Bug:
The SQLite rewrite used seqNo directly for field_id, but appendField was not assigning seqNo when creating new fields during disc mapping. Some fields were appended with the default seqNo=0, which translated to field_id=-1 after the seqNo-1 conversion. Because (capture_id, field_id) is a primary key, the -1 rows collided, causing dropped/overwritten rows, shrinking field_record (missing fields) and triggering out-of-bounds errors in consumers like ld-analyse. JSON output was fine because it didn’t rely on field_id PKs.

Fix:
Ensure every appended field receives a contiguous seqNo (fields.size()+1) before being stored. Applied in both the TBC library (lddecodemetadata.cpp) and the json-converter copy (lddecodemetadata.cpp). This guarantees field_id stays 0..N-1 with no negatives or duplicates, so all fields persist in SQLite and downstream tools no longer see missing/invalid field numbers.