Three weeks’ notice. The on-call is in her head.

Priya Nair is leaving for a competitor. She is the incident commander for payments and wrote most of the cloud infrastructure - and the reasons behind it live nowhere but her memory. Here is how a capture runs when the clock is short.

Subject

Priya Nair

Lead Platform Engineer

Engagement

Resignation

Transition

Last day

3 weeks

Coverage · day 0

0/100

The situation

The person you page at 3 a.m. just gave notice

Northwind, a 200-person B2B SaaS company. Priya has owned payments reliability and the cloud platform for eight years. Her resignation gives three weeks - and a competitor is the destination, so goodwill time is thin.

What they own

  • PaymentsThe billing and payments service, its webhooks, and the idempotency logic that keeps customers from being double-charged.
  • On-callIncident command - the runbooks and the manual moves for a payments outage.
  • CloudThe Terraform that defines production, split across two state backends nobody else has touched.
  • FailoverActive-passive multi-region - a failover only she has ever run.

What the manager is afraid of

“Priya is our incident commander. When payments breaks, she’s who we page. Half our infra is Terraform she wrote, and there’s tribal knowledge about why prod is configured the way it is. Three weeks isn’t much.”

Successor: Sam Okafor, a mid-level engineer who will hold the pager. Strong on the app, new to the infra.

Phase 01 · Set up

Before a single question, the system maps the risk

It reads the role and the manager’s intake and predicts where the tacit knowledge concentrates, so the first interview walks in aimed, not cold.

Predicted high-risk tacit areas

  • CriticalPayments incident response - the manual rollback and webhook triage under fire.
  • CriticalMulti-region failover - documented nowhere; run only by Priya.
  • HighTerraform state operations - split backends with an ordering trap.
  • HighThird-party processor quirks - retry storms and idempotency workarounds.

A probe it generated for session 1

“When payments pages you, what is it usually - and what do you do in those first ten minutes that a runbook wouldn’t tell someone?”

Phase 02 · Capture

3 interviews, 3 layers

Each session has a different job. After every one, the transcript is analyzed and coverage is re-scored, so the number climbs as the tacit knowledge comes out.

UnexploredExploredSuccessor-ready
1The system mapGeneral capture~50 min · AI interview

Transcript · excerpt

Facilitator

When payments pages you, what is it usually?

Priya

Nine times out of ten it’s webhooks backing up, and everyone assumes it’s our queue. It almost never is. The processor has a retry storm after their own blips, and our idempotency check absorbs it - unless the check is bypassed on a hotfix path, which has happened. First thing I look at is their status page and our dedupe metric, not our queue depth.

Facilitator

Is that written down?

Priya

The runbook says "check the queue." That’s the wrong first move. It’s just what I do.

Knowledge captured

  • Payments & webhook serviceSystem
  • Incident first-response orderProcess
  • Terraform / cloud topologySystem
  • Deploy pipelineProcess

Tacit knowledge surfaced

Diagnoses webhook backups by the processor status page and the dedupe metric first - not queue depth, which the runbook wrongly leads with.

HighIncident first-response runbook points at the wrong signal.
HighNo documented multi-region failover procedure.
Coverage38▲ +38
2Under fireCritical decisions~55 min · AI interview

Transcript · excerpt

Facilitator

Walk me through the last real payments outage.

Priya

A bad deploy double-fired billing events. The fix isn’t "roll back" - roll back naively and you re-emit the same events. You flip the billing flag off first, then roll back, then replay from the last good offset with dedupe on. Order matters or you double-charge people. I’ve never written that ordering down.

Priya

And if it’s regional, failover is manual: promote the passive replica, repoint the two Terraform backends in the right order, warm the cache. Apply them out of order and you corrupt state. That’s the scariest thing I’m leaving behind.

Knowledge captured

  • Billing rollback orderingDecision
  • Manual region failoverProcess
  • Terraform apply orderingDecision

Tacit knowledge surfaced

The exact order to disable the billing flag, roll back, and replay with dedupe - to avoid double-charging during recovery.

Multi-region failover promotes the passive replica and repoints two Terraform backends in a specific order; the wrong order corrupts state.

AutoAutomation candidate: the manual failover should be scripted and rehearsable.
HighTerraform state ordering trap is undocumented and destructive.
Coverage61▲ +23
3Handover under a clockHandover checklist~46 min · AI interview

Transcript · excerpt

Facilitator

What does Sam need in his hands before your last day?

Priya

The deploy-freeze rules and the one IAM role everything assumes - if that role’s policy changes, half the pipeline stops and the error is misleading. And the nightly billing reconciliation job: if it fails, don’t rerun it blind, check the checkpoint table first. I want Sam to run a real deploy and a rollback with me watching, twice.

Facilitator

Could he hold the pager today?

Priya

For the common stuff, after one supervised week, yes. A regional failover - not alone, not yet. That one needs to be scripted before I go.

Knowledge captured

  • Deploy-freeze rulesProcess
  • Critical IAM role dependencySystem
  • Nightly reconciliation jobProcess

Tacit knowledge surfaced

A single IAM role that the whole pipeline silently depends on; changing its policy fails deploys with a misleading error.

ClosedIncident first-response and rollback ordering now documented.
MedRegion failover scripted as a ticketed task before the last day.
Coverage77▲ +16 · gate cleared

Phase 03 · Analyze

What the conversations became

47

Knowledge nodes

7

Tacit items saved

2

Automation candidates

77

Quality score

The knowledge map

The conversations become a structured, connected graph - one node per area of knowledge, linked by relationship and coloured by how transfer-ready it is. This is the same view you get inside the product; drag a node, or zoom in.

Loading graph…

Coverage by area

8
coverage (% captured) avg depthout-of-scope excluded
Successor-readyExploredUnexplored

Even on a three-week clock, coverage reached 77/100 with no open critical gaps. Two areas remain Priya-dependent and are carried forward with owners: the multi-region failover (now a scripting task) and Terraform state operations. The rest is in runbooks Sam can hold.

Phase 04 · Deliverables

The artifacts, generated from the capture

Every document is built from the same structured findings, so they agree with each other. Numbers are code-computed; the writing is grounded in what was actually said. Here is the flagship in full.

Flagship deliverable · shown in full

Continuity Report

Knowledge transfer - Lead Platform Engineer

Subject · Priya NairSuccessor · Sam OkaforSessions · 3Coverage · 77/100Open critical gaps · 0

Executive summary

Priya Nair holds concentrated ownership of payments reliability and the cloud platform, and is leaving in three weeks for a competitor. Three capture sessions raised coverage from 0 to 77/100 and cut single points of failure from 6 to 2. The judgment behind incident response - which signal to trust first, and the exact order to recover billing without double-charging - is now written down. Two destructive-if-wrong operations remain Priya-dependent and carry dated plans below.

01 · Critical systems & current understanding

  • Payments incident response - was Priya only; now a runbook that leads with the right signal.
  • Multi-region failover - documented; scripting ticket filed to remove the manual risk.
  • Terraform state ops - ordering trap documented; still expert-dependent.
  • Deploy pipeline & reconciliation - written and transferable to Sam.

02 · Tacit knowledge captured

  • First move: diagnose webhook backups by the processor status page and dedupe metric, not queue depth.
  • Billing recovery: disable the flag, roll back, replay with dedupe - in that order - to avoid double-charging.
  • Failover: promote the passive replica and repoint two Terraform backends in a specific order or state corrupts.
  • Hidden dependency: one IAM role the whole pipeline assumes; a policy change fails deploys with a misleading error.

03 · Risks & mitigations

  • High - region failover run only by Priya. Script and rehearse it. Owner: Platform, wk 2-3.
  • High - Terraform state ordering is destructive. Document + a guarded apply wrapper. Owner: Priya + Sam, wk 2.
  • Med - Sam has not held the pager solo. One supervised on-call week. Owner: Sam, wk 2-3.
  • Med - reconciliation job reruns are risky. Checkpoint-check step added to the runbook. Owner: Priya, wk 1.

04 · Successor readiness - Sam Okafor

Sam can already ship and operate the application. After one supervised on-call week and two watched deploy-and-rollback drills, he is ready for common incidents. A regional failover is deliberately held back until it is scripted - the plan converts a 3 a.m. hero move into a rehearsed procedure. Formal readiness sign-off sits with the engineering manager.

The rest of the pack

Successor Briefing

Sam’s day-one on-call manual: the first-response order, the recovery sequences, and the hidden dependencies.

Transition Task List

Every handover task with an owner and a date, front-loading the destructive-if-wrong operations.

Risk Heatmap

Payments, failover, Terraform and pipeline on one chart with the last-day clock running.

Gap Report

What is runbook-ready, what is shallow, and what became a scripting ticket - with the remediation plan.

Knowledge Risk Assessment

The bus-factor case: six single points of failure before capture, two after.

Quick Reference

The incident one-pager: trust this signal first, recover in this order, never change that role.

Want the finished documents to hold? Download real sample deliverables from a full engagement.

Phase 05 · Readiness

The difference in three weeks

Day 0

  • One person who can steady payments at 3 a.m.
  • Failover and recovery order "in her head."
  • 6 single points of failure across payments and infra.
  • Coverage 0/100.

Handover ready

  • 77/100 coverage, no open critical gaps.
  • On-call runbooks that lead with the right signal.
  • Bus-factor down to 2; failover now a scripting ticket.
  • Sam ran a supervised on-call week and two deploy drills.

Run one on your own critical role.

Start free: the role analysis, the interview scripts, and a full Knowledge Risk Assessment on your own expert cost nothing.

Start freeSee the other examples