The _fp_auto_create_job grouping key was (recipe, part, coating).
Lines that shared all three but differed in thickness (or serial)
silently collapsed into one fp.job — the second line's thickness/SN
was lost, and any downstream cert printed the first line's values
across both batches. Silent mis-attestation = compliance hole.
Extended the key tuple to (recipe, part, coating, thickness, serial).
Single-line SOs and same-(thickness, SN) multi-line SOs collapse
identically to before. Only lines that previously merged when they
shouldn't have now split into their own fp.jobs.
TDD via test_so_confirm_splits_by_thickness:
- seeds the part with default_process_id so both lines hit the
`if recipe:` branch (where the bug lived — the no_recipe branch
already split correctly per line)
- confirms 2 jobs after action_confirm with each carrying its
own thickness via the linked SO line
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
e4681a58c6
