Result: partial
Cause: none found. Only red: arm64 job, dsl_grammar::text_the_grammar_refuses_is_refused_at_the_block (2886 passed, 1 failed; run 41746282250). findings came back [] where NK1178 was expected.
Fixed: nothing; no commit (nothing half-made).
Tests: that test passes here at HEAD, alone and with the whole dsl_grammar module on a cold NIKAIA_CACHE_DIR; it passed on arm64 in the two develop runs before (41733893156, 41733702585). bootstrap_compiler and changes were green on this run.
Ruled out: the 9 unpushed commits do not touch the path. Their diff is parser async forms, NK2115 in parse_findings.nika, NK2120 in check_places/check_core_expr (Expr::Spawn only), tests and briefs. The path itself is check_core_a_block_over_a_grammar -> check_core_comptime_binding -> Host::run_grammar (crates/nikaia/src/build_time.rs) -> Workshop::run.
Suspect: an arm64 CI flake in the grammar workshop, a shared build-time-parsers dir under load. I did not confirm it. A refused run should give NK1178 in any case, so empty findings suggest a silent path (e.g. bt_not_folded when ran.found is false, or comptime_binding returning early).
Probe: rerun the arm64 job. If it fails again, run cargo test -p nikaia --test it dsl_grammar there with RUST_BACKTRACE=1 and print found and wall in Host::run_grammar.
Found: possible flaky silent skip of a grammar run (above), not reproduced; not from these commits. Note: my own cold-cache run of dsl/grammar tests filled the disk (2.3G free, 95%) and caused unrelated “No space left” failures; I deleted the scratch cache.
Files: none
Commit: none
