ai-value-chain-monitor

AI Value Chain Monitor — Pilot

What’s in this folder

• report.html — the front end. Open it directly; it’s fully self-contained (no server needed) and seeded with this week’s real edition.
• generate_report.py — the backend job. Calls Claude with web search once per segment, returns structured JSON, and writes report_data.json + a dated snapshot in /history.
• .github/workflows/weekly-report.yml — runs generate_report.py every Monday and commits the refreshed report_data.json back to the repo.
• requirements.txt — the one dependency (anthropic).

To wire up the weekly cadence

1. Push this folder to a GitHub repo.
2. In the repo’s Settings → Secrets and variables → Actions, add a secret named ANTHROPIC_API_KEY with your API key.
3. That’s it — the workflow runs every Monday at 13:00 UTC, or you can trigger it manually from the Actions tab (workflow_dispatch).
4. Every run writes a fresh report_data.json in pending_review status, plus a dated copy in /history so you have a full version trail.
5. Host report.html wherever you like (GitHub Pages is the easiest zero-cost option) and point it at report_data.json in the same repo — right now it uses embedded demo data, so the one code change left is swapping the embedded SEGMENTS array for a fetch('report_data.json') call.

The editorial gate

report_data.json always comes out of the automated job as pending_review — nothing is auto-published. The pilot’s approve/reset buttons in report.html currently write that status to the artifact’s own storage as a stand-in; connecting that button to actually update report_data.json (e.g. via a small API endpoint or a second GitHub Action triggered by a repo dispatch event) is the natural next step once you’re happy with synthesis quality.

Validating the pilot

Before expanding to more reports, the things worth checking over the next few weekly cycles:

• Are the momentum scores/tags directionally right, or does the model over/under-react to noisy weeks?
• Are the sourced signals actually the most consequential ones, or is it picking up SEO filler?
• Does the bull/bear sentiment section reflect real commentary, or drift toward blandness?