The lab has its own falsifier
Every Hypothesis this lab registers must state, before it runs, the result that would refute it (R3). The lab itself states no such result. It has three gates that can stop it before M1 — data unusable (M0.6), programme unaffordable in money (M0.14), unaffordable in trades (M0.15) — and once those close, nothing. There is no condition under which a built, running lab concludes that its own premise was wrong.
That is the discipline applied to every question except the one that paid for the apparatus.
The lab therefore declares a falsifier before its first Hypothesis resolves: a count of resolved mechanism verdicts, and an outcome, at which the lab closes. It is recorded in configuration, evaluated mechanically against the Register, and never judged in the moment — the same construction as a Strategy's retirement rule, for the same reason.
This is not alpha exhaustion
The two stops look alike and are opposites.
| Alpha exhaustion | Lab falsifier | |
|---|---|---|
| Stops | the search | the enterprise |
| Says | "no budget left to look" | "looking here was the wrong idea" |
| Repaired by | new data — replenishment accrues against unconsumed observations (ADR-0011) | nothing |
| Status | terminal until new data | terminal |
Alpha exhaustion is a resource running out. A lab falsifier firing is a result: the premise that a detectable edge exists — in this instrument class, at this cost structure, reachable by this apparatus — has been measured and found wanting. It is the most valuable thing the lab can produce short of an edge, and it is the one outcome nobody builds a lab to hear.
Why it must be declared before the first verdict
A threshold set once results are in is a threshold set with knowledge of the results. The lab would be doing to itself precisely what the sealed reserve, the spec hash and the pre-declared floor exist to stop it doing to a Hypothesis. M0.18 therefore sits inside M0, beside the alpha split (M0.11) and the money (M0.13), and the number is the author's on the same grounds: a default is a recommendation, and no one else's threshold for abandoning this is meaningful.
Considered options
- No falsifier; stop when it feels wrong. The status quo. Rejected on the design's own argument against judged retirement (ADR-0010): the moment of deciding is the moment least able to decide, and sunk cost is at its maximum exactly when the evidence against is strongest.
- Alpha exhaustion is the stopping rule. Rejected by ADR-0011's own mechanism — exhaustion is repaired by new data, so it stops nothing permanently. A lab that replenishes each quarter runs forever by construction.
- A time box — stop after N months. Measures effort, not evidence. It fires on a lab that is working slowly and spares one that is failing fast, which is backwards.
- A money box — stop at the cap. Already exists as N5, and does useful work, but it stops the spending, not the believing. A lab can run at near-zero marginal cost on data already bought and still be wrong.
Consequences
The count is over resolved hypotheses, including superseded ones. Supersession is the obvious escape hatch: a null approaching the threshold could be relabelled superseded and the count reset. A resolved Hypothesis counts against the falsifier permanently, whatever later happens to it.
Both tiers count, and not equally. A null mechanism hypothesis (ADR-0027) says the effect is not there. A null implementation hypothesis says one apparatus could not reach an effect that may be. The threshold is stated over mechanism verdicts; implementation nulls are recorded and not counted, unless the author states otherwise at M0.18.
The lab refuses to start without it, on the R9 pattern. A missing falsifier is not a default of infinity; it is a refusal.
Firing it is a publication, not a deletion. The Register is publishable as it stands (ADR-0008). A lab closing on its own declared threshold publishes the count, the floors and every verdict — the negative result this apparatus was built to be able to produce honestly.