I found our FR10 reporting startup resubmitting every pending bracket-verification entry. Pending covers two things: never sent, and sent before Python crashed without saving the result. Or possibly sent; that is exactly what the file cannot establish. I have stopped using that automatic restart path while we sort out the bookkeeping.
What evidence survives outside that file for the affected attempt? The fixture job may retain an identity and state you can compare, but do not infer never started just because Python failed to save a result.
The installed job has an attempt identity. The old application file saved only the bracket label, which gets reused. We cannot match this particular pending entry confidently.
Who is handling the actual bracket while you investigate? The screen change needs to match that handover, not leave someone choosing whether to load it again
Have the integrator and job owner define reconciliation for that held case, including what the operator sees and who can decide the next step; this is not a decision the parser should make from an empty field.
Shift lead owns the held bracket and its identified location; integrator owns the job-state comparison. New draft saves an attempt identity before submission and distinguishes never submitted, unresolved submission and a reconciled result.
What happens if the process stops after saving that identity but before it can establish whether the request reached the other side? That is the gap worth testing deliberately.
Offline interruptions on both sides of submission keep that same attempt unresolved. A second restart does too. We can still distinguish a queued entry which never entered the submission step, but no old pending row is being silently recategorised that way.
Cover could identify it from the draft screen, but read result saved as permission to continue the physical job. We changed that to name the recorded verification result and left physical continuation to the agreed job response. Another walkthrough is due.