A Galahad Labs Product
Every departure is a data event. Chronicle turns exit interviews into structured, searchable institutional knowledge before it walks out the door.
Used inside Galahad for engineering, operations, and executive handovers.
The Problem
70%
of organisational knowledge is undocumented, existing only in the heads of individual employees.
6 months
Average time for a replacement to reach the productivity of the person who left, if they ever do.
42%
of role-specific knowledge is permanently lost after an employee departs, according to internal knowledge audits.
The standard exit interview produces a two-page PDF about culture fit. It captures nothing about the systems, decisions, or relationships that made the departing person effective. Six months later, the replacement is reinventing workarounds that were already solved in 2021. Chronicle fixes the interview, not the offboarding form.
How It Works
01
Create a session for a departing employee. Choose a role-specific template — engineer, executive, or general — that targets the right knowledge domains.
02
Run the interview using guided question frameworks, or import an existing transcript. Upload audio for automatic transcription with speaker diarisation.
03
The pipeline processes the transcript through the four extraction layers and tags every artefact with confidence, decay estimate, and linked entities.
04
Structured artefacts flow into your organisational knowledge base. Generate handover briefs, flag single-source risks, and track knowledge decay over time.
What it looks like
Three views from inside the product. Knowledge browser, risk map, and the extraction view that turns transcript segments into structured artefacts.
Knowledge browser
Every interview produces a stack of artefacts. Filter by type, sort by decay window, search by entity. The decay bar on each card shows how long the artefact stays useful before it needs revalidation.
chronicle · knowledge
4 of 142 · sorted by decay window
Shadow Process
Production deploy runs cleanly except cache layer — init script does not restart memcache. The platform engineer SSHes into the box and runs sv restart cache. Documented nowhere.
6mo
Decision Rationale
Aurora was rejected during the 2022 migration after a week of testing showed 400ms+ lag on the reporting replica during batch jobs. Self-managed Postgres on RDS gives predictable lag.
24mo
Risk · Single Source
The finance reconciliation cron has six branch conditions added incrementally over three years. Logic is held by the senior backend engineer; no second reader on the code.
3mo
Tribal Knowledge
Tuesday is the lowest-traffic day post-newsletter. Rebuild was moved off Sunday after a 2023 incident where a slow rebuild collided with the weekly billing run.
12mo
chronicle · risk map
Items where only one person holds the knowledge. Sorted by blast radius.
Platform engineering
7/11 single-sourceCriticalPayments
4/9 single-sourceCriticalData & analytics
3/7 single-sourceHighCustomer ops
2/6 single-sourceMediumMarketing systems
1/4 single-sourceMediumFinance ops
0/3 single-sourceLowImminent decay window
Platform engineering has 7 single-source artefacts with a senior engineer departure scheduled in 19 days. 4 are tagged critical. Recommended: schedule shadowing sessions on the cache layer and the deploy pipeline this week.
Risk map
Single-source-of-knowledge items, grouped by team and ranked by blast radius. When a departure is scheduled, Chronicle surfaces the items most at risk in the handover window so shadowing time goes where it matters.
Extraction view
Transcript on the left, extracted artefacts on the right. Every artefact links back to the transcript span it came from — claims are auditable to source text, not opaque model summaries.
chronicle · interview · extraction
Layer 02 · Shadow Process Mapping
Interviewer
Walk me through how code gets from your machine to production. Not the docs — the actual process.
Departing engineer
Platform · 6 years
OK so the pipeline does most of it. But the cache layer — it doesn't restart cleanly. Init script was written before we moved to memcache and never updated.
Interviewer
So how does the cache come back up?
Departing engineer
Platform · 6 years
I SSH in after every deploy and run sv restart cache. Takes about thirty seconds. Everyone on the team thinks it's automated.
3 artefacts · linked to 2 entities
Shadow Process
6mo
Risk · Single Source
3mo
Undocumented Dependency
6mo
+ 4 candidate artefacts pending review
The Framework
The interview guide and extraction prompts are tuned per layer. Questions probe different parts of memory, and the Claude pipeline is instructed to look for different kinds of evidence at each stage.
01
Layer
Why things are done the way they are.
Architectural choices, vendor selections, process rules. The reasoning that produced the current system, traced back to the options considered and the constraints at the time.
02
Layer
The real workflow, not the documented one.
How work actually moves through the team. Which step in the Confluence page everyone skips. Which Slack DM replaced the approval form. The gap between the diagram and the job.
03
Layer
Who knows what. Who really decides.
Names (anonymised to roles) tied to systems, tools, and authority. The informal network that routes around the org chart. Who gets called at 2am when billing breaks.
04
Layer
What breaks when this person leaves.
Single points of failure, undocumented credentials, one-person workflows, knowledge with no second holder. Ranked by blast radius, flagged before the handover window closes.
Extraction Pipeline
Each artefact is tagged with confidence, decay estimate, and linked entities. Reviewers validate; the system tracks drift over time.
The reasoning behind architectural, strategic, and operational choices.
Actual workflows versus documented ones. The real way work gets done.
Who knows what. Who really decides. The informal influence network.
Single points of failure, undocumented access, critical dependencies.
Institutional memory that exists nowhere in writing.
Clever fixes for systemic problems. The duct tape holding production together.
Systems and processes coupled in ways the architecture diagrams do not show.
Unwritten rules about how the organisation actually operates.
What has gone wrong before and why. Lessons that stop the repeat.
What works and why. Repeatable approaches, captured before the person leaves.
Who It Is For
The problem
A senior engineer resigns. You have three weeks to extract fifteen years of architectural context before the laptop is returned.
With Chronicle
A searchable record of decisions, workarounds, and undocumented dependencies. The replacement reaches context in weeks, not quarters.
The problem
Exit interviews produce a PDF nobody reads and zero operational insight. Institutional memory leaves with the person.
With Chronicle
Every departure produces structured artefacts tied to roles, teams, and risk. Knowledge loss becomes measurable.
The problem
Founders, early engineers, or long-tenured operators are leaving. Their context is the business and it lives in their heads.
With Chronicle
A durable record of the decisions that built the company. Future operators inherit the why, not just the what.
Pricing
Per-interview pricing covers the full pipeline. No platform fee, no minimum commitment. Annual licences trade volume for a fixed cost and a dedicated reviewer.
Pay for what you capture. Best for organisations running a handful of senior departures a year.
Volume discount kicks in at 10+ interviews per year.
Unlimited interviews, dedicated reviewer, and integrations with your existing knowledge tools.
Best for teams running 10+ interviews per year or with strict procurement requirements.
All prices exclusive of VAT. Self-hosted and air-gapped deployments available on the annual licence.
Security & Data
Exit interviews contain some of the most sensitive material in the company. Chronicle treats that as a design constraint, not an afterthought.
Artefacts attribute knowledge to roles, not names. "The CTO" rather than "Sarah". Protects individuals and keeps the record useful after reorganisations.
High, medium, or low. Reviewers see which items were strongly stated versus hedged. Low-confidence items flow to a validation queue.
Every artefact carries an estimated months-until-stale. Knowledge about a legacy system decays faster than a cultural norm. You see what needs refreshing.
Self-hosted Postgres. Transcripts, artefacts, and risk maps stay inside your infrastructure. The Claude extraction call is stateless.
Departing employees can flag material for removal before extraction. Final artefact set is reviewed by the interviewer before it enters the knowledge base.
Every artefact links back to the transcript span it came from. Claims are traceable to source text, not to an opaque model summary.
FAQ
No. Surveys capture sentiment. Chronicle captures knowledge. The output is a structured record of decisions, processes, and risks, not a satisfaction score.
Interviewees see the artefact set before it is published. Items can be redacted. Names are replaced with roles. Legal and HR retain the same review rights they have over any interview transcript.
They live in your Postgres database. They are not used to train external models. Chronicle calls the Claude API statelessly per extraction; no vendor-held session state, no shared memory across customers.
Ninety minutes for a senior role is typical. The framework has core questions plus role-specific probes. You can split across multiple sessions if needed.
Yes. Upload audio or paste a transcript. Automatic transcription with speaker diarisation is built in via Deepgram.
Artefacts export to Merlin (agency knowledge base), and the JSON schema is documented for export into Confluence, Notion, or a homegrown wiki. Integration is at the artefact layer, not the transcript layer.
Get Started
Schedule a structured interview, run it through the extraction pipeline, and see the structured artefacts in your knowledge base the same day. No platform migration. No training programme.