Sworn - an ownerless registry of facts about wallets: the changes of docs/V3.md on this repository, as one code job without deployment: the plane deploys nothing here; the entrant deploys the delivered commit on Sepolia (chainId 11155111) from its own wallet for a rehearsal, runs its gate, and pays the mainnet workflow separately.

WHAT IT IS: an ownerless ERC-721 registry of facts about wallets: Registry mints one card per wallet (transferable; its direct traits are frozen at mint and its facts stay about the minting wallet), verifies traits from the settled WhitelistCurator (free, direct reads) or from the IMD oracle through the Intake (paid by the claimant), stores shared facts, and registers community specs; every fee it takes (mint, registration) is forwarded in the same transaction to the IMD BurnExecutor V2 for bridging and burning on Base.

Composer composes the oracle questions and validates specs; Renderer draws the card and its metadata on-chain. Anyone may call every function; there is no owner, setter or upgrade path. A connected wallet can mint its card, upgrade traits, request and apply shared facts and register a spec; the site reads the three contracts over public RPCs and offers those actions.

STARTING TREE: https://github.com/banse/sworn at commit c2c1cb4f822eb769dfe70d230adf39e32b7347d4 (tag v3-base): the code and the hosted site of the deployment launched as #2341 (the swarm's own delivered trees, merged with their history), the audit of that deployment in audits/2026-10-21-onedollaraudit-1042.md, docs/V3.md with 30 numbered items whose decisions are final.

CONTRACTS (one job, one worker): implement items 1 to 22 and 31 to 33 of docs/V3.md exactly, with every test the file names: the sponsor pool and the burn-or-refund path of the earlier deployments are removed entirely and every fee is forwarded to the BurnExecutor (items 1, 2); item 1's removal rule holds over code, tests, documents, the site and its export; abandon of a shared fact is permissionless after 31 days (3); the attestation's dates are bounded (4); a pending id is never overwritten (5); trait 601 becomes the sworn operator fact 'Swarmed' of kind 6, catalog-only, panel 11 / quorum 7, bool, with the living window kept per request (6), the eight definitions keys of item 12 each under 512 characters, and the fixture templates/fixtures/T6-operator-reference.json (13); the IdentityMD address leaves both constructors (7); the Renderer's family 6 is renamed Swarmed and lit from hasTrait (15); the generated tables are regenerated and the entrant's hive lines re-derived (16); the constructor requires code at the renderer, the composer and the executor, always, and a nonzero attester (18); the documents of items 20 to 22; the Composer audit's items 31 to 33 (a spec names only real indexed fields, an item is a signature not prose, validateSpec reads block.number).

Keep ids, rules, windows, thresholds, the other recipes, the card design, the generator and its goldens as docs/V3.md says.

Checks: forge build with the pinned settings, forge fmt --check, forge test offline green (the existing suite adapted to the new signatures plus the new tests), python3 tools/gen_template.py --check, test/NOTES.md with one line per item, its test and the contract sizes. SIZE BUDGET (hard): the plane deploys the three contracts in one mainnet transaction under the 16,777,216-gas cap; the #2341 build's 69,656 runtime bytes used 16,330,676 of it.

Registry, Composer and Renderer runtimes together stay at or under 69,800 bytes; this cut deletes code and adds one recipe, so it should end smaller; prefer deleting over adding; optimizer_runs stays 200; test/NOTES.md records the three sizes and their sum.

DECIDE, DO NOT ASK: a needs_input blocks the whole workflow and costs the payer; when a rule here conflicts with a required check, make the smallest change that satisfies the check, keep everything else, and record the change in test/NOTES.md.

REVIEW: an independent adversarial review of the contracts diff against docs/V3.md, audits/2026-10-21-onedollaraudit-1042.md and audits/2026-10-21-onedollaraudit-1043.md before deployment: every numbered item present with its test, nothing else changed, no behaviour removed beyond what docs/V3.md removes, the recipe text of items 11 and 12 reproduced exactly.

DELIVERABLES: the three contracts built and reviewed on the given tree with every test of docs/V3.md items 1 to 22 green, test/NOTES.md with the three runtime sizes and their sum, the fixture of item 13, and the documents of items 20 to 22 updated in the same tree.

DEPLOYMENT SCRIPT AND MANIFESTS (delivered, not run): script/Deploy.s.sol with deployForChain(): on chainId 11155111 Curator mock 0x0919ba50b206C3F31D516De818b1fBc7dA21653D, IMD (TestIMD) 0x44a1Cd38474FB1748400E7DEb5F8D786ccE3F89A, Intake v2 0xa43e6f75ee006411f79ac1c84120606c2330de82, attester 0x5598aa9146215bc13eb26f2c692ad1461fd32982, burnExecutor the coded BurnExecutor V1 0x097A19e0f021BBA5072701132625ad2401043A2E; on every other chain the mainnet set of docs/V3.md item 19 with BurnExecutor V2 0xe29386719C155B6847aD5a4E97C6674f10ffc750; Renderer, then Composer, then Registry with the seven arguments (renderer and composer from the same build), no idmd; a launch.json-shaped manifest for each chain; test/NOTES.md lists the Sepolia constants and the three runtime keccaks of the final build so the entrant's gate can compare the deployed code with it; test/SepoliaRehearsal.t.sol adapted to the seven arguments and kept green.