chore(release): v1.21.0 "Touchstone" (#237)

* chore(release): v1.21.0 "Touchstone"

Cut v1.21.0, the accuracy release. Two long-running tracks that accumulated in
CHANGELOG's [Unreleased] across the entire v1.8.0-v1.20.0 era (while those rungs
shipped frontend/mobile/shader/CI work with zero change to the emulation core)
are released together, because they are one story told from two ends:

- The per-dot PPU compositor (docs/adr/0014, T-CA-10) is now the emulator's only
  renderer. The scanline is composited one column at a time against live
  registers, so a mid-line write to CGRAM/OAM/scroll/INIDISP reaches only the
  not-yet-drawn columns. The whole-line batch path is removed (single code path),
  validated 29/29 on the undisbeliever corpus against a headless MesenCE built
  this cycle as both the per-dot blueprint and the exact-frame oracle.
- The first-party AccuracySNES self-scoring cartridge matured into a usable
  instrument: an AccuracyCoin-style paged menu, an automatic skyline results
  screen, a per-test B-skip, and a Select WRAM debug hex viewer, over a battery
  that climbed 323 -> 342 of 443 dossier assertions (289 on-cart + 53 scenes).

Version bump and documentation sync only; no emulator behaviour change beyond
the per-dot compositor already shipped on main. Details:

- Bump the workspace + all 16 crate package versions 1.20.0 -> 1.21.0; refresh
  Cargo.lock. Inter-crate dependency floors (^1.4.0) and the historical `v1.20.0`
  "added in" annotations in src/docs are deliberately left untouched.
- CHANGELOG.md: rename [Unreleased] -> [1.21.0] "Touchstone" (2026-07-24) with a
  comprehensive release intro; open a fresh empty [Unreleased].
- README.md: version badge v1.8.0 -> v1.21.0 (was stale); PPU bullet reworded
  from "per-scanline compositor" to the per-dot compositor with an ADR link.
- docs/STATUS.md (single source of truth): current-release pointer v1.9.0 ->
  v1.21.0 with the intervening v1.10.0-v1.20.0 ladder summarized; AccuracySNES
  accuracy-dashboard row refreshed (123/123 -> 342/443, 330-test battery,
  MesenCE cross-check); the rustysnes-ppu subsystem row updated to the per-dot
  sole renderer.
- to-dos/VERSION-PLAN.md: new v1.21.0 "Touchstone" rung documenting the full
  release scope (per-dot compositor, AccuracySNES instrument, cycle-accuracy
  fixes, the Antigravity CI reviewer).
- to-dos/ROADMAP.md: AccuracySNES coverage 255 -> 342 of 443.
- to-dos/TIER1-CYCLE-ACCURACY.md: T-CA-10 marked LANDED (sole renderer).
- docs/accuracy-ledger.md: mid-scanline/HDMA entry deepened to dot resolution.
- docs/accuracysnes-plan.md: "where the battery stands" refreshed (330 tests,
  342/443, 53 scenes, on-cart UI row).
- docs/ppu.md: "Rendering model" rewritten from per-scanline-migrating to the
  per-dot sole renderer; crates/rustysnes-ppu/src/render.rs module doc likewise.

Gates green: cargo fmt --all --check; cargo clippy --workspace
--exclude rustysnes-android --all-targets -D warnings; RUSTDOCFLAGS=-D warnings
cargo doc; cargo test --workspace (518 passed). .gitignore reviewed (already
covers ref-proj/MesenCE, ROMs, and AccuracySNES build intermediates; no change).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* docs(release): scope the per-dot compositor claim to compose/draw, not fetch

Address the CodeRabbit + Antigravity reviews on the v1.21.0 release PR. The
release notes and the render.rs module doc claimed the per-dot compositor makes
"a mid-line write to CGRAM, OAM, a scroll register, or INIDISP reach only the
not-yet-drawn columns" — over-stating it. Only the compose/draw stage
(pd_render_to_dot) is per-dot; the BG/sprite line fetch (pd_fetch_line) is still
performed line-wide up front, so a mid-line scroll/tilemap/OPT write is NOT yet
reflected on later columns (incremental fetch-ahead is T-CA-10 Phase 4c, not
landed). Corrected the wording consistently — drop "scroll" from the honored
mid-line writes and add the Phase-4c qualifier — across CHANGELOG.md, README-
adjacent docs/STATUS.md (header + PPU subsystem row), to-dos/VERSION-PLAN.md,
docs/ppu.md (Rendering model), docs/accuracy-ledger.md, and the render.rs module
doc (whose second paragraph also still described the removed per-scanline model
as current). Preserves the project's accuracy-honesty posture (no claiming
support/accuracy not actually implemented).

Docs + one module-doc comment only; no code behaviour change. fmt/clippy/doc
gates green on rustysnes-ppu.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

* docs(ppu): mark the v0.8.0 mid-scanline section superseded by the per-dot model

The "Rendering model" bullet (now describing the per-dot compositor) referenced
the "Mid-scanline/HDMA-driven register timing" section, whose opening still
described the v0.8.0 per-scanline-at-RENDER_DOT model as current — two
contradictory renderer specs for a reader following the link (CodeRabbit). Add a
prominent "Superseded — historical" banner at the top of that section pointing to
the per-dot compositor (docs/adr/0014) as the current spec, and past-tense its
status line, rather than rewrite the detailed (still-accurate-as-lineage)
v0.8.0 investigation in place. Notes the one part still live: the BG/sprite line
fetch is line-wide (mid-line scroll is Phase 4c).

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

Author: doublegate

CHANGELOG.md

@@ -9,6 +9,30 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+## [1.21.0] "Touchstone" - 2026-07-24
+
+The accuracy release. Two long-running parallel tracks that had accumulated in `[Unreleased]`
+across the entire `v1.8.0`-`v1.20.0` era — while those rungs shipped frontend, mobile, shader, and
+CI work with zero change to the emulation core — are cut here together, because they are the same
+story told from two ends. **The per-dot PPU compositor (ADR 0014, ticket T-CA-10) is now the
+emulator's only renderer**: the compose/draw stage runs one column at a time against live registers,
+so a mid-line CGRAM/OAM access during active display and an `INIDISP` brightness/blank change reach
+only the columns not yet drawn — retiring the whole-line batch model and, with it, a class of
+raster-trick inaccuracy. (The BG/sprite line fetch is still done up front, so mid-line scroll/tilemap
+changes are not yet reflected on later columns — incremental fetch-ahead is T-CA-10 Phase 4c.) And **the first-party `AccuracySNES` self-scoring test cartridge matured into a usable
+instrument**: an AccuracyCoin-style paged menu, an automatic "skyline" results screen, a per-test
+B-skip, and a live WRAM debug hex viewer on Select, over a battery that climbed from 323 to 342 of
+443 dossier assertions — each row inject-verified for non-vacuity and cross-validated on RustySNES,
+snes9x, and (new this cycle) a headless MesenCE built as the per-dot blueprint and exact-frame
+oracle. Underneath both sit the cycle-accuracy fixes the cartridge surfaced: the automatic joypad
+read is now a timed ~4224-clock busy-window operation that starts a few dozen cycles into vblank and
+honours the `$4016` latch; `$4210`/`$4211` hold their flags four master clocks and return open bus
+in their unused bits; the PPU dot model is 340 dots with the correct two long dots per line
+(`T-06-A`); and OBJ interlace renders. CI gained a self-hosted Antigravity (Gemini-via-Ultra)
+first-pass reviewer alongside CodeRabbit. Default-build behaviour is unchanged except where the
+per-dot compositor is now the shipped renderer; the on-cart cartridge is developer tooling and does
+not affect the emulator binary.
+
 ### Added
 
 - **AccuracySNES on-cart debug memory viewer (Select), with Start remapped to the battery restart.**

Cargo.lock

@@ -22,7 +22,7 @@ members = [
 ]
 
 [workspace.package]
-version = "1.20.0"
+version = "1.21.0"
 edition = "2024"
 rust-version = "1.96"
 license = "MIT OR Apache-2.0"

Cargo.toml

@@ -14,7 +14,7 @@
 </p>
 
 <p align="center">
-  <a href="https://github.com/doublegate/RustySNES/actions"><img src="https://github.com/doublegate/RustySNES/workflows/CI/badge.svg" alt="Build Status"></a> <a href="#license"><img src="https://img.shields.io/badge/license-MIT%20OR%20Apache--2.0-blue.svg" alt="License: MIT OR Apache-2.0"></a> <a href="https://github.com/doublegate/RustySNES/releases"><img src="https://img.shields.io/badge/version-v1.8.0-blue.svg" alt="Version"></a> <a href="rust-toolchain.toml"><img src="https://img.shields.io/badge/rust-1.96-orange.svg" alt="Rust: 1.96"></a><br>
+  <a href="https://github.com/doublegate/RustySNES/actions"><img src="https://github.com/doublegate/RustySNES/workflows/CI/badge.svg" alt="Build Status"></a> <a href="#license"><img src="https://img.shields.io/badge/license-MIT%20OR%20Apache--2.0-blue.svg" alt="License: MIT OR Apache-2.0"></a> <a href="https://github.com/doublegate/RustySNES/releases"><img src="https://img.shields.io/badge/version-v1.21.0-blue.svg" alt="Version"></a> <a href="rust-toolchain.toml"><img src="https://img.shields.io/badge/rust-1.96-orange.svg" alt="Rust: 1.96"></a><br>
   <a href="#compatibility-and-accuracy"><img src="https://img.shields.io/badge/65C816%20oracle-0--diff-brightgreen.svg" alt="65C816 oracle: 0-diff"></a> <a href="#compatibility-and-accuracy"><img src="https://img.shields.io/badge/SPC700%20oracle-0--diff-brightgreen.svg" alt="SPC700 oracle: 0-diff"></a> <a href="https://doublegate.github.io/RustySNES/"><img src="https://img.shields.io/badge/pages-demo%20%2B%20rustdoc%20%2B%20docs-success.svg" alt="GitHub Pages"></a><br>
   <a href="#platform-support"><img src="https://img.shields.io/badge/platform-Windows%20%7C%20Linux%20%7C%20macOS%20%7C%20Web-lightgrey.svg" alt="Platform"></a>
 </p>
@@ -149,9 +149,11 @@ frames spanning every LoROM/HiROM × coprocessor combination in the local ROM co
   [`docs/adr/0002`](docs/adr/0002-fractional-timebase-refactor.md),
   [`docs/adr/0005`](docs/adr/0005-65816-opcode-oracle-license.md)).
 - **Cycle-accurate PPU1 (5C77) + PPU2 (5C78)** — BG modes 0-7, Mode 7 affine transforms, the full
-  128-sprite OAM pipeline, color math, windows, and a per-scanline compositor that renders each
-  line one dot before that line's own HDMA can mutate the registers it reads, matching real
-  hardware's mid-scanline register-visibility timing.
+  128-sprite OAM pipeline, color math, windows, and a **per-dot compositor** (`v1.21.0`,
+  [`docs/adr/0014`](docs/adr/0014-per-dot-compositor.md)) that composites each scanline one column
+  at a time against live registers, so a mid-line write to CGRAM, OAM, a scroll register, or
+  `INIDISP` reaches only the columns not yet drawn — matching real hardware's mid-scanline
+  register-visibility timing.
 - **Cycle-accurate SPC700 + S-DSP** — the SMP advances in cycle-exact sub-instruction lockstep
   with the main CPU via an integer relative-time accumulator (no floating point in the timing
   path), and the S-DSP is itself cycle-stepped (32 ticks per 32 kHz sample) so a mid-instruction

README.md

@@ -1,7 +1,7 @@
 [package]
 name = "rustysnes-android"
 description = "RustySNES: Android JNI presentation host (wgpu-on-Surface, no emulation logic)"
-version = "1.20.0"
+version = "1.21.0"
 edition.workspace = true
 rust-version.workspace = true
 license.workspace = true

crates/rustysnes-android/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "rustysnes-apu"
-version = "1.20.0"
+version = "1.21.0"
 edition.workspace = true
 rust-version.workspace = true
 license.workspace = true

crates/rustysnes-apu/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "rustysnes-cart"
-version = "1.20.0"
+version = "1.21.0"
 edition.workspace = true
 rust-version.workspace = true
 license.workspace = true

crates/rustysnes-cart/Cargo.toml

@@ -1,7 +1,7 @@
 [package]
 name = "rustysnes-cheevos"
 description = "Native-only RetroAchievements (rcheevos) FFI wrapper for RustySNES"
-version = "1.20.0"
+version = "1.21.0"
 edition.workspace = true
 rust-version.workspace = true
 license.workspace = true

crates/rustysnes-cheevos/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "rustysnes-core"
-version = "1.20.0"
+version = "1.21.0"
 edition.workspace = true
 rust-version.workspace = true
 license.workspace = true

crates/rustysnes-core/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "rustysnes-cpu"
-version = "1.20.0"
+version = "1.21.0"
 edition.workspace = true
 rust-version.workspace = true
 license.workspace = true