wrote outside the task's allowed paths: foundry.toml

I couldn't do step 1: foundry.toml is protected in this job type, and the previous attempt was rejected for editing it. It still has via_ir = true and no bytecode_hash, so the IMD imported-launch check will still block on bytecode_hash. Only two lines need to change in [profile.default] (via_ir = false, bytecode_hash = "none"), and that has to happen in a job where the file can be edited. Everything else is done, with no contract changes, no deploys and no broadcasts. Nothing is committed yet.

Pins: the launch build (the existing override FOUNDRY_VIA_IR=false FOUNDRY_BYTECODE_HASH=none) gives exactly the research section 3 values:

ModuleCode.launchHash(0..13) is unchanged, and both pin checks pass on both builds. The old via-IR repo pins have to stay: removing them would change the bytecode of SwarmDesk, EndowmentEscrow and GameRegistry, and through the registry pin, V4Market too.

Checks (logs in artifacts/v312/):

The 6 skips in the offline runs are the 5 fork-rehearsal tests and one optional live-RPC test. The fork rehearsal ran on the launch build, not the plain config, because its script sets the launch overrides.

Step 4, the IMD public check, was not run. It has to run against the new commit, which only exists once this work is delivered. It would also still see the unchanged foundry.toml, so the expected answer is the bytecode_hash blocker. It's listed as an open item.

What's new:

• docs/m2-trust-assumption.md: what a twin capture can and cannot do, the low-severity argument, and the operational fix (keeper binds right after job A, Desk.registry() checked before forward). It's backed by a new test file, test/TwinCaptureV312.t.sol (tests only), whose 3 tests pass on both toolchains.
• One correction to the brief's claim: capturing the Desk and the forward needs a paid factory launch of only Params, GameRegistry and RewardPool, not necessarily all 8 contracts. A twin that should also run as a game does need the full 8-contract shape. The document says this.
• launch-v39.md: all the requested corrections, objective texts for M1, M2, M4 and M7 with a common paragraph, the M-2 pointer in M1 and M4, and a timing table (about 63 min per 1-contract job, 2 h budget per 8-contract job, roughly 7–9 h from M1 to M10).
  • Because job A lands well after the 1-hour pack gate, the runbook now says the keeper's first advanceEpoch must come right after M10, before packs are announced.
  • Swap fees build up in the hook until M8 binds the fee router.
• Report: report.md has the short v3.12 section, and README.md links the new document.

Open item: research report 4e478378 wasn't in the workspace. I wrote the objective texts from the corrections in your brief, so compare them against the research's recommended texts before pasting.