6502tinymachines

N0 report: the contract exists, and two repositories hold their goldens through it

Run stamp: 2026-09-02, rustc 1.97.1. The console sketch (v0.2, in the 6502 repository until the nes repo exists) put N0 first: nes-bus as its own repository, the frame types every chip crate speaks, and a gate that is not a compile check but a golden check.

What closed

  • The crate: FrameParity, the NES frame constants and DotFrame (moved from ntsc-crt, which re-exports them); PpuPins, CpuPins and the two 40-pin DIP tables; CartEdge and the 72-pin table; the Cartridge trait with NROM; AudioSamples. No dependencies. Seven tests hold it to itself (encodings pinned by value, tables closed and positionally complete, NROM's CIRAM pins per the solder options, the /RD mutation flipping exactly one pin).
  • ntsc-crt adopts (its 051ef9d, pinned tag v0.2.2): the moved types became re-exports and nothing else changed. Its whole suite (58 tests) and its 54 self-counted doc claims stay green.
  • 2c02 adopts (its 58dc408): DotFrame is consumed from here, and the harness's CHR bus service reads the chip through a PpuPins frame extracted every half-step. The P1 node golden (3,624 states, the whole register program and 3,000 rendering half-steps) replays green through the frame with REQUIRE_GOLDEN_P1=1, all five workspace tests passing, goldens byte-identical to the P1 close.
  • The mutation gate, both directions: MUTATE=rd routes the CHR service through PpuPins::mutated_rd_for_proof (the contract lying about one pin's polarity) and the P1 replay goes red; MUTATE=1 (the CHR byte off by one bit) still goes red at _db0, state 623, the same divergence the P1 report stamped, proving the refactored path kept the old proof alive.

Decisions made en route

  • Analog pins are Option<f32> (vout, ad1, ad2): a digital extraction cannot sample them, and filling them with a number would launder a fake reading into a frame. None means not sampled; a consumer that needs one must refuse None by name. This was changed before any consumer existed (v0.1.0 to v0.1.1).
  • One revision of the contract per build. Every repository pins the same nes-bus tag, because two git sources of the same crate are two incompatible DotFrame types in one build. The bump sequence is: tag here, bump ntsc-crt and tag it, bump 2c02's both pins together.
  • Frames are instantaneous. No half-step counter in the structs; trace formats and comparison drivers belong to each chip repo's own golden machinery, the way v6502-pins wraps its frames in a Trace.

Authored, awaiting their gates

  • CpuPins and CartEdge have no consumer yet; their first gates are N1 (the 2A03 pin lockstep) and N4/N5 (the glue and the console). Until then the DIP and edge tables are claims from the nesdev wiki, dated in the source.
  • The ad/a_hi extraction in the 2c02 harness reads the internal address bus as the pad view, faithful to the reference's own handleChrBus; the full AD0..7 pad mux (address phase versus data phase) is the console's problem at N5, and the harness says so where it does it.
  • The sketch wants a check-self-counts.py row per milestone; this repository does not carry the tool yet. The claims above quote test counts from the runs they describe, and the tool arrives when the repo has enough prose to drift.

Pulled at build time from nes-bus/docs/n0-report.md; the repository is the one copy.