Oninit® BLT — printers (cross-source corroboration)

The same source application is run twice. In the first run only printers.4gl is configured; the schema DDL is held back on purpose. With no schema, BLT has nothing to bind the embedded INSERT / UPDATE / DELETE statements to — it can read the source, but it can’t prove the table exists. So the three corresponding database-operation findings drop one notch from high to medium.

Run A — source only, no DDL (baseline)

mode: recovery
inputs:
  source_4gl:
    paths: [examples/printers.4gl]
output:
  directory: ./report/

Without the schema, BLT can’t confirm the printers table exists. The three database-operation findings (the INSERT, UPDATE, DELETE in add_det, update_det, delete_det) sit at medium confidence — BLT saw them in source but couldn’t corroborate them against the schema.

Run B — source plus binary, no DDL (lift)

mode: recovery
inputs:
  source_4gl:
    paths: [examples/printers.4gl]
  binary:
    paths: [examples/Printers.4ge]
output:
  directory: ./report/

Adding the compiled binary as a second input changes nothing about the source. But BLT now extracts the embedded SQL strings the compiler baked into the executable. Those strings independently mention printers for the same three operations. That parallel evidence overrides the resolution gap, and every database-operation finding returns to high confidence with an explicit cross-source-corroborated marker on its payload.

Why the lift fires — and where it stops

BLT’s confidence model has a hard ceiling at high. Cross-source corroboration overrides the unresolved-cap rule (which would otherwise drop the finding to medium) but does not push past high; a high finding with corroboration stays high, with the marker set so the reviewer can see the parallel evidence.

Corroboration is also discriminating: the match is keyed on (table, operation), not just table. A binary SELECT on a table does not corroborate a source INSERT on it; only the matching operation counts. Two independent derivations of the same fact are stronger than one resolved derivation — and BLT makes that visible rather than collapsing it.