Retro report — window 2026-10-25T23:38:20Z → 2026-10-28T02:24:42Z

Window covered

• Digest source: boucle/metrics, fetched: yes.
• Metrics rows in window: 0. Metrics rows on branch: 0. Last metrics row: never.
• Issue rows considered: 0 (bounded to 40).
• Closed issues in the window: 1 — "Landing page boucle.dev — afro-futuriste, Boucle d'or, 'construire un produit est un jeu d'enfant' #1 Second consecutive metrics blackout: retro blind, collector never produced a row".
• metrics_coverage: 0/1. The digest carries the verbatim warning: "WARNING: metrics blackout — 0 metrics rows for 1 closed issue(s) in window; the retro evidence layer is missing (collector unwired, or its rows dropped before the window query)."

Counts

• Per-issue rows: none (metrics blackout — the empty aggregates are absence of evidence, not evidence of no activity).
• Outcomes: unknown (no rows). Verdicts: none. Escalations: none. Tokens/cost: null (no run reported a number).
• Upstream posts this window: 1 enrichment (#170), 1 new proposal (see below).

Findings

1. Third consecutive blackout, and the collector has never produced a row. The window's only closed issue is itself a blackout report, and the digest states last metrics row: never. This is no longer a transient outage: the publish path has never delivered a row on this consumer. The actionable question is no longer "why did the collector stop" but "why has it never started" — wiring, schedule, or permissions. Evidence for upstream #170 (collector liveness / persist the publish outcome): a consumer whose collector has never emitted a single row still shows a healthy-looking loop (issues close, retros run), so the failure is invisible to every signal except the retro's own coverage line. Enriched #170 with this case.
2. The blackout is now self-referential: the loop's only closed issue is the blackout. The retro's coverage denominator (1) is the retro's own bookkeeping issue. Even a perfectly wired collector could not have produced a row for it (no agent work ran), so the digest's 0/1 coverage line can never reach 1/1 while blackout reports are the only traffic. This is a digest defect, not a collector defect: the denominator should count issues that ran agent work, or the digest should annotate bookkeeping-only issues. Filed as a new upstream proposal (Size: S).
3. Lesson titles confirm the wiring theme is known and still open. Recent lessons Quiet retro window 2026-10-17 20:48-21:08Z: no activity, no findings #117 ("A documented switch nobody wired is worse than no switch"), Second consecutive metrics blackout: collector never produced a row (up-to-date path never converges it) #118 ("A GitHub cron asks; GitHub decides"), Retro window 2026-10-19/21: metrics blackout (collector never published), loop liveness unverifiable #120 ("Subscribe to the events you route, not the events that exist") and chore(deps): bump astro from 7.2.1 to 7.3.3 #123 ("A measurement that omits what it could not measure reads as a measurement of zero") all describe exactly this failure family. The class is documented; the instance persists. No new lesson is warranted — the admission test fails (already covered, and the fix is wiring, not documentation).

What this window could NOT see

With zero metrics rows: outcomes (terminal done vs human), iterations taken, verdicts, escalations, tokens, cost, and need-vs-delivered for every issue that closed in the window. Per the digest legend, the absence is reported as a gap, never read as a zero.

Anonymisation note

All consumer identifiers in the digest were already replaced by retro_scrub ([consumer], [host], [user], [email], boucle/N-). This report quotes only the digest and generic shapes ("a consumer repo"); no project name, host, username, path, e-mail or IID appears. Upstream posts were made through bin/retro-post, which scrubs before posting.

Upstream posts made

• Enriched #170 (metrics collector liveness) with the never-emitted-row consumer case.
• Created a new proposal: metrics_coverage denominator counts retro bookkeeping issues (Size: S).