handle_contact_message's copy of the payload/metadata/signal-cache lookup becomes a _message_signal call with its own SNR keys (it also accepts signal_to_noise and signal_noise_ratio), so DMs and channel messages share one implementation. The helper loses a comment that only made sense inline.
A DM replay (1,500 corpus packets as RF rows, 1,500 DMs with varied SNR/RSSI key layouts, metadata, cached signal values keyed by pubkey prefix, with and without radio contacts) gives byte-identical processed messages and logger calls; passing the channel keys instead changes them. The channel replay is unchanged.
handle_channel_message stays on MessageHandler (the plan keeps the handle_* entry points there) and its blocks move to modules/channel_message.py, unchanged apart from returning what they used to assign: _split_channel_sender, _message_signal (SNR/RSSI from payload, metadata or the signal cache; the DM handler can share it), _channel_route_from_rf (signal values, decoded packet, hops and path from the correlated RF row, still refusing a fallback row's route, #80), _decode_scope_packet, _channel_reply_scope, _channel_scope_allowed (the flood_scopes allowlist, returning False where it returned) and _full_sender_pubkey. Two copies of a stale comment are gone.
CHANNEL_SENDER_FALLBACK and _signal_value move with them, since channel_message can't import from message_handler; message_handler imports both back, so existing imports keep working.
Checked with a new replay: the 4,009-packet corpus plus TRACE and truncated packets as RF log rows, interleaved with channel messages (120 decrypted with the public channel key, the rest synthetic, some arriving with an empty RF cache) in four configurations (no allowlist, flood_scopes keys, '*' only, keys plus '*' with old cached messages). Processed messages, flood-scope observations, every logger, mesh graph and web viewer call and the caches are byte-identical, as is the RF log replay. Mutations in six of the helpers each change the output; replacing the raw-hex path guess with None and dropping int() from cached RSSI don't, because the replay's undecodable packets never yield a guessed path and its cached RSSI values are already ints.
agessaman