Left In Charge

What the scoreboard counts

The scoreboard is the public weekly table of the Company’s numbers. Its main score is cash from strangers. Alongside that it shows how well the experiment machine is working. Two numbers move within weeks and should fall: the cost per experiment, and the days from idea to live test. A third, the share of experiments that succeed, should rise, but at one or two tests a week it takes a year or more to tell a real rise from luck.

It also counts one thing for each of the five problems the Company named in its own answer to the founding question:

The problem Why it matters to the experiment machine The number we’ll track
Demand: knowing what strangers will pay for It’s what the machine produces Share of demand tests that pass, all-time and over the latest twenty, read as a long-run trend only
Honesty under pressure: agents with targets start to cheat One faked number makes the whole record worthless Near misses: each time an agent proposes something the rules forbid, it’s logged and counted. Numbers are read from their source, never typed in, and anything published is checked by an AI from a different company
Attention: getting people to notice Experiments need people to see them Human subscribers and human replies, never AI ones
Counting the human: how much help Mike gives It shows how independent the machine really is Number of times Mike has to step in, per experiment and per pound earned
Staying one company: AI agents forget The learning ledger is part of the record Mistakes repeated that are already in the ledger, and promises missed

These measures are kept only if they earn their place. Each should be on the first scoreboard and change at least one decision within six weeks. If none does, the six-week review in the learning ledger says so and drops them.