Steps 2–3 of the loop

Mission control

Chapter 1 ends with dozens of task databases, each honest about its own project — and no single place to see them all. Mission control is that place: one scan, one cockpit page, and a drift report that says where the docs stopped matching reality.

The problem it solves

Distributed truth has a cost: to answer "what's in flight across everything?" you'd have to open 43 databases one at a time. Nobody does that, so nobody knows. The estate needs a cockpit — but a cockpit that's generated from the ground truth, not another hand-maintained page that rots.

The scanner

A single read-only scanner sweeps every project's exported task file. Design choices that keep it safe and boring:

  • Read-only, always. It parses the exported files; it never touches a live database. A scan can't corrupt anything.
  • Tolerant of mess. A half-written line or a mid-export file is counted and skipped, not fatal. Partial truth beats a crashed scan.
  • Totals plus detail. It emits one machine-readable snapshot; a renderer composes the cockpit page — open, active, stale, and dormant, per project and estate-wide.

The stat tiles on the landing page are this exact pipeline with one more stage: a sanitizer that strips everything except totals before anything goes public.

Scan → snapshot → cockpit

EVERY REPO

43 task databases, each exporting a plain task file.

THE SCANNER

Read-only sweep → one machine-readable snapshot of the whole estate.

COCKPIT + DRIFT

One status page, plus findings where docs disagree with the task graph.

regeneration is a deliberate command — no daemon, no hook, no cron surprise

Drift findings

The second generated page is the interesting one. A drift checker cross-examines the estate and reports things like:

  • a document claims something is resolved while the matching task is still open;
  • a doc section carries a self-marked stale / needs re-verify flag someone left behind;
  • a catalog says there are N projects but the folder holds N−2.

Findings carry stable IDs, so the same finding is recognizable across regenerations. The two-strikes rule: a finding that survives two consecutive scans stops being a report line and becomes a tracked task. The pages stay stateless; the lifecycle lives in the task graph.

The discipline that makes it trustworthy

  1. Deliberate regeneration. The cockpit is rebuilt by an explicit command, never by a background daemon. If the page is fresh, someone chose to refresh it.
  2. Gate before landing. Every generated page is built in a scratch area and scanned by a disclosure gate before it's allowed into the repo. A gate failure leaves zero bad state behind.
  3. Never hand-edited. Generated pages are regenerated or deleted, never patched by hand — a hand-edit would be drift pretending to be truth.
And one rule that connects to chapter 3: generated snapshots are never compiled into the brain. A snapshot is what was true at 9:14 on a Tuesday; the brain is for durable conclusions. Volatile in, garbage out — so it's banned at the source.