Blanks, controls and reference samples inside the slice

1 · Learn the move · Verified-data override

Your run already contains its own truth-tellers: blanks with a known ceiling, a positive control with a certificate range, a reference sample your lab has measured forty times. QC through a model works when those knowns travel inside the slice, labeled, with their expectations attached - because then every claim the model makes about the data has judges sitting in the same table. The override rule does the rest: your expected values outrank anything the model computes or believes, and a control outside its range stops the analysis rather than getting narrated past. The failure mode this kills is the plausible explanation - a model, asked why the positive control is low, will produce three fluent reasons and an offer to proceed; the discipline is that a failed control is a failed run until a human decides otherwise, and no explanation, however good, converts FAIL to PASS from inside the chat.

My dataset includes labeled control rows with expectations I have verified independently: [control: expected range/value, source of expectation]. Rules: (1) evaluate every control against its expectation FIRST, before any other analysis; (2) my stated expectations override anything you compute or recall - if they conflict, the run is in question, not my numbers; (3) any control outside its range = RUN FLAGGED: stop, report which control failed and by how much, and do NOT analyze the samples or explain the failure away; deciding whether the run stands is mine. Data: [paste]