[MendeMatthias]

The gap. Every health signal easyNode has is about blocks: synced, peers, bytes served, archive census, stall verdict. A node can pass all of them and never receive a transaction, and nobody would know.

What the operator sees now. "Your node's role" ends with a Transactions line, same shape as the lines above it (mark, value, one plain note), sentence rendered in Rust:

verdict	value	note (example)
receiving	Exchanging transactions	14 of 44 peers have sent this node transactions, so payments reach it and it passes them on.
unknown	Not known yet	Connected for 5 hours. Transactions are rare on BTX, so hearing none yet proves nothing. This answers after a day. (or: only N peers / still catching up)
isolated, no open peer	Not receiving any	None of the 8 peers takes transactions: they pass only blocks, or are still catching up themselves. In 30 hours no transaction has reached this node. Restart the node to find other peers.
isolated, open peers	Not receiving any	5 of 8 peers take transactions, but in 30 hours not one has sent this node any. Either nobody is sending, or those peers do not pass them on. Restart the node to find other peers, and if it still says this a day later, ask in the BTX Community group.

unknown reads grey (helps = null), never green.

How it works. summarize_tx_relay reads the per-peer message byte counters (bytesrecv_per_msg.tx) from the getpeerinfo the refresher already runs each tick, so it costs no extra RPC. judge_tx_relay(summary, initial_block_download) turns it into receiving, isolated or unknown; the refresher caches it (AppState::tx_relay_cache, cleared on stop/start like the archive census), get_node_status returns it as tx_relay (TS type TxRelayHealth in main.ts) and appends tx_relay_line to role_lines. The local service report carries it as tx_relay, schema 3 (main was still at 2, so 3 was free).

PeerInfo gained conntime, relaytxes and minfeefilter. A peer object without conntime is unknown rather than connected since 1970.

Re-measured today, and two things changed from September. 2026-10-29 10:46 and 10:58 UTC, the two btxd v0.34.15 nodes behind btxscan.io (btx-esplora-1, /data/btx and /data/btx2), read-only getpeerinfo, getnetworkinfo, getmempoolinfo only:

	node 1	node 2
peers	44 (27 in / 17 out)	31 (16 / 15)
peers that had delivered a transaction	14	16
peers that want transactions (relaytxes)	33	23
peers at the 0.09170997 "fee"	2 (0.32.3, 0.32.12)	3 (0.31.0, 0.33.2, 0.34.10)
minfeefilter values seen	0, 0.00001, 0.09170997	same
mempool	0 tx, mempoolminfee 0.00001	0 tx, 0.00001

Open peers (want transactions at the relay floor) by how long connected, open / delivered:

connected	node 1	node 2
under 6 h	17 / 0	4 / 2
6 to 24 h	4 / 2	7 / 5
a day or more	13 / 12	9 / 9

1. The window is now a day, not six hours. Relay works today, unlike September, but traffic is thin: 17 open peers under six hours old on a node that relays fine had delivered nothing between them, and two peers 6.5 h and 14.8 h old had each delivered the same 65.6 KB. Six hours would have called a healthy node isolated. After a day, 21 of 22 open peers had delivered.
2. The 9,171 sat/byte peers are not pricing anything. 0.09170997 BTX/kvB is exactly MAX_MONEY rounded to the top fee-filter bucket, which is what a node sends while it is in initial block download. Every peer showing it ran an old engine. The wall threshold (0.01 BTX/kvB) still separates the three values peers actually send by orders of magnitude, so it stays; the wording now says "still catching up" instead of "asks a fee nobody pays", and a new open_to_tx count (wants transactions and not walled) picks the reason sentence. For the same reason the judge now answers unknown while this node itself is in initial block download, since it asks its peers for nothing then.

Measurements and dates are in comments at TX_WALL_BTX_PER_KVB and TX_RELAY_WINDOW_SECS in crates/btx-core/src/node_api.rs.

EasyBTX site: nothing reads service-report.json there (searched EasyBTX/site); the report is a local file and checkin.rs has its own schema. No site change needed.

Tests. 12 tx-relay tests in node_api.rs (the September peer shapes, plus: six quiet hours are not isolation, a node in initial download is not judged, the four line wordings, the wall threshold against the values peers send). Ran locally what CI runs, on this branch merged with main: web tsc --noEmit, npm test (280 passed), vite build; btx-core cargo check --locked --all-targets, cargo test --locked (1228 passed, 2 ignored, before the last main merge; after it, node::tests::an_adopted_node_is_found_by_name_and_lifted_before_its_stop hung past 60 s on this Mac, the known local-node case, and platform::unix::tests::process_cpu_time_reads_a_live_process failed once under load and passes alone; neither is touched here, CI is the judge); tauri shell cargo check --locked --all-targets and cargo test --locked (297 passed, 1 ignored).

Not in this change: no attempt to fix relay. This makes the failure visible, which is the precondition for that.

🤖 Generated with Claude Code