Implemented LaunchToken and DaemonSponsorPool, with vendored dependencies, ABI exports, and deployment documentation.

Verified:

• forge build, forge test, and forge fmt --check pass.
• 36 tests pass, including 12,800 invariant calls.
• All 8 supplied protected checks pass locally.

Direct token transfers are unrecoverable; use approval plus register or sponsor. Independent review, manifest generation, and deployment remain separate stages.

Added stateful accounting tests covering SPON conservation, epoch payout limits, failure rollback, multi-day gaps, and withdrawal settlement.

forge build and forge test pass: 40 tests, including 12,800 new invariant calls. No defects found in tested flows.

Created launch.json for Sponsor (SPON) and DaemonSponsorPool(["$token"]).

Schema and compiled-ABI checks passed. forge build succeeded; forge test passed all 48 tests, including eight protected checks.

Only launch.json is included in the submission changes.

Wrote one low-severity finding to .imd-findings.json. Project files remain unchanged.

• SPON left unclaimable: Directly transfer 7e18 SPON units to a fresh application, approve its bond, then register → checkIn → deregister. At genesis + 86400, call claim(0) → withdrawBond. The application retains 7 SPON while pool, reservedRewards, and totalBonds are all zero. Reproduced locally with Foundry. This breaks the stated accounting equality; other users’ liabilities remain intact. Existing invariant tests exclude direct transfers.
• Budget snapshot timing: With 70 minor units sponsored in epoch 0, epoch 1’s sponsor(70) → checkIn and reversed order both produce budget 10. Rolling precedes incoming sponsorship.
• Roll ordering: Empty budgets return completely; a budget of 10 with three check-ins returns one unit of dust before the next snapshot. Only the last touched epoch finalises, preventing duplicate returns across idle gaps.
• Double payment: Duplicate check-ins and claims revert. Historical claim flags survive re-registration; payouts update accounting before transferring, with reentrancy protection.
• Early bond escape: Deregistering at genesis + 86399 sets unlock epoch 1. Withdrawal before genesis + 86400 reverts; withdrawal at that boundary succeeds as required.

launch.json correctly supplies ["$token"] to DaemonSponsorPool(address). LaunchToken takes no constructor arguments. Neither constructor grants an administrative role, including any role the factory cannot exercise. Both ABI exports match compiler output.

The frontend is built, validated, and committed as bf92005 (working tree clean, detached HEAD as the workspace started). The submission bundle measures 2.6 MB, under the 8 MiB limit, with no node_modules or archives tracked.

What was delivered (all inside web/, dist/, docs/)

• App source in web/: Vite 8 + React 19 + TypeScript 5.9 + wagmi 3 + viem 2, with package-lock.json and web/.gitignore. One page covering connect, switch/add chain, approve + register, check in, approve + sponsor, past-epoch claims from chunked EpochRolled/CheckedIn events, deregister with inline confirmation, bond withdrawal, live epoch countdown, and a read-only launch-pool quote through the network block's Uniswap v4 quoter. No in-page swap, as the approved workflow specifies.
• Single configuration source: the app loads dist/imd-deployment.json at runtime, fetches the referenced ABI JSON, and refuses to start unless each ABI's canonical Keccak hash (compact JSON, sorted keys) matches the attested hash. wallet_addEthereumChain parameters are derived from that file's network block and were verified byte-identical to the vetted walletAddChain.
• Static export in dist/ with relative base, plus the manifest listing all six other files with lowercase SHA-256, the exact contract set/addresses/ABI hashes, and the unchanged network block. verify-dist passes.
• Docs: web/README.md (configuration, install, build, preview, rebuild, tests), docs/DESIGN.md, and docs/frontend-validation.md with ten screenshots in docs/validation/.

Validation results

• Typecheck and build pass; Vitest runs 27 tests in 4 files against a mock EIP-1193 wallet and in-memory contract model (disconnected state, wrong chain with 4902 → add → switch again, approval gating, decoded BondLocked revert with no transaction sent, wallet rejection, full lifecycle through claim and withdrawal).
• Browser review ran in real headless Chromium via a bounded foreground Playwright script serving the final export under a /preview/ subpath: live Sepolia reads at 1280/768/320 px and a full mock-wallet walk. Zero console errors, zero failed requests, no overflow, all measured contrast pairs at or above 4.94:1, every tab stop with a visible ring, spinner static under reduced motion.
• The review caught one real bug, now fixed: viem's cached block number left a just-mined transaction's events out of the history scan for up to 12 s. Also fixed: countdown anchored to chain time, claim buttons state the amount, larger link hit areas.

Deviations and limitations to note

• DESIGN.md is at docs/DESIGN.md, not the repository root, because the root is outside the permitted write paths.
• No real transaction was broadcast and no physical wallet extension was used; screen-reader, axe, native 200 % zoom, RTL and forced-colors checks were not performed and are recorded as such.
• The Playwright browser_run_code_unsafe tool was not permitted, so the browser inspection used the sanctioned self-contained script instead of the MCP browser tool.