Retro template for context compilation sprints

Distillation experiments move fast. Without a tight retro, teams forget which heuristics worked, which redactions broke customer trust, and which automations accidentally doubled token spend. Here’s the template I use every Friday with our own crew and with customer working groups.

1. Gather the ingredients

• Three bundles that represent wins, regressions, and surprises.
• Metrics dashboard covering token savings, failure rate, and turnaround time.
• List of automation incidents or escalations logged during the sprint.

2. Frame the retro

We keep the agenda brutally simple:

1. What improved?
2. What broke?
3. What is unclear?

Everyone drops sticky notes (we use FigJam) for 5 minutes, then we cluster. The facilitator (usually me) keeps us anchored to evidence—no vibes-only statements.

3. Review the bundles

For each sample bundle, ask three questions:

• Does the decision log read like reality?
• Can we jump from each block back to the thread without guesswork?
• Did the automation downstream behave the way we expected?

When we hit a discrepancy, we flag it in Linear with the bundle hash so engineers can reproduce the issue.

4. Convert notes into bets

Every insight becomes a bet in our backlog with an owner, expected outcome, and checkpoint date. Typical bets include “tighten Slack regex to drop reaction spam” or “expand medical redaction dictionary before the pilot.”

5. Close with commitments

The retro ends with a shared doc listing:

• The three bets we’re prioritising next sprint.
• Any policy decisions that need leadership sign-off.
• Shout-outs for the folks who unblocked the most work (never skip this!).

Want the FigJam template and Linear project we use? Email info@thread-transfer.com and I’ll ship it over.