaves may skip the §7.2.5 PR_TAG_WINDOW body cache — step 4's dedup table already collapses identical-tag retransmits, and a leaf isn't fanning the same body to multiple downstream relays the way a transit node does, so the 30-second cache offers no additional dedup-convergence benefit. The cache exists upstream because Destination.announce runs the same code path for both leaves and transit nodes; on a leaf, the cache is incidental.

For a chronological walk-through of the full request → response → path-table cycle, see flows/path-discovery.md.

7.3 Ratchet rotation (forward-secrecy hygiene, not dedup)

The 32-byte ratchet_pub field in announces is meant to rotate periodically. The purpose is forward secrecy: rotating the ECDH key on a regular cadence limits the plaintext window an adversary can decrypt if a single ratchet privkey leaks. It is not what makes your announces visible to the mesh.

The actual replay-and-loop defence in upstream is keyed on random_hash, not on ratchet_pub — see §4.5 step 6.3 (path-table replacement check not random_blob in random_blobs at RNS/Transport.py:1710, 1735, 1748). Verified by tools/verify_ratchet_dedup.py: two announces sharing a ratchet_pub but differing in random_hash[:5] are both accepted by upstream's replay machinery.

⚠️ Spec correction: Earlier revisions of this section claimed transit nodes dedup announces on (destination_hash, ratchet_pub) tuples and that a non-rotating client becomes invisible to the mesh after one announce. That was wrong on the mechanism: upstream's RATCHET_INTERVAL = 30 min × ANNOUNCE_INTERVAL = 5–15 min means most upstream announces share a ratchet across 2–6 emissions, so if relays really dropped on ratchet_pub equality, upstream wouldn't function. The actual win observed in the bootstrap test (per agent.md §5) was incidental — the fix that rotated ratchets per announce also rotated random_hash, and it was the latter that mattered.

7.3.1 Rotation cadence

Upstream Destination.rotate_ratchets() (RNS/Destination.py:227-235) runs on every announce but is a no-op unless RATCHET_INTERVAL = 30*60s has elapsed since the last rotation:

def rotate_ratchets(self):
    if now > self.latest_ratchet_time + self.ratchet_interval:
        new_ratchet = Identity._generate_ratchet()
        self.ratchets.insert(0, new_ratchet)
        ...

So a Sideband emitting an announce every 10 minutes generates a new ratchet at most every 30 minutes (3 announces per ratchet). Path-response announces and periodic announces both call rotate_ratchets() and both go through this no-op-if-recent gate.

7.3.2 What MUST be unique per announce

For your destination to remain visible across multiple announces, what MUST change between back-to-back emissions is random_hash, not ratchet_pub. Per §4.1, random_hash is constructed as:

random_hash = get_random_hash()[:5] + int(time.time()).to_bytes(5, "big")

So as long as you regenerate the first 5 random bytes per announce (which any sensible implementation does), upstream's replay defence accepts each announce as fresh regardless of whether the ratchet rotated. A clean-room client that hard-coded random_hash to a constant value would be invisible after the first announce; one that uses fresh random bytes per announce is visible regardless of ratchet rotation cadence.

7.3.3 Per-announce ratchet rotation is fine but not required

Implementations MAY rotate the ratchet on every announce — the only cost is more frequent ratchet-ring growth (capped by §7.4 RATCHET_COUNT = 512) and slightly more CPU. They MAY also follow upstream's at-most-every-30-minutes pattern. Either is interop-correct.

What MUST be stable across all rotations: the long-term encryption / signing keys and the identity_hash / destination_hash. Rotating those means contacts have to re-discover you (different dest_hash, no path table entry).

7.3.4 Path-response announces SHOULD reuse the current ratchet

When fulfilling a path? request via Destination.announce(path_response=True, tag=tag) (§7.2.4), implementations SHOULD reuse the current ratchet rather than rotate. Rotation cadence is governed by §7.3.1 (the 30-minute window), not by inbound path? arrivals — a leaf burst-rotating on a flood of identical-target path? requests would burn through ratchet-ring slots without any forward-secrecy benefit, since the announces are all going to the same in-flight requester. Upstream's rotate_ratchets() no-op-if-recent gate enforces this implicitly; a clean-room implementation should mirror the behaviour explicitly.

7.3.5 Ratchet-less announces are always accepted

Emitting context_flag = 0 (no ratchet_pub field, body layout per §4.5 step 1, second branch) is interop-correct against every RNS 1.x receiver. validate_announce parses both layouts unconditionally; there are no upstream peers that strict-reject ratchet-less announces. The trade-off is forward secrecy only: a ratchet-less destination encrypts every opportunistic message to its long-term X25519 key (§3.2 step 2), so a future leak of that long-term privkey decrypts every prior message. Path-table population, signature verification, dest_hash routing, and Link establishment all work unchanged. A new implementation that defers ratchets to v2 will interop fine; the missing forward secrecy should be called out in its README. A future RNS major may make ratchets mandatory, but that would be a wire-incompatible change announced ahead of time.

7.4 Ratchet ring (inbound decrypt tolerance)

Senders cache the most recent ratchet they've seen for each destination. If you rotate your ratchet faster than relays propagate the announce, in-flight messages may arrive encrypted to your previous ratchet. To decrypt these, keep a ring of recent ratchet privkeys and try each in order during decrypt. The fallback to the long-term identity privkey is the ultimate safety net.

Upstream's default ring size is Destination.RATCHET_COUNT = 512 (RNS/Destination.py:85 in RNS 1.2.4), with a minimum rotation interval of RATCHET_INTERVAL = 30*60 seconds (line 90) and per-ratchet RATCHET_EXPIRY = 60*60*24*30 seconds (RNS/Identity.py:69). A new ratchet is generated on each rotate_ratchets() call and prepended to the in-memory list; _clean_ratchets truncates back to RATCHET_COUNT. The 512 figure is generous and not a hard interop requirement — it's an in-memory bound on the inbound-decrypt try-list.

A minimal client may keep just the current ratchet privkey, accepting that the brief window between rotation and announce-propagation will lose some messages. Mention the trade-off in your implementation notes.

7.5 Periodic re-announce

Transport node path tables expire entries after a few minutes. Clients should re-announce on a 5–15 minute cadence as a baseline so cached paths stay fresh. Without this, even peers who saw your initial announce will be unable to reach you after path TTLs lapse.

7.6 `TCPServerInterface.OUT` is True by default in practice

RNS/Interfaces/TCPInterface.py line 522 sets self.OUT = False in the constructor. This is overridden to True by RNS/Reticulum.py post-init at line 771-772 for any interface declared in the rnsd config:

if "outgoing" in c and c.as_bool("outgoing") == False: interface.OUT = False
else:                                                  interface.OUT = True

Spawned client interfaces (one per connecting TCP client) inherit OUT from their parent. So in practice, every TCPServerInterface CAN forward unless the operator explicitly opted out. Do not waste time chasing the constructor's OUT = False default; it doesn't hold post-init.

7.7 Source

RNS/Transport.py outbound, inbound, request_path, announce. RNS/Reticulum.py interface_post_init for the OUT-flag override.

8. Transport framing

8.1 KISS (BLE / serial / RNode link)

FEND  = 0xC0    // frame delimiter
FESC  = 0xDB    // escape
TFEND = 0xDC    // escaped FEND  → 0xDB 0xDC
TFESC = 0xDD    // escaped FESC  → 0xDB 0xDD

frame = FEND || cmd_byte || escaped(data) || FEND

cmd_byte for received/transmitted Reticulum packets is CMD_DATA = 0x00. RNode firmware prefixes each received CMD_DATA frame with CMD_STAT_RSSI = 0x23 (one byte payload, signed value = byte − 157) and CMD_STAT_SNR = 0x24 (one byte payload, signed Q6.2 → divide by 4 for dB).

Over BLE, KISS frames are split across BLE notifications. A streaming parser MUST accumulate bytes across notifications and emit complete frames only on FEND boundaries.

8.2 HDLC (TCP / `rnsd TCPServerInterface`)

FLAG = 0x7E
ESC  = 0x7D
ESC_MASK = 0x20

frame = FLAG || escaped(data) || FLAG
escape: 0x7E → 0x7D 0x5E   (FLAG ^ ESC_MASK)
        0x7D → 0x7D 0x5D   (ESC  ^ ESC_MASK)

No command byte, no RSSI/SNR sidecar — the HDLC payload IS the raw Reticulum packet. Source: RNS/Interfaces/TCPInterface.py::HDLC.

8.3 RNode air-frame header and split-packet protocol

The 1-byte header described here lives between RNodes on the LoRa air-frame, not on the KISS host channel. The upstream RNode firmware adds it on every TX and strips it on every RX before forwarding the payload to the host as CMD_DATA. KISS hosts (RNS, NomadNet, Sideband, etc.) NEVER see this byte. Two RNodes that talk LoRa to each other use it to glue two LoRa frames into one Reticulum packet of up to 508 bytes; an alternative implementation that talks LoRa to an RNode (e.g. a clean-room repeater firmware) MUST construct and parse this header bit-exactly, or its TX will be invisible and its RX will mistake the header byte for the first payload byte.

Header byte layout

From markqvist/RNode_Firmware/Framing.h:105-108:

bit 7..4 : seq         (NIBBLE_SEQ   = 0xF0) — random sequence id, set on each TX
bit 3..1 : reserved    (currently always 0)
bit 0    : FLAG_SPLIT  (NIBBLE_FLAGS = 0x0F, FLAG_SPLIT = 0x01)
SEQ_UNSET = 0xFF                            — sentinel: "no first half buffered"

Helpers (Utilities.h:1218-1224):

inline bool    isSplitPacket(uint8_t h) { return (h & FLAG_SPLIT); }   // 0x01 mask
inline uint8_t packetSequence(uint8_t h){ return h >> 4; }             // 0..15

Constants (Config.h:59-61):

#define MTU         508    // max reassembled Reticulum packet payload (2 × 254)
#define SINGLE_MTU  255    // max LoRa frame size (header + up to 254 payload bytes)
#define HEADER_L    1      // header overhead per LoRa frame

Transmit (`RNode_Firmware.ino:716-742`)

uint8_t header = random(256) & 0xF0;                      // fresh random seq nibble
if (size > SINGLE_MTU - HEADER_L) header |= FLAG_SPLIT;   // split iff payload > 254
LoRa->beginPacket();
LoRa->write(header);
for (i=0; i < size; i++) {
    LoRa->write(tbuf[i]);
    if (written == 255 && isSplitPacket(header)) {        // first frame full
        LoRa->endPacket();
        LoRa->beginPacket();
        LoRa->write(header);                              // SAME header byte on frame 2
        written = 1;
    }
}
LoRa->endPacket();

Behavioral facts that matter for interop:

Sequence nibble is randomized on every fresh TX, not incremented. Two consecutive split packets from the same node will have different (random) seq nibbles. This is the trick a memory-fading reader might recall as "the header rotates between transmissions" — it's per-packet randomization, not a per-retransmit byte rotation. There is no retransmit-driven byte rotation or rechunk; LoRa transmission is fire-and-forget at this layer, and a higher-layer retransmit (e.g. an RNS PROOF timeout firing again) just re-enters this function and gets a fresh random seq nibble.
Both frames of a split share the same header byte byte-for-byte — same seq nibble, same FLAG_SPLIT bit. The receiver pairs them by exact equality of the seq nibble.
The split point is at exactly 255 bytes total in the LoRa frame (1 header + 254 payload). The second frame is header || remainder, where remainder is whatever is left after 254 bytes of payload have been emitted. Maximum reassembled packet payload is 2 × 254 = 508 bytes — Reticulum's HW_MTU for the RNode interface is set to match.
Single-frame packets (payload ≤ 254) still carry the 1-byte header but with FLAG_SPLIT == 0. The seq nibble is still random per TX.

Receive / reassembly (`RNode_Firmware.ino:359-446`)

State on the receiver: seq (default SEQ_UNSET = 0xFF) tracks the seq nibble of any buffered first-half. Per inbound LoRa frame:

Inbound `FLAG_SPLIT`	Buffered `seq` state	Inbound seq	Action
1	`SEQ_UNSET` (none)	any	Buffer this frame as the first half. Store its seq.
1	matches inbound seq	== buffered	Append. Reassembly complete. Reset buffer.
1	doesn't match	!= buffered	Discard buffered first-half. Replace with this frame as a new first-half.
0	`SEQ_UNSET` (none)	n/a	Deliver this single-frame packet directly.
0	first-half present	n/a	Discard the buffered first-half; deliver this single-frame packet.

In other words: the receiver holds at most one in-progress first-half, keyed by its random seq nibble. Any inbound frame that doesn't match (different seq, or non-split, or simply a long enough silence) replaces or discards it.

Reassembly timeout — implementation-defined

Upstream RNode firmware does not have an explicit time-based timeout for a buffered first-half — it relies on subsequent traffic (any inbound frame) to clear stale state via the table above. The clean-room repeater at thatSFguy/reticulum-lora-repeater/src/Radio.cpp:189-194 adds a defensive 500 ms timeout: if no second half arrives within that window, the buffered first-half is discarded. This is implementation-private: a packet that takes longer than 500 ms to fully transmit (very low SF + large payload) would be lost on a repeater following the clean-room timeout but would survive against an unbounded upstream RNode receiver as long as no other LoRa traffic landed in between.

A new alternative implementation should either match upstream's "no explicit timeout" or pick a value tied to the worst-case airtime of two SINGLE_MTU frames at the configured SF/BW, not a flat 500 ms.

Sequence-collision airtime ceiling

Because the seq nibble is 4 bits of randomness chosen per TX, two unrelated split packets from the same sender that overlap in time at any receiver will collide with probability 1/16 per pair. At sane LoRa duty cycles this is a non-issue, but it bounds the protocol — a sender that emits split packets back-to-back faster than the air can ferry them risks a reassembled packet that mixes halves of two distinct senders' outputs. The receiver has no way to detect this short of validating the resulting Reticulum packet (which a corrupt mix would fail at the HMAC step). Don't burst.

Source map

File	What it pins down
`RNode_Firmware/Framing.h:105-108`	`NIBBLE_SEQ`, `NIBBLE_FLAGS`, `FLAG_SPLIT`, `SEQ_UNSET` constants
`RNode_Firmware/Config.h:59-61`	`MTU`, `SINGLE_MTU`, `HEADER_L`
`RNode_Firmware/Utilities.h:1218-1224`	`isSplitPacket`, `packetSequence` accessors
`RNode_Firmware/RNode_Firmware.ino:716-742`	TX-side header construction and split logic
`RNode_Firmware/RNode_Firmware.ino:359-446`	RX-side reassembly state machine
`reticulum-lora-repeater/src/Radio.cpp:35-45, 188-316, 351-405`	Clean-room reimplementation; adds 500 ms reassembly timeout

8.4 RNode KISS configuration handshake

§8.1 covers the KISS framing between a host and an RNode. This section covers the commands a host issues over that framing to bring up an RNode and configure its radio. Before any Reticulum DATA packet can flow, every command listed here must succeed in approximately the order listed.

The canonical reference is RNS/Interfaces/RNodeInterface.py (the Python host driver) plus RNode_Firmware/Framing.h (the firmware-side command codes).

8.4.1 Command-byte inventory

Each KISS frame is FEND || cmd_byte || escaped_payload || FEND (§8.1). The cmd_byte values used during configuration:

Hex	Name	Direction	Purpose
`0x00`	`CMD_DATA`	both	Reticulum packet payload (the steady-state command after configuration is done)
`0x01`	`CMD_FREQUENCY`	host → RNode	Set radio frequency in Hz; payload is 4-byte big-endian uint32
`0x02`	`CMD_BANDWIDTH`	host → RNode	Set radio bandwidth in Hz; payload is 4-byte big-endian uint32
`0x03`	`CMD_TXPOWER`	host → RNode	Set TX power in dBm; payload is 1 byte
`0x04`	`CMD_SF`	host → RNode	Set spreading factor; payload is 1 byte (typically 7..12)
`0x05`	`CMD_CR`	host → RNode	Set coding rate denominator; payload is 1 byte (typically 5..8 for `4/5`..`4/8`)
`0x06`	`CMD_RADIO_STATE`	host → RNode	`0x00 = OFF`, `0x01 = ON` (turn the modem on after config)
`0x07`	`CMD_RADIO_LOCK`	host → RNode	Lock the radio against further config changes
`0x08`	`CMD_DETECT`	both	Hardware detect ping/pong (see §8.4.3 below)
`0x09`	`CMD_IMPLICIT`	host → RNode	Toggle implicit-header LoRa mode (advanced)
`0x0A`	`CMD_LEAVE`	host → RNode	Tell the RNode the host is disconnecting; payload `0xFF`
`0x0B`	`CMD_ST_ALOCK`	host → RNode	Short-term airtime limit; payload is 2 bytes big-endian uint16 of (limit × 100)
`0x0C`	`CMD_LT_ALOCK`	host → RNode	Long-term airtime limit; same encoding as ST_ALOCK
`0x0F`	`CMD_READY`	RNode → host	Boot complete signal
`0x21`	`CMD_STAT_RX`	RNode → host	RX-counter status
`0x22`	`CMD_STAT_TX`	RNode → host	TX-counter status
`0x23`	`CMD_STAT_RSSI`	RNode → host	RSSI sidecar for the next CMD_DATA frame; `dBm = byte − 157`
`0x24`	`CMD_STAT_SNR`	RNode → host	SNR sidecar; signed Q6.2 → `dB = byte / 4`
`0x26`	`CMD_STAT_PHYPRM`	RNode → host	Reports current frequency/bandwidth/SF/CR back; used for verification
`0x40`	`CMD_RANDOM`	both	Get random bytes from the RNode's hardware RNG
`0x47`	`CMD_BOARD`	RNode → host	Reports board model code
`0x48`	`CMD_PLATFORM`	RNode → host	Reports MCU platform code
`0x49`	`CMD_MCU`	RNode → host	Reports MCU variant code
`0x50`	`CMD_FW_VERSION`	RNode → host	Reports firmware version (2 bytes: major, minor)
`0x55`	`CMD_RESET`	host → RNode	Hard-reset the RNode; payload `0xF8` (`CMD_RESET_BYTE`)

Full inventory in RNode_Firmware/Framing.h:24-95. The configuration handshake uses the bolded subset.

8.4.2 Bring-up sequence

Adapted from RNodeInterface.initRadio (RNS/Interfaces/RNodeInterface.py:470-481):

1. Open serial port (or BLE GATT, or whatever bearer)
2. Optionally: hard_reset()           — CMD_RESET 0xF8 (2.25s wait afterwards)
3. detect()                            — CMD_DETECT + CMD_FW_VERSION + CMD_PLATFORM + CMD_MCU
4. (RNode replies asynchronously with CMD_DETECT 0x46, CMD_FW_VERSION, CMD_PLATFORM,
    CMD_MCU, CMD_BOARD over multiple frames — host correlates by command byte)
5. setFrequency()                      — CMD_FREQUENCY + 4B big-endian Hz
6. setBandwidth()                      — CMD_BANDWIDTH + 4B big-endian Hz
7. setTXPower()                        — CMD_TXPOWER + 1B dBm
8. setSpreadingFactor()                — CMD_SF + 1B
9. setCodingRate()                     — CMD_CR + 1B
10. setSTALock() / setLTALock()        — optional airtime limits
11. setRadioState(RADIO_STATE_ON)      — CMD_RADIO_STATE + 0x01
12. (RNode now in operational state; CMD_DATA frames flow in both directions)

The order matters: most firmwares accept config commands only while the radio is OFF (steps 5-10 must precede step 11). Setting parameters after RADIO_STATE_ON either silently fails or requires a RADIO_STATE_OFF round-trip first depending on firmware version. A clean-room driver should always set the radio OFF (or be in initial-boot state where it's OFF by default) before reconfiguring.

8.4.3 The `CMD_DETECT` exchange

host → RNode :  FEND CMD_DETECT(0x08) DETECT_REQ(0x73) FEND
RNode → host :  FEND CMD_DETECT(0x08) DETECT_RESP(0x46) FEND

DETECT_REQ = 0x73 and DETECT_RESP = 0x46 are at RNode_Firmware/Framing.h:99-100. The two-byte exchange tells a host "yes, this thing on the other end of the serial port is an RNode and it's awake". The host follows up immediately with CMD_FW_VERSION, CMD_PLATFORM, CMD_MCU queries — those queries each have a single 0x00 placeholder byte payload (per RNodeInterface.detect() line 484) and the RNode replies asynchronously with the same command code carrying the actual answer.

A host driver should accumulate replies for ~1-2 seconds after sending detect() before assuming detection failed. The replies arrive in unpredictable order because the firmware fires them off as it produces each value.

CMD_FW_VERSION's payload format is 2 bytes: [major, minor]. RNS rejects RNode firmware older than its REQUIRED_FW_VER_MAJ / REQUIRED_FW_VER_MIN constants and aborts the bring-up. A clean-room driver should at minimum log the version for diagnostics.

8.4.4 4-byte big-endian numerics

CMD_FREQUENCY and CMD_BANDWIDTH payloads are unsigned 32-bit integers in big-endian byte order:

c1 = self.frequency >> 24
c2 = self.frequency >> 16 & 0xFF
c3 = self.frequency >> 8  & 0xFF
c4 = self.frequency       & 0xFF
data = KISS.escape(bytes([c1, c2, c3, c4]))

The byte values are KISS-escaped before transmission per §8.1 (e.g. a frequency of 0xC0... would have its leading 0xC0 byte escaped to 0xDB 0xDC).

CMD_TXPOWER, CMD_SF, CMD_CR, CMD_RADIO_STATE payloads are single bytes, also subject to KISS escaping.

8.4.5 Receive sidecar metadata

Every CMD_DATA frame from the RNode is preceded by two short metadata frames in the same byte stream (§8.1 already mentions this; the encoding):

FEND CMD_STAT_RSSI(0x23) <rssi_byte>  FEND
FEND CMD_STAT_SNR(0x24)  <snr_byte>   FEND
FEND CMD_DATA(0x00)      <data...>    FEND

Decode:

RSSI in dBm = rssi_byte - 157 (e.g. rssi_byte = 50 means -107 dBm).
SNR in dB = (signed)snr_byte / 4 — snr_byte is interpreted as signed two's-complement Q6.2 fixed-point. So 0x10 (16) = 4 dB, 0xF0 (-16) = -4 dB, etc.

A host driver must cache the most recent RSSI/SNR pair and apply it to the next CMD_DATA frame. If it processes CMD_DATA before the sidecars arrive (e.g. the byte stream re-ordered them across an unreliable link), RSSI/SNR will be from the previous packet. In practice the firmware emits them in a tight sequence within microseconds, so reordering is only a concern over BLE notification boundaries (§8.1 closing paragraph).

8.5 RNode CSMA / airtime accounting

Real LoRa networks need carrier-sense and airtime budgets to avoid stepping on each other. The RNode firmware implements both server-side; the host is mostly told what's happening via CMD_STAT_CHTM (channel-time-metric, 0x25 in Framing.h:45) and chooses whether to inform the application.

8.5.1 Airtime caps (`CMD_ST_ALOCK` / `CMD_LT_ALOCK`)

The host can set per-channel airtime limits via:

CMD_ST_ALOCK (0x0B): short-term airtime lock. Payload is 2 bytes big-endian uint16 of (limit_percent × 100) — so 0x0B B8 = 3000 = 30.00%. Default in RNS/Reticulum.py is Reticulum.ANNOUNCE_CAP = 2.0 (= 2% airtime cap on transmissions, encoded as 0x00C8).
CMD_LT_ALOCK (0x0C): long-term version, same encoding. Long-term window length is firmware-private (typically 1 hour).

Once the cap is exceeded the firmware simply refuses to transmit and reports CMD_ERROR ERROR_QUEUE_FULL (0x04) if the host queues additional packets. A clean-room driver should treat these errors as backpressure and queue at the application layer rather than retry-spinning at the KISS layer.

8.5.2 Pre-TX carrier sense

Before transmitting, RNode firmware listens on the configured frequency for a short window and aborts the TX if it detects an in-progress LoRa preamble — Listen-Before-Talk. The exact CSMA windowing is firmware-private; a clean-room implementation that talks LoRa via RadioLib (rather than via an RNode) needs to implement its own LBT to avoid stepping on RNodes and other peers. The reference implementation in markqvist/RNode_Firmware/RNode_Firmware.ino:683-712 (the add_airtime accumulator and channel-utilisation tracking) is the canonical algorithm.

For host-side use cases — i.e. a Reticulum client driving an RNode — the firmware handles all CSMA invisibly and the host should not attempt its own. Host-side rate limiting at the announce-cap layer (§4.5 SHOULD-rule for ingress, and Reticulum.ANNOUNCE_CAP for outbound) is sufficient.

8.6 AutoInterface multicast discovery (LAN auto-detect)

AutoInterface is the IPv6-multicast-based protocol Reticulum nodes use to discover each other on a LAN with zero configuration. Drop a node on any IPv6-capable network, configure [[Default Interface]] type = AutoInterface, and it finds peers automatically — no static IPs, no rendezvous server.

The reference implementation is RNS/Interfaces/AutoInterface.py (~700 lines). This section specifies the wire-visible bits a clean-room implementation needs.

8.6.1 IPv6 multicast group derivation

AutoInterface.py:202-212. Each AutoInterface mesh is identified by a group_id (default b"reticulum"); the actual multicast address is derived from a SHA-256 of the group_id:

group_hash = SHA256(group_id)                     # 32 bytes

# Build the lower 7 hextets from group_hash bytes [2..14]:
gt  = "0"
gt += ":" + f"{(g[3]+(g[2]<<8)):02x}"            # bytes [2:4]
gt += ":" + f"{(g[5]+(g[4]<<8)):02x}"            # bytes [4:6]
gt += ":" + f"{(g[7]+(g[6]<<8)):02x}"            # bytes [6:8]
gt += ":" + f"{(g[9]+(g[8]<<8)):02x}"            # bytes [8:10]
gt += ":" + f"{(g[11]+(g[10]<<8)):02x}"          # bytes [10:12]
gt += ":" + f"{(g[13]+(g[12]<<8)):02x}"          # bytes [12:14]

mcast_discovery_address = "ff" + multicast_address_type + discovery_scope + ":" + gt

Where:

Field	Default	Value bits
`multicast_address_type`	`"1"` (temporary) — alt `"0"` (permanent)	4 bits, `RFC 4291` flags
`discovery_scope`	`"2"` (link-local)	4 bits, IPv6 scope: `"2"=link, "4"=admin, "5"=site, "8"=org, "e"=global`

So with the default group_id = b"reticulum", default multicast_address_type = "1", default discovery_scope = "2", and SHA256(b"reticulum")[2:14] filling the lower hextets, every default-config Reticulum node on the same link-local subnet finds the same multicast address.

8.6.2 UDP ports

AutoInterface.py:47-48:

Port	Default	Use
`discovery_port`	`29716` (UDP)	Periodic discovery announces from each peer; receivers learn other peers' link-local IPv6 addresses by their incoming source addr.
`unicast_discovery_port`	`discovery_port + 1` = `29717`	Per-interface unicast probes, used to disambiguate which physical interface a peer is on.
`data_port`	`42671` (UDP)	Once peers know each other's addresses, actual Reticulum packets flow as plain UDP datagrams between them on this port.

A clean-room implementation MUST listen on the discovery port for inbound multicast packets and the data port for inbound unicast packets, and emit periodic announces to the multicast address+port.

8.6.3 Discovery cadence

AutoInterface.py:61-64:

Constant	Value	Meaning
`ANNOUNCE_INTERVAL`	`1.6s`	Each AutoInterface emits a discovery announce on this cadence.
`PEERING_TIMEOUT`	`22.0s`	A peer not heard from within this window is dropped.
`PEER_JOB_INTERVAL`	`4.0s`	Cadence of the per-interface peer-management job (eviction, rediscovery).
`MCAST_ECHO_TIMEOUT`	`6.5s`	If our own multicast emissions are not echoed back within this window on a given physical interface, the multicast routing on that interface is presumed broken.

The 1.6s + 22s pairing means a new node is discovered within ~1.6s of join (as soon as one announce cycle completes from any existing peer); a departing node is forgotten within ~22s of last contact. Both bounds are implementation-private, but a clean-room with radically different values may have visible interop quirks (peer flap if you announce too rarely; bandwidth waste if too often).

8.6.4 Discovery announce body format

The discovery announce is a plain UDP datagram on the multicast address. The body is implementation-private — upstream uses a small msgpack blob containing the peer's group_hash (so peers from a different group on the same link don't accidentally peer), interface MTU, and an optional IFAC seal (Interface Authentication Code; if present, peers without the matching IFAC key reject the announce). Specific bytes of the discovery announce body aren't part of the wire spec for Reticulum; they're part of the wire spec for AutoInterface peering, and a clean-room AutoInterface implementation needs to mirror upstream's format. Read AutoInterface.py::announce_handler and AutoInterface.peer_jobs for the full layout.

8.6.5 Once peers are discovered

After the discovery handshake establishes that two nodes are in the same Reticulum group on the same physical link, all subsequent Reticulum packet traffic flows as plain UDP datagrams on the data port (42671 by default), unicast between the discovered link-local IPv6 addresses. There's no per-packet framing beyond the UDP envelope — each datagram body is one complete Reticulum packet (§2). Out-of-order delivery is handled by Reticulum's normal dedup and (where present) Link sequencing; UDP packet loss is masked by Reticulum's PROOF receipts and Resource sliding-window retransmits.

Effective HW_MTU is 1196 bytes (AutoInterface.HW_MTU, line 44) — chosen to fit comfortably within standard Ethernet MTU minus IPv6/UDP overhead.

8.6.6 IFAC integration

If the AutoInterface is configured with an ifac_identity (out-of-band-shared key), every Reticulum packet on the data port is IFAC-sealed and unmasked using the standard Transport-level IFAC mechanism (RNS/Transport.py:1338-1390). Peers with mismatched IFAC keys can see each other's discovery announces but can't decode each other's data — a pragmatic privacy boundary on a shared LAN.

8.6.7 Source map

File	What
`RNS/Interfaces/AutoInterface.py:43-72`	Default ports, group_id, scope/address-type constants, intervals
`RNS/Interfaces/AutoInterface.py:202-212`	Multicast address derivation from `SHA256(group_id)`
`RNS/Interfaces/AutoInterface.py:108+`	Per-interface socket setup (multicast join, unicast bind)
`RNS/Interfaces/AutoInterface.py::announce_handler`	Discovery announce emission
`RNS/Interfaces/AutoInterface.py::peer_jobs`	Peer aging, multicast-echo loop, IFAC sealing

9. Implementation gotchas

The findings here cost the most debugging hours per insight ratio. They're not in the upstream manual.

9.1 LXMF `source_hash` is the destination hash, not the identity hash

The 16-byte source_hash field in an LXMF body is the sender's destination hash (SHA256(name_hash || identity_hash)[:16]), NOT the raw 16-byte identity hash. Sending the identity hash here means the recipient can't look you up in their contacts (which are keyed by destination hash) and the conversation gets orphaned.

9.2 Web Crypto and JCA AES-CBC auto-pad PKCS#7 — do not pad manually

Both browser window.crypto.subtle.encrypt({name:"AES-CBC", iv}, key, plaintext) and JCA's Cipher.getInstance("AES/CBC/PKCS5Padding") apply PKCS#7 padding automatically. Manually padding before calling them produces double-padded ciphertext (16 garbage bytes added) that decrypts to plaintext + a trailing PKCS#7 block which the receiver can't strip cleanly.

9.3 RNS bundles `umsgpack` — encode display names as `bytes`, not `str`

RNS/vendor/umsgpack.py is locked to behaviors regardless of system msgpack:

_pack_string (Python str) → 0xa0|len/0xd9/0xda/0xdb (fixstr/str8/str16/str32)
_pack_binary (Python bytes) → 0xc4/0xc5/0xc6 (bin8/bin16/bin32)
_unpack_string decodes to Python str via bytes.decode("utf-8")
_unpack_binary returns raw Python bytes

The downstream parser at LXMF/LXMF.py:131 does dn.decode("utf-8") on the unpacked first element. This works only when dn is bytes. If a producer wrote a str-encoded name (fixstr), umsgpack returns Python str, .decode() raises AttributeError, the parser swallows it and returns None → no display name.

Implementation rule: encode the display name field as msgpack bin (Python bytes equivalent), never str. Upstream LXMRouter does this correctly via display_name.encode("utf-8") before packing.

9.4 Display name preservation across re-announces

Inbound announce ingestion code that uses

new_name = extracted ?? known_label ?? ""
merged   = (new_name).ifBlank { existing.name ?? "" }

clobbers a real cached name with the placeholder known_label (e.g. "LXMF delivery") whenever a minimal re-announce arrives without app_data. The next full announce restores it. Symptom: contacts blink to placeholder names briefly during/after activity.

Correct priority order: extracted ?? existing ?? known_label ?? "". The known label fallback is for completely unknown destinations only.

9.5 Self-announce echo

If the operator runs both an originating client and a transport node on the same machine (or the same RNode loops back its own emissions), a client will receive its own announce and may add itself to the contact list. Filter announces whose dest_hash == our_dest_hash before ingestion.

9.6 Clockless sender timestamps

LoRa devices without an RTC will populate the LXMF timestamp field with seconds-since-boot (small integers like 30, 90720). Treat any timestamp before 2020-01-01 (1577836800) as "no clock" and substitute the local receive time. Otherwise messages from clockless devices appear at January 1 1970 in the inbox.

9.7 Periodic re-announce is non-optional

Even after a successful initial announce, paths in the mesh expire within minutes. Without a 5–15 minute re-announce loop, the second message any peer tries to send you will fail because the relay's path table has aged out. (See also §7.5.)

There is no upstream-mandated default — RNS/Reticulum.py:764 uses 6*60*60 (6 h) for interface-level discovery announces and RNS/Transport.py:192 uses 2*60*60 (2 h) for transport-management announces, but those are not the cadence end-user destinations announce at. Sideband emits roughly every 30 minutes; the upstream manual recommends 30–60 minutes for a desktop client. Practical guidance for application destinations:

Deployment	RECOMMENDED cadence
Low-MTU LoRa node, mostly-on radio	5–10 min — short enough to outpace path-table TTL, sparse enough not to dominate airtime
Always-on rnsd-on-IP relay	15–30 min — faster doesn't help (peer caches stay fresh between announces)
Mobile / power-constrained client	5–10 min while radio active, suppress while suspended

AVOID < 60 s — short intervals trigger ingress rate limiting (§4.5 step 8) and burn ratchet-ring slots without benefit, since ratchets only rotate every 30 min anyway (§7.3.1). AVOID > 30 min on lossy links — the longer the gap, the more likely your next outbound message lands during a window when no relay holds a path back to you.

9.8 The destination hash uses the bare app-name string

An earlier-vintage bug in several implementations was to include the identity's hex hash in the name_hash input. expand_name in upstream Python takes an identity parameter and conditionally appends the identity hex IF the identity is non-None — but the Destination construction path passes identity = None. The name_hash MUST be SHA256(plain_app_name_string)[:10], nothing more. (See also §1.2.)

9.9 Diagnostic: rx-log every inbound packet at the engine entry

A single line of the form

rx <size>B H<1|2> <PT> dest=<hex> ctx=0x<hex> hops=<n>

logged before any filtering converts hours of "messages aren't arriving" debugging to seconds. Without it, packets dropped by if (dest != ours) return vanish silently and look identical to "the bytes never arrived". Symmetric tx logging on outbound is similarly cheap insurance.

9.10 microReticulum `random_hash` lacks the timestamp half

Real interop bug to plan around: attermann/microReticulum's Destination::announce emits 10 fully-random bytes for the announce random_hash field rather than the upstream Python form of 5 random bytes || big-endian uint40 unix_seconds (see §4.1). The Python form is preserved as a comment in the C++ source with a TODO add in time to random hash next to it; the timestamp half was never implemented.

Effect on a mixed-vendor mesh: a Python RNS receiver parses random_hash[5:10] of a microReticulum announce as a far-future timestamp (median ~year 19403 AD because the random uint40 is uniformly distributed across 0..2^40-1). The path-table replacement rule at RNS/Transport.py:1723-1755 rejects subsequent real-timestamped announces from Python sources as "stale" until the path TTL expires.

Symptom: a microReticulum repeater works fine when it's the only path; in a mesh that also has Python relays, paths "stick" to the microReticulum side even when shorter / fresher Python paths come up, until natural TTL expiry. First-contact path-table population is unaffected — the bug only surfaces on path replacement.

Workarounds when building a clean-room implementation that talks to a microReticulum mesh:

Emit the upstream form yourself (you have a clock — even seconds-since-boot is preferable to random bytes; the path-table comparison only cares about ordering, not absolute time).
If you receive a uint40 timestamp that's more than, say, 24 hours in the future, treat it as suspect — but be cautious because legitimate Python senders with skewed clocks could trip this.

The repeater repo's pre_build.py patches several other microReticulum protocol bugs (ratchet announce parsing, identity hash length 16→32, DATA/PROOF forwarding) but does not patch this one. Filing an upstream issue against attermann/microReticulum to land the original Python timestamp form is the durable fix.

10. Resource fragmentation protocol

A Resource transfers a payload that exceeds the per-packet content limit of an established Reticulum Link. It is the only way to carry an LXMF body, NomadNet page, or file larger than ~360 bytes (LINK_PACKET_MAX_CONTENT) over a Link. Resource is built on top of an active Link — it relies on the Link's session key for encryption (§3.1 link-derived form) and on the Link's bidirectional DATA channel for control traffic.

The complete reference is RNS/Resource.py (1380 lines in RNS 1.2.4); RNS/Packet.py:74-79 defines the context constants. This section describes the wire-level invariants a clean-room implementation must respect; many implementation choices (window sizing heuristics, watchdog timers, EIFR computation) are private and listed only when their absence would cause an interop break.

10.1 When Resource runs

Three triggers in upstream:

LXMessage.send() for DIRECT method with representation == RESOURCE. Set automatically when the encrypted-form LXMF body exceeds LINK_PACKET_MAX_CONTENT (LXMF/LXMessage.py:415-421).
NomadNet page request fulfillment — a server returning a page whose body exceeds the link MTU.
Direct file transfers via rncp and similar utilities.

10.2 Initiator-side preparation

Given input data and an RNS.Link in ACTIVE state (RNS/Resource.py:248-478):

Optional metadata prefix. If the caller supplied a metadata dict, msgpack-pack it and prepend length(3 bytes, big-endian uint24) || packed_metadata to the body. The has_metadata (x) flag in the advertisement signals this. Receivers strip the prefix during reassembly (line 699-707).
Optional bz2 compression. If auto_compress is true and the data fits within auto_compress_limit (default 64 MiB), the body is bz2-compressed and the compressed (c) flag is set. If compression doesn't shrink the data, the uncompressed form is sent and c is cleared.
Random hash prefix. A 4-byte (Resource.RANDOM_HASH_SIZE) random hash is prepended to the (compressed-or-not) body — Resource.py:405/412, a fresh RNS.Identity.get_random_hash()[:4] call. This prefix is not the r field, and is not part of the hash / expected_proof input. It is a separate throwaway value that travels inside the encrypted blob; the receiver strips and discards it (§10.8 step 3). The advertisement's r field carries a different value — self.random_hash, generated by its own get_random_hash()[:4] call at Resource.py:440 — which is the actual integrity-hash and hashmap salt.
Link encryption. The full random_hash || (compressed?) data blob is encrypted using link.encrypt(...) — i.e. the link-derived Token form (§3.1), no ephemeral_pub prefix. The encrypted (e) flag is set.
Hash and proof material (Resource.py:440-443). All three are computed over the original uncompressed plaintext — the caller's input, including any metadata prefix from step 1 (Resource.py:332) — not the compressed body, and not the random-prefixed wire blob from step 3:
- random_hash = RNS.Identity.get_random_hash()[:4] — the value the advertisement's r field carries.
- hash = SHA256(plaintext || random_hash) (32 bytes)
- truncated_hash = hash[:16]
- expected_proof = SHA256(plaintext || hash) (32 bytes) — what the receiver will eventually return in the RESOURCE_PRF packet.
The 4-byte prefix from step 3 is not in any of these inputs. The receiver strips the prefix and bz2-decompresses before hashing (§10.8 steps 3-5), so the sender must hash the uncompressed, unprefixed plaintext for the two sides to agree. A receiver that includes the prefix, or hashes the compressed form, rejects every legitimate Resource as CORRUPT.
Part split. The encrypted body is sliced into parts of size SDU = link.mtu - HEADER_MAXSIZE - IFAC_MIN_SIZE. Each part becomes a packed RNS.Packet(link, part_data, context=RESOURCE); the packed wire bytes are stored in parts[i] for later sending.
Hashmap. Each part is fingerprinted to MAPHASH_LEN = 4 bytes. The full hashmap is b"".join(map_hashes). Hash collisions within the COLLISION_GUARD_SIZE = 2 × WINDOW_MAX + HASHMAP_MAX_LEN window are detected at construction time — if two parts hash to the same 4-byte map_hash within that window, the random hash is regenerated and the whole hashmap is recomputed. Without this guard, the receiver can't disambiguate which part it just received from a part-request that named a colliding map_hash.

After preparation: total_parts = ceil(size / SDU); total_size includes metadata; total_segments = ceil(total_size / MAX_EFFICIENT_SIZE) where MAX_EFFICIENT_SIZE = 1 MiB - 1 = 1_048_575.

10.3 Wire packet contexts used during a Resource transfer

All of these are sent on the established Link and use the Link's session key for encryption (or are unencrypted PROOF-type, depending on context):

Context	Direction	Type	Body
`RESOURCE_ADV (0x02)`	initiator → receiver	DATA	msgpack dict (§10.4)
`RESOURCE (0x01)`	initiator → receiver	DATA	one part of the encrypted body, raw
`RESOURCE_REQ (0x03)`	receiver → initiator	DATA	request bytes (§10.5)
`RESOURCE_HMU (0x04)`	initiator → receiver	DATA	hashmap continuation (§10.7)
`RESOURCE_PRF (0x05)`	receiver → initiator	PROOF	`resource_hash(32)
`RESOURCE_ICL (0x06)`	initiator → receiver	DATA	resource_hash(32) — initiator cancel
`RESOURCE_RCL (0x07)`	receiver → initiator	DATA	resource_hash(32) — receiver reject/cancel

10.4 RESOURCE_ADV — the advertisement

The first packet in the transfer. Body is umsgpack.packb(dict) with these keys (RNS/Resource.py:1336-1358):

Key	Type	Meaning
`t`	int	Transfer size — encrypted byte length on the wire
`d`	int	Data size — original uncompressed plaintext byte length
`n`	int	Number of parts in this segment
`h`	bytes(32)	Resource hash — `SHA256(data
`r`	bytes(4)	Random hash prefix
`o`	bytes(32)	Original hash of the first segment (= `h` if single-segment)
`i`	int	Segment index (1-based)
`l`	int	Total segments
`q`	bytes(?) or None	Request id if this Resource carries the response to a Link REQUEST
`f`	int	Flags byte (see below)
`m`	bytes	Hashmap fragment for THIS advertisement segment — up to `HASHMAP_MAX_LEN = ⌊(LINK_MDU - 134)/4⌋` 4-byte map_hashes

The flags byte f packs six booleans (Resource.py:1310, 1377-1382):

bit 0 : e — encrypted
bit 1 : c — compressed
bit 2 : s — split (multi-segment)
bit 3 : u — is_request (this Resource is the body of a Link REQUEST)
bit 4 : p — is_response (this Resource is the body of a Link RESPONSE)
bit 5 : x — has_metadata

HASHMAP_MAX_LEN matters: the entire hashmap may not fit in one ADV. If n > HASHMAP_MAX_LEN, the receiver reconstructs subsequent map segments via RESOURCE_HMU packets after exhausting the first slice (§10.7).

The advertisement is sent once on Resource.advertise(); if no part requests arrive within the watchdog timeout, it is retransmitted up to MAX_ADV_RETRIES = 4 times before the resource is cancelled (Resource.py:573-590).

Security: cap t and d at receive time. t and d are the sender's claims about how big the resource will be. A misbehaving or hostile peer can advertise multi-gigabyte values that a naïve receiver will then try to allocate buffers for. Two attack shapes matter:

Direct allocation bomb. Receiver pre-allocates an output buffer sized from t or d and OOMs before any chunk arrives.

Decompression bomb (when c = 1). A small (~tens of KB) bz2 input legitimately expands to gigabytes. The chunk-count cap from HASHMAP_MAX_LEN (§10.4) bounds raw on-wire chunks but does NOT bound the post-decompression buffer.

Implementations SHOULD enforce a per-application cap (a few MiB is reasonable for NomadNet pages and propagation /get blobs; file downloads MAY allow more if the receiver has the budget) and reject advertisements with t or d over the cap before responding with the first RESOURCE_REQ. When c = 1, the decompressor MUST also abort if the running output total exceeds the cap (defense in depth — a sender that lies about d would otherwise bypass the parse-time check). Reference: a receiver implementing delivery_resource_advertised(resource) returning False (§5.8.3 / §16.9) is the upstream-blessed way to refuse oversized advertisements.

Upstream RNS adopted this cap in 1.1.9 after a CVE-class report: Resource.assemble uses bz2.BZ2Decompressor.decompress(data, max_length=self.max_decompressed_size) and rejects the resource if decompressor.eof is False after the bounded read (RNS/Resource.py:686-691). The Channel-mode counterpart is Buffer.RawChannelReader which caps each chunk at RawChannelWriter.MAX_CHUNK_LEN via the same max_length mechanism (RNS/Buffer.py:95-97). Clean-room implementations should mirror this — a bz2.BZ2Decompressor.decompress(data, max_length=N) plus eof check is the minimum. Do not use the one-shot bz2.decompress() API for resource bodies — it has no output bound and will allocate as much memory as the input legitimately expands to.

10.5 RESOURCE_REQ — receiver requests parts

Sent by the receiver to ask for a window's worth of specific parts (Resource.py:934-983). Body layout:

hashmap_exhausted_flag(1)  || [last_map_hash(4) if exhausted]
|| resource_hash(32)
|| requested_map_hashes(N × 4 bytes)

Where:

hashmap_exhausted_flag is 0x00 (HASHMAP_IS_NOT_EXHAUSTED) if the receiver still has unrequested map_hashes from the most-recently-known hashmap segment, or 0xFF (HASHMAP_IS_EXHAUSTED) if it has consumed all of them and needs the next hashmap segment.
If exhausted == 0xFF, the request continues with the last map_hash the receiver knows from the current segment (4 bytes). The sender uses this to determine which segment of the hashmap to send back via RESOURCE_HMU.
resource_hash is the 32-byte h from the advertisement.
The trailing requested_map_hashes is a concatenation of N × 4-byte map_hashes the receiver wants delivered. N is at most WINDOW (initial 4, dynamically grown — see §10.10).

Receivers who already have the part for a requested map_hash don't issue requests for it; the request is constructed only from parts[search_start:search_start+window] where parts[i] is None (Resource.py:944-960).

10.6 RESOURCE part packets

For each map_hash in a RESOURCE_REQ, the sender locates the matching pre-packed part within parts[receiver_min_consecutive_height : receiver_min_consecutive_height + COLLISION_GUARD_SIZE] and emits it as a regular Link DATA packet with context = RESOURCE (0x01) (Resource.py:1011-1023). The body is just the part's encrypted data — no metadata, no sequence number. The receiver matches the inbound part to its hashmap by recomputing its 4-byte map_hash and inserting it into parts[i] at the position where hashmap[i] matches (Resource.py:866-885).

Two interop traps:

Map_hashes are not guaranteed unique across the whole resource — only within COLLISION_GUARD_SIZE of any sliding-window position. A receiver that searches the entire hashmap for a matching part-hash can mis-place a part if two distant parts collide. The reference receiver searches only hashmap[consecutive_completed_height : consecutive_completed_height + window].
Parts are link-encrypted but otherwise opaque — the receiver has no way to validate a part beyond its 4-byte map_hash until the whole resource assembles and the SHA-256 over the reassembled data matches h.

Implementation gotcha: chunks are NOT individually encrypted — they are raw slices of an already-encrypted whole. Per §10.2 step 4, the entire random_hash || (compressed?) data blob is link- encrypted ONCE, then split into MTU-sized parts at step 6. Each wire chunk is just outerToken[i*sdu : (i+1)*sdu] — a fragment with no Token-form header (no IV, no HMAC) of its own. Receivers MUST hand inbound chunk bytes directly to the hashmap match (SHA-256(chunk || random_hash)[:4]) without attempting per-chunk Token decrypt. The single decrypt step happens once over the concatenated assembly inside assemble() (§10.8), not per packet.

A receiver that calls link.decrypt(chunk) on each inbound RESOURCE part will fail with HMAC verification errors on every chunk — each slice is missing the Token header bytes the decrypt expects. This is a common implementer mistake and the spec text "parts are link-encrypted" reads ambiguously enough that several clean-room ports have made it. Verbatim from Resource.py:607-625:
for i in range(0, hashmap_entries):
    data = self.data[i*self.sdu : (i+1)*self.sdu]   # slice ciphertext
    map_hash = self.get_map_hash(data)              # hash the SLICE
    part = RNS.Packet(link, data, context=RNS.Packet.RESOURCE)
    part.pack()
    self.hashmap += part.map_hash
    self.parts.append(part)
The body of each RESOURCE packet is data here — a raw slice of the already-encrypted self.data. No re-encryption.

10.7 RESOURCE_HMU — hashmap update

When the sender receives a RESOURCE_REQ with exhausted == 0xFF and a last_map_hash, it locates the position of last_map_hash in its full hashmap, advances to the next HASHMAP_MAX_LEN window, and emits the hashmap continuation (Resource.py:1030-1064):

body = resource_hash(32) || umsgpack.packb([segment_index(int), hashmap_segment_bytes])

The segment_index is part_index // HASHMAP_MAX_LEN. The receiver applies this with Resource.hashmap_update(segment, hashmap) to extend its known hashmap and continues issuing RESOURCE_REQ for the new range.

If the part_index doesn't land on a HASHMAP_MAX_LEN boundary, the sender treats it as a sequencing error and cancels the resource (Resource.py:1043-1046).

An exhausted RESOURCE_REQ MAY still carry parts — a conformant sender fulfils them and sends the RESOURCE_HMU. When the hashmap_exhausted_flag is 0xFF, the REQ body may still end with a non-empty requested_map_hashes trailer (§10.5). The sender MUST emit the requested part packets and the hashmap continuation; the two are independent. Serving the HMU is not a substitute for fulfilling the bundled part requests.

In the RNS reference (Resource.py:982-1071, request() — verified against RNS 1.2.9, the current release), the part-fulfilment loop runs for every REQ regardless of the flag, and the if wants_more_hashmap: HMU branch runs afterward, in addition. The reference receiver (request_next, Resource.py:931-981) routinely produces this packet shape: as its window scan reaches the end of the known hashmap, it has already accumulated the still-outstanding part-hashes from the known region into requested_hashes, then sets the exhausted flag and stops — emitting 0xFF || last_map_hash(4) || resource_hash(32) || requested_map_hashes.

A receiver MAY keep part requests and hashmap pulls in separate REQ packets — emitting a part-less exhausted REQ (body 0xFF || last_map_hash(4) || resource_hash(32), no trailing map_hashes) purely to pull the continuation. This interoperates with every conformant sender. But it is a receiver-side simplification only: a sender MUST NOT assume peers do this, and MUST NOT skip part fulfilment for an exhausted REQ. A sender that does drops every bundled part silently — see playbook.md §7 (2026-05-19).

10.8 RESOURCE_PRF — final proof

When the receiver has assembled the full resource (received_count == total_parts), it runs assemble() (Resource.py:672-726):

Concatenate parts[0..n] to a single buffer.
link.decrypt(...) to plaintext.
Strip the 4-byte random_hash prefix — discard, do NOT compare to advertisement.r (see callout below).
If compressed: bz2-decompress.
Recompute SHA256(plaintext || random_hash) — over the prefix-stripped, decompressed body — and compare to h.
If match: peel off metadata if x is set, write data to the destination; status = COMPLETE.
If mismatch: status = CORRUPT; cancel.

Implementation gotcha: the leading 4 bytes are NOT advertisement.r. Step 3 reads "strip the 4-byte random_hash prefix" — sender-side Resource.py:567 writes those bytes via RNS.Identity.get_random_hash()[:4], a fresh random call. They are deliberately distinct from self.random_hash (the value the advertisement's r field carries — used only for the hashmap formula SHA256(chunk || r)[:4] and the integrity formula SHA256(data || r)). A receiver that does assert prefix == advertisement.r will reject every legitimate Resource as corrupt. Just strip and discard. Integrity is proven exclusively by step 5's SHA256(plaintext || random_hash) against h — that's the only check that matters; the prefix bytes are scaffolding.

On COMPLETE, the receiver emits the proof:

proof_data = resource_hash(32) || full_proof(32)
where full_proof = SHA256(plaintext || resource_hash)

sent as RNS.Packet(link, proof_data, packet_type=PROOF, context=RESOURCE_PRF) (Resource.py:755-766). The full_proof is exactly what the initiator pre-computed as expected_proof in §10.2 step 5 — it can validate the proof bytewise without re-running the SHA-256.

The initiator's validate_proof (Resource.py:785-824) checks proof_data[32:] == self.expected_proof and transitions status to COMPLETE. If the resource is multi-segment (s == True), the next segment's advertisement is sent immediately upon proof of the current segment.

10.9 RESOURCE_ICL / RESOURCE_RCL — cancellation

Either side can cancel; the body is just resource_hash(32):

RESOURCE_ICL (0x06) — initiator cancel. Sent when the initiator decides to abort (e.g. the user kills the upload, the link MTU shrinks below the resource's pre-packed parts, the watchdog gives up after MAX_RETRIES = 16).
RESOURCE_RCL (0x07) — receiver reject / cancel. Sent on advertisement reject (Resource.reject(adv_packet) at line 155-163, e.g. resource too large per app callback) or on receiver-side abort.

Either form transitions the resource to FAILED, releases the parts, and notifies the link's resource-concluded callback.

10.10 Sliding window and rate adaptation

The receiver controls request-pacing via a sliding window:

WINDOW          = 4    # initial outstanding requests
WINDOW_MIN      = 2
WINDOW_MAX_SLOW = 10   # default cap
WINDOW_MAX_FAST = 75   # cap once link is observed to be fast
WINDOW_MAX_VERY_SLOW = 4
WINDOW_FLEXIBILITY = 4

After each successful round (every requested part arrived), window += 1 up to window_max; window_min += 1 once window - window_min > WINDOW_FLEXIBILITY - 1 (Resource.py:902-906). The window cap is promoted to WINDOW_MAX_FAST after FAST_RATE_THRESHOLD consecutive rounds at observed throughput > RATE_FAST = 50 kbps / 8, and demoted to WINDOW_MAX_VERY_SLOW after VERY_SLOW_RATE_THRESHOLD = 2 rounds below RATE_VERY_SLOW = 2 kbps / 8 (Resource.py:917-927). These are receiver-private — they're not negotiated, so two implementations with different rate-detection cutoffs interop fine but may emerge with different effective throughput on the same channel.

10.11 Multi-segment resources

For payloads larger than MAX_EFFICIENT_SIZE = 1 MiB - 1, the resource is split into multiple segments at MAX_EFFICIENT_SIZE boundaries (Resource.py:299-314). Each segment is its own Resource with its own RESOURCE_ADV; the i (segment_index) and l (total_segments) fields disambiguate. The o (original_hash) field carries the first segment's h so the receiver can correlate segments belonging to the same logical transfer.

The sender doesn't pre-prepare every segment up front — it builds segment N+1 in __prepare_next_segment while segment N is still being delivered, and sends segment N+1's advertisement only after it has received the proof for segment N (Resource.py:768-783, 822-824). This caps memory usage; a 100 MiB transfer doesn't materialize 100 segments simultaneously.

The 3-byte big-endian uint24 metadata length encoding (§10.2 step 1) is what limits per-resource metadata to METADATA_MAX_SIZE = 16 MiB - 1.

10.12 Compression and encryption layering

Encryption layering is outermost — the wire bytes look like:

wire_blob           = prefix(4) || maybe_compressed       # the body that gets encrypted
prefix              = fresh get_random_hash()[:4]         # NOT `r`; receiver strips & discards
maybe_compressed    = compressed_body iff `c` flag, else uncompressed
parts[i]            = link.encrypt(wire_blob)[i*SDU : (i+1)*SDU]   # encrypt whole, then slice

hash                = SHA256(uncompressed_body || random_hash)    # integrity; random_hash = adv `r`

Critically, the link encryption is applied to the WHOLE concatenated data first, then sliced into parts — not to each part individually. This means part boundaries don't align with cipher block boundaries; a missing part can't be decrypted in isolation. The receiver must accumulate all parts before calling link.decrypt() (Resource.py:676-679).

This also means swapping in a new link session key mid-transfer would break decryption — the encryption happened with the link's key as it was when the resource was constructed.

10.13 Source map for §10

File	What it pins down
`RNS/Resource.py:43-156`	Class header, constants, state machine values, `reject` / `accept`
`RNS/Resource.py:248-478`	`Resource.__init__` — preparation, hashmap construction, collision guard
`RNS/Resource.py:520-596`	`__advertise_job`, watchdog, advertisement retransmit
`RNS/Resource.py:672-726`	`assemble` — receiver reassembly, decrypt, decompress, hash-match
`RNS/Resource.py:755-829`	`prove` and `validate_proof`
`RNS/Resource.py:831-932`	`receive_part` — receiver-side part insertion + window adjust
`RNS/Resource.py:934-983`	`request_next` — receiver-side RESOURCE_REQ construction
`RNS/Resource.py:985-1064`	`request` — initiator-side fulfillment + RESOURCE_HMU emission
`RNS/Resource.py:1237-1383`	`ResourceAdvertisement` — pack/unpack of the ADV msgpack dict
`RNS/Packet.py:72-78`	RESOURCE_* context constants

11. REQUEST/RESPONSE protocol (NomadNet pages, propagation `/get`, custom RPC)

A generic over-Link RPC mechanism. NomadNet uses it for page fetches; LXMF propagation uses it for offline-message retrieval; any application can register handlers for arbitrary paths. There is no separate "NomadNet wire format" — NomadNet is just one consumer of this protocol.

This section specifies the wire bytes; the application-layer paths (e.g. NomadNet's /page/index.mu) are caller-defined.

11.1 Wire form — REQUEST (initiator → server)

RNS/Link.py::request line 478-527. After an active Link is established (§6), the initiator builds:

request_path_hash = SHA256(path.encode("utf-8"))[:16]
unpacked_request  = [time.time(), request_path_hash, data]
packed_request    = umsgpack.packb(unpacked_request)

Then dispatches based on size:

`len(packed_request)`	Wire form
`≤ link.mdu`	One Link DATA packet, `context = REQUEST (0x09)`, body = `packed_request`
`> link.mdu`	Resource transfer (§10), with `request_id = SHA256(packed_request)[:16]`, `is_response = False` (sets `u = True` in the Resource advertisement flags per §10.4)

The msgpack array layout:

[0]  timestamp            float (seconds since unix epoch, requester's clock)
[1]  request_path_hash    bytes(16) — SHA-256 of the requested path string, truncated
[2]  data                 application-defined value, encoded directly into the
                          outer msgpack list — NOT a pre-msgpacked byte blob

Implementation gotcha: element [2] is encoded once, not twice. data is whatever the application wants to send: None (msgpack nil) for plain GETs, a dict for NomadNet form posts (§11.6), a list for LXMF propagation /get rounds (§11.6), or bytes for opaque application blobs. The whole [time, path_hash, data] list is msgpacked exactly once. Element [2] is NOT a pre-encoded byte blob wrapped as msgpack bin — that's a common implementer mistake (see below) and it silently corrupts every form submission and every propagation poll because server-side handlers do isinstance(data, dict) / isinstance(data, list) and the bin form is bytes, falling through to the no-op branch.

Concrete example for a NomadNet form post field_message=hello:
data = {"field_message": "hello"}      # native Python dict
envelope = [time.time(), path_hash, data]
packed = umsgpack.packb(envelope)      # ONE pack call
# → on the wire, element [2] decodes back to a {} map, NOT to bytes
Pre-pack callers (umsgpack.packb(data) then passing the bytes as element [2]) produce a wire envelope where decode yields bytes for [2] — looks structurally similar but is semantically a different type, and every NomadNet Node.py:109 / LXMF LXMRouter.__get_handler drops the request silently with no error response.

For single-packet REQUESTs, request_id = SHA-256(packet.get_hashable_part())[:16] — i.e. the 16-byte truncation of the packet hash, computed over the on-the-wire bytes (low nibble of flags || raw[2:] for HEADER_1 / raw[18:] for HEADER_2). NOT a hash of the inner plaintext or of the msgpack-encoded packed_request blob. The server side at Link.handle_request:1286 literally calls packet.getTruncatedHash(). Both sides MUST hash the same bytes to match. For Resource REQUESTs the request_id is carried explicitly in the advertisement's q field (§10.4) and the initiator MUST set it to the truncated SHA-256(packed_request)[:16] of the inner plaintext per Resource.py::__init__ line 478 (Resource path uses the plaintext-hash form because there is no single packet to hash). The receiver uses this id to correlate the inbound RESPONSE with this REQUEST.

11.2 Wire form — RESPONSE (server → initiator)

RNS/Link.py::handle_request line 853-904. The server's response generator returns a value, and the dispatcher picks the wire form by size:

packed_response = umsgpack.packb([request_id, response])

if len(packed_response) <= link.mdu:
    RNS.Packet(link, packed_response, DATA, context = RESPONSE).send()
else:
    response_resource = RNS.Resource(packed_response, link,
                                     request_id=request_id, is_response=True,
                                     auto_compress=auto_compress)

Wire form	Trigger
Link DATA packet, `context = RESPONSE (0x0A)`, body = `umsgpack([request_id, response])`	response fits in `link.mdu`
Resource transfer, `request_id` field set, `is_response = True` (advertisement flag `p`)	response too large

The request_id in element [0] of the response msgpack lets the initiator match the response to the original outbound REQUEST in Link.pending_requests even when several requests are in flight on the same Link (Link.handle_response line 906-925).

Security: initiators MUST verify element [0]. The request_id check isn't decorative — without it, a misbehaving or compromised transit relay can replay a stale RESPONSE from a prior request and the initiator accepts it as the answer to whatever's currently pending. An implementation that drives only one in-flight request per link at a time is "lucky" today (the wrong-id RESPONSE just happens to carry sane bytes for the application to display), but as soon as it adds link reuse, partials, or any kind of pipelining the bug becomes a silent confused-deputy.

Compute expected_id correctly. Server-side Link.handle_request:1286 is:
request_id = packet.getTruncatedHash()
i.e. SHA-256(packet.get_hashable_part())[:16] where get_hashable_part() (Packet.py:332-338) is:
hashable = (raw[0] & 0x0F) || raw[2:]              # HEADER_1
hashable = (raw[0] & 0x0F) || raw[18:]             # HEADER_2 (skips transport_id slot)
NOT a hash of the inner plaintext. Compute the same on the initiator from your outbound REQUEST packet's wire bytes; on every inbound RESPONSE, drop the packet (and log) if decoded[0] doesn't match. Many clean-room implementations have read this section's prior wording ("16-byte truncated hash of packed_request") as "hash the inner plaintext bytes" and produced a formula that never matches what the server sent — every RESPONSE gets dropped, every page-fetch and /get round times out silently. The hashing is over the on-the-wire packet bytes, not the encrypted-then-decrypted payload.

File responses

If the server's response generator returns a (file_handle, metadata) tuple, the response goes out as a Resource carrying the file's bytes with optional msgpack metadata in the Resource advertisement's metadata slot — RNS/Link.py:888-895:

if type(response) == tuple and isinstance(response[0], io.BufferedReader):
    file_handle = response[0]
    metadata    = response[1] if len(response) > 1 else None
    response_resource = RNS.Resource(file_handle, link,
                                     metadata=metadata, request_id=request_id,
                                     is_response=True, auto_compress=auto_compress)

This is how NomadNet ships large pages with attached MIME-type / size hints — the file goes through the §10 Resource pipeline; the metadata hits the advertisement's m slot reserved for the resource hashmap but also gets a separate metadata-prefix slot per §10.2 step 1 (the 3-byte length-prefixed msgpack-packed metadata blob inserted before the random_hash).

11.3 Path hash collision avoidance

request_path_hash is the 16-byte truncation of SHA256(path) — collision space is 2^128, effectively no collisions in practice. The server's request_handlers dict is keyed by this hash:

# RNS/Destination.py::register_request_handler
request_path_hash = SHA256(path.encode("utf-8"))[:16]
self.request_handlers[request_path_hash] = (path, response_generator, allow, allowed_list, auto_compress)

A server registers a path string; clients hash the path and look it up. The path string itself is not on the wire — only its hash. This means the server can publish opaque path tokens that resist enumeration: a client must already know the path string to fetch the resource at it. NomadNet uses human-readable paths like /page/index.mu because the clients (Sideband, the NomadNet client) need them to be discoverable; a private file-server use case can use random tokens for security-by-obscurity.

11.4 Authorization (`allow` modes)

Registered via Destination.register_request_handler(path, response_generator, allow=...):

Mode	Constant	Effect
`ALLOW_NONE`	`0x00`	Reject every request (handler is a stub for testing).
`ALLOW_LIST`	`0x01`	Accept iff the requester has identified themselves on the link (via `link.identify(identity)`) AND their identity_hash is in `allowed_list`.
`ALLOW_ALL`	`0x02`	Accept any request that arrives on this Link, regardless of caller identity.

Link.identify(identity) runs LINKIDENTIFY (context = 0xFB) packets; this is how the requester proves which long-term identity is making the request without re-running a fresh Link handshake. Most public NomadNet pages use ALLOW_ALL; private pages and propagation-node operator commands use ALLOW_LIST.

11.5 RequestReceipt — initiator-side state machine

RNS/Link.py:1335-1530. When Link.request() returns a RequestReceipt, the initiator can attach:

response_callback(receipt) — fires when the response has fully arrived (single packet OR resource concluded).
failed_callback(receipt) — fires on timeout or link teardown.
progress_callback(receipt) — fires each time more bytes arrive (for Resource responses; reports receipt.progress 0.0..1.0).

Default timeout is link.rtt × link.traffic_timeout_factor + Resource.RESPONSE_MAX_GRACE_TIME × 1.125 — typically a few seconds plus a generous response-side grace. Caller can override via the timeout= kwarg.

11.6 NomadNet specifics (informational, not normative)

Click to expand — NomadNet-layer conventions on top of §11 (form data env vars, link target syntax, micron page headers, /file/ downloads, ALLOW_LIST, partials). Skip if you're not implementing a NomadNet client; the §11 wire form is the protocol layer.

NomadNet pages are served over this protocol with these conventions. Source-of-truth for all of these is upstream markqvist/NomadNet: nomadnet/Node.py (server) and nomadnet/ui/textui/Browser.py (client).

11.6.1 Paths and the `nomadnetwork.node` aspect

Server: hosts a destination at nomadnetwork/node aspects (name_hash = 213e6311bcec54ab4fde). Pages are registered as register_request_handler(path="/page/<name>.mu", ...).
Client: default path is /page/index.mu (Browser.py:67 DEFAULT_PATH).
Path format: /page/<name>.mu for micron pages, /file/<name> for static file downloads (§11.6.5).
Path hash on the wire is the §11.1 SHA-256(path)[:16] truncation — /page/index.mu and /page/help.mu are distinct request_handler keys.

11.6.2 Form data and env-var convention

When a client tap on a micron link with form fields fires a REQUEST, element [2] of the envelope is a msgpack dict (NOT pre-msgpacked bytes — see §11.1). Two key prefixes are conventional and special-cased server-side:

Prefix	Source	Server treatment
`field_<name>`	Form-input values typed by the user	Exported as env var `field_<name>=<value>` to the page's executable handler
`var_<name>`	URL-query-style parameters embedded in the link itself	Exported as env var `var_<name>=<value>`

Node.py:109-111 (upstream master, fetched 2026-05-04):

if data != None and isinstance(data, dict):
    for e in data:
        if isinstance(e, str) and (e.startswith("field_") or e.startswith("var_")):
            env_map[e] = data[e]

The field_ vs var_ distinction is purely cosmetic on the wire (both become env vars), but in micron syntax they have separate origins:

Form fields (field_<name>) come from <flags|namevalue>` widgets that render as text inputs / checkboxes / radios. The Browser collects current widget state into a dict at submit time.
URL parameters (var_<name>) come from key=value entries in the third backtick component of a link: `[label`/page/foo.mu`username=alice|active=true|message] produces {"var_username": "alice", "var_active": "true", ...} PLUS field_message from a widget named message (Browser.py:198-205). Entries with = are var-params; entries without are field-widget names whose current values get included.

Checkbox semantics (Browser.py:226-241)

For checkboxes specifically:

Unchecked: the field key is omitted from the dict entirely (NOT sent as empty string).
Multi-select (multiple checkboxes sharing the same field name): values are comma-joined ({"field_topics": "weather,radio"}).

Implementations that always send {"field_<name>": ""} for unchecked boxes will break server-side handlers that test if "field_subscribe" in env: ....

11.6.3 Link target syntax (parsed by `Browser.py` `expand_shorthands` + `link_request`)

A micron link's target string (the second component of [labeltarget]or third of[labeltargetfields]`) is one of:

Form	Meaning	Browser.py ref
`/path/to/page.mu`	Same-node nav: load `path` on the currently-selected destination.	implicit
`<32hex>` (bare 16-byte truncated identity hash, hex-encoded)	Cross-node nav to `nomadnetwork.node` at that hash, default path `/page/index.mu`.	255-259
`<32hex>:/page/x.mu`	Cross-node nav with explicit path.	255-259
`nnn@<32hex>[:/path]`	Same as bare-hash form; `nnn` is a shorthand for `nomadnetwork.node`.	184-189
`lxmf@<32hex>` / `lxmf.delivery@<32hex>`	Open a conversation in the LXMF (messaging) layer, NOT a page fetch.	184-189, 266-322

expand_shorthands (lines 184-189):

def expand_shorthands(self, destination_type):
    if destination_type == "nnn":   return "nomadnetwork.node"
    elif destination_type == "lxmf": return "lxmf.delivery"
    else: return destination_type

Implementations should normalize hash hex to lower case before keying any cache / repo lookup, and reject inputs with embedded separators (dead:beef:…) — the wire form is plain bytes, accepting forgiving variants creates aliases for the same destination and risks cache-poisoning.

11.6.4 Page-level header conventions

A .mu page MAY begin with one or more single-line headers prefixed #!. These are stripped by Browser.py before micron rendering and are NOT part of the page body:

Header	Effect	Ref
`#!c=<seconds>`	Cache-TTL hint. `0` = "do not cache." Default cache is 12 h.	Browser.py:1315-1335
`#!bg=<3hex or 6hex>`	Page-wide background color.	Browser.py:1282-1302
`#!fg=<3hex or 6hex>`	Page-wide foreground color (overrides theme default).	Browser.py:1282-1302

The #!c=N header is widely used; the color headers are rare. A client that doesn't honor any of them still renders pages correctly.

11.6.5 File downloads (`/file/...`)

Pages whose path starts with /file/ are static downloads, not micron content. The server's response generator returns:

return [open(file_path, "rb"), {"name": file_name.encode("utf-8")}]

— a (file_handle, metadata_dict) pair. The transport-layer file response shape per §11.2 §"File responses": the file bytes go through the §10 Resource pipeline, AND the metadata is also embedded as a length-prefixed msgpack blob in the Resource advertisement's metadata-prefix slot (§10.2 step 1). Clients receive [filename_bytes, file_data_bytes] after Resource assembly (Browser.py:1437-1441).

A client that hasn't implemented file downloads can detect /file/ paths and either show a "downloads not supported" message or just discard the response.

11.6.6 Authorization: `ALLOW_ALL` vs `ALLOW_LIST`

Pages are registered with one of three allow modes (Destination.py:35-40):

ALLOW_ALL — anyone with a Link can fetch. Used for public NomadNet pages, the propagation node's /get, etc.
ALLOW_LIST — caller's identity hash must appear in the page's .allowed file. Server checks remote_identity.hash against the list at request time (Node.py:152-154).
ALLOW_NONE — registered handlers that exist but reject all requests (rare; debug only).

For ALLOW_LIST the client MUST call link.identify(identity) immediately after the link transitions to ACTIVE and BEFORE issuing the REQUEST. This sends a LINKIDENTIFY (context = 0xFB) packet whose 128-byte payload is public_key(64) || signature(64), with the signature computed over link_id || public_key — proving the long-term identity hash to the responder. The wire format is specified in §6.7.6. Without it, remote_identity is None server-side and every ALLOW_LIST page returns DEFAULT_NOTALLOWED. See Browser.py:1245-1250 for the upstream call site:

def link_established(self, link):
    if self.app.directory.should_identify_on_connect(self.destination_hash):
        self.link.identify(self.app.identity)

Privacy note for client implementers. Calling link.identify on every link reveals the user's long-term identity hash to any node they browse — including pages on hostile public hubs. Implementations SHOULD make identify opt-in per destination (or per session), only firing it when the user has affirmatively decided to authenticate. Anonymous browsing of ALLOW_ALL pages should not pin identity.

11.6.7 Partial pages (server-side includes)

A micron page may embed `{<path>[`<refresh_seconds>[`<fields>]]} placeholders. The Browser tracks each placeholder, opens / reuses a Link to the partial's destination, fetches <path> as a sub-REQUEST, and substitutes the response bytes into the rendered output. If a <refresh> is set, the partial is re-fetched periodically.

Implementation reference: Browser.py:493-606 (__load_partial, start_partial_updater). Partials are how live "chat tail" / "status" panels work on real NomadNet community pages. A client without partial support sees the literal placeholder text and the page renders as a static snapshot.

11.6.8 Source map (NomadNet ↔ wire)

Concept	Upstream Python file:line
Default path	`nomadnet/ui/textui/Browser.py:67`
Form-field collection	`Browser.py:198-241`
`field_` / `var_` env-var mapping	`nomadnet/Node.py:109-111`
Shorthand expansion (`nnn`/`lxmf`)	`Browser.py:184-189`
Cross-node link routing	`Browser.py:248-322`
Identify-on-connect	`Browser.py:1245-1250`
Cache-TTL header `#!c=N`	`Browser.py:1315-1335`
Color headers `#!bg=` / `#!fg=`	`Browser.py:1282-1302`
`/file/...` download dispatch	`Browser.py:781-785, 1420-1462` + `Node.py:128-141`
Partial placeholders	`Browser.py:493-606`
Allow modes / `ALLOW_LIST` enforcement	`Node.py:152-154`

None of these are wire-spec — they're caller conventions layered on top of §11. A Reticulum client that can't render micron markup or doesn't implement the form/cache/partial conventions can still fetch pages and display the raw bytes; the protocol layer doesn't care about content.

11.7 Source map

File	What
`RNS/Link.py:478-527`	`Link.request()` — initiator-side packing and dispatch by size
`RNS/Link.py:853-904`	`Link.handle_request()` — server-side path lookup + auth + response dispatch
`RNS/Link.py:906-925`	`Link.handle_response()` — initiator-side response correlation
`RNS/Link.py:1335-1530`	`RequestReceipt` — callback machinery
`RNS/Destination.py::register_request_handler`	Server-side handler registration
`RNS/Destination.py:35-40`	`ALLOW_NONE/ALLOW_LIST/ALLOW_ALL` constants
`RNS/Packet.py:81-82`	`REQUEST = 0x09`, `RESPONSE = 0x0A` context constants

12. Transport-relay behaviour

Everything in §1-§11 applies to both leaf clients and transport-mode nodes. This section covers what additionally runs on a node configured with enable_transport = Yes in the [reticulum] config — i.e. a node whose role is to forward traffic for others. Reticulum's relay is host-routed (no broadcast flooding except for path-discovery), keyed by the path_table populated from announces.

A leaf client can ignore §12 entirely. Implementations that target the rnsd-replacement or repeater use case need every sub-section.

12.1 The `transport_enabled` toggle

Reticulum.transport_enabled() returns the value of the enable_transport config option (default False). Setting it to True:

Allows the node to populate path_table, announce_table, link_table, reverse_table, and tunnels for non-local destinations (a leaf only populates path entries it personally needs).
Enables the §12.2 DATA forwarding branches in Transport.inbound.
Enables the §12.3 ANNOUNCE rebroadcast branch.
Enables Transport.identity — the transport node's own identity, used for transport_id insertion in HEADER_2 packets (§2.3) and as the requesting_transport_instance field in path requests (§7.1).

A clean-room implementation testing forwarding without operating as a real transport node SHOULD respect the same flag: ignoring the toggle and unconditionally forwarding turns every implementation into a network-flooding hazard.

12.2 DATA forwarding rules

For an inbound DATA packet (packet_type == DATA, destination_type not LINK) where:

packet.transport_id == Transport.identity.hash (the originator picked us as the next hop), AND
packet.destination_hash is in Transport.path_table,

the relay rewrites the wire bytes according to path_table[dest][HOPS] and re-transmits on path_table[dest][RVCD_IF]. From RNS/Transport.py:1500-1580, three cases by remaining_hops:

12.2.1 `remaining_hops > 1` — forward as HEADER_2

Increment hops (already done by Transport.inbound line 1395), replace the transport_id with the next-hop transport_id from the path table, keep the rest of the packet:

new_raw = packet.raw[0:1]                        # flags byte unchanged
new_raw += struct.pack("!B", packet.hops)        # incremented hops byte
new_raw += next_hop                              # 16B transport_id (new next hop)
new_raw += packet.raw[18:]                       # original dest_hash + ctx + body

The flags byte high nibble is unchanged — the packet stays HEADER_2 with the TRANSPORT bit set. Final wire form is flags(1) || hops+1(1) || new_transport_id(16) || dest_hash(16) || ctx(1) || body.

12.2.2 `remaining_hops == 1` — strip transport headers, forward as HEADER_1 broadcast

The destination is one hop away on the next-hop interface; no further transport_id is needed. Convert to HEADER_1 with BROADCAST transport type:

new_flags = (HEADER_1 << 6) | (BROADCAST << 4) | (packet.flags & 0x0F)
new_raw = struct.pack("!B", new_flags)
new_raw += struct.pack("!B", packet.hops)
new_raw += packet.raw[18:]                       # original dest_hash + ctx + body (transport_id stripped)

This is the inverse of the §2.3 originator HEADER_1→HEADER_2 conversion: the relay strips the transport_id when the packet has reached its last hop.

12.2.3 `remaining_hops == 0` — local destination, just bump hops

The destination is registered on the relay itself (it's both our path-table next-hop AND a local destination). Just increment hops and pass through unchanged for local processing — the standard Destination.receive path takes over from there.

12.2.4 LINKREQUEST forwarding extras

When the forwarded packet is a LINKREQUEST, the relay also writes a link_table entry keyed by the link_id (computed via §6.3's link_id_from_lr_packet). Entry contents (Transport.py:1556-1565):

[ now,                            # 0  IDX_LT_TIMESTAMP
  next_hop,                       # 1  IDX_LT_NH_ID — next-hop transport_id
  outbound_interface,             # 2  IDX_LT_NH_IF
  remaining_hops,                 # 3  IDX_LT_REM_HOPS
  packet.receiving_interface,     # 4  IDX_LT_RCVD_IF
  packet.hops,                    # 5  IDX_LT_TAKEN_HOPS
  packet.destination_hash,        # 6  IDX_LT_DSTHASH
  False,                          # 7  IDX_LT_VALIDATED
  proof_timeout ]                 # 8  IDX_LT_PROOF_TMO

This entry is what lets the relay forward the eventual LRPROOF back to the initiator on the reverse path (§12.5) and forward subsequent Link DATA in both directions.

The relay also performs the §6.6 MTU clamp at this point: if the LINKREQUEST carries signalling and the next-hop interface's HW_MTU is smaller than the requested value, the signalling bytes in new_raw are rewritten in place with the clamped MTU before transmission.

12.2.5 Non-LINKREQUEST DATA — reverse_table entry

For any other forwarded DATA (the much-more-common opportunistic LXMF case), the relay writes a reverse_table entry keyed by packet.getTruncatedHash() (Transport.py:1570-1574):

[ packet.receiving_interface,    # 0  IDX_RT_RCVD_IF — interface to send PROOF back through
  outbound_interface,            # 1  IDX_RT_OUTB_IF — interface forward was sent on
  time.time() ]                  # 2  IDX_RT_TIMESTAMP

The reverse_table is what lets the eventual PROOF receipt (§6.5) trace its way back to the originator without consulting the path_table again — see §12.5.

12.3 ANNOUNCE rebroadcasting

When an inbound ANNOUNCE validates (per §4.5) AND the destination is non-local AND transport_enabled OR is_from_local_client, the relay queues a rebroadcast. From Transport.py:1810-1895:

if (Reticulum.transport_enabled() or is_from_local_client) and packet.context != PATH_RESPONSE:
    if not rate_blocked:
        Transport.announce_table[packet.destination_hash] = [
            now, retransmit_timeout, retries,
            received_from, announce_hops, packet,
            local_rebroadcasts, block_rebroadcasts, attached_interface,
        ]

The announce_table entry queues a delayed retransmit; the actual emission happens in the periodic Transport.jobs cycle which scans the table for entries whose retransmit_timeout has elapsed and fires them on each suitable interface.

12.3.1 Announce cap (`ANNOUNCE_CAP`)

Reticulum.ANNOUNCE_CAP = 2.0 (default 2% of airtime, configurable via [reticulum] announce_cap). Each interface tracks its outbound announce airtime and when the rolling-window utilization exceeds the cap, further announces are queued in interface.announce_queue rather than transmitted immediately. process_announce_queue (RNS/Interfaces/Interface.py:232-272) drains the queue at a rate the cap permits, picking the lowest-hop-count entry first.

The cap is per-interface, not global — a relay with multiple interfaces budgets each one independently, which lets a fast TCP backbone interface announce freely while the same node throttles announces on a slow LoRa interface. Without per-interface caps, a single high-rate interface would starve every other.

12.3.2 `random_blob` replay defence

§4.5 step 6.3 already documents this from the receiver's perspective; for the rebroadcast logic, the relay only queues an announce if the new random_blob (the 10-byte random_hash field, treated as an opaque blob for routing purposes) is not already in the cached random_blobs list for this destination. The list is capped at Transport.MAX_RANDOM_BLOBS (default 32) entries, sliding-window. This prevents an announce from looping through a multi-relay topology because each relay only forwards each unique blob once.

12.3.3 Path-response announces don't rebroadcast

packet.context == PATH_RESPONSE short-circuits the rebroadcast branch (line 1822). Path-response announces travel back along the reverse path from the responder to the requester (see §7.2 and flows/path-discovery.md), and the relay's job is to forward them on a single specific interface (attached_interface), not re-broadcast to the whole mesh. Mishandling this would multiply path-response traffic by the relay fanout.

12.4 Path table management

Transport.path_table[destination_hash] entry shape (Transport.py:3457-3464):

[ timestamp,                # 0  IDX_PT_TIMESTAMP — when last refreshed
  next_hop,                 # 1  IDX_PT_NEXT_HOP — 16B transport_id of next hop
  hops,                     # 2  IDX_PT_HOPS — distance to destination
  expires,                  # 3  IDX_PT_EXPIRES — unix-seconds eviction time
  random_blobs,             # 4  IDX_PT_RANDBLOBS — sliding window of recent blobs
  receiving_interface,      # 5  IDX_PT_RVCD_IF — interface to forward on
  packet ]                  # 6  IDX_PT_PACKET — cached announce packet for path-? response

12.4.1 TTLs

Three different expiry constants based on the attached_interface.mode:

Mode	TTL constant	Default seconds	Used for
`MODE_ACCESS_POINT`	`Transport.AP_PATH_TIME`	1 hour	Hub-and-spoke topologies (TCP servers, BLE gateways)
`MODE_ROAMING`	`Transport.ROAMING_PATH_TIME`	4 hours	Mobile devices that disappear and reappear
(default)	`Transport.PATHFINDER_E`	30 days	Stable backbone interfaces

The wide spread of defaults reflects expected churn rates: AP-mode interfaces have many short-lived clients; roaming devices come and go; backbone TCP relays are essentially permanent.

12.4.2 Eviction

Transport.jobs runs a stale_paths accumulator that walks path_table and pops entries whose expires timestamp has passed (Transport.py:750-770). Eviction is silent — no notification to the application; the next outbound message to the destination just re-discovers it via request_path per §7.1.

A relay also evicts path entries whose underlying interface has been removed (receiving_interface not in Transport.interfaces). This handles the case where a TCP client disconnects.

12.4.3 Persistence

If [reticulum] persist_paths = Yes, the path_table is serialized to {storagepath}/paths (a pickled dict in upstream RNS) so it survives restarts. The repeater repo's pre_build.py adds a "skip redundant path writes" patch to avoid hammering the on-board flash on nRF52 — for clean-room implementations, the persistence cadence is implementation-private.

12.5 Reverse-table link transport

Once a Link's LINKREQUEST has been forwarded by a relay (§12.2.4 wrote the link_table entry), every subsequent Link packet — DATA, KEEPALIVE, PROOF, LINKCLOSE — must be forwarded by the same relay in the appropriate direction. Transport.inbound uses link_table and reverse_table for this:

12.5.1 LRPROOF forwarding

When an LRPROOF arrives whose dest_hash (= link_id) is in the relay's link_table AND the proof arrives on the next-hop interface (packet.receiving_interface == link_entry[IDX_LT_NH_IF]), the relay validates the signature against the destination's known long-term public key (recalled via Identity.recall(link_entry[DSTHASH])) and forwards on the receive interface (Transport.py:2110-2145):

new_raw = packet.raw[0:1] + struct.pack("!B", packet.hops) + packet.raw[2:]
Transport.transmit(link_entry[IDX_LT_RCVD_IF], new_raw)
link_table[packet.destination_hash][IDX_LT_VALIDATED] = True

After validation, the link_table entry is marked validated, and from now on the relay forwards Link DATA in both directions transparently.

12.5.2 Link DATA forwarding

For a DATA packet with destination_type == LINK whose dest_hash is in link_table, the relay forwards on the appropriate direction's interface. The link_table entry remembers both sides via IDX_LT_NH_IF (toward initiator end) and IDX_LT_RCVD_IF (toward responder end); the relay picks based on which interface the inbound packet arrived on.

Unlike the path_table forwarding in §12.2 — which strips transport_id (HEADER_2 → HEADER_1) at the last hop — link_table forwarding does NOT touch the header. The relay bumps hops and emits packet.raw[0:1] + struct.pack("!B", packet.hops) + packet.raw[2:] (Transport.py:1618-1620). Whatever header bytes the initiator emitted reach the destination verbatim. This is why the originator's wire conventions for post-handshake link DATA are constrained — see §6.4.3: senders MUST emit HEADER_1 with no transport_id for every link-addressed packet, because a HEADER_2 link packet would arrive at the destination with transport_id intact and be dropped by the destination's packet_filter as "for another transport instance".

12.5.3 PROOF receipt forwarding via `reverse_table`

Transport.py:2199-2208. When a PROOF arrives whose dest_hash is in reverse_table (i.e. an opportunistic-DATA proof being routed back to its originator), the relay pops the entry, checks the proof arrived on the correct outbound interface (receiving_interface == reverse_entry[IDX_RT_OUTB_IF]), and forwards on the originally-receiving interface:

new_raw = packet.raw[0:1] + struct.pack("!B", packet.hops) + packet.raw[2:]
Transport.transmit(reverse_entry[IDX_RT_RCVD_IF], new_raw)

Reverse_table entries are popped on use (one-shot routing) and aged out by Transport.jobs after Transport.REVERSE_TIMEOUT (default 30s). This bounds the relay's memory regardless of whether the proof ever arrives.

12.6 Tunnels and shared-instance protocol

Two related state mechanisms a transport node maintains:

12.6.1 `discovery_path_requests`

When a transport-enabled relay receives a path? for a destination it doesn't know AND doesn't have a local client to forward to, it records a discovery_path_requests[dest_hash] entry (Transport.py:2955-2970):

pr_entry = {
    "destination_hash": destination_hash,
    "timeout": time.time() + PATH_REQUEST_TIMEOUT,    # 15s
    "requesting_interface": attached_interface,
}
Transport.discovery_path_requests[destination_hash] = pr_entry

Then forwards the path? to every other interface preserving the original tag. This is recursive transport-mode discovery — the relay is acting as a search proxy. When the response announce eventually arrives back, the relay forwards it on requesting_interface (the one the original path? came from), and the entry is aged out.

12.6.2 `tunnels`

A tunnel is an interface-level path mechanism for handling temporarily-disconnected interfaces (e.g. a mobile peer that comes and goes). The tunnels[interface_tunnel_id] state lets the relay reconstruct paths through the interface when it reconnects, without requiring all paths to be re-discovered from scratch. The shape (Transport.py:792-832):

[ now,                                       # 0  IDX_TT_TIMESTAMP
  expires,                                   # 1  IDX_TT_EXPIRES — TUNNEL_TIMEOUT
  paths_dict,                                # 2  IDX_TT_PATHS — dest_hash → path-entry
  ... ]

Each path inside the tunnel's paths_dict mirrors a path_table entry. When the tunnel's interface returns, the relay re-installs every path from the tunnel into the active path_table, jump-starting connectivity. Without this, every reconnection would require a full announce flood across the mesh.

TUNNEL_TIMEOUT defaults to substantially longer than path TTLs because tunnels persist across interface flap.

12.6.3 Shared-instance protocol

When multiple processes on one host share a single Reticulum stack (via share_instance = Yes in the rnsd config), one process owns Transport and the others connect to it as local clients via a small TCP loopback interface. The shared instance treats local-client traffic specially:

from_local_client and for_local_client are computed on every inbound packet (Transport.py:1453-1456).
Path-table entries with IDX_PT_HOPS == 0 mean "destination is a local client" — the §2.3 originator-side HEADER_1 conversion applies for hops==1 too, so the shared instance gets a transport_id-tagged packet (Transport.py:1097-1108).
Local-client originated path? requests are forwarded to every external interface, fanning out the search across the shared mesh (§7.2 dispatch branch 3).

The wire protocol for shared-instance loopback is just the same Reticulum packets over a TCP loopback interface — no special framing or commands. What's "shared" is the path_table and announce dispatch, not the wire format.

12.7 Source map for §12

File	What
`RNS/Transport.py:1500-1580`	DATA forwarding (HEADER_1↔HEADER_2 conversion for relay)
`RNS/Transport.py:1556-1565`	`link_table` entry shape
`RNS/Transport.py:1570-1574`	`reverse_table` entry shape
`RNS/Transport.py:1810-1975`	ANNOUNCE rebroadcast queue and per-interface dispatch
`RNS/Transport.py:2110-2145`	LRPROOF forwarding via `link_table`
`RNS/Transport.py:2199-2208`	PROOF receipt forwarding via `reverse_table`
`RNS/Transport.py:3457-3464`	`path_table` entry shape (IDX_PT_*)
`RNS/Transport.py:750-832`	stale-path / stale-tunnel eviction
`RNS/Transport.py:2955-2970`	`discovery_path_requests` recursive search
`RNS/Interfaces/Interface.py:232-272`	per-interface `announce_queue` and `ANNOUNCE_CAP` enforcement

13. Threading and concurrency model

The wire spec is silent on threading, but a clean-room client built single-threaded mostly works for opportunistic LXMF and starts breaking on Resource transfers and Link keepalives. This is consistently the #1 cause of "my client compiles and almost works but is flaky." Everything below is implementation-private — there's no wire requirement to use threads, only to satisfy the timing guarantees that upstream's threading provides. But the upstream Python implementation is highly concurrent; an alternative implementation that wants to interop has to provide the same guarantees, however it achieves them.

13.1 Long-running threads

Upstream RNS spawns the following persistent daemon threads at Transport.start():

Thread	Source	Cadence	Purpose
`Transport.jobloop`	`RNS/Transport.py:283, 486-489`	every `job_interval = 0.250s`	Runs `Transport.jobs()` — the catch-all maintenance pass: link state checks, announce-queue drain, stale-path eviction, hashlist cleanup, reverse-table cleanup, tunnels housekeeping.
`Transport.count_traffic_loop`	`RNS/Transport.py:284, 452-483`	every 1s	Snapshots per-interface RX/TX byte counters into rolling-window deques for bandwidth/airtime accounting.
`Link.__watchdog_job`	`RNS/Link.py:751-828`	per-link, RTT-driven	One per active Link. Drives keepalive emission (initiator side), STALE→CLOSED transitions, and link-establishment timeouts. Sleeps `min(WATCHDOG_MAX_SLEEP=5s, RTT-derived)` between iterations.
`Resource.__watchdog_job`	`RNS/Resource.py:564-670`	per-resource	One per in-progress Resource. Detects retransmit timeouts, advertisement retries, and PRF-wait timeouts.
`AnnounceHandler` callbacks	`RNS/Transport.py:1995-2025`	per inbound announce	Each accepted announce fires its registered handler on a fresh daemon thread — the dispatcher does not serialize. Two announces from the same destination back-to-back run two handler threads concurrently.
Per-interface RX threads	`RNS/Interfaces/*Interface.py`	always	Each interface (TCP, KISS, RNode, AutoInterface) has its own blocking-read RX thread that calls `Transport.inbound(raw, self)` on each complete frame.
`process_announce_queue`	`RNS/Interfaces/Interface.py:266-267`	one-shot timer per drain	Per-interface `announce_queue` drain uses `threading.Timer` to schedule the next emission at the airtime-cap-derived wait time. Not a long-running thread but a chain of one-shots.
`Resource.__advertise_job`	`RNS/Resource.py:520-542`	per-resource	One-shot daemon thread that performs the resource hashmap construction (which can take seconds on a large body) so the calling thread doesn't block.

A clean-room implementation with cooperative scheduling (e.g. asyncio, embedded RTOS task model) needs to provide equivalent behavior for each row. The key invariants — not the exact thread inventory — are what matter for interop:

The watchdog must run independently of the calling code, or links go stale silently when the application is busy.
Announce-handler callbacks must NOT block subsequent inbound packet dispatch. If your handler runs synchronously on the receive thread, a slow handler stalls every other inbound traffic.
The job loop must run regardless of inbound traffic; otherwise path_table doesn't evict stale entries, discovery_path_requests doesn't time out, and the announce_table doesn't drain its queued retransmits.

13.2 Lock inventory

Upstream uses about 30 named locks. The shared-state ones a clean-room implementation must guard equivalently (or substitute single-threaded equivalent):

Lock	Guards
`Transport.path_table_lock`	`Transport.path_table` reads and writes
`Transport.announce_table_lock`	`Transport.announce_table` reads and writes
`Transport.link_table_lock`	`Transport.link_table` (transit-relay link forwarding state)
`Transport.reverse_table_lock`	`Transport.reverse_table` (PROOF reverse-routing state)
`Transport.active_links_lock`	`Transport.active_links` list
`Transport.pending_links_lock`	`Transport.pending_links` list
`Transport.tunnels_lock`	`Transport.tunnels`
`Transport.destinations_map_lock`	`Transport.destinations_map` (local destinations registered for receive)
`Transport.announce_handler_lock`	`Transport.announce_handlers` list
`Transport.path_requests_lock`	`Transport.path_requests` rate-limiting cache
`Transport.discovery_pr_tags_lock`	`Transport.discovery_pr_tags` dedup
`Transport.jobs_lock`	held for the entire `jobs()` body — long-held, blocking
`Identity.known_destinations_lock`	`Identity.known_destinations` dict reads/writes
`Identity.ratchet_persist_lock`	ratchet persistence file I/O
`Link.watchdog_lock`	per-link gate; the watchdog `wait`s on this when the link is in the middle of a state change
`Link.receive_lock`	per-link inbound packet processing
`Resource.assembly_lock`	per-resource gate around assemble()
`Destination.ratchet_file_lock`	per-destination ratchet file I/O

Transport.jobs_lock is the most aggressive — it's held for the entire jobs() execution (which can include I/O for path persistence, announce queue draining, etc.). This is what bounds how often jobs() can run; you can't pile up parallel jobs() invocations even if job_interval elapses while one is running.

13.3 Callback-thread guarantees (and lack thereof)

What upstream guarantees to application-level callbacks:

Destination.set_packet_callback — fires once per inbound DATA, on the receive thread. Synchronous. A slow callback stalls subsequent inbound packet dispatch on the same interface.
Link.set_link_established_callback — fires once when a link transitions PENDING → ACTIVE. On the receive thread.
Link.set_link_closed_callback — fires once when a link transitions to CLOSED, regardless of cause (timeout, peer close, local teardown). On the watchdog thread or the receive thread depending on which path triggered the close.
PacketReceipt.set_delivery_callback — fires once when a PROOF arrives matching this receipt. On the receive thread.
AnnounceHandler.received_announce — fires once per accepted announce, on a fresh daemon thread. This is the only callback that's NOT on the receive thread (Transport.py:1995-2025).
Resource.callback — fires once on resource conclude, on the assembly thread.

Implications for a clean-room implementation:

Don't block on the receive thread. A set_packet_callback that does I/O or PoW work blocks every other inbound packet on the same interface until it returns. The standard pattern is: copy the data out, hand it to a worker queue, return immediately.
Announce handlers race. Two callbacks for the same destination can run concurrently; if your handler mutates shared state (a contacts list, a UI), use a lock or single-thread the writes.
Link-closed can fire from two paths. Watchdog timeout or peer LINKCLOSE both call link_closed_callback. Make the callback idempotent.

13.4 Implementation-private constants

These are not on the wire but affect timing-sensitive interop. A client that uses radically different values may diverge from upstream's behavior in subtle ways:

Constant	Default	Notes
`Transport.job_interval`	`0.250s`	Quarter-second cadence of `jobs()`.
`Transport.links_check_interval`	`1.0s`	Throttles inside `jobs()`; links are scanned at most every 1s.
`Transport.tables_cull_interval`	`5.0s`	Throttles path/reverse/link table eviction inside `jobs()`.
`Transport.hashlist_maxsize`	`1000000`	Packet-hash dedup ring; once full, half is purged on next `jobs()`.
`Link.WATCHDOG_MAX_SLEEP`	`5s`	Cap on link watchdog sleep regardless of RTT.
`Resource.WATCHDOG_MAX_SLEEP`	`1s`	Resource watchdog cadence cap.
`Resource.PROCESSING_GRACE`	`1.0s`	Grace before a resource is considered timed out.
`Resource.SENDER_GRACE_TIME`	`10.0s`	End-of-transfer grace if some parts haven't been requested.

A client running on a constrained device (less RAM, slower CPU) can scale all of these up — at the cost of slower path-table responsiveness and slightly later timeout decisions. Don't scale them down unless you've actually measured your platform; below ~100 ms job_interval upstream Python burns measurable CPU just on the bookkeeping passes.

13.5 Source map

File	What
`RNS/Transport.py:283-284`	top-level thread spawn at startup
`RNS/Transport.py:132-148`	the lock inventory (Transport-side)
`RNS/Transport.py:174, 176, 187`	`job_interval`, `links_check_interval`, `tables_last_culled`
`RNS/Transport.py:486-489`	`jobloop` — the periodic driver
`RNS/Transport.py:492+`	`jobs()` body (held under `jobs_lock`)
`RNS/Transport.py:1995-2025`	announce-handler dispatch (fresh thread per callback)
`RNS/Link.py:751-828`	per-link `__watchdog_job`
`RNS/Resource.py:564-670`	per-resource `__watchdog_job`
`RNS/Resource.py:520-542`	one-shot `__advertise_job`
`RNS/Interfaces/*Interface.py`	per-interface RX thread

14. Failure modes — symptom → root cause

§9 lists gotchas keyed by cause ("here's a thing that's true"). This section is the inverse index, keyed by symptom — what you're observing, and where to look. Each entry names the symptom, points at the section that explains why, and (where useful) names a tools/verify_*.py script that locks in the fix.

Identity / announce

Symptom	Likely cause	Fix / verifier
Generated identity files don't load on upstream `rnsd`	§1.3 — on-disk byte order is `X25519_priv \|\| Ed25519_priv` (NOT the opposite as some old docs claim)	`tools/verify_destination_hash.py` round-trips `to_file`/`from_file`
Sideband shows you as "Anonymous" or random hex instead of your display name	§9.3 — display name was msgpack-encoded as `str` instead of `bytes`. Upstream's `dn.decode("utf-8")` raises silently	`tools/verify_msgpack_quirk.py`
Announces validate locally but upstream peers reject as "Destination mismatch"	§1.2 — `name_hash` recipe wrong; the `identity=None` branch of `expand_name` does NOT include the identity hex in the hash input	`tools/verify_destination_hash.py`
Upstream announces with ratchets get rejected by my validator	§4.5 step 1 — body parser didn't branch on `context_flag` bit; ratchet-bearing announces shift `signature` 32 bytes deeper	§4.5 step 1
First contact with a peer works, but path table never refreshes from a Python source after a microReticulum announce arrives	§4.1 / §9.10 — microReticulum emits 10 fully-random bytes for `random_hash` instead of 5-random + 5-uint40-timestamp. Python receivers parse `random_hash[5:10]` as far-future and lock the path against fresher Python announces	§9.10
Periodic re-announce works locally but peers can't reach me after a few minutes	§7.5 / §9.7 — re-announce loop isn't running. Transit relays evict path entries within minutes regardless of TTL	§9.7
Announces propagate fine but my client populates its contact list with itself	§9.5 / §4.5 step 8 — self-announce echo. Filter `dest_hash == our_dest_hash` before ingesting any inbound announce	§4.5 step 8

Token crypto / opportunistic LXMF

Symptom	Likely cause	Fix / verifier
Decrypted plaintext is correct but has 16 garbage bytes appended	§9.2 — manual PKCS#7 padding on top of platform's automatic padding (Web Crypto / JCA `AES/CBC/PKCS5Padding`)	§9.2
HMAC validates but AES decrypt produces gibberish	§3.2 — HKDF salt is wrong. Salt MUST be the recipient's 16-byte `identity_hash`, not the destination hash, not the ratchet pub	`tools/verify_token_crypto.py`
Decrypt works for the first message after announce but fails for subsequent ones	§3.3 / §7.4 — recipient rotated their ratchet, you're still using the cached `ratchet_pub`. Re-fetch the latest announce or use the long-term encryption key as fallback	§3.3
Tampered packets are accepted as valid	§3.3 — verifying HMAC AFTER AES decrypt (or not at all). Encrypt-then-MAC: verify HMAC FIRST	`tools/verify_token_crypto.py`
LXMF decrypts cleanly but signature validation fails	§5.6 — try both raw `packed_payload` AND a stripped-and-re-encoded form (with the optional 5th `stamp` element removed)	§5.6
`source_hash` lookup returns nothing even though I just received an announce from that peer	§9.1 / §5.4 — `source_hash` is the SENDER's destination hash (`SHA256(name_hash \|\| identity_hash)[:16]`), NOT the raw 16-byte identity hash	§9.1

Link establishment / proof receipts

Symptom	Likely cause	Fix / verifier
LINKREQUEST goes out but no LRPROOF arrives	§6.1 — body length wrong. 64 (no signalling) or 67 (with §6.6 signalling); anything else is rejected	`tools/verify_link_handshake.py`
LRPROOF arrives but signature validation fails	§6.2 — body order wrong. Actual upstream is `signature \|\| responder_X25519_pub \|\| signalling`; the `link_id` is in the packet header, not the body	`tools/verify_link_handshake.py`
Link handshake fails specifically when MTU signalling is present on one side but not the other	§6.6.5 — signalling bytes (when present) are part of the LRPROOF `signed_data`. A mismatch means signed_data differs and signature fails	§6.6.5
Link establishes but tears down within 5 minutes of inactivity	§6.7 — KEEPALIVE not implemented. Initiator sends `0xFF` ping every `keepalive` seconds; responder replies with `0xFE` pong	§6.7.1
Sender sees DATA bursts repeatedly retransmitted, link dies	§6.5 — receiver isn't emitting the mandatory PROOF receipt for each CTX_NONE Link DATA packet	`tools/verify_proof_packet.py`
Some peers work, others reject every PROOF I send	§6.5.2 — wrong proof body length. Upstream default emits 64-byte implicit proofs (`signature` only) but your peer expects 96-byte explicit (`packet_hash \|\| signature`). Validator dispatches on length	`tools/verify_proof_packet.py`
Initiator and responder complete LRPROOF but every Resource ADV / link DATA the initiator sends is silently dropped at the responder	§6.4.2 — initiator never emitted LRRTT after LRPROOF. Responder stays in `HANDSHAKE`, `link_established` callback never fires, LXMF's `set_resource_strategy(ACCEPT_APP)` never installs, and `Link.receive` hits the silent `ACCEPT_NONE` branch on every RESOURCE_ADV	§6.4.2
Single-hop link works, but the same flow over a multi-hop link silently drops every link DATA / Resource part at the destination	§6.4.3 — link-addressed packets emitted as `HEADER_2` with `transport_id` set to the next-hop relay. link_table forwarding doesn't strip `transport_id`, so the destination's `packet_filter` rejects it as "for another transport instance" (LOG_EXTREME). Use `HEADER_1` and `transport_id=None` regardless of hop count	§6.4.3

Resource transfers (large bodies)

Symptom	Likely cause	Fix / verifier
Resource advertisement arrives, but my receiver never asks for parts	§10.5 — RESOURCE_REQ shape: `exhausted_flag(1) [\|\| last_map_hash(4)] \|\| resource_hash(32) \|\| requested_map_hashes(N×4)`	§10.5
Resource transfers but assemble fails with hash mismatch	§10.12 — encryption is applied to the WHOLE concatenated body BEFORE part splitting. Accumulate all parts, then run `link.decrypt()` once	§10.12
Resource hash collisions during construction	§10.2 step 9 — collision-guard must regenerate `random_hash` and recompute the hashmap when any 4-byte map_hash collides within `COLLISION_GUARD_SIZE` window	§10.2 step 9
`ADV` for >1MiB body never resolves	§10.11 — multi-segment cutover at `MAX_EFFICIENT_SIZE = 1 MiB - 1`. Each segment is a separate Resource; sender only sends segment N+1's ADV after PRF for segment N	§10.11

Path discovery

Symptom	Likely cause	Fix / verifier
Path? requests sent but no announce response	§7.2.1 — tagless requests are dropped. Body must be `target_dest_hash(16) [\|\| transport_id(16)] \|\| tag(≥1)`	`tools/verify_path_request.py`
Path? requests accepted by responder but I get no announce back	§7.2.6 — leaf clients only respond when `target_hash == our_destination_hash`. Don't respond for destinations you don't OWN	§7.2.6
Spurious double-announces in response to one path request	§7.2.2 — `discovery_pr_tags` dedup table missing on responder. Without it, every retransmitted path? produces another announce	§7.2.2
Sending opportunistic LXMF triggers a path? on every send, never converges	§7.1 — path? is gated by `not has_path() AND method == OPPORTUNISTIC`. If your `has_path()` always returns False, you're storming the network	§7.1

Transport / framing

Symptom	Likely cause	Fix / verifier
LoRa packets > 254 bytes drop entirely on RNode	§8.3 — RNode air-frame split protocol not implemented. Random seq nibble + FLAG_SPLIT bit; both halves share the same header byte	`tools/verify_rnode_split.py`
RNode receives correctly but TX is silent	§8.4.2 — KISS configuration handshake incomplete. CMD_RADIO_STATE = 0x01 must be the LAST step	§8.4.2
Received RSSI/SNR values are garbage	§8.4.5 — wrong sidecar decode. `RSSI = byte - 157`, `SNR = signed Q6.2 / 4`. Sidecar frames precede each `CMD_DATA` frame	§8.4.5
Multi-hop packets arrive but local-destination packets don't	§2.3 — originator HEADER_1→HEADER_2 conversion not applied for hops > 1. Originators must do this conversion themselves when path table reports `hops > 1`	`tools/verify_packet_header.py`
Sending to multi-hop peers fails silently after path table populated	§7.6 — `TCPServerInterface.OUT` is True by default in practice (constructor's `False` is overridden at runtime). Don't waste time chasing a stuck OUT flag	§7.6

LXMF specifics

Symptom	Likely cause	Fix / verifier
Messages from clockless devices appear at January 1, 1970	§9.6 — substitute. Treat any timestamp before `1577836800` (2020-01-01) as "no clock"; substitute local receive time	§9.6
Modern Sideband marks my messages as spam / drops them	§5.7 — recipient requires a stamp (announced via `stamp_cost` in app_data) and your client doesn't compute one. PoW is 3000-round HKDF over `message_id`, target_cost leading zero bits	§5.7
Display name disappears after a re-announce	§9.4 — wrong name-priority order. Use `extracted ?? existing ?? known_label ?? ""`, NOT `extracted ?? known_label ?? existing ?? ""`	§9.4
Propagation node accepts messages but my client never retrieves them	§5.8.3 — `/get` request needs the link to be `identify()`-d first; otherwise it returns `ERROR_NO_IDENTITY`	`flows/receive-propagated-lxmf.md`
Custom propagation node implementation rejects all client `/offer` requests	§5.8.5 — element [5] of the propagation announce app_data is a 3-element list `[stamp_cost, stamp_cost_flexibility, peering_cost]`, NOT a single integer	§5.8.5

Concurrency

Symptom	Likely cause	Fix
Links go stale even though my application is actively using them	§13.1 — your watchdog runs on the same thread as your application. Move it to a daemon thread	§13.1
Slow announce handler stalls subsequent inbound packets	§13.3 — packet callback runs synchronously on the receive thread. Queue and return; don't do I/O or PoW on the receive thread	§13.3
`link_closed_callback` fires twice for one link	§13.3 — callback fired from both watchdog timeout AND inbound LINKCLOSE paths. Make idempotent	§13.3
Two announces from the same destination produce duplicate UI rows	§13.3 — handler callbacks race on fresh threads. Lock or single-thread the writes to your contacts list	§13.3

When all else fails

§9.9 — add a single one-line rx <size>B H<1\|2> <PT> dest=<hex> ctx=0x<hex> hops=<n> log at the top of your Transport.inbound equivalent. The number of debugging hours this saves is hard to overstate. Symmetric tx logging on outbound is similarly cheap.

15. Time and clock requirements

Reticulum has time-sensitive behaviour scattered across many sections. This is the consolidated reference for what kind of clock you need where, what tolerance the protocol gives you, and what fails on a no-RTC device (Faketec, RAK4631 stock, Heltec_T114, generic nRF52 LoRa boards).

15.1 Three clock kinds

Kind	What it tells you	Typical embedded availability
Absolute wall time	Current Unix-seconds (e.g. `1714780800`)	Only with NTP sync, GPS, or hand-set clock
Boot-relative monotonic seconds	Seconds since the device booted	Always, via `millis()` / `time.monotonic()`
High-resolution monotonic	Sub-second timing for RTT and watchdog	Always, but precision varies

Most upstream Python code assumes wall time is available because it runs on hosts with NTP. Embedded clean-room implementations need to be careful about which kind each call site needs.

15.2 Required: monotonic seconds (every implementation)

These break a single-clock implementation if missing. All can be satisfied by boot-relative seconds — they only need order, not absolute value.

Use	Section	What fails without it
Link RTT measurement	§6.7.1	`keepalive` interval can't adapt; defaults to `KEEPALIVE_MAX = 360s` worst case
Link watchdog (last_inbound, last_outbound, last_keepalive timestamps)	§6.7	Link can't detect staleness; lingers forever or tears down spuriously
Resource transfer watchdog (last_activity, advertisement retry timing)	§10	Resource transfers stall without retry; SENDER_GRACE_TIME never triggers
`Transport.path_requests` rate-limit (`PATH_REQUEST_MI = 20s` minimum interval)	§7.1	Path? storms — repeats faster than rate limit allows
`Transport.tables_last_culled` periodic eviction trigger	§13.4	Path/reverse/link tables grow without bound
`Transport.discovery_pr_tags` aging (`PATH_REQUEST_GATE_TIMEOUT = 120s`)	§7.2.2	Path-request dedup table never evicts old entries
`Interface.ic_burst_freq` rolling deque for ingress rate limiting	§4.5 step 8	Per-interface ingress limiter can't compute Hz

15.3 Recommended: monotonic-with-no-skew across announces (timestamp encoding)

The §4.1 random_hash carries a 5-byte big-endian uint40 timestamp:

random_hash = get_random_hash()[:5] + int(time.time()).to_bytes(5, "big")

Transit relays read random_hash[5:10] as a unix-seconds value and use it for path-table replay-ordering decisions (§4.5 step 6.3). Two requirements:

Monotonic across announces from the same destination. A new announce should have a higher timestamp than older ones from the same destination, or relays will reject it as "older than what we have cached" in the equal-or-greater-hop branch.
Comparable to other peers' timestamps. If all your announces always look like "year 1970" (boot-relative seconds presented as unix), you'll consistently lose path-replay comparisons against peers with real wall time. That's actually fine — your announces just won't replace cached entries from real-time peers — but the inverse case is the §9.10 microReticulum bug: random random_hash[5:10] looks "far future" and freezes the path table.

No-RTC strategy: emit boot-relative seconds. You'll always look stale to wall-time peers (their announces win in path-replace decisions, which is correct because their data is fresher), and you'll get monotonic-from-boot ordering between your own announces (correct).

Wrong strategy: emit fully-random bytes (the §9.10 microReticulum bug). Locks you in as "latest" forever.

15.4 Recommended: wall time (LXMF-level)

These use absolute Unix-seconds. A device without wall time can substitute, with caveats:

Use	Section	Substitution if no wall time
LXMF body `timestamp` (`payload[0]`)	§5.3	Use boot-relative seconds. Recipients per §9.6 should treat any timestamp before `1577836800` (2020-01-01) as "no clock" and substitute their local receive time.
Outgoing message `LXMessage.timestamp` for sender-side ordering	§5.3	Same as above.
Stamp ticket expiry (`fields[FIELD_TICKET][0]`)	§5.7.3	You can't substitute here. Tickets you issue with boot-relative seconds will appear to have already-expired-or-already-distant-future expiries to recipients. If your device has no wall time, don't issue tickets — fall back to PoW stamps (§5.7.2).
Propagation node `timebase` field in `/offer` requests	§5.8.5	Same as random_hash strategy: boot-relative is fine; you'll appear "stale" but your peers' state stays consistent.

15.5 Optional: high-resolution monotonic for diagnostics

These are nice-to-have; missing them just degrades observability:

Per-packet RX timestamp for RTT decomposition.
Airtime accounting (sub-second precision improves ANNOUNCE_CAP enforcement; integer seconds is fine).
Resource transfer establishment_rate calculation.

Use whatever monotonic source your platform provides; even 1 ms resolution from millis() is plenty.

15.6 What fails on a no-RTC, no-NTP-sync device

A device that boots with no clock at all (time.time() returns a small integer, RTC chip absent or empty) and never syncs:

✅ Sending and receiving opportunistic LXMF works fine. The §9.6 receiver-side fix-up (substitute local receive time when timestamp < 2020) handles your "year 1970" timestamps cleanly.
✅ Receiving propagated LXMF works. The propagation node tags messages with its own timestamp; you don't need yours.
✅ Establishing Links works. RTT is measured locally and only used for relative cadences.
⚠️ Periodic re-announces work, but your random_hash[5:10] will always look stale to wall-time peers. Your announces propagate fine; they just don't win path-table replacement races against fresher peers (which is correct — they ARE fresher).
⚠️ Path-table updates from your own announces work the first time (no cached entry to compare against), but subsequent re-announces may not replace stale cache entries on transit relays. Practical effect: your destination is reachable but transit relays keep trying older paths longer than ideal.
❌ Issuing LXMF tickets doesn't work — the expiry timestamp in FIELD_TICKET is meaningless without wall time. Don't issue tickets; rely on PoW stamps.
❌ Sending propagated LXMF with ticket-based stamp shortcuts doesn't work for the same reason.

A single one-time clock sync (BLE config, web flasher, manual button-press at known time, GPS, rnstatus peer query) flips most of the ⚠️ items to ✅. The repeater repo's BLE config protocol can carry a clock value in the connection handshake; that's the simplest fix.

15.7 Source map

Section	What relies on time
§4.1	`random_hash[5:10]` emission timestamp
§4.5 step 6.3	Path-table replacement using `random_blob` timestamps
§5.3	LXMF body timestamp
§5.7.3	LXMF ticket expiry
§5.8.5	Propagation node timebase field
§6.7.1	Link KEEPALIVE / RTT cadence
§7.1	`Transport.path_requests` rate limit
§7.2	`discovery_pr_tags` aging
§7.5	Periodic re-announce cadence
§9.6	Clockless sender LXMF timestamp fix-up
§10	Resource watchdog timeouts
§13.4	All `Transport.jobs` periodic intervals

16. Bounded-state inventory (memory limits at a glance)

Embedded clean-room implementations need to know up front which data structures grow with traffic and which are bounded by protocol-level caps. This section is a single-table reference for every bounded structure across §1-§15.

16.1 Per-node state caps

Structure	Cap	Where	Notes
`Transport.path_table`	(unbounded — count grows with mesh size)	§12.4	Grows with the number of distinct destinations the node has heard about. Bounded effectively by TTL eviction (§12.4.2): AP_PATH_TIME (1h), ROAMING_PATH_TIME (4h), PATHFINDER_E (30d). On a tiny LoRa mesh this is dozens of entries; on a global Reticulum mesh routed through a TCP backbone it can be thousands.
`path_table[dest][IDX_PT_RANDBLOBS]` (per-destination random_blob history)	`Transport.MAX_RANDOM_BLOBS = 64` (RNS 1.2.4)	§4.5 step 6.3, §12.3.2	Sliding window. Caps memory growth from one destination's announce stream.
`Transport.announce_table`	(unbounded — populated only for in-flight announces awaiting rebroadcast)	§12.3	Drains via `Transport.jobs` retransmit timer, capped at `PATHFINDER_R = 4` retries each. Effective cap: number of announces seen × time.
`Transport.discovery_pr_tags`	`Transport.max_pr_tags = 32000`	§7.2.2	Path-request dedup table. Older entries aged out by `Transport.jobs`.
`Transport.path_requests`	(unbounded — one entry per recently-issued path? request)	§7.1	Aged out at `Transport.PATH_REQUEST_GATE_TIMEOUT = 120s`.
`Transport.discovery_path_requests`	(unbounded)	§7.2.3, §12.6.1	Aged out at `Transport.PATH_REQUEST_TIMEOUT = 15s`.
`Transport.link_table` (transit-relay link state)	(unbounded)	§12.2.4, §12.5	One per Link the relay is forwarding for; cleared on link teardown or stale aging.
`Transport.reverse_table`	(unbounded)	§12.5.3	One entry per in-flight DATA→PROOF round-trip; popped on use, aged at `Transport.REVERSE_TIMEOUT = 30s`.
`Transport.tunnels`	(unbounded)	§12.6.2	One per tunnel-able interface; aged at `Transport.TUNNEL_TIMEOUT`.
`Transport.packet_hashlist` (dedup ring)	`Transport.hashlist_maxsize = 1,000,000`	§13.4	Half is purged on next `Transport.jobs` after the cap is hit.
`Transport.active_links`	(unbounded — one per active Link the node owns or relays)	§6
`Transport.pending_links`	(unbounded — one per Link in PENDING/HANDSHAKE state)	§6.7	Aged out at `Link.ESTABLISHMENT_TIMEOUT_PER_HOP × hops + KEEPALIVE`.

16.2 Per-interface state caps

Structure	Cap	Where
`Interface.held_announces`	`Interface.MAX_HELD_ANNOUNCES = 256`	§4.5 step 8
`Interface.announce_queue`	`Reticulum.MAX_QUEUED_ANNOUNCES` (default ~64; configurable)	§12.3.1
`Interface.ia_freq_deque` (incoming announce rate)	`Interface.IA_FREQ_SAMPLES` rolling sliding window	§13.1
`Interface.oa_freq_deque` (outgoing announce rate)	`Interface.OA_FREQ_SAMPLES` rolling sliding window	§13.1

16.3 Per-destination state caps

Structure	Cap	Where
`Destination.ratchets`	`Destination.RATCHET_COUNT = 512`	§7.4
`Destination.path_responses`	(per-tag, aged at `Destination.PR_TAG_WINDOW = 30s`)	§7.2.4
`Destination.links` (responder-side active links)	(unbounded — one per established Link to this destination)	§6

16.4 Per-Link state caps

Structure	Cap	Where
`Link.outgoing_resources`	(unbounded — one per in-flight outgoing Resource on this link)	§10
`Link.incoming_resources`	(unbounded — one per in-flight incoming Resource on this link)	§10
`Link.pending_requests`	(unbounded — one per outstanding REQUEST on this link)	§11.5

16.5 Per-Resource state caps

Structure	Cap	Where
`Resource.window`	runtime: between `WINDOW_MIN = 2` and `window_max`	§10.10
`Resource.window_max`	one of `WINDOW_MAX_VERY_SLOW = 4`, `WINDOW_MAX_SLOW = 10`, `WINDOW_MAX_FAST = 75`	§10.10
`Resource.parts`	`Resource.total_parts = ceil(size / SDU)`	§10.2 step 7
`Resource.hashmap`	4 × `total_parts` bytes	§10.2 step 8
`Resource.req_hashlist`	(unbounded per resource — one entry per RESOURCE_REQ packet seen)	§10.6

16.6 Identity/cryptography caches

Structure	Cap	Where
`Identity.known_destinations`	(unbounded)	§4.5 step 6 — main growth vector. Persisted across restart; aged out via `Identity.clean_known_destinations` based on `Identity.RATCHET_EXPIRY = 30 days` for unused entries.
`Identity.known_ratchets`	(unbounded — one per `known_destinations` entry that has ever announced a ratchet)	§4.5 step 6, §7.4
`Transport.blackholed_identities`	(operator-controlled; empty by default)	§4.5 step 5

16.7 LXMF-level caps

Structure	Cap	Where
`LXMRouter.locally_delivered_transient_ids`	(operator-bounded)	§4.5 step 6 / §5.7 dedup
`LXMRouter.outbound_stamp_costs`	(per peer — grows with peer count)	§5.7.4
`LXMRouter.available_tickets`	(per peer / direction)	§5.7.3
`LXMRouter.propagation_entries`	(operator-bounded — propagation node only)	§5.8
`LXMRouter.peers`	(operator-bounded — propagation node only)	§5.8

16.8 Channel state caps

Structure	Cap	Where
`Channel.window`	`Channel.WINDOW = 2` initial, growth like §10.10	§6.8.4
`Channel.message_factories`	(per-Link, application-defined)	§6.8.3
`Channel.outbound_queue` / inbound	(unbounded — one entry per in-flight message)	§6.8.4

16.9 What this means for embedded targets

A typical nRF52 / RAK4631 / Heltec_T114 client carrying ~64KB of usable RAM should:

Not run as a transport node (skips most of §16.1's largest structures: link_table, reverse_table, tunnels, large path_table). Leaf clients only populate path_table for destinations they personally need, dramatically smaller.
Cap Identity.known_destinations at a sane size (e.g. 50-200 entries) and drop older ones when full. Upstream's unbounded growth is fine on a desktop; embedded clients need explicit eviction. Loss of an entry just means re-discovering via §7.1 path? on next outbound to that destination.
Bound Resource.hashmap size — a 1 MiB resource has 1024 parts at SDU=1024, so a 4 KiB hashmap. Reject incoming Resources whose advertised n would exceed your memory budget; the receiver's delivery_resource_advertised callback can return False to reject (§5.8.3 / NomadNet pattern).
Stick to WINDOW_MAX_SLOW = 10 rather than WINDOW_MAX_FAST = 75 for any Resource transfer to bound part-buffer memory.
Avoid registering Channel message types with large pack() outputs.

For comparison: a desktop rnsd typically settles around 50-200 MB of memory in steady state on a moderately-busy mesh, dominated by path_table and known_destinations growth.

17. Implementation taxonomy: who needs which sections

Reticulum applications fall into three categories. Most of this spec only matters if you're in category 3. Categories 1 and 2 inherit upstream Python RNS's protocol implementation and pick up most of the wire-level correctness for free.

This section exists to save category-1/2 readers from over-engineering, and to flag for category-3 readers exactly which spec sections are theirs to implement vs. theirs to verify against.

17.1 The three categories

Category	Description	Examples
1: Upstream-RNS-based	Python application that does `import RNS` and uses upstream's `Reticulum` / `Transport` / `Identity` / `Destination` / `Packet` / `Link` directly. Inherits all wire-level behavior from upstream.	Sideband (Mark Qvist's flagship), NomadNet, `liamcottle/reticulum-meshchat`, `rncp`, `rnsh`, `rnstatus`, anything in `pip show rns` example code
2: Wrappers / language bindings	Non-Python application whose Reticulum protocol layer is a wrapper around upstream Python RNS — typically via FFI, subprocess, or a network bridge to a co-resident `rnsd`. Inherits wire correctness from the wrapped layer.	Future native iOS / Android / desktop apps that embed CPython, `rnsh` clients in shell scripts, anything that uses `rnsd`'s socket interface
3: Clean-room implementations	Application that re-implements the Reticulum protocol layer in another language without calling into upstream. Bears full responsibility for wire-level correctness.	`attermann/microReticulum` (C++), `thatSFguy/reticulum-lora-repeater` (C++ via microReticulum), reticulum-mobile-app (Kotlin), reticulum-lora-webclient (JavaScript), any from-scratch implementation in Rust / Go / Swift / etc.

17.2 Section relevance by category

Three classes of section in this spec:

Class	What it tells you	Cat 1	Cat 2	Cat 3
Wire format (§1-§8, §10, §11)	What bytes appear on the wire and in what order	Reference only — you emit and parse correctly because upstream does	Reference only — your wrapper does the work	You implement these. Bug here → can't talk to anyone
Implementation gotchas (§9)	Things upstream does that surprise you when reading the manual	Yes — because the gotchas often manifest as application-layer behaviour you need to explain to users	Yes — same reason	Yes — and you need to reproduce the gotchas faithfully or peers reject your traffic
Behavioural guidance (§7, §12, §13, §14, §15, §16)	Threading, timing, transport-relay, memory caps, failure debug	Mostly informational — upstream handles it	Mostly informational	Critical — you implement everything here from scratch
Test vectors / verifiers (§18, `tools/`)	Round-trip-able byte sequences	Reference for understanding	Reference for understanding	Required — these are your regression suite

17.3 Worked example: §2.3 originator HEADER_1→HEADER_2 conversion

A reader hitting §2.3 might wonder "do I need this?" Three different answers:

Cat 1 (e.g. MeshChat, Sideband): No — RNS.Transport.outbound at lines 1074-1083 does the conversion automatically when you call Packet.send() to a destination with path_table[dest][HOPS] > 1. Your app just calls LXMessage.send() or Packet.send() and §2.3 happens invisibly. You can read §2.3 to understand WHY some captures show HEADER_2 with a transport_id, but you have no code to write.
Cat 2 (wrappers): Same as Cat 1 — the wrapped Python RNS does the conversion. Your wrapper is just relaying API calls.
Cat 3 (clean-room): Yes, you implement §2.3 yourself. Failure to do so means your packets aren't forwarded by transit relays — they're processed and dropped silently per Transport.py:1500 (only HEADER_2 packets with transport_id == relay.identity.hash enter the forwarding branch). The symptom is "messages I send through a relay never arrive, but direct-link messages do." Sideband works in shared-instance and direct-TCP modes both because upstream does the conversion; a clean-room app working only via shared-instance is masking the missing §2.3.

17.4 Pragmatic implication

If you're a category-1 or category-2 developer reading this spec for operational understanding — debugging an interop issue, writing a deployment guide, explaining behaviour to users — read §1-§9 and §13-§16; skip the implementation depth.

If you're a category-3 developer building from scratch, you need everything. Verify each section as you go using tools/verify_*.py, and treat §14 (failure modes) as your fault-finding entry point when integration testing reveals a discrepancy.

If you're not sure which category you're in: grep -r "import RNS" your_codebase is a quick check. Any hit means cat 1 (or cat 2 if it's behind an FFI wall). No hits means cat 3.

17.5 Application protocols layered over Reticulum

Reticulum is a transport substrate; user-facing features are application protocols built on top. This spec documents LXMF (§5) in depth because its wire format is published nowhere else. Other application protocols carry their own authoritative specs and are out of scope here — but a clean-room author building a client for one still needs the RNS-layer sections below.

Reticulum Relay Chat (RRC) — a live, IRC-style chat protocol. Authoritative spec: https://rrc.kc1awv.net/ (its CBOR envelope, message types, and state machines live there, not here). RRC is a category-3-style consumer of RNS: it defines its own wire format but relies entirely on Reticulum for transport. An RRC client built clean-room needs:

RRC depends on	Sections in this spec
Hub destination (`rrc.hub` aspect) + client identity	§1.1, §1.2, §9.8
Identity hash as the canonical sender id (RRC envelope key 4, "opaque, do not re-encode")	§1.1; §9.1 — the identity-hash-vs-destination-hash pitfall RRC's rule guards against
All traffic over a single Reticulum Link	§6.1–§6.4, §6.7
Single-packet CBOR frames sized to the link MTU	§2.1–§2.2, §2.4, §6.6 (MTU is negotiated — see note below)
Ordered/reliable delivery, if RRC layers over Channel	§6.5, §6.8

RRC does not use opportunistic packets, Resource transfer (§10), REQUEST/RESPONSE (§11), or LXMF (§5) — an RRC-only client can skip those entirely.

Two things RRC's spec leaves implicit that this spec makes explicit: link MTU is negotiated (§6.6), not a fixed 500 bytes, so frames sized to the default can overflow a smaller link; and a bare Link does not itself guarantee ordered/reliable delivery — that comes from Channel (§6.8) or packet receipts (§6.5).

18. Test vectors

See test-vectors/. Currently populated:

identities.json — Alice and Bob private-key inputs plus their derived public_key, identity_hash, and lxmf.delivery destination_hash. Verified by tools/verify_destination_hash.py; regenerated by tools/regen_identities.py. Covers SPEC.md §1.1 and §1.2.

⚠️ UNVERIFIED: The remaining vector categories — signed announce packets, encrypted opportunistic LXMF DATA, and Link handshake (LINKREQUEST + LRPROOF + derived session keys) — are not yet populated. See agent.md §5 and todo.md for the remaining bootstrap work.

An implementation that round-trips every test vector — both directions — should be wire-compatible with upstream Reticulum and LXMF for the covered operations.

19. Source map

Upstream Python sources, in rough order of frequency-of-reference:

File	What lives here
`RNS/Identity.py`	Key generation, `to_file`/`from_file`, `validate_announce`, `recall`
`RNS/Destination.py`	`expand_name`, `name_hash`, destination hash construction
`RNS/Packet.py`	Header pack/unpack, packet types, contexts, `prove`
`RNS/Transport.py`	`outbound`, `inbound`, `request_path`, path table, HEADER_1↔2
`RNS/Link.py`	Link establishment, LRPROOF, session-key derivation
`RNS/Cryptography/Token.py`	The Fernet-style Token format
`RNS/vendor/umsgpack.py`	The bundled msgpack with locked bin/str semantics
`RNS/Interfaces/TCPInterface.py`	TCPClient/TCPServer, including HDLC framing
`LXMF/LXMessage.py`	LXMF body pack/unpack, opportunistic vs link methods
`LXMF/LXMF.py`	`display_name_from_app_data`, `stamp_cost_from_app_data`, etc.
`LXMF/LXMRouter.py`	Delivery destination registration, announce-app-data assembly

When upstream code changes such that this document drifts, please open a PR.

274 KiB Raw Blame History Unescape Escape

Reticulum Wire Specifications

1. Identity and destination hashes

1.1 Identity composition

1.2 Destination hash

1.3 Private key on-disk format

File-system facts

Constructing from raw bytes — from_bytes HAZARD

Cross-implementation portability

1.4 GROUP destinations (symmetric-key alternative to SINGLE)

1.4.1 Key generation

1.4.2 Wire format

1.4.3 Destination hash for GROUP

1.4.4 On-disk format

1.4.5 Why most clients don't bother

2. Packet header

2.1 Flag byte layout

2.2 Two header forms

2.3 Originator HEADER_1 → HEADER_2 conversion

2.4 Hop count

2.5 Context byte

2.6 Source

3. Token cryptography (modified Fernet)

3.1 Wire format

3.2 Encrypt steps (opportunistic)

3.3 Decrypt steps

3.4 Source

4. Announce wire format

4.1 Packet body

4.2 Signed data

4.3 app_data format for LXMF delivery destinations

4.4 Announce filtering by name_hash

4.5 Announce validation rules (receive side)

1. Body parse — branch on context_flag

2. Signature verification

3. destination_hash recomputation

4. Public-key collision rejection

5. Blackhole list check

6. Caching the announce contents

7. PATH_RESPONSE distinction

8. Implementation-private behavior (SHOULD)

9. Source map for §4.5

4.6 rrc.hub announce app_data (Reticulum Relay Chat)

5. LXMF wire format

5.1 Opportunistic delivery (single Reticulum DATA packet)

5.2 Direct delivery (over an established Reticulum Link)

5.3 msgpack_payload

5.4 Source/destination semantics

5.5 Signed data

5.6 Signature verification — msgpack variant tolerance

5.6.1 Canonical encoder for senders

5.7 LXMF stamps and tickets (anti-spam)

5.7.1 Stamp wire format

5.7.2 Stamp generation (proof-of-work)

5.7.3 Tickets — pre-shared shortcut around proof-of-work

5.7.4 The full stamp-cost field inventory

5.7.5 Source map

5.8 Propagation node protocol (offline message store-and-forward)

5.8.1 The lxmf.propagation destination

5.8.2 Peer-to-peer sync via /offer

5.8.3 Client retrieval via /get

5.8.4 Peering keys (PoW for peer-to-peer auth)

5.8.5 Propagation node announce app_data

5.8.6 Source map

5.9 LXMF field constants and helper specifiers

5.9.1 Top-level fields dict keys

5.9.2 FIELD_IMAGE (0x06) value shape

5.9.3 FIELD_AUDIO (0x07) value shape

5.9.4 FIELD_RENDERER (0x0F) value shape

5.9.5 Propagation-node metadata keys

5.9.6 Functionality signalling keys

5.9.7 FIELD_FILE_ATTACHMENTS (0x05) value shape

5.9.8 fields[16] (0x10): tap-back reactions (app extension)

5.9.9 fields[0x30] + optional fields[0x31]: reply-to threading (app extension)

5.10 Source

6. Reticulum Link protocol

6.1 LINKREQUEST (initiator → responder)

6.2 LRPROOF (responder → initiator)

6.3 link_id derivation

6.4 Session keys and link activation

274 KiB

Raw Blame History

Constructing from raw bytes — `from_bytes` HAZARD

4.3 `app_data` format for LXMF delivery destinations

4.4 Announce filtering by `name_hash`

1. Body parse — branch on `context_flag`

3. `destination_hash` recomputation

7. `PATH_RESPONSE` distinction

4.6 `rrc.hub` announce app_data (Reticulum Relay Chat)

5.3 `msgpack_payload`

5.8.1 The `lxmf.propagation` destination

5.8.2 Peer-to-peer sync via `/offer`

5.8.3 Client retrieval via `/get`

5.9.1 Top-level `fields` dict keys

5.9.2 `FIELD_IMAGE` (`0x06`) value shape

5.9.3 `FIELD_AUDIO` (`0x07`) value shape

5.9.4 `FIELD_RENDERER` (`0x0F`) value shape

5.9.7 `FIELD_FILE_ATTACHMENTS` (`0x05`) value shape

5.9.8 `fields[16]` (`0x10`): tap-back reactions (app extension)

5.9.9 `fields[0x30]` + optional `fields[0x31]`: reply-to threading (app extension)

6.5 Packet receipts (regular `PROOF` packets)

6.7.1 KEEPALIVE (`context = 0xFA`)

6.7.3 LINKCLOSE (`context = 0xFC`)

6.7.6 LINKIDENTIFY (`context = 0xFB`)

6.8 Channel mode (`CHANNEL = 0x0E`)

7.1 Path requests: peers send `path?` before opportunistic LXMF when no path is known

7.2.3 The five-way dispatch in `Transport.path_request`

7.2.5 Timing: `PATH_REQUEST_GRACE` and roaming

7.6 `TCPServerInterface.OUT` is True by default in practice

8.2 HDLC (TCP / `rnsd TCPServerInterface`)

Transmit (`RNode_Firmware.ino:716-742`)

Receive / reassembly (`RNode_Firmware.ino:359-446`)

8.4.3 The `CMD_DETECT` exchange

8.5.1 Airtime caps (`CMD_ST_ALOCK` / `CMD_LT_ALOCK`)