Kerth's Four Key Questions (The Original 4)

When to use

End of project, end of quarter, post-release — Kerth wrote the format for end-of-project review and it still fits milestones best. On sprint cadence it's a reset: run it when the retro has decayed into complaint triage, because 'what did we learn' and 'what still puzzles us' can't be answered with a gripe. Skip it when the team needs sprint-level commitments in hand by the end — Start/Stop/Continue frames commitments; this frames understanding.

How it runs

1. Write the questions up verbatim

   Four columns, Kerth's exact wording: 'What did we do well, that if we don't discuss we might forget?' 'What did we learn?' 'What should we do differently next time?' 'What still puzzles us?' Long headers, yes. The length is the format — trim question one and you're running What Went Well under a different name.

2. Silent write, five minutes

   Standard silent write; senior person writes last. The unfamiliar wording helps here — people actually read the questions instead of pattern-matching to the board they filled in last sprint.

3. Walk the columns in order

   Well, learned, differently, puzzles. On question one, push each card to name the practice: what exactly do we keep doing, and where does it get written down? On puzzles, capture — don't solve. The room will try.

4. Close each column's loop

   Questions one and two go somewhere that outlives the meeting — a runbook, a decision log. Question three becomes one or two commitments, not ten. Every puzzle gets an owner to investigate, or an explicit 'we're leaving this open.'

Why it works

The first question's subordinate clause does the work. 'What did we do well' invites a victory lap; 'that if we don't discuss we might forget' turns the same column into knowledge capture — the answer isn't applause, it's a practice worth writing down before it evaporates. And 'what still puzzles us' is the only classic column that treats not-knowing as a legitimate finding rather than a failure to have an opinion. Keep, learn, change, investigate — every later format is a subset of these four.

Variations

• Pair it with the Prime Directive read aloud at the top — same book, same meeting. Kerth designed them together.
• The original context: Kerth's retrospectives were multi-day end-of-project affairs, and the four questions organized the whole event. The 45-minute board version is the compressed descendant — fine, as long as you know which one you're running.
• If the team can't stop solving in the room, timebox the puzzles column to capture-only and schedule the investigations for afterwards.

Facilitator notes

Question one is the one to facilitate hardest — the clause is your instruction. When a card says 'the release went great,' ask what specifically we'd forget: the practice, the trick, the runbook step. No answer means it's applause, and applause doesn't need a column. Hold the line on puzzles too: the column exists so 'I don't know why that happened' is an admissible card. The moment someone answers a puzzle from across the room, the people holding the real ones stop writing.

Pitfalls

• Paraphrasing the questions. 'What went well' is not question one — the forgetting clause is the difference between knowledge capture and a victory lap.
• One-word column headers. Tools truncate the questions to Well / Learned / Differently / Puzzles, and within three sprints the wording erodes back to the generic board. Keep the full sentences visible.
• Solving puzzles in the room. It turns the retro into a debugging session and teaches the room to stop writing them.
• Ending with an ownerless puzzles column. A puzzle nobody investigates is a complaint with better manners.

Remote tips

Open the board a day early — the learned and puzzles columns improve with thinking time. Keep the full question wording as the column headers even though it's long; the one-word labels retro tools default to are exactly how Kerth's wording eroded in the first place. Anonymous cards help the puzzles column most — 'I don't understand why we do X' is a hard card to sign.

Example outputs

• Did well (would forget it): the feature-flag rollback on day two saved the release. Runbook step written up while we still remember the order of operations.
• Learned: the vendor sandbox's rate limits don't match production. Cost us a week of debugging the wrong layer.
• Differently next time: involve support from the first beta week, not after the ticket spike.
• Still puzzles us: the staging migration was clean, prod failed twice. Nobody knows why yet — Sam owns finding out by Thursday.

FAQ

Four Key Questions or What Went Well / What Didn't?

What Went Well is this format with the clauses amputated and the puzzles column deleted — two columns, no learning question, no licence to be confused. If the familiar board has gone stale, running the original four is the cheapest reset there is: same effort, twice the surface area.

Do I have to use Kerth's exact wording?

Yes. The wording is the format. 'What did we do well, that if we don't discuss we might forget?' assigns the column a job — capture what would otherwise evaporate. Trim it to 'what went well' and the job disappears; you're back to the retro every team already runs on autopilot.

Won't 'what still puzzles us' just collect complaints?

A puzzle with an owner is an investigation; a puzzle without one is a complaint in better clothes. Spend the last five minutes assigning them. Some stay open — that's fine, and it's the point: this is the one column where 'we don't know yet' is a legitimate outcome.

Related activities

Recommended use cases