Check Byte Order

Check Byte Order

Decode one sample into a complete candidate table.

Follow ../../references/interaction-contract.md.

Process

1. Read references/sample-identity.md and references/layouts.md.
2. Require raw 16-bit words and the point, route, unit, area, and protocol-offset identity.
3. Match the requested datatype width to the sample before decoding. Reject a 32-bit request when only one register was captured.
4. Run python3 <skill-dir>/scripts/run.py --input <capture.json> --types uint32,int32,float32 --output <evidence.json> with the applicable datatype family. When the source already establishes the datatype, pass only that type (for example --types float32). Enumerate every layout without adding unrelated datatype interpretations to the user's answer.
5. Evaluate every supported layout for the sample width. For one-register integers, state that byte and word order are not applicable and report protocol order AB without presenting a misleading BA interpretation.
6. Apply scaling only after raw decoding.
7. Report NaN, infinity, subnormal, range, and stability evidence.
8. Eliminate candidates contradicted by explicit engineering constraints. If exactly one remains, present the proof and one scoped confirmation; otherwise ask once with the complete shortlist and distinguishing evidence needed.
9. If the question is coil or packed-bit numbering, stop. That is a map hold for check-map, not a byte-order candidate table.

Output files

• evidence.json - Open this candidate table. It says whether word order applies, then shows only layouts supported by the sample width. It records evidence only; it does not change the map.

Completion requires every candidate to share one sample_id; the evidence selects no winner.

Stop

• Stop when no raw words or complete sample identity exist.
• Incomplete sample identity routes back to capture-sample, not to applying a layout.
• Stop for coil or packed-bit numbering; do not treat it as ABCD/BADC/CDAB/DCBA.
• Do not select a winning layout.
• Stop for writes, broadcasts, discovery, or unbounded polling.

Handoff

• No raw sample exists: suggest capture-sample.
• The uncertainty is coil or packed-bit numbering: suggest check-map.
• The user confirms a layout: suggest apply-review, then plan-reads.