Context
The 2026-05-06 audit found that 52 of 83 remaining TC_FAIL files are actually parse failures. `home ast` reports success on these files; `home check`'s internal re-parse then rejects them with parse errors (often surfacing as `error(DebugAllocator) ... leaked` because the parser bails without freeing).
This makes the audit data confusing — files appear in the typecheck-fail bucket but actually fail at parse. It also means we've been undercounting parse coverage in every audit since #16 fixed the hangs.
Examples
Sample files where this happens (check via the audit TSV `~/Desktop/home-os-kernel-audit-2026-05-06.tsv`, filter `first_error` matching `DebugAllocator` or `Undefined variable`):
- `drivers/pci.home` — likely parse error on `var x: [u8; 32] = undefined` at module scope (the #35 sibling case with explicit `= undefined`)
- `drivers/ahci.home` — same shape
- ~50 more
Hypothesis
`home ast` and `home check` use different parse paths (or one short-circuits more aggressively than the other). Bringing them into agreement means:
- `ast` should reject what `check` rejects (so audits classify cleanly)
- OR `check` should accept what `ast` accepts (less likely — `check` is the stricter one)
The right fix is probably to have `ast` use the same parser code path `check` uses, surfacing the same errors.
Why this matters
Highest single-fix leverage on the post-#34 board. Reclassifies ~50 files cleanly, surfaces real bugs, makes audit data actionable.
Approach
- Find where `home ast` and `home check` invoke the parser. Compare code paths.
- Identify the divergence — is it different parser entry points? Different error filters? Different walk?
- Align: have `ast` propagate parse errors the way `check` does.
- Probably also fix the `DebugAllocator leaked` symptom: when the parser bails on an error, it should free its allocations cleanly.
Reference
- 2026-05-06 audit: `~/Desktop/home-os-kernel-audit-2026-05-06.tsv`