Evidence model and rules

Evidence tiers

TierMeaningProduced today?
MEASURED_ON_DEVICEExecuted on a named target with logs and samplesNo
MEASURED_HOST_ONLYTimed on the machine running the toolYes (toy benchmark)
IMPORTED_MEASUREMENTSupplied by someone else with provenancePlanned
STATIC_MATCHDeclared manifests agreeYes
MODELED_ESTIMATEFrom explicit assumptions and formulasNo
NOT_RUN / UNKNOWN / BLOCKED / FAILED / HUMAN_REVIEW_REQUIREDHonest absence or stop statesYes

Language rules

  1. Never write "runs on", "supported on" or "optimized for" a device without MEASURED_ON_DEVICE.
  2. INT8 / INT4 are precision labels, not speed or power claims.
  3. Missing values are NOT_MEASURED, never zero.
  4. Same file name with different bytes is a different artifact (SHA-256 identity).
  5. Latency, throughput, quality, memory, energy and cost are separate metrics.

Governance rules (in rules/)

compatibility-tier, evidence-tier, input-trust, measurement-truth, privacy-offline, quality-parity, quantization, release-control, reproducibility, scope-and-permissions, security.

Finding catalogue

IDSeverityTrigger
ES-COMPAT-MODEL-FORMATblockerRuntime does not list the model format
ES-COMPAT-OPERATORSblockerDeclared operators missing from runtime
ES-COMPAT-PRECISIONblockerRuntime lacks model precision
ES-COMPAT-MEMORY-ESTIMATEblockerDeclared memory exceeds device availability
ES-COMPAT-ARCHITECTURE / ES-COMPAT-OSblockerTarget mismatch
ES-UNKNOWN-*reviewA check could not be decided from declarations
ES-NOT-RUN-DEVICEreviewAlways present: no device execution
ES-NOT-MEASURED-ENERGYinfoAlways present
ES-SYNTHETIC-INPUTinfoFixtures or declarations, not observations