xln

Current public testnet update and web rollback

Checked and deployed source: efb6f5d4181114457cf3d62077c5d624b9fef9b3. Previous live source: 33260898b968f1b6c70ea0b702048f1997598466. The public RPC reports Anvil chain31337. This is testnet validation, not mainnet activation or a public Sepolia financial proof. This commit records evidence; it does not change production code, financial semantics, formats or contracts.

Update and current-format recovery

The canonical deploy-platform.sh --remote xln.finance --frontend-only --production built and installed the published artifact in28.894s. Its existing remote-main guard required a clean main and an exact published fast-forward. No reset, nginx change, credential change or service restart occurred in that command. The previous compiled web roots were preserved before installation, together with23 Account views and3 historical Runtime frame hashes/state hashes (7.900s).

A separate exact-owner PM2 restart used the existing parent-only shutdown policy, --no-treekill --kill-timeout 60000, without --update-env. It completed36.624s; managed recovery took31.824s. The server PID changed1014368→1045083. Anvil, watchtower, the separate Sepolia watchtower and logrotate owners were unchanged. Protected file hashes/modes were unchanged. All7 recorded crash/quiesce/shutdown failure counters were zero. Existing current JDB/RDB stores were preserved, not regenerated.

The initial equality assertion on entire latest Account views failed1.414s. The first mismatch was height7→8 with the same financial root and every balance/limit. The existing historical comparison then passed for all3 hubs in10.628s: all23 historical Account states, all3 historical Runtime frames, and every current financial root/balance/limit matched exactly. Each hub’s one advanced Account linked the preserved frame hash, contained only set_credit_limit, and was timestamped after restart. This explains the new credit-policy frames without weakening historical recovery assertions. Both the failed command and complete comparison evidence remain in the archive. Account views are not unique economic operations.

Exact delivered artifacts and reversible web installation

Initial public HTTPS verification passed8/8 in25.734s with a fresh embedded wallet, real Account workers, relay traffic, deep-route reload/unlock, compiler preview with financial admission unavailable, and graph observer. Console errors/warnings, page errors, HTTP failures and request failures were all zero.

The existing web-artifacts.sh installer then restored the preserved previous web archive (16.211s,1338 exact files). The same real browser verification passed8/8 in 25.762s. Finally, the saved current archive was reinstalled (8.718s,1385 exact files), and browser verification passed8/8 in29.810s. Both phases checked actual public /ui/ HTML and its main JavaScript digest against their selected archive. Runtime and Account-worker bytes/MIME were checked against the current canonical local build. All browser issue counters remained zero. Every web-install phase preserved process identities and protected file hashes/modes; no backend restart or data reset occurred.

Preserved archives are outside source on the host at /root/xln-deployment-evidence/efb6f5d41/current-update/:

Final exact-source operator health passed1.196s:3 hubs, self-relay,1 tower,0 open incidents. The public backend and current web artifact remain on the checked source. The financial engine, storage, contract, dependency and startup implementations were unchanged between these two sources; the core production difference is an export of an existing public stack-identity parser. This specific compatibility does not imply that an arbitrary prior backend can read a later checkpoint/WAL.

Limits and reproduction

This proves a web artifact rollback against the current backend, not backend binary rollback. It does not replace loaded in-flight crash tests, ordered-outbox replay, uninterrupted1hsoak, full integrated release/browser gates or public Sepolia proofs. The public signed-broadcast403 policy decision remains open. Existing MacBook access still failed at21:02UTC (2.095s); no service/access settings were changed.

results.json records exact commands, exits and durations. proof.json contains safe public roots, comparisons, artifact digests and browser results. command-evidence.tar.gz retains bounded procedures and actual outputs, including failures; sha256.json identifies these artifacts. Private environments, tokens, keys and databases are excluded. Procedures contain exact host/source/PID guards; inspect current owners before adapting them, and never blindly reuse an old PID.