Lexery - Maintenance Runbook

Operational guide for keeping the Lexery second brain aligned with repos, Linear, and production reality — without boiling the ocean every week.

Cost guardrails: Lexery - Cost Ledger. What we ingest: Lexery - Source Registry. Automation design (target state): Lexery - Automation Architecture.

Principles

1. Delta-first: summarize what changed, not the entire monorepo, unless a monthly audit demands it.
2. Human gate for P0: production, security, and customer-facing claims need eyes — Tier 2 models propose, humans confirm.
3. Single chronology: every automated pass leaves a trace in Lexery - Log.

---

Tier 1 — Daily (automated, no model)

Goal: catch motion early; cost ~$0.

Steps

1. Per tracked repo (see Lexery - Source Registry), run:
   • git fetch --all --prune (if network allowed)
   • git log --oneline --since="1 day ago"
2. If new commits exist:
   • Append a one-line digest per repo to Lexery - Log (date, branch, hash prefix, author if useful).
3. Branch inventory:
   • git branch -a (or scripted equivalent) — flag new remote branches for human review in Lexery - Drift Radar or Lexery - GitHub History.

Outputs

• Updated Lexery - Log.
• Optional: JSON/CSV drop under _system/state/ with last-seen HEAD per repo (implementation detail).

Failure modes

• Fetch blocked: log “offline pass” and use local-only git log.
• Dirty worktree: note in Log — may explain skew in Lexery - Branch Divergence.

---

Tier 2 — Weekly (automated, cheap model via OpenRouter)

Goal: structured delta for maintainers; target cost pennies per run (see Lexery - Cost Ledger).

Steps

1. Monorepo stat delta:
   • git diff HEAD~20..HEAD --stat (adjust window by velocity).
2. GitHub PRs (if gh authenticated):
   • gh pr list --state all --limit 10
3. Linear (via MCP or export):
   • New/changed issues touching Brain, Portal, LLDBI.
4. Model pass (cheap model):
   • Input: the above artifacts + previous week’s Lexery - Log tail.
   • Output: bullet summary: themes, risks, suggested wiki page updates (file names only).
5. Human or scripted patch:
   • Apply edits to canon pages: Lexery - Current State, Lexery - Linear Roadmap, stage notes if behavior changed.
6. Drift:
   • Append any new contradictions to Lexery - Drift Radar; do not duplicate resolved items.

Outputs

• Short “week in Lexery” block in Lexery - Log.
• Targeted wiki patches + unresolved link list (if any).

---

Tier 3 — Monthly (manual or premium model)

Goal: consistency and architecture refresh; budget in Lexery - Cost Ledger.

Steps

1. Full consistency audit:
   • Orphan wikilinks, pages with status: inferred that should graduate to observed, duplicated narratives between Lexery - Open Questions and Drift and Lexery - Drift Radar.
2. Lexery - Current State rewrite pass — one screen executive truth.
3. Canvases (Lexery - Master Map.canvas, Lexery - Runtime Graph.canvas, etc.) if repo topology or pipeline edges changed.
4. Lexery - Cost Ledger — reconcile estimated vs actual spend if bills available.
5. Drift retirement:
   • Move fixed items out of Lexery - Drift Radar into Lexery - Log or archive section.

Outputs

• Updated updated: frontmatter on touched pages.
• Explicit list of new unknowns for Lexery - Unknowns Queue.

---

File system conventions (planned)

Path	Purpose
_system/scripts/	Cron-friendly shell/TS for Tier 1–2
_system/state/	Last-processed SHAs, hashes, dedupe keys
_system/exports/	Optional Linear/JSON snapshots

If these folders do not exist yet, create them when first script lands — document the addition in Lexery - Log.

Escalation

• Security or credential exposure in logs → stop automation, rotate secrets, fix redaction before next Tier 2 run.
• Contradiction between prod behavior and docs → P0 row in Lexery - Drift Radar + owner in Lexery - Team and Operating Model.

Links

• Lexery - Automation Architecture
• Lexery - Cost Ledger
• Lexery - Log
• Lexery - Index
• Lexery - Source Registry