The Baker Challenge
Two kitchens run the same plan. The only difference is who decides.
In Atlanta, a baker reads the board and decides what goes in the oven. On a second machine — same release of the software, same settings, same day — the advice is followed exactly as it is given, every time, without argument. Everything else is held as still as we can hold it. What is left over is the price of the decision itself, and until this morning nobody had a number for it.
Atlanta · Tuesday 11 August 2026 · day one
There are no results in this article. The experiment is one morning old. What follows is how it is built, what it counts, the first two pictures it produced — and what today already showed us about our own instruments.
How it is set up
Two kitchens. One plan. One difference.
The first kitchen is the shop: a working patty bakery in Atlanta, a wall tablet by the oven, a person reading it. The board tells him what to bake, in which oven, and when to start. He decides. Sometimes he decides something else — that is the whole reason he is standing there.
The second kitchen runs the same board on its own machine. The same release, down to the version number, read off the running machine rather than taken from a release note. The same freshness target, the same buffers, the same opening hours, the same closing taper. The one thing it does not have is a person. In his place stands something that does exactly one thing: it reads the advice, waits two minutes, and loads what was advised — unchanged and unrounded.
The two minutes are not ours to choose. They are the shop's own grace period: the line past which the app itself calls an advice ignored and raises it on the board. A stand-in that waits exactly that long is, by the shop's own definition, never late and never early.
And it presses the buttons. It goes through the same call the START button goes through — no side door into the database, no shortcut written specially for the experiment. If it had one the comparison would be worth nothing, because the point of the whole exercise is that both sides travel the same code. The only difference left in the building is who decided.
What a followed plan looks like
The stand-in's first bake this morning: the advice had stood 2.11 minutes — 127 seconds, by the app's own stamp — when the load started at 8:25:33 AM. What went in was, piece for piece, what had been advised: the two fast flavors carrying the bake, Chicken and Spinach a sheet apiece. Nothing rounded up, nothing left off, nothing added because it felt thin. Ready at 8:56:03 AM — a stocked case, four minutes before the doors opened.
The two boards at that minute are not pictured here. What they showed was the difference in the case, flavor by flavor, and the counts are the whole of it — so it is described above rather than shown.
What we are counting
Four numbers, in the order the shop already ranks them
Both kitchens are measured the same way, every day, per flavor and across the whole case. Each number is money in a form a bakery recognizes — a customer turned away, a customer kept waiting, a patty that sat too long, a tray that goes in the bin.
Minutes at zero
For each flavor, how many minutes of the trading day it stood at nothing. Not a percentage — minutes, and each stamped with the hour it happened in, because minutes without Spicy over lunch and the same minutes at four in the afternoon are not the same loss.
Every one of those minutes is a customer who wanted spicy and was told there is none. That is the most expensive outcome the shop has, which is why it is rule one.
The wait, when it is in the oven
A flavor can be at zero in the case and still be minutes away from coming out of the oven. That customer can be kept — if he is told. So the wait is counted on its own: how many minutes short a customer was of the flavor he came in for.
It is the number the board facing the line answers, and it is the whole difference between "a few more minutes" and "sorry".
Age at sale
How old each patty was when it sold — per flavor, and across the case. The shop's freshness target is sixty minutes. Measured on Saturday, before any of this started, the flavors sat between fifty and seventy-one minutes: some inside the target, some past it.
Never running out is easy if you bake early and let it sit. This is the number that stops that from counting as a solution.
Left at close
Pieces in the case when the doors lock, against the owner's target of twenty or fewer across the whole case. It is the only one of the four you can carry to the bin and count.
Which is exactly why it never stands alone: a clean close bought by running out at four o'clock is a worse day, not a better one.
→ the order settles every tie: never run out · then almost nothing left at close · then as fresh as possible
Day one
The first two pictures, and how to read them
At one moment this afternoon the two boards did not agree, and they did not agree by a wide margin: the shop's case held well over half again what the stand-in's held, and the shop was a full oven session further into its day. The same minute, both boards.
That gap is not a verdict on anybody. The second kitchen stopped receiving sales at 9:59 AM, shop time, and has logged next to nothing since. With no sales arriving, nothing became due, and the stand-in went quiet — not because it failed, but because it had nothing to answer. So the first thing this experiment measured was its own instrument. A comparison between two kitchens is never better than the feed into both of them, and we would rather find that out on day one, in public, than three weeks from now inside a chart.
The second picture is from Monday, and it is the reason the experiment exists at all. The baker raised nearly two thirds of that day's oven loads — every one of them by exactly one sheet, and every one of them between 9:55 AM and 2 PM. Not one load lowered. That is not somebody ignoring a board. That is a person telling the plan, the same way over and over, inside the same stretch of the day, that it runs light through the middle.
Whether he is right is precisely what the second kitchen is there to answer, and the answer will arrive in pieces sold, pieces binned and minutes at zero — not in anybody's opinion, his or ours.
The interesting part
What else this makes measurable
Holding everything still except the decision does more than price the decision. It puts a number on a short list of things bakeries normally argue about — and it makes plain which of them we cannot measure yet.
Measurable today
What the judgment is worth, in pieces
Monday's raised loads are already counted to the piece: the shop knows exactly how much more went into its ovens than the plan asked for. What that extra bought has never had a number — sales that would otherwise have been stockouts, or trays that went in the bin. Run both kitchens through the same day and the difference between them is that number, in the shop's own products.
Measurable today
How fast advice is acted on, and what waiting costs
Every advice carries the moment it was given and the moment START was pressed. The stand-in's is fixed at two minutes; a person's is whatever the day allows — a customer at the counter, a delivery at the door. A load started at 12:05 sells from 12:35 — half an hour from the decision to the first sale, and that is measured, not rounded. So ten minutes of delay does not cost ten minutes of nothing: it moves the whole rhythm of the afternoon one bake late. Both sides stamp it already. Pricing it is arithmetic on top.
Measurable today
Oven sessions per patty sold
One kitchen was a full oven session ahead of the other at the same minute. Over a full day the honest form of that is sessions per hundred patties sold, which is a line on an energy bill and exactly the kind of number that gets asserted in a franchise pack and never proven. Both kitchens stamp every session start and every sale, so it needs no new instrument at all.
Needs another instrument
What a customer buys instead
When Spicy is out, does he take Mild or does he leave? One of those costs almost nothing and the other is a lost sale, and they are nowhere near the same size. The register records what was sold, never what was asked for. We can watch a flavor stand at zero and watch what rang up in those same minutes, and the shape of it is suggestive — but suggestive is not measured. To call it a swap, the ask has to be recorded where it happens, at the counter.
Needs another instrument
Whether a visible countdown changes what a waiting customer does
The board facing the line went up on the shop wall on Monday: one line per flavor, how long until the next batch is out, a time and never a count. The reasonable expectation is that somebody who can see a time on the wall waits, and somebody who can see nothing leaves. Nothing in either kitchen counts the person who looks at the case and walks out, and the second kitchen has no line to read the board at all. This one the challenge cannot answer. It needs a count at the door or a tap at the counter, and until it has one there will be no number here.
Straight about it
What today does not prove
Three things, and none of them belongs in small print.
The two boards did not open on identical terms. One setting — how heavily today's own sales weigh against the day the plan was built from — was still being aligned when the doors opened, and it was aligned within the first trading hour. The correction that setting feeds cannot act at all until half an hour of trading has been recorded, so the window in which the shop actually traded on the old value is about half an hour, right at the start of the day. Short. Not nothing. Today is a shakedown of the rig, not a clean day.
The two kitchens keep their own histories. The second one is not the shop run twice — it is a second kitchen, on the same day, on the same release, deciding differently. So what this compares is a baker against a stand-in, and any difference we report is a difference between two kitchens that have each been keeping their own record. Where a difference cannot be traced to a decision, we will say so rather than bank it.
And there are no results yet. The experiment is one morning old. Anything on this page that looked like an outcome today would be an invention, and this site does not print numbers it has not measured — not for the pilot, and not for this.
Who this is for
The same experiment, read two ways
If you run one shop
Nothing here is a report card on your baker. When he raises load after load by a sheet he is telling the plan something it did not know, and the plan's job is to learn it — so that on the day he is busy at the counter, or not in at all, the board already knows it too.
What is on offer is a quieter day, not a stricter one: less arithmetic in front of a hot oven, fewer customers told no, and a case that runs itself down to almost nothing by the time the doors lock.
If you run fifty
Every group already has a best baker and a newest one, and the distance between them is worth money nobody has ever counted. This is how it gets counted: the same board, the same release, the same four numbers per location, measured the same way and reported with their exceptions named.
What travels to a new location is the method — not fifteen years in one person's hands, and not a claim in a franchise pack that a new franchisee has no way to check.
A busy Saturday forgives everybody. The experiment is for the slow Monday and the rainy afternoon, where the difference between a plan and a decision is the whole margin.
Talk to us
See it on your own numbers
A walkthrough is thirty minutes: your products, your hours, your ovens, and what the board would have said last Tuesday. Franchise operator? Mention how many locations — running this across a group is a conversation of its own.