This directory holds post-hoc audit reports that capture why a particular investigation
went the way it did, the hypotheses that were tried and ruled out, and the exact state an
unresolved bug was left in — detail too long-form for the per-subsystem spec docs. Modeled on
the same directory in ../../../RustyNES/docs/audit/ (a different project; only the
organizing convention is borrowed, not any NES-specific content). It is intentionally distinct
from its siblings:
| Directory | Captures | Audience |
|---|---|---|
docs/ (top level) |
WHAT the system does — per-subsystem specs (cpu.md, ppu.md, cart.md, etc.). |
Future maintainers building against the spec. |
docs/audit/ (this dir) |
WHY an investigation reached its current state — root-cause trail, ruled-out hypotheses, exact reproduction steps for an open bug. | Future Claude / future maintainers resuming a stalled investigation. |
docs/adr/ |
DECIDED cross-cutting architecture choices in Michael Nygard ADR form. | Same audience as this dir, but for settled decisions rather than open investigations. ADRs are short and decision-focused; audits are long and provenance-focused. |
spc7110-boot-crash-2026-07-08.md— the SPC7110 coprocessor's boot-crash investigation: thev0.4.0-landedbus_mirroraddressing fix (root cause #1, confirmed and fixed) and root cause #2 (the boot-blocking one) — closed as of the audit file's own conclusion: the local Far East of Eden Zero dump is an English fan-translation ROM hack (confirmed by SHA256 mismatch against the genuine cartridge dump), not a RustySNES bug; a correctly-sourced original-cartridge dump is the remaining, purely ROM-sourcing, gap (docs/rom-test-corpus.md). Seedocs/cart.md's SPC7110 entry for the current summary; this file is the full trail behind it.fractional-timebase-go-no-go-2026-07-11.md—v1.1.0's evidence-based assessment of whetherdocs/adr/0002's fractional-timebase refactor is warranted yet. Conclusion: no — every currently-documented accuracy residual is either a ROM-sourcing/validation gap or, in the one case that looked timing-adjacent (the CPU oracle'se1.edivergence), an inter-reference disagreement rather than a proven bug — none require sub-cycle resolution to close.