CF-1.1 / G1 GUARDED ANALYSIS
Physical measurements: 0. This kit contains no instrument evidence.

Use this runner for new records:
  python analyse_guard.py measurements.jsonl run-metadata.G1.json --output result.json

Copy run-metadata.G1.template.json to run-metadata.G1.json and fill it from
actual records. Run verify_guard.py to reproduce the synthetic software checks.
Empty measurement input still returns NO_EVIDENCE. Synthetic tests remain
synthetic and cannot create a physical-evidence result.

The original CF-1.1 analyse.py, protocol, CAD and KIT-SHA256.json are preserved
byte for byte. G1 adds finite-value checks for leakage, prestabilisation and
absolute pressure. Valid-input paired numeric results are unchanged. The new
KIT-G1-SHA256.json covers the complete combined kit, except itself.

LABORATORY INTERPRETATION BEFORE PHYSICAL ATTRIBUTION
Record actual outlet absolute pressure and inlet temperature with the raw
readings. A fixed differential head does not itself hold absolute density or
inlet conditions constant. A laboratory reviewer must quantify whether their
paired differences and calibration bounds could explain the flow contrast,
and account for those effects before attributing the result to cooling.
G1 reports the recorded absolute-pressure range. It does not automatically
normalise away these conditions or authenticate a calibration/review identifier.
Set boundary_condition_review_record to the identifier of that actual review;
without it a supported numeric contrast stays REQUIRES_LABORATORY_BOUNDARY_REVIEW.

KEEP REJECTED ATTEMPTS
Keep every raw attempt, unchanged, in an immutable raw-data archive. The primary
analysis file is a documented selection of the FIRST quality-valid complete
attempt for each paired block, using the frozen AB/BA order and canonical run
IDs. Preserve a selection manifest mapping each primary pair to its raw attempt
and listing all earlier quality failures and their reasons. Repeat only a
quality-invalid pair. Never replace a valid pair because its gain is small,
negative or unfavourable. Include that manifest in the boundary review record.
Submitting duplicates/interleaved retries directly as the primary file remains
invalid; retaining raw attempts separately does not mean deleting them.

Read START-HERE.txt, protocol.json and bench-review-notes.txt before obtaining
or fabricating equipment. Calibrate the actual installed chain and retain the
controls, as-built dimensions and raw source records. The software checks
consistency and the stated numeric decision; it cannot certify authenticity,
complete an experimental interpretation or qualify vehicle performance.
