Skip to content

Wire React dashboard to real APIs + self-improvement surface (31n) - #28

Open
cosmicz wants to merge 3 commits into
mainfrom
agent/drv-31n1
Open

Wire React dashboard to real APIs + self-improvement surface (31n)#28
cosmicz wants to merge 3 commits into
mainfrom
agent/drv-31n1

Conversation

@cosmicz

@cosmicz cosmicz commented Jun 28, 2026

Copy link
Copy Markdown
Collaborator

React dashboard wired to the real h7w API endpoints (incidents SSE with named provenance event handled separately, telemetry/fleet/model), surfaces the self-improvement model (registry version + keep-if-better + honest incumbent scope/feature labels), honest Aug-17 hero with badged fixture fallback. Rebuilt bundle. nav-approved at 5c878ee: full pytest 244, dashboard build, live verify.mjs against :8000 all passed, merge-tree clean. Tracking: aieng26hack-5fq.8co.31n.

🤖 Generated with Claude Code

cosmicz and others added 3 commits June 28, 2026 11:46
…ace (31n)

Connect the React mission-control dashboard to the live FastAPI endpoints
served by h7w, with honest provenance badging end-to-end.

- data.ts: loaders call /api/fleet|meta|telemetry|model|monitor|learning-curve
  instead of static /fixtures; static fixtures kept only as an explicit, badged
  `offline` fallback when the API is unreachable.
- App.tsx: consume the NAMED `provenance` SSE event via addEventListener (NOT
  onmessage — onmessage stays Incident-only, preserving the h7w back-compat
  contract); track the dashboard data-source badge in state.
- ProvenanceBadge: real / fixture / synthetic / offline / unavailable, with the
  provenance note on hover, surfaced in the TopBar.
- SelfImprovement: new live surface for /api/monitor (per-horizon recall/miss at
  a chosen alert budget + model version/onsets) and /api/learning-curve (v0->vN
  held-out ROC-AUC over the no-skill baseline). Explicitly badged illustrative.
- AgentTriage model card: rigorous-vs-provisional and fixture honesty flags.
- backend: add GET /api/learning-curve (repo-root artifact, packaged fallback,
  honest available:false floor) + committed in-package copy.
- rebuild + redeploy the dashboard bundle; prune stale asset bundles.

Tests: 21 passed (3 new /api/learning-curve cases). ruff format+check clean;
vite build green.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…ion (31n)

- verify.mjs: rewrite to run against the live FastAPI backend and derive every
  expected value from /api/* (no stale hardcoded 2344/882 fixture numbers).
  Asserts: real fleet cell count, REAL provenance badge, SSE incident stream +
  telemetry, learning-curve points, per-horizon recall table, model honesty
  flags, and a clean console (benign best-effort feedback 404 excluded).
- README: replace the "static fixtures" data section with the live /api/* wiring
  and the provenance-badge honesty model.

Live verify: ALL CHECKS PASSED against uvicorn on :8000.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…mprove panel (31n)

Address AC: the dashboard must label the incumbent honestly as a per-GPU
early-warning scorer applied fleet-wide — NOT a topology/cascade model (bead 1b4
hasn't produced a topology-augmented incumbent). Until bead 4sz exposes learned
parameters/contributions, only the feature set (card metadata) is shown. Rendered
from /api/monitor.features. Rebuild + redeploy bundle.

UI smoke: ALL CHECKS PASSED.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant