Operator: manually test the WebUI in a browser (#277)

[CheeryProgrammer]

The WebUI can't be exercised in a headless sandbox — it needs a real browser, which is your machine. This is a manual UX pass to confirm the product surface and file gaps.

Bring it up

docker compose up -d
# open the app in a browser (default http://localhost:8080 or your HTTP_PORT)

Walk through and check

• Search / recall returns results with citations; untrusted content is visibly flagged (Provenance/trust tagging + mandatory review for untrusted ingested content #273)
• Logs tab shows app logs (feat: in-app log capture + WebUI log viewer with copy buttons #166)
• Proposals tab — only populated if the local-LLM extractor is enabled (EXTRACTOR=local-llm); approve/reject flows work
• Write actions (merge / retire / approve) are disabled by default; enabling needs WEBUI_MODE=interactive + AUTH_TOKEN=... (paste the same token in the UI). Confirm the gate.
• Node detail view: entity + edges + rationale render; trust surfaced (Node-level trust + per-call add_document trust (follow-up to #367) #375)
• Capture/browse flow feels usable

Report

#277 asks to make the WebUI a real product surface (capture + browse + node-detail). As you click through, note:

• What's missing or confusing (screenshots welcome)
• Which flows you actually want vs. what's there
  Drop findings on WebUI: capture + browse + node-detail (make it a product surface) #277 so the WebUI work can be scoped to real needs.

Related: #277, #166, #120 (auth), #375 (trust display).

🤖 Generated with Claude Code