Found reviewing #260.
distinct is sorted least-similar-first and deduplicated — then used solely to print fifteen lines. The TSV is written from rows: corpus order, no deduplication, with the 1241 OK_* rows interleaved. Compare reports/research_grounding_drift.tsv:16 (similarity 0.67) with :22 (0.20).
So #260's claim that "the output sorts least-similar first, so ectoine leads and fumarate trails" describes the console summary, not the artifact. The 490-item backlog the PR is named for exists nowhere on disk — which is the whole deliverable, since nobody triages from a console scrollback.
Fix: sort and deduplicate the TSV itself, or emit distinct as a second file alongside the full row dump.
Found reviewing #260.
distinctis sorted least-similar-first and deduplicated — then used solely to print fifteen lines. The TSV is written fromrows: corpus order, no deduplication, with the 1241OK_*rows interleaved. Comparereports/research_grounding_drift.tsv:16(similarity 0.67) with:22(0.20).So #260's claim that "the output sorts least-similar first, so ectoine leads and fumarate trails" describes the console summary, not the artifact. The 490-item backlog the PR is named for exists nowhere on disk — which is the whole deliverable, since nobody triages from a console scrollback.
Fix: sort and deduplicate the TSV itself, or emit
distinctas a second file alongside the full row dump.