Swarm Sweep — Fix Wallet Token Discovery

Fixed discovery and rebuilt dist/. The old flow depended on SwarmExit’s empty sale registry.

Verified 163 real holdings across 220 contracts. Typecheck, production build, 35 regression tests, 101 interaction checks, and 16 design checks passed.

Selling still requires the owner’s token/route configuration. Publication remains pending; an updated remote deployment was not verified.

Results and limitations

Not delivered: artifacts/live-discovery.json. Under artifacts/, only this step's named outputs leave the machine.

Saved five reproducible findings to .imd-findings.json. Implementation files are unchanged.

1. High — bug: held refunds are falsely reported as returned tokens. Actual contract event order overwrites the held-refund outcome.
2. High — bug: precise IMD quotes cause horizontal overflow. At 320px, an ordinary 18-decimal quote expands the page to 358px.
3. Medium — bug: rejected multicalls bypass isolated balance reads. Individual reads succeed, but the app reports discovery failure without attempting them.
4. Medium — bug: desktop route labels are clipped. The final “IMD” disappears from a supported two-leg route.
5. Medium — bug: mobile detection hides discovered wallets. An available EIP-6963 provider disappears when window.ethereum is absent.

Each finding includes exact source lines, reproduction inputs, measured evidence and the proposed correction.

Discovery itself works for the supplied holder: 220 contracts checked, 163 positive holdings, with independently matching raw balances and decimals at Ethereum block 26,202,610. Rewinding discovery recovered two subsequent launches. The original cause was using the owner-managed sale registry as the token-discovery universe.

Validation passed: production build, typecheck, 35 regression tests, 101 Chromium interaction assertions, 12 legacy tests, ABI integrity and the WebKit connection/discovery/sale-fixture flow. Sales used fixtures; no mainnet transaction was signed.

Fixed all five reproduced findings and rebuilt dist/.

• 42 regression tests and 129 browser checks passed.
• Typecheck, production build and 16/16 design checks passed.
• Mainnet verification detected 163 holdings with matching balances.

Recorded every verdict in .imd-responses.json and evidence in VALIDATION.md. Submission size is about 3.56 MB.

Remote publication and deployed-version verification remain pending the project publisher.

All five earlier findings are fixed and verified. The findings file stays empty.

Settled findings, each re-run against the rebuilt export

• Held refunds misreported (high): fixed. The reducer in web/src/lib/sweep.ts now groups the receipt events per token and gives RefundHeld precedence over TokenRefunded. The unit tests for both event orders pass, and the browser run shows BETA as Unsold with "Token refused the refund. Claim it below.", zero wallet balance, the exact held amount and an enabled Claim refund control.
• 320px horizontal scroll from a precise quote (high): fixed. With expected proceeds of 100.123456789012345678 IMD the page measures scrollWidth 320 in both the quote and the sold-outcome states. All 18 decimals are visible in the screenshot, and the mobile grid uses shrinkable tracks.
• Isolated balance fallback skipped on rejected multicalls (medium): fixed. With Multicall3 rejected on both RPC URLs through a real viem transport, discovery completes with four holdings and five isolated balanceOf calls. Raw and formatted ALPHA balances are correct.
• Desktop route pill clipped (medium): fixed. At 1440 the BETA pill wraps to two lines and reads "Uniswap V2 to ETH, then IMD" in full. The geometry assertion confirms the pill ends inside its cell.
• Mobile sheet hid an EIP-6963 wallet (medium): fixed. With the iPhone user agent and no window.ethereum, the sheet lists Mock Wallet and Coinbase Smart Wallet, connects, and discovers the four holdings.

Checks run this round

• Production build and type check: pass.
• Unit tests: 42 passed, 0 failed.
• Browser interaction script against the export in Chromium at 320, 390 and 1440: all checks passed, no console or page errors.
• taste@1 probe: 16 of 16 passed. One control is under the 44px comfortable size, which is within the bar.
• WebKit was not available on this machine, so the fixes are confirmed in Chromium only.

Brief requirements

• "Support Ethereum mainnet first": met, chain pinned to mainnet in the wallet config and sale guards.
• "Discover all eligible Swarm-launched token contracts using verified launch records ... or on-chain factory/deployment events": met, registry snapshot plus on-chain registry reads in the discovery module.
• "Do not rely on token names, symbols, hardcoded sample tokens": met, discovery matches contract addresses only.
• "Read ERC-20 balances and decimals correctly, including tokens with unusual decimals": met, DELTA with 6 decimals and the 24-decimal unit test.
• "Include tokens with positive balances even if they have no available sell route": met, GAMMA and DELTA rows shown disabled.
• "Display clear loading, empty-wallet, RPC failure, and partial-discovery states. Never silently report zero tokens": met, each state screenshotted in the interaction run.
• "Add automated tests covering launch discovery, wallet balance matching, zero balances, RPC failures, chain switching, and tokens without sell routes": met, the two test files plus the browser script.
• "Test against known Swarm-launched token contracts and a wallet known to hold them": met per VALIDATION.md and the live verification script, not rerun against mainnet this round.
• "Run the production build, type checks, and all relevant tests": met, rerun above.
• "Display each discovered token's balance, USD value when available, and sell-route status": met, token table.
• "Keep tokens without sell routes visible but disabled for selling": met.
• "Preserve sell-all and sell-selected functionality, transaction previews,