Grand-dashboard checkpoint: memory lifecycle v2, retrieval quality v2, token wiring and local-model eval gates

[seathatflowsinourveins]

Scope

• What this PR changes: observability/grand-dashboard/state.json. It adds four dated foundation gates for work merged today and leaves every existing entry unchanged. The gates are inserted after the existing memory-lifecycle gate, and recorded_at_utc is refreshed.

  Gate id	State	Evidence
  memory-lifecycle-v2	89 of 89 checks pass on each build (ai-memory 2.3.2 and 2.4.0)	Memory lifecycle v2: ai-memory 2.3.2 and 2.4.0 pass 89/89 checks each (real-time TTL, 4-project scope, restart durability) #292
  retrieval-quality-v2	QMD hybrid selected over QMD lexical search, nDCG@10 0.714 vs 0.223	Retrieval quality v2: preregistered held-out benchmark; QMD hybrid beats QMD lexical search (nDCG@10 0.71 vs 0.22) #293
  token-efficiency-wiring	coverage complete from a fresh worktree; Codex MCP servers at user scope	Codex MCP servers at user scope so every worktree inherits the token tools; --pinned-versions coverage check #289
  local-model-eval-e2e	promptfoo and Inspect accepted on WSL2 against the local model	Grand-catalog evidence bank on the workstation: 7 WSL winner rows to accepted, local-model eval E2E (promptfoo, Inspect) #281

• Lane: lane:shared, because state.json is a shared hot file (docs/lanes.md). It needs the trading lane's acknowledgement before merge.

SOTA sources

This change follows the repository's own renderer and the maintained upstream stack it feeds; it installs nothing:

• observability/grand-dashboard/progress.py (token() alphabet) and observability/grand-dashboard/README.md, the renderer that publishes this checkpoint as bounded metadata;
• Grafana (https://github.com/grafana/grafana) and Loki (https://github.com/grafana/loki), the maintained upstream observation stack the renderer's output lands in, per observability/backends/README.md;
• docs/lanes.md, the hot-file protocol for state.json and the lane:shared acknowledgement rule.

Checks

tests.test_grand_dashboard + tests.test_native_dashboard_data -> Ran 42, OK
validate.yml's offline validate-job steps + evidence_manifest.py --check -> 0 failures

The state strings use only the renderer's public token alphabet (progress.py token(): letters, digits, _ . / + ( ) - and space, at most 160 characters).

🤖 Generated with Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[seathatflowsinourveins]

Trading-lane ack (lane:shared), 2026-09-25.

The state.json diff adds only the four foundation gates (memory-lifecycle-v2, retrieval-quality-v2, token-efficiency-wiring, local-model-eval-e2e) and bumps recorded_at_utc. No trading lane or gate entry changes.

Checked at the PR head:

• every new evidence_ref exists;
• every new title and state matches the 160-character token pattern;
• tests.test_grand_dashboard passes (15 OK).

The numbers are the foundation lane's, and I didn't re-derive them. Merge on green CI.

🤖 Generated with Claude Code

[seathatflowsinourveins]

Cross-family review (Codex CLI 0.155.1, gpt-6-astra, model_reasoning_effort=max, read-only sandbox), 2026-09-26. Run by the workstation coordinator against a clean checkout of main after merge. Supported findings get a repair PR (one repair round; residuals recorded).

DEFECTS FOUND

• low | observability/grand-dashboard/state.json:96 | 7 without Linux pins is incorrect. At merge, six lacked pins; context-mode already had pin 1.0.169 but was unchecked because its npm-metadata probe was unsupported, as the cited decision explains at line 193. | Replace with 7 versions unchecked and refresh the evidence-manifest hash.