FPC conformance: triage the ~237 untriaged skips into gap: / wontfix:

• Type: task (testing infra / frontend audit) — Track T files it; the per-test verdicts are P/B/A knowledge but the tagging lives in T's skip list.
• Status: done
• Opened: 2026-07-11 (follow-up to [[feature-testmgr-fpc-compare-and-web-dashboard]])
• Owner: —

Why

The dashboard's conformance page groups FPC-suite skips by tag, and the adjusted pass rate excludes wontfix: (tests that probe FPC internals / intentional divergence and can never pass). The taxonomy + rendering shipped, but the ~237 existing test/pascal-conformance/pxx.skip entries are all untriaged — no tag yet, so today they all read as "open". The honest picture (what fraction of the suite is a real pxx gap vs never-applicable) needs each entry classified.

What

Agentic pass over pxx.skip: for each untagged entry, read the test source + its recorded reason and decide:

• wontfix: <reason> — probes FPC-internal behaviour: RTTI/typeinfo layout, FPC-specific units (fgl, variants as FPC ships them), internal-error message text, FPC bootstrap intrinsics, pointer-value printing, anything where matching FPC would mean copying an implementation detail rather than a language feature.
• gap: <reason> — a real Pascal feature pxx doesn't implement yet (already clustered in [[task-pascal-conformance-long-tail]]: RTL gaps, generics holes, case-label edges, custom enumerators, dynarray ops…). Cross-link the cluster ticket where one exists.

Batch it (e.g. the investigator subagent per category prefix), commit the retagged skip list, and the dashboard reflects the split on the next idle bench.

Notes

• Pure skip-list edits — Track T lane, no compiler change, no gate beyond the conformance runner staying green.
• Deprioritized like its parent ([[task-pascal-conformance-long-tail]] prio 12): the burndown is paused, but this makes the reporting honest, which is cheap.

Log

• 2026-07-11 — filed.
• 2026-07-12 — resolved, commit 9114ea42.