The repairs are done and committed. All 36 tests pass, forge fmt --check passes, and the build completes.

What was wrong. None of the failures were contract defects. All eleven failing entries came from six test-harness mistakes, five of them duplicated across the two token orderings.

• Three "did not revert" failures came from a view call placed directly after vm.expectRevert. The helper that reads the pool key, and one balance read, were the "next call" and succeeded. I hoisted those reads above the cheat.
• The replay test funded the copycat wallet with 10 IMD and then had it buy with 100. It now gets 110 IMD.
• The rebalance test assumed nothing would be trimmed on the first call. The cap decays by 300,000 PLEA per day, so the 90-minute warp before the first rebalance legitimately trims 18,750 PLEA. The test now asserts that trim and allows the small recovered IMD in the wall total.
• The fuzz mining test hit the expected tail of the mining lottery. With 14 flag bits a salt matches with probability 2^-14, so about 0.2% of deployer addresses have no match under the 100,000-try cap. I verified the counterexample address by mining past the cap: its first match is salt 107,554. The test now pre-mines for the predicted launch address and excludes addresses that cannot be mined within the cap, which the separate try-cap test already covers. It passes with the previously failing seed and two fresh ones.

Verification.

Operational note for the launch. The README already states that the on-chain salt mining can fail for roughly 0.2% of deployer addresses. If PleaLaunch reverts with MiningFailed during the launch rehearsal, the launch must be retried from a different deployer nonce so the launch contract lands at another address.

Done. The PLEA system is built, tested and committed on main (commit 4ecaf34). Clean rebuild, forge test (55 passing) and forge fmt --check all pass with solc 0.8.26, optimizer 200 runs, no via-IR, bytecode_hash = "none".

What was delivered

• src/PLEA.sol: 1e9 token, minted once in init, which only works in the deploy transaction via an EIP-1153 transient flag. Cabal transfer rules, add-only allowlist, firstReceivedAt, killCabal from the gate.
• src/CabalGate.sol: submitSell with the UTF-8/zero-width/Bidi/tag/variation-selector and [PLEA filters, fact score table, oracle body and canonical questionHash, EIP-712 deliverVerdict (domain "IdentityMD Oracle" v2, signer 0x5598…), 7-minute executeSell with full-fill check, appeals, 2h pending expiry, 48h dead-man. It is also the buy router, settling IMD before the swap.
• src/PleaLaunch.sol: on-chain CREATE2 salt mining in fixed-memory assembly (100,000-try cap), hook deploy, PLEA.init.
• src/PleaHook.sol: seed at 5,700 IMD cap with single-sided PLEA, exact-input buys delivered by the hook in afterSwap (swapper PLEA delta 0), fee split, Stacker cashback with 300k gas reserve and plain-IMD fallback, 90-minute launch fee and 5M cap, cap/trim/ratchet, IMD buy wall, tips only for ≥100 IMD moved, 25-slot checkpoints, cost basis.
• src/PleaDistributor.sol, relayer script (relayer/relayer.ts, Bun + cast, no dependencies), static site (site/index.html), README with launch parameters, operator steps and assumptions.
• Vendored forge-std, OpenZeppelin, Uniswap v4-core and solmate Owned as plain files; no submodules.

Previous failure avoided. The earlier attempt was killed during build (exit 137). This build uses no via-IR, lint-on-build off and v4-core's own non-IR path, and compiles in about 3 seconds.

Measured in tests. Whole launch ≈ 10.6M gas, PleaLaunch ≈ 4.9M, mining ≈ 0.42M for 2,472 tries. Expected tries are 16,384, so an unlucky salt may need up to ~17M gas; the cap fails with ≈0.2% probability and the launch simply retries.

Design points to know

• Uniswap v4 skips hook callbacks when the hook is the caller, so buys go through the gate, not the hook.
• When a third-party router settles after the swap on early trades, the PoolManager holds no IMD yet. The hook then books fees as ERC-6909 claims and defers owner fee and cashback; settleClaims() and claimCashback() redeem them. The gate's buy path never hits this.

Assumptions flagged in the README

• Stacker interface assumed as credit(address,uint256) pulling IMD from the caller.
• The body's consumer key follows the brief; the relayer may need to strip it if the HTTP door rejects unknown keys. Canonical question hashing is tested for self-consistency, not against a live PLEA attestation.
• OracleAttestation.sol uses isValidSignatureNow because the vendored OpenZeppelin lacks the calldata variant; the protocol vector and live attestation f7af4af1 still verify and recover 0x5598….

The work is committed and verified. forge build, forge test (44 tests across 5 suites, all passing, both token orderings) and forge fmt --check pass with solc 0.8.26, cancun, bytecode_hash = "none", no via-IR.

What was built

• src/PLEA.sol: 1e9 supply minted only in init, which works solely inside the deploy transaction via an EIP-1153 flag that is spent on first use. Transfer rules exactly as specified, firstReceivedAt, add-only allowlist, killCabal only from the Gate.
• src/PleaLaunch.sol: constructor mines a CREATE2 salt in fixed-memory assembly (cap 100,000 tries, NoSaltFound after), deploys PleaHook{salt} and calls PLEA.init. Initcode 22 KB, well under the 49 KB probe limit.
• src/PleaHook.sol (20.2 KB runtime): exact-input buys only, PLEA delivered by the hook in afterSwap so no ERC-6909 PLEA claims can exist, sells only from the Gate, 0.25% PLEA burn per trade, IMD fees 0.5/0.5/0.25%, launch window fee 70%→0% and 5M PLEA cap, cashback via stacker.credit{gas: gasleft()-RESERVE} with plain-IMD fallback, cap/trim/ratchet with burn and IMD buy wall, hourly 25-slot checkpoint ring, cost basis, hourly tip from the stream only.
• src/CabalGate.sol: UTF-8 validation with all forbidden classes, fact score table, full oracle body and canonical questionHash recomputation, EIP-712 verification in the Gate's own domain, pending expiry, 7-minute execution window, denied wait, single appeal, dead-man switch.
• src/PleaDistributor.sol, src/PleaRouter.sol (post-launch buy router for the site), script/relayer/relay.mjs (dependency-free, uses cast), site/ (Buy, Plead, Wall, Seed button), README.md with manifest values, operations and assumptions.

Tests cover the full launch in one call, init outside the deploy tx, fuzzed deployer addresses yielding exact 0x28CC flag bits, the try cap reverting, seed once with others unable to initialize or add liquidity, every bypass path (transfer, Permit2-style spender, direct Gate send, rogue sell router, hookless pool, claim-minting router, no-hookData router), fee splits, stacked cashback with paused and gas-starved fallbacks, trims/burns/wall, tip rules, checkpoints, text rules, amount limits, verdict checks and replay onto copied pleas, appeal, dead-man, the protocol's vector digest, and recovery of signer 0x5598… from live attestation f7af4af1.

Open items, documented in the README

• Stacker and TestSIMD on Sepolia are unverified and no Sepolia RPC was reachable here, so credit(address,uint256) pulling IMD from the caller is an assumption; a mismatch only degrades cashback to plain IMD.
• The canonical question JSON follows the brief and has not been checked against a live verdict for one of these bodies.
• The 100,000-try cap leaves a 0.2% chance per deployer address that mining fails; the launch would simply be retried.
• Measured mining cost is about 160 gas per try (3.47M for 21,568 tries), so a 30M launch gas limit is advised.