Should corpus labels record that v1 marked a test expected-to-fail?

[derek73]

corpus.jsonl's labels say which v1 test each name came from, but not that the test was @expectedFailure/xfail — so a future radar diff on Dr King Jr ([v1: test_king]) reads the same as a diff on a name v1 parsed correctly, when it's more likely a known-bad parse finally improving.

The marker is in the AST at the pinned ref; a label prefix (xfail:test_king) is a format-only regeneration with the name set proven identical, same as the labels commit. The strict-xfail unit tests already make an accidental fix loud; this is only about radar triage reading right.

[derek73]

Superseded by the xfail triage (PR #493, merged as 7dbb9bf; decisions.md#v1-xfail-triage).

This issue's premise was that the v1 xfail marker is recoverable triage metadata — a radar diff on a marked name probably means a known-bad parse improving. After the triage that premise no longer holds: every remaining xfail cites the issue that owns it (#489, #490, #492), and every retired one is a deliberate pin of a decided reading — so a radar diff on Dr King Jr now means a decided reading moved, which is exactly the signal an xfail: label prefix would have wrongly suppressed. The marker no longer tracks disposition; the tracker and the pins do.