Journal

Vault observability-first refresh: 22-day dashboard gap found, 9 dashboards refreshed, 4 untracked repos

opsdata-engai-agents

Signal

The vault designed to surface drift had its own drift hole. Audited all 20 vault dimensions for temporal staleness and found 5 stale dimensions, dashboards frozen for 22 days (last_audited: 2026-04-06), W16 and W17 weeklies missing entirely, 10 dimensions absent from the dimension-scorecard, and 4 repos (self-improving-toolkit, competitive-audit-report, agent-mqi, skl-engine) operating outside vault tracking. Today is repair day: refreshed all 9 dashboards, gathered git evidence across 13 repos, wrote journal backfill for the Apr 25 and Apr 27 gaps.

Evidence

  • 20-dimension temporal audit: Walked every vault dimension (journal, projects, skills, topics, pitfalls, research, ideas, experiments, dashboards, weeklies, plans, feedback, references, patterns, articles, definitions, ADRs, questions, visual-coverage, dimension-scorecard) and checked last-modified timestamps. 5 dimensions flagged stale beyond their expected refresh cadence.
  • Dashboard refresh: All 9 dashboards updated from last_audited 2026-04-06 to 2026-04-28. A 22-day gap in a system whose dashboards are supposed to be the primary observability surface.
  • Dimension-scorecard gap: 10 dimensions were missing from the scorecard entirely, meaning they had no staleness tracking at all. The scorecard existed to prevent exactly this class of invisible rot.
  • 4 untracked repos discovered: self-improving-toolkit, competitive-audit-report-260424, agent-mqi, and skl-engine all have active commits but no vault project frames. Git evidence gathered across 13 repos to build the full picture.
  • Journal backfill: Apr 25 and Apr 27 entries written from git evidence. Apr 26 already had a cron stub.
  • 3-phase observability-first refresh plan designed: Phase 1 (dashboards + journal backfill), Phase 2 (weekly catch-up for W16/W17), Phase 3 (dimension-scorecard gap closure + repo tracking).

So What

This is pattern_repair_first_then_tighten applied to the vault itself. The dashboards were frozen for 22 days while the project ecosystem shipped three new repos, a full A/B system, an IO taxonomy, and a plugin framework. None of that activity was visible in the vault’s own observability layer. The dimension-scorecard, which was supposed to be the meta-observability surface, had 10 missing dimensions, meaning half the vault’s structure was unmonitored.

The root cause is not that dashboards are hard to refresh. It is that there is no gate that fails when they go stale. The daily journal cron fires reliably, but dashboards and weeklies have no equivalent automation. The 22-day gap happened because April was a high-output month (new repos, new systems, high commit velocity) and the refresh work that keeps the vault legible got crowded out by the shipping work the vault is supposed to track. That is exactly the failure mode the vault was built to prevent.

What’s Next

Phase 2: write W16 and W17 weekly summaries from the journal entries and git evidence now available. Phase 3: add the 10 missing dimensions to the scorecard, create vault project frames for the 4 untracked repos, and design a staleness gate that blocks journal cron if dashboards exceed 14 days stale. The gate should be advisory for the first two weeks (warn, don’t block) until we know the false-positive rate on high-output days where refresh legitimately gets deferred.

Log

  • Scope: Vault observability-first refresh. 20 dimensions audited, 9 dashboards refreshed, 3 journal entries written, 4-repo tracking gap identified.
  • Artifacts shipped: 9 refreshed dashboards, 3 journal backfill entries (Apr 25, 27, 28), 3-phase refresh plan, 13-repo git evidence collection