# Inclusion rule for pre-21-October runs in "A Readout Is Not an Idle"
Written 22 Oct 2026, BEFORE the history inventory (scratchpad/history_inventory.md) was read. Fixed by its SHA-256 below.

A run enters the paper's EVIDENCE if and only if all three hold, judged from its design files, not its results:
1. Design: a mid-circuit readout arm and an idle arm of equal scheduled length, on the same untouched qubits, in the same job.
2. Quantity: the untouched qubits' phase, coherence or concurrence is read (the same estimators the paper uses, or ones that can be recomputed with them from the saved data).
3. Record: job IDs and saved outputs (raw or committed) exist on disk.

Every run that passes is reported with its outcome WHATEVER IT SHOWS — including runs that disagree with the 21-22 Oct readings, are consistent with zero, or cannot tell — and labelled "not pre-registered" unless a timestamped plan exists. The month between those runs and 21 Oct is reported as a reading of its own (does the effect hold across weeks, and at what size), never pooled with the October numbers.

A run that fails any of 1-3 enters only the TIMELINE: date, processors, job count, what it measured, in one line, with no numbers carried into the claims. Framework vocabulary in old records is not carried into the paper.

No run is included or excluded because of its result. If the design files are ambiguous on 1 or 2, the run is CANNOT TELL and goes to the timeline with that label.