[dash0-dev]

Why

/insights reads as a stack of sentences, and over the last 30 days the only section anyone interacted with was the utility grid. This is the first visual version — the learning loop plus when memory moves — behind insights-visual (default off), so it can ship without changing what anyone sees.

What changed

• insights-visual flag and an InsightsView copy-and-suffix resolver: off renders the original InsightsPage untouched, on renders InsightsVisualPage.
• The visual page, on ONE window: cost line first, a Learn → Recall → Promote → Prune strip, Pulse (written / chosen / bulk as small multiples on a shared axis), a recall flow (client → read tool → scope type), and runs as a weekday × hour rhythm plus swimlanes; the utility grid, runs list, scope leaderboard and health diagnostics stay.
• Pure, spec-covered models (insights-loop, flow-layout, recall-flow, runs-rhythm) over existing REST routes only — no backend change.
• Series that reach PostgREST's 1000-row max_rows are shown as lower bounds rather than drawn as a cliff; the rhythm uses paginated runs for the same reason.
• Bounded load: the cost line, utility grid and strip share ONE hour-aligned /memories/utility request — this also changes the flag-off page (one request instead of two, no all-history cost) — and below-the-fold sections fetch on scroll; /utility hit statement timeouts under the old fan-out.

How to verify

• cd packages/web && npx vitest run src/lib/{insights-loop,flow-layout,recall-flow,runs-rhythm}.spec.ts
• Storybook Pages/Insights/Visual (+ /Tests) and Pages/Insights/InsightsView/Tests; or flip the flag on /settings/developer.

Notes for reviewers

The two new visual stories need /update-baselines (Chromium was not reachable where this was authored). User-facing docs and llms.txt are untouched on purpose — nothing changes for users until the flag flips (docs/otel.md lists the new insights.* analytics: the page range section and insights.recall_flow_view_changed) — and the lifecycle flow and prune trend wait on an audit-log aggregate endpoint.

[dash0-dev]

/update-baselines

[github-actions]

The dashboard preview for this PR — redeployed on each push that changes the web app.

Project	Deployment	Actions	Updated
lorekit	Ready	Preview	Oct 21, 2026 1:14pm UTC

_{Preview always points at this PR's latest commit; Deployment is the immutable build of 05bb828. · Comment /web-preview to force a redeploy. · Workflow logs}

[dash0-dev]

✅ No issues found

Adds the visual /insights page behind the default-off insights-visual flag: pure spec-covered models over existing REST routes, one shared hour-aligned /memories/utility request, and below-the-fold sections fetched on scroll.

Progress: open review threads 2 → 1 → 2 → 0 → 0 across the last 5 reviews

Review details

Gate		Details
Description vs. code	✅	The description matches what the diff does.
Prior review feedback	✅	Earlier review comments are resolved.
Documentation	✅	The change is documented well enough to follow.
Self-review signals	✅	No debug logs, leftover TODOs, or unreviewed stubs.
Code review	✅	The multi-lens review found no blocking issues.

Run	Details
Mode	incremental · 0 lines in delta · tier deep · depth checkout · no finder ran (zero delta) — Gates 1, 4, 5 and 6 carried from the full review of e9ab8ec; description and code unchanged
Reach	54 changed exports · 28 dependent files · 0 checked · 28 not checked · 4 changed files imported by 28 files, not traced TrendChip — 3 consumer files · 0 checked InsightsRangeSection — 2 consumer files · 0 checked FlagKey — 1 consumer file · 0 checked MemoryArchiveAction — 1 consumer file · 0 checked useLoreUtilityRows — 1 consumer file · 0 checked +49 more packages/web/src/lib/analytics/track.ts — imported by 21 files packages/feature-flags/src/registry.ts — imported by 4 files packages/web/src/components/lore/LoreUtilityGrid.tsx — imported by 4 files packages/web/src/components/lore/LoreCostHeadline.tsx — imported by 3 files
Found	produced 0 → posted inline 0 · cleared 0 · carried forward 0 · deferred 0 · below-bar 0
CI	red: CI Summary, Web — Storybook interaction + visual tests (changed) · 10 passed
Memories	skipped (zero-delta fast path)

^{Nothing to report — standards (skipped), measurability (skipped), integrations (skipped), severity, 0 files skipped.}

^{pr-reviewer · commit e9ab8ec · no code changes since e9ab8ec, gate checks only · how these findings are produced · updated Oct 21, 2026 12:36pm UTC}

[mthines]

/update-baselines

[github-actions]

✅ Update visual baselines: committed 41 updated baseline(s) as 332d493283de0f41c5982988c9c55c57af68b61e.

ℹ️ This commit was pushed with GITHUB_TOKEN, so web-test will not re-run automatically — push a follow-up commit if you need CI to re-verify.

[dash0-dev]

Darkplane Automerge did not approve this pull request. It scored 60.76 of 100, and 75 was required.

• 🔴 Change footprint scored 30.
  This is a large hand-written feature of about 3,400 lines across 45 files. It adds a new insights-visual flag in registry.ts (default off) and a new visual page with layout and data-derivation logic in flow-layout.ts, recall-flow.ts and runs-rhythm.ts. Mitigating: the flag gates the page body and most of it is new files with specs and stories. It also edits shared code, including StatCard.tsx, the query cache key in lore-utility.ts, and the analytics types in track.ts.

The other 2 criteria

• 🟢 Blast radius scored 70.
  The new page sits behind the insights-visual flag (defaultVariant: 'off'), is added through the InsightsView resolver, and uses only existing read routes, so reverting is clean. It is not fully isolated, though. It also edits shared code that runs with the flag off: LoreCostHeadline.tsx and lore-utility.ts (the shared query key and since window), StatCard.tsx, track.ts, and Badge screenshots.
• 🟢 Evidence quality scored 82.
  On the evaluated head, CI Summary, Storybook interaction and visual tests, Typecheck/Test/Lint, Playwright e2e and the Vercel preview all pass. The PR adds specs for insights-loop, flow-layout, recall-flow, runs-rhythm, lore-cost and track, plus Storybook interaction tests, and the review threads were addressed. Remaining gaps: the automated reviewer's last report listed Storybook and CI Summary as red before baselines were regenerated, no human approval is visible, and the new UI is behind a default-off flag.

_{An AI evaluator produced these scores from the diff, the repository history, and the pull request text. The pull request text is author-controlled, so treat the scores as signals rather than proof. Reviewed for commit 05bb828.}