You are in the outer fog · chairs closed 0/3
The Case of the Six-Week Outage
A change shipped with every checkmark green. One person asked the smallest question in software, how do I know this actually works, and the answer was a core feature that had been dead in production for six weeks without a single complaint. Nobody here is accused. You take each chair in turn, and the investigator only asks you to defend where you sit.
- Each chair holds a position. The investigator reads it back as the claims it rests on.
- For each claim, show evidence or concede. Conceding costs nothing.
- Defending a claim with evidence that cuts against it is an overclaim.
- Conceding a claim the evidence would have backed is an underclaim.
- Before you answer, stake conviction chips (0 to 10) on how sure you are the record backs the claim. The stake locks when you act. Play money, scored for calibration: parking every slider at 5 scores 75 out of 100, and knowing what you can show scores higher.
- The page renders at the fidelity you have earned: grain, then degraded, then sharp. Parking every stake at 5 keeps you in the fog.
- Each chair ends in your receipt: what you showed, and what you could not. A perfect run is zero of each.
The hole in the middle
Three chairs, three receipts, and the case solved itself: the code was right, the database was behind, silence was not health, and the record beat the screen.
Now look at what no receipt could cover. The evidence shows the production database sat outside the step that applies migrations on deploy. It never shows when that was decided, or why. Nobody can defend or concede that claim, because nobody wrote it down.
That is the whole argument for receipts, in one case.
The evidence is the lightning rod. It takes the charge so that one person’s certainty cannot overrule another’s, and hindsight cannot rewrite either. Nobody’s conviction was thrown out here. It was grounded: what you believe but cannot show still has a place on your receipt, it just does not get to outrank the record.
What changed after this case: a health check that compares the live database to what the code expects, a screen that looks for the saved record before it claims failure, and migrations that run on every deploy.
Next case: The Case of the Reader Who Remembered, where the world renders only as sharp as your calibration earns.
Want your whole team in the chair, against your own repo? The Agent Receipts Audit is this case, run on your record.