Cadence infrastructure — knowledge handoff

What's safe, what's still in one head — told honestly, before the expert leaves.

PREPARED WITH CADENCE (SYNTHETIC) · EXPERT: DEPARTING SR. INFRA ENGINEER · 5-DAY SPRINT · OUSIOS LLC

Coverage Captured owner-independent, handoff-tested Partial documented, gaps flagged At-risk still in one head

The sprint & the decisionWhy we ran this

Cadence's senior infrastructure engineer is leaving in six weeks. Most of what they hold is in their head. Over five days we triaged their domains by risk × opacity, captured the highest-risk ones into runbooks a different engineer can run, and produced an honest map of what's safe and what's still a single point of failure. We deliberately did not try to document everything — we targeted the red.

Coverage at a glance — up top, by designThe bus-factor verdict

Verdict

Two of three critical systems are now owner-independent. The deploy pipeline and the incident response path are captured and handoff-tested. One remains a single point of failure: the legacy billing reconciliation job — and we flagged it honestly rather than fabricate steps no one could confirm.

If the engineer left tomorrow: deploys and incidents are safe; billing reconciliation is not.

Coverage: 2 CAPTURED · 1 PARTIAL · 1 AT-RISK — closing plan on the red below.

The bus-factor mapEvery system, rated honestly

System	Coverage	Status — and what it'd take to close
Deploy pipeline	Captured	Full runbook; a second engineer ran a deploy unaided on Day 5. Owner-independent.
Incident response	Captured	On-call runbook + escalation path documented and walked through. Handoff-tested.
Monitoring / alerts config	Partial	Documented, but two alert thresholds are the engineer's judgment, not written rules — flagged below.
Legacy billing reconciliation	At-risk	A nightly cron only this engineer fully understands. Needs a dedicated half-day session to capture safely; we refused to guess the steps.

What got capturedThe durable runbooks

Deploy pipeline runbook — prerequisites, steps in order, rollback path, the three decision points, and every dependency named.
Incident response runbook — triage, escalation, the "who to wake and when" path, and the two failure modes that actually recur.
Forward spec: the queue migration — the in-flight project, documented from day one so it never becomes next year's at-risk item.

The flagged gaps — honest, not fabricatedWhat we deliberately did not invent

Still at risk — flagged, not filled

An AI told to "just document it" would have written confident steps for all of these. We didn't — because the engineer couldn't confirm them, and a guessed runbook is worse than a known gap.

Billing reconciliation cron: the failure-recovery procedure exists only as the engineer's intuition. Needs a half-day capture session before they leave — booked for next week.
Two monitoring thresholds: set by feel, not rule. Flagged for the engineer to write down the actual logic, not for us to reverse-engineer a plausible number.
One vendor account: recovery access path unconfirmed. Listed as unknown, not assumed.

The flagged-gap list is part of the deliverable. Honest "we couldn't verify this" beats a confident guess that fails the day it's needed.

The discipline, made visibleHow this sprint works — and why it isn't "ask AI to document it"

AI accelerates the drafting. It cannot know what it wasn't told — so the discipline is verification and honest gaps, not speed.

Triage by risk × opacity Target what hurts most if it walks and is least written down — not the easy wins.
Extract & structure Facilitated capture; AI drafts runbooks, marking anything inferred as a gap to confirm.
Spec the forward work Document what's being built next, so it's never retro-captured later.
Verify — expert signs off, gaps flagged Every step confirmed by the person who holds it; nothing invented.
Handoff + the bus-factor map A different engineer runs a green runbook unaided; red is shown as red.

Ask AI to document from a transcript	This sprint
Invents the steps it wasn't told	Flags the gap, never fabricates it
Gaps look filled — confidently wrong	Gaps shown honestly as at-risk
No idea what it doesn't know	Expert confirms every procedure
A doc only its author can follow	Handoff-tested by someone else
A clean-looking liability	An honest coverage verdict

A confident doc that fabricates the parts no one checked is worse than no doc. The product is knowing, honestly, what's still at risk.

Get it out of their head before they walk out the door

Five bounded days: facilitated capture, verified runbooks, and an honest bus-factor map. Pair it with the AI Onboarding Sprint to install the habits that keep docs alive. Book a Documentation Sprint →

SAMPLE / SYNTHETIC · The Cadence infrastructure handoff is invented to demonstrate the method; no real person, company, or system is referenced. Flagged gaps are illustrative of the discipline (flag, never fabricate). · Method: Ousios AI Documentation Sprint (spec 22 + SOP-12) · The Architect (secondary The Craftsman) · Aspen Grove brand · Ousios LLC · 2026-06-23