fix(catalog): harden evaluators against correct-answer phrasing variance

Both reviewers found evaluators that scored correct answers as zero:
hardcoded flag order, quote-style rigidity, missing multiline anchors,
comma-grouped numbers, ratio notation, and numeric strings in JSON.
A shared ExactToken evaluator now normalizes fences, quotes, and
trailing periods for exact-answer tests. The obsolete tmutil snapshot
reference now accepts the current localsnapshot verb.
This commit is contained in:
2026-08-17 02:48:05 +01:00
parent 8ab83a57ca
commit 7264fe81b7
32 changed files with 815 additions and 152 deletions
+4
View File
@@ -479,6 +479,10 @@ func TestGoContextCancellationTraceTest_Eval(t *testing.T) {
{"exact correct", "context canceled", 1},
{"correct case-insensitive", "CONTEXT CANCELED", 1},
{"correct with surrounding whitespace", " context canceled ", 1},
// A7 regression: fenced/quoted decoration on the correct answer
// must still score full credit.
{"fenced", "```\ncontext canceled\n```", 1},
{"quoted with trailing period", `"context canceled".`, 1},
{"wrong: assumes select falls to default", "not done", 0},
{"wrong: wrong sentinel error", "context deadline exceeded", 0},
}