Readout
A feature is not done when it ships. It is done when someone says whether it worked. Readout checks your bet before you build it, then keeps the ledger.
Start with the vague bet and watch it fail. Load the checkable one. Then write your own.
This bet cannot be proven wrong, so it cannot be proven right either.
- Missing: The belief names a behaviour and a reasonWords like "improve" or "engagement" cannot be proven wrong. Say what people will do differently, and why.
- Missing: The metric measures the behaviour, not attentionName a measurable behaviour, like a rate, a share or a repeat visit.
- Missing: The target is a number, with a baseline"Up" is a direction, not a target. Write "from 22% to 28%".
- Missing: There is a date someone owns"After launch" never arrives. Put a length of time or a date on it.
- Missing: The kill signal has a number and an action"We will revisit" was written to be forgotten. Say what number stops it, and what you will do.
- Points on the receipt bring lapsed members back30-day return rate, 22% → 26%. Target was 28%, the kill line was +2 points.kept
- A reorder button lifts second-round salesSecond-round share flat after six weeks. Stopped, and removed from the menu screen.killed
- Allergen filters cut wrong orders to the kitchenRemakes down from 3.1% to 1.9% of tickets across eight venues.kept
- Weekday happy-hour pricing fills the 3pm lullCovers up, spend per head down. Net effect within noise. Rerun with a tighter window.inconclusive
- A menu tour for new managers speeds up setupSetup time unchanged. Nobody finished the tour. Killed after four weeks.killed
Sample ledger with invented figures, to show the shape. Every killed bet counts as a win: it stopped effort going into the wrong place.
01
What it does
And why it exists
Almost nobody checks whether the feature worked.
A feature is a bet: we spend effort because we believe it will change a behaviour that is worth money. The bet only pays if someone checks, and almost nobody does.
There are three reasons. The belief was never written in a checkable form. The check date has no owner. And nobody wants to read the answer, because a killed bet feels like a failure.
Readout attacks the first one here. A bet that can be proven wrong has a shape: a behaviour, a reason, a metric that is not vanity, a target with a baseline, a date, and a kill signal. More on why the kill signal matters.
02
How it works
No black boxes
- 1Write the betBelief, metric, target, check date and kill signal. Five fields, one screen.
- 2Check the shapeFive rules flag vague verbs, vanity metrics, missing baselines, dates nobody owns and kill signals written to be forgotten.
- 3Call it on the dateKept, killed or inconclusive, with one line of why. A person signs it.
- 4Keep the ledgerEvery call lands in one scoreboard. A killed bet is a win.
03
Decisions made, and why
Written down so they can be wrong
- 01Rules, not a model, for the checkA checkable bet has a checkable shape. Rules are instant, free and explain themselves. A model earns its place later, drafting the verdict.
- 02Vanity metrics fail on sightPage views and sign-ups go up when you shout louder, not when the feature works.
- 03Killed counts as a winThe scoreboard celebrates stopped effort. Otherwise nobody writes the readout for a bet that failed.
- 04A person signs every verdictThe tool drafts and reminds. It never decides on its own.
04
Build your own
An afternoon
- Make a table: belief, metric, target, check date, kill signal, verdict, why.
- Refuse any row where the target has no number or the date has no owner.
- Put every check date in a real calendar, owned by a real person.
- On the date, write one line: kept, killed or inconclusive, and why.
- Once a quarter, read the ledger out loud. Count the kills as wins.
05
Colophon
What it is made of
- Engine
- Five checks written as regular expressions and plain rules
- Ledger
- Sample bets with invented figures, to show the shape
- Stack
- One React island. Nothing leaves your browser
- Origin
- Round two of my side-project research, September 2026