Skip to content

Drift Detector — roadmap

Kept out of the README to keep the front page focused on using the tool.

What's next

  • Trend history — the dashboard shows the latest run; week-over-week burn-down needs a multi-run archive (a real persistence layer, not faked from one run).
  • Broader fleet access — the scanner only covers repos its token can read; the rest are flagged blind. Giving the bot read access across the fleet unlocks full coverage.
  • More integration shapes — each new vendor/API idiom is a reviewed catalog contribution through the absorb gate (the reviewed adaptation mechanism).
  • GitLab-native CI — move the scheduled run from GitHub Actions to GitLab CI (kills the cross-host token + egress, makes the private Cockpit free on GitLab Pages). Deferred pending a self-hosted runner — details in TECH_DEBT.md.
  • AI-surface consolidation — done. The three AI report surfaces (a separate probabilistic.html, a separate adhoc.html, and the certified dashboard) are now one dashboard: dashboard.html's AI Frontier tab, with leads/shapes/research kept in their own blobs and badged by tier (UNVERIFIED LEAD, GATE-VALIDATED, SOURCED). The CLI subcommand that used to render the side-car page is now leads (writes <state>/leads.json, a drift-leads/v1 document; it no longer renders HTML). The certified/unverified firewall is enforced by verify's ai-firewall invariant, which asserts no AI-derived record reaches the certified payload.