docs(ledger): characterise the silent cohort — they halt, each differently

[doublegate]

Docs-only. Characterises the titles issuing zero RDP commands, and refutes a lead by measurement.

They are not stalled — they have stopped

Dumping what actually reaches WB at one-cycle granularity:

Title	Retiring instruction	Status	Reading
Banjo-Tooie	B -1 @ 0x800136A0	IE=1, IM=0xFF, IP=0	self-loop, interrupts enabled, nothing pending — nothing can release it
Donkey Kong 64	BNE r3,r2,-1 @ 0x80000A04	IE=0	spin-poll with interrupts disabled, delay slot is a NOP so neither register is reloaded — unexitable by construction
Jet Force Gemini	B -1 @ 0x8003F5D8	IE=0, IP=0x04	halted with an RCP interrupt pending but masked; still servicing the odd exception (SD t0,88(k0) retires)

This is the same shape Super Mario 64 had before the PIF fix — a game that hit an assert and gave up. DK64 is the most distinctive and the best next target: a loop that cannot exit on its own terms.

osMemSize refuted as a lead

RI_SELECT was seeded to 0x14 — the documented IPL3 configuration (TSEL = 0b0001, RSEL = 0b0100, N64brew RDRAM Interface §RI_SELECT) — so IPL3 would take the RDRAM-already-up branch.

Nothing changed. osMemSize stayed 0, the silent titles stayed silent, SM64 was unaffected. Combined with SM64 rendering its title screen with osMemSize == 0, the lead is now doubly weakened. Refuted by measurement rather than left standing as a plausible story that later work would build on.

What this means for the plan

The remaining silent titles are not one shared fix. Each reached its halt for its own reason, exactly as SM64 reached its assert because osContInit miscounted controllers. The method that worked there is the one to repeat: find the branch into the halt, read what it tested, trace that value back.

Gates

cargo fmt --all --check · cargo test --workspace · markdownlint — green. Docs only, no behaviour change.

[coderabbitai]

Important

Review skipped

Review was skipped as selected files did not have any reviewable changes.

💤 Files selected but had no reviewable changes (1)

• docs/accuracy-ledger.md

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: ASSERTIVE

Plan: Pro Plus

Run ID: 78ee71f3-f090-4ec9-9e4a-bfc306b012dc

📥 Commits

Reviewing files that changed from the base of the PR and between fb81c26 and 0615d75.

📒 Files selected for processing (1)

• docs/accuracy-ledger.md

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

_{Comment @coderabbitai help to get the list of available commands.}

[doublegate]

Adjudication — Antigravity review

No blocking issues.

Register naming — adopted, and it improved the finding

Correct: the rest of the row names registers by ABI (t0, k0, s3) while the DK64 snippet used the raw r3/r2 my probe's disassembler emitted.

Fixing it turned out to be more than consistency. r3/r2 are v1/v0 — the return-value pair, so the loop is comparing two returned values, which points straight at the way in: trace what last wrote them. The ledger now reads:

Donkey Kong 64: BNE v1, v0, -1 at 0x8000_0A04 with IE = 0 (registers given by ABI name, as elsewhere in this row; the raw encoding is r3/r2, and that they are v0/v1 — the return-value pair — is itself a hint: the loop is comparing two returned values, so tracing what last wrote them is the way in)

The raw encoding is kept alongside so the disassembly can still be matched against the probe output.

1,800 characters onto an already long single-line row — acknowledged, tracked

Agreed, and this is the third independent time it has been raised (also #188, #193). It is task #49: extract all long residuals to docs/residuals/R-NN.md in one pass, preserving every superseded claim verbatim, with the table linking to them.

Not doing it as a rider on a docs change about halt diagnosis — and worth noting the concrete harm is now demonstrated rather than theoretical: CodeRabbit found on #194 that unescaped pipes in this row were corrupting the table (MD056: extra data will be missing), silently dropping content from the end of the row, which is exactly where the newest appends land.

[github-actions]

Antigravity review (Gemini via Ultra)

This pull request updates docs/accuracy-ledger.md with low-level CPU trace findings for three non-rendering titles (Banjo-Tooie, Donkey Kong 64, and Jet Force Gemini), characterising the specific self-loops where each halts.

Blocking issues

None found.

Suggestions

• docs/accuracy-ledger.md: Appending ~1,700 characters to a single markdown table cell on line 413 makes raw file maintenance and git diffs difficult to parse. Consider breaking extensive investigation logs into a dedicated document under docs/ and referencing it from the ledger table.

Nitpicks

• The PR title uses British spelling (characterise), whereas the ledger entries mix American and British variants (characterised, rasterised, color); standardise spelling rules across documentation for consistency.

_{Automated first-pass review by agy on a self-hosted runner -- not a human review.}