[karczuRF]

The closing PR of task 0027 (decision 68). It fills in what's left for criteria 12, 16 and 17. 0027 stays active for criterion 1, which needs a real testnet settlement (none since the alarms went on). A small follow-up will tick it and archive the task.

AC 1 is still open

AC 1 needs pool events, per-route counts and settle outcomes as live metrics. Per-route counts are live (http_requests_total for /health, /supported, /discovery). But there has been no /settle through the deployed facilitator since the alarms went on, so settlement_outcomes_total and the real pool_events_total types have no series yet. The only pool_events_total point is the drill's fake sequence-resync. Next: one real testnet payment through x402-facilitator.sorobanscan.com, then a small commit that ticks AC 1 and moves 0027 to archive.

docs/monitoring.md

• Every metric: the catalog's 61 metrics and the 5 AWS metrics the alarms read, each with its namespace, kind, unit, dimensions as CloudWatch sees them, and the alarms that read it.
• The alarms that are off: channel-held, wal-failing, wal-stale, backup-age, snapshot-age, account-state, account-effects and backup-secret-read, each with the task that turns it on (0089, 0026 or 0127).
• The uptime figure: 99.56% from 2026-11-06 11:16 to 2026-11-09 15:33 UTC (4,557 of 4,577 minutes up). All 20 down minutes are the drill's planned stop, so it's 100% without it. The command to read it again is included.
• The disk drill: after the 15:11:44 deploy (fire-test at 15:13:28), fake points at 85% fired disk-root at 15:26:16 and it went back to OK at 15:36:16. audit-disk (90%) stayed OK. memory was not fired: a Minimum alarm can't be faked while the agent writes real samples.
• The rows for disk, memory and emf-validation are updated.

A test that keeps the page complete

infra/test/alarms.test.ts fails when the page lacks a catalog metric, or an alarm that has neither a section (on) nor a row (off). Mutation-checked: removing fee_charged_stroops, pool_queued or wal-stale from the page fails it.

0027

• Criterion 12 ticked (disk and audit-disk fired, memory with its exception).
• Criterion 16 ticked: 37 of the 44 alarm names that are on fired. The 7 that didn't each have their reason, and the 8 that are off are fired by the task that turns each on.
• Criterion 17 ticked.
• New sections: Implementation Notes, Issues Encountered (the exit-130 stop, the catalog's semantic merge, Minimum alarms, fake points in the history, no settlement yet) and Future Work.
• A history entry.