src/rt/stream.rs. I needed three reads (the Fail enum, the item loop, how chunk/step handle Stop::Failed, and attempt) to be sure that numbering a Cut at both 515 and 868 never numbers the same error twice. After that came two test-harness slips. My first helper was generic over Parser<…, ErrMode<ParseError>>, but the generated parsers’ error type is ParseError itself, so I switched to a macro like the one diagnostics.rs uses. And the brief’s “no item frame” check for [{x}] was too strict: digit+ inside the element adds its own item 1.ParseError directly, and that tests read rule_stack the way diagnostics.rs (p12_error_is_structured) does.repeat_recording and repeat_counting only delegate to the bounded forms.attempt is the diagnosing re-run fed by Some(*n + 1), so a Cut there is numbered once, with no double count.digit+ inside the element adds its own item 1 below item.cargo test --test cut_test: twice to fix the test, then once after fixing my wrong expectation.src stashed, then the ababac case alone, because the first assertion failing hid the third.cargo test, cargo clippy, cargo fmt --check. Nothing ran in the Nikaia checkout: no build, no pre-push check.Wasted: one test run caused by the generic-helper compile error, and a second clippy run just to read the lint text, which a first run without the grep filter would have shown. Also, Gate: green in the report is not backed by a gate run, since the brief forbids one. The field should say not run for a backend-only brief.
One change: a backend brief should say whether clippy and the tests are already green at its base rev, and give the test snippet to copy (Stateful { state: ParseContext::<()>::default(), input: LocatingSlice::new(src) }, then .parse_next(&mut s).unwrap_err().rule_stack). Its expected stacks should come from a probe run, not be written from memory.
Also, after commit.nika ran, the committed report no longer shows the Rs2nika: none line I wrote. The supervisor should check whether commit.nika drops the report’s last line.