4990aab7-4d6a-424e-a524-8b5bd50ab5f0IMD JOB

Wrote [`.imd-findings.json`](/root/.identitymd/work/4990aab7-4d6a-424e-a524-8b5bd50ab5f0/e21d2002-e218-4072-97a9-30c35fa5a43f/.imd-findings.json). - **Low:** Constructor-allocation fuzz test incorrectly fails when `deployer == factory`. Reproduced with seed `2`; remains unfixed. - No critical/high findings. Confirmed fixed token supply, registry independence, matching ABIs, and no duplicate token deployment in the manifest. - Limitations: forced ETH and directly transferred tokens are unrecoverable; evidence authenticity and availability remain external. Existing project files were unchanged.

The manifest is done and committed. The reported check failure was not caused by launch.json. **What failed and why.** The failing test lives in an accepted test file, test/OracleChallengeToken.allowances.t.sol, outside my write scope. Its fuzz filter does not exclude the test's own factory address from the fuzzed deployer. When the fuzzer samples that labelled address, the test deploys a second token as the factory and then asserts the factory holds nothing, which contradicts its own setup. The token itself behaved correctly in that run. With the reported seed it fails deterministically. With eight fresh seeds it passed every time, so the verifier's run is seed-dependent on that one test. The one-line fix is an extra assume in the test, which only a source assignment may apply. **What I verified for the manifest.** - launch.json validates against the LaunchManifest schema and its refinements with a scratch validator in /tmp. - Token contract name, symbol, decimals and the empty constructor match src/OracleChallengeToken.sol and its ABI export. - The single application contract IMDOracleDisputeRegistry has no constructor, so its constructorArgs are empty, matching the ABI export. - Pool fields are native ETH, fee 3000, tickSpacing 60, and the standard sqrtPriceX96 with no hook. **What I changed.** I appended manifest check evidence and the test-suite observation to the notes field so the independent reviewer sees it. Notes remain well under the limit. **Final checks.** From a clean build in an environment with no variables, all 132 tests passed. The only changed path is launch.json, committed as 25355bf on the imd-submission branch.

Done. Three new test files are committed under `test/`, the full suite passes offline with an empty environment, and the working tree is clean. **Result** | Check | Outcome | | --- | --- | | `forge clean && env -i forge build --offline` | passes | | `env -i forge test --offline` | 132 passed, 0 failed (65 pre-existing + 67 new) | | New fuzz tests at 2000 runs | all pass | | `forge fmt --check test/`, `forge lint` on new files | clean | | `git status --porcelain` after commit | clean, only `test/**` touched | **What the new tests cover** - `test/IMDOracleDisputeRegistry.adversarial.t.sol` (33 tests). Deploys from an explicit pranked deployer and uses five locally derived wallets. Covers cross-wallet withdrawal attempts in every pairing, repeat withdrawal with a frozen timestamp, responses and withdrawals against withdrawn and nonexistent ids, all zero-field combinations with their declared error order, empty and oversized strings on both write paths, multi-byte UTF-8 exactly at and one character over the 280 and 512 byte limits, verbatim round-trips of arbitrary bytes, pagination edges plus property fuzz over arbitrary offset and limit, ETH rejected on every selector including a 2300-gas send, and fuzzed proofs that another account cannot alter a prior author's dispute or response and that random withdrawal attempts never touch any other record. - `test/OracleChallengeToken.allowances.t.sol` (25 tests). Metadata, exact 10^27 supply minted to whichever caller runs the constructor (fuzzed EOAs and a deploying contract), transfer chains, approve overwrite and clear, transferFrom with exact ERC-6093 arguments for insufficient allowance and insufficient balance, owner-as-spender still needing an allowance, max allowance not decremented, and fuzzed 24-step transfer and approval sequences that keep supply and the balance sum at 10^27. - `test/RegistryTokenIndependence.t.sol` (9 tests). Full open, respond and withdraw lifecycle from wallets asserted to hold zero OCTEST with no approvals, the sole token holder having no registry privilege, an accidental max approval never being spent, token transfers leaving registry records byte-identical, and a bytecode check that the registry runtime contains no call opcode and no ERC-20 selector or token address. **Defects** I read the registry and token adversarially and found no incorrect behaviour, so I did not write a findings file. Two behaviours are worth knowing but are documented design choices rather than defects: the registry stores invalid UTF-8 and NUL bytes verbatim since it only counts bytes, and a stranger calling withdraw on an already withdrawn dispute gets the challenger error rather than the status error. Both are pinned by tests as observed behaviour.

Committed as 18d4ca3 with a clean working tree. Both contracts, tests, ABIs and docs are in place. **What was delivered** - `src/IMDOracleDisputeRegistry.sol`: ownerless, non-upgradeable, append-only registry. Anyone opens a dispute (bytes16 request UUID, sourceChainId, snapshot hash, evidence hash, URI, rationale) with challenger recorded as msg.sender and sequential ids from 1. Anyone appends responses to an Open dispute with author recorded as msg.sender. Only the challenger can withdraw, and withdrawal changes only status and withdrawnAt. Zero commitments, zero chain id, zero request id, text outside 1-280 bytes, URI outside 1-512 bytes, bad ids and page limits outside 1-50 all revert with named custom errors. No payable, receive or fallback function, no external calls, no token reference. - `src/OracleChallengeToken.sol`: plain OpenZeppelin v5.2.0 ERC-20, name Oracle Challenge Test, symbol OCTEST, 18 decimals, no constructor arguments, 10^27 minor units minted once to msg.sender. No owner, mint, admin, upgrade, fee or hook. - `docs/abi/IMDOracleDisputeRegistry.json` and `docs/abi/OracleChallengeToken.json`, exported from the build artifacts and checked by `scripts/export_abis.py --check`. - README plus `docs/abi.md`, `docs/deployment.md`, `docs/security.md` and `docs/dependencies.md`. They cover hash encoding, public-data semantics, spam and Sybil limits, every state transition, the manifest facts (OCTEST is the launch token, the registry is the sole application contract with empty constructorArgs, deploy the token once, no LP or MerkleDistributor here) and operational responsibilities. - forge-std and the minimal OpenZeppelin ERC-20 closure vendored as ordinary files with a SHA-256 lock file. No submodules, no deployment script. **Verification** | Check | Result | | --- | --- | | `forge clean` then `env -i` offline build and test | 65 passed, 0 failed | | `forge fmt --check` | pass | | ABI export check and vendor hash check | pass | | Pinned protected token floor, run in test/scratch against real creation code | 6 of 6 passed | | Pinned protected project floor, same method with CREATE2 predictions | 2 of 2 passed | The test suite includes unit, fuzz and a stateful invariant test that mirrors every stored field across random open, respond and withdraw sequences and asserts nothing is ever edited or lost. The scratch copies of the protected tests were removed after the rehearsal, so the committed suite does not depend on any environment variable. **Left to later stages**: the manifest assignment writes launch.json from `docs/deployment.md`, the independent review follows, and services deploy on Sepolia and hand addresses to the frontend. No addresses are asserted anywhere in the repository.