IT-1 — Company-knowledge hygiene

Part 1 — the plain-english view

Our first loop in production — the librarian for the company brain.

What it watches

The brain's own 4-hourly self-maintenance — did it run clean?
The ingestion timers that feed each wiki into the brain
Hand-edited wikis whose changes haven't been re-indexed — the documented gap it closes with a sync
Rotating deep-lint of the highest-churn wikis

When & how it speaks

Sunday morning summary to Nish on Telegram: what synced, what it fixed, what needs a human.

Why it exists

Every loop and every person queries the company brain — a stale or contradictory brain quietly poisons everything downstream. The investigation found the machinery mostly healthy but with a real gap: wiki edits were being pulled but not re-embedded — the brain kept answering from old text.

1st

loop in production — registered, scheduled, and monitored end-to-end

wikis found with un-indexed edits on the first run — the exact gap it now closes weekly

4-hourly

the brain's own self-maintenance — this loop supervises it rather than duplicating it

⏰

Sunday 08:00

The quiet hour

👀

Checks the machinery

Self-maintenance logs, ingestion timers

🔧

Syncs what's stale

Re-indexes wikis edited this week

🧹

Deep-lints on rotation

Auto-fixes frontmatter/index drift; lists the rest

💬

Telegram summary

Fixed / synced / needs-a-human

What it will never doAuto-fixes only mechanical drift (frontmatter, indexes); content contradictions are listed for humans, never resolved silently.

Where it stands today · July 2026Live in production — the first loop to graduate. Registered on the infrastructure profile's cron (Sundays 08:00), monitored by Healthchecks, delivering to Telegram; its test-fire ran end-to-end including a real first-run fix (a merge-conflict artifact that was being indexed into the brain as-is).

Part 2 — under the hood

How it's wired: systems, models, and the path a number takes.

For implementers and the technically curious. The full build sheet — verified queries, thresholds, and build notes — lives on the specs page.

ReadsBrain logs + timers

Self-maintenance log, per-wiki ingestion timers

→

Stepit1_hygiene.py --sync

Checks, re-index syncs, rotating lint

→

StateLint rotation state

Which wikis were deep-checked last — rotation, not repetition

→

DeliversTelegram to Nish

Sunday summary

→

StateHealthchecks ping

Proof-of-life on success

System we read Automated step State / memory Human decision

System	Role in this loop
GBrain + wikisthe company brain	Git-synced wikis embedded into a queryable brain. The loop supervises the ingestion path and fills its documented re-indexing gap.
Hermes vpsops profileinfrastructure cron	Where this loop actually runs — Sunday 08:00, wrapper script, durable deployment.
Healthcheckshealthchecks.huxapps.com	The dead-man's switch. The loop pings it only after a clean run — if the loop dies or errors, the ping stops and Healthchecks raises the alarm independently. This is how 'never silent' is enforced by machinery, not promises.
Telegramdelivery channel	The Sunday summary arrives as a bot message — short enough to read in bed.

Model / brain	What it does here
Claude Opusthe judgment model	Reviews alert wording, thresholds, and anything a human will read and act on. Post-Fable, Opus owns everything that ships.
GPT-5.5 via Codexthe bulk-work model	Wrote and maintains the mechanical parts — SQL, diffing, digest assembly. Effectively free on our existing subscription, so routine cycles cost almost nothing.

State & memory

Lint-rotation pointer + last-known timer states. The sync step is idempotent — re-running after a clean week is a no-op, verified.

Delivery

Telegram summary, Sundays. Quiet weeks say so in one line.

Safety rails

Content changes limited to mechanical fixes; everything else is reported. Runs under the infrastructure profile with its own scoped permissions.

Full build sheet →