One of the quietest failure modes in operational software is silent zero-coercion: turning an uninspected numeric input into `0` by default. When rain shortens an apiary visit and an unperformed Varroa mite wash is stored as `0.00`, subsequent inspections misread absence of measurement as verified absence of infestation.
Under SPEC-04 (Epistemic Data Standard v1.0), every observation field across Praksis relational tables and JSON contracts carries an explicit five-state provenance envelope: unmeasured in that window (`NULL`), structurally inapplicable to the current stage or category (`N/A`), unverifiable historical origin (`UNKNOWN`), noisy or anomalous reading held in quarantine (`SUSPECT`), and schema-validated observation with stated margin (`VERIFIED`).
This single invariant governs F01–F10 hive cross-sections in Artvin, the 28-case Clinical Apidology Atlas, and high-altitude GPX telemetry on Mount Sandras. Whenever required inputs are `NULL` or `SUSPECT`, decision calculators refuse synthetic averages and surface the exact epistemic gap to the operator.