Created launch.json with the accepted token metadata, Game arguments ["$owner", "$token"], and required pool values.

Validation passed: manifest and ABI checks, forge build, and all 53 tests. Only launch.json is changed for submission.

The review is complete and the tracked tree is untouched. Only the findings file and my scratch probe, both git-ignored, were added.

Outcome: the math in Game, Battlefield and CommandPower holds against the Math Precision, Boundary and Numerical Gap guides. I found one low-severity seam and no loss-of-funds defect. The findings file holds that one finding plus 22 coverage rows.

What I verified, by trace and by executed probes in test/scratch:

• Pro-rata settlement uses 512-bit mulDiv with floor rounding, the divisor is non-zero whenever a player has pending power, and a closed epoch's budget cannot change until every winner is credited. Sum of floors never exceeds the budget across a 512-run fuzz with three winners and mixed sponsor, cosmetic and recycled budgets.
• Pending power from an earlier epoch is credited at that epoch's budget before a later rewarded win relabels the account. A same-day second rewarded win is impossible under the 24-hour cooldown, so the relabel path cannot corrupt an open epoch.
• Refund, recycle and claim each move liabilities by exactly the amount transferred, with exact balance-delta checks. Refund subtracts only the caller's contribution, which is always a disjoint component of the budget, so no underflow is reachable.
• Battlefield integer fields are bounded well inside uint16 and uint8, the turn and deadline gates complement each other at the exact boundary, and the cosmetic price splits exactly 75 to 25.
• The launch token is unmodified OpenZeppelin ERC20 with the exact 10^27 supply.

The one finding (low): recycling carries rounding dust of a few wei into the next day, and the funding gate at src/Game.sol:266 only tests for a non-zero budget. A day whose only budget is 1 wei therefore marks battles as rewarded, every winner's share floors to zero, and the wallet's 24-hour cooldown is still consumed. The reproduction is executed in the scratch probe and listed step by step in the findings file, with a minimal fix that leaves the design intact.

Not reached: no launch.json exists in the tree yet, so the manifest row is marked unreached. The README's declared constructor order matches the Game constructor and the exported ABI.

The review is complete. The findings file is at the repository root, both snippets match the tree line for line, and no tracked file was changed.

Outcome. The permission model of Game is sound in the Access Control, Trust Gap and Asymmetry areas. I found one low-severity trust-gap defect and one informational documentation-to-manifest conflict. No critical, high or medium issue survived tracing.

What I verified

• The owner's only power is recovering forcibly sent ETH to the owner. No owner path touches deposits, credits, retired tokens or results. The vendored Ownable2Step is genuine OpenZeppelin v5.
• Every value-moving entry point pays or pulls only from the caller. The permissionless ones, accrue, expire and recycle, cannot change what any player or sponsor is owed, because a closed epoch's budget is immutable and dust recycling needs full accrual first.
• The paired surfaces are symmetric: fund and refund, claim and claimPrize, refundToken and refundPrize, the view functions and their mutating counterparts. Liabilities are decremented before every push, and exact balance checks make failed pushes roll back cleanly.
• The constructor assigns no role from the sender or origin, makes no external call, and the ABI exports match the compiled source.

Findings reported

1. Low. Any nonzero budget, even 1 wei, makes a day "rewarded". A rewarded win then burns the wallet's 24-hour cooldown while paying zero. Anyone can plant 1 wei on each of the next 30 days, and dust recycling creates such days on its own. My scratch test shows a victim who wins on a 1-wei day is locked out of the next, genuinely funded day until 23:35.
2. Info. The README names a literal owner address that the approved workflow lists only as a fee recipient. The manifest must write $owner, not that literal.

Coverage. All 17 listed entry points have a row, 16 hold and act and fund point at finding 1, with four extra invariant rows. The one unreached row is the manifest itself, which does not exist in the tree yet. The hard-coded IMD prize address agrees with the launch's paired currency in the supplied reference, but I could not check it against the chain table, which was not among my inputs.

Review complete. The findings file is written and valid, and no tracked file changed.

Outcome. One low-severity finding; all 17 listed entry points plus 7 extra invariant rows are covered, 21 holds and 3 finding, none unreached.

What I checked (Execution Trace, Periphery, First Principles). I traced every act branch through decode, Battlefield.step, _finish, CommandPower.record and _accrue, and every funding, refund, recycle and claim path against the epoch-freeze and liability invariants. The build and all 53 project tests pass, the ABI exports match compiled source, OpenZeppelin is vendored as plain files, and the token meets the protected floor. The pro-rata accounting is sound: a closed epoch's budget and total power cannot change until every share is credited, and every liability change is bracketed by an exact balance-delta transfer.

The finding (low). rewardAvailable at src/Game.sol:266 treats any nonzero budget as a funded day, and a rewarded win always starts the 24-hour cooldown. A griefer can fund 30 future days with 1 wei each, and recycled rounding dust does the same for free. The scratch test confirms the worst case: a player who wins at 23:44:59 on a 1-wei day earns 1 wei and is locked out of the entire following, well-funded day, because rewarded starts need 20 minutes before midnight. Suggested minimal fix: a minimum per-asset budget for a day to count as rewarded, or a player-chosen practice flag at start.

Not found. No fund-loss, wrong-recipient or stuck-state path in my area. The epoch check at src/Game.sol:291 is unreachable but harmless. The README's owner address could not be verified against a launch section I was not given, and no launch.json exists yet, so I left both out rather than report unsubstantiated.

The findings file parses, covers all 17 listed entry points plus three invariant rows, and the working tree has no tracked changes. Review complete.

Scope

Files read in full: src/Game.sol, src/lib/Battlefield.sol, src/lib/CommandPower.sol, src/LaunchToken.sol, all four test files and the fixture, README, RULES and INTERFACE docs, the protected deployment floor, the Pashov Mindset and the Economic Security, Invariant and Flow Gap guides. The project's 53 tests pass. I added a scratch probe under test/scratch/ (gitignored) that exercises two sponsors, split refunds, chained recycling across five days, late accrual through a second victory, and double-accrue attempts. No launch.json exists yet.

Severity counts: 0 Critical · 0 High · 0 Medium · 1 Low · 1 Info

Findings

[L-1] Reward flag cannot be declined; a 1-wei funded day burns the winner's cooldown. src/Game.sol:281 marks every session rewarded whenever today's budget is non-zero. Anyone can fund a day with 1 wei, a cosmetic funds tomorrow with 25 tokens, and recycled dust carries 1 wei forward daily. A rewarded win finishing after 23:40 UTC locks the player out of the whole next day, which may hold the real budget. Reproduced in the probe: 1 wei on day D, 10,000 IMD on D+1, player wins late on D, gets 1 wei and is ineligible for every hour of D+1. Suggested fix is a practice bit on the start command.

[I-1] README line 38 names the brief's 0.25% fee address as the owner. The constructor is correct and INTERFACE says $owner. The risk is only a manifest author copying the literal.

What held

The ledger is sound. Token balance equals liability plus retired, and IMD balance equals prize liability, with exact equality through every sequence I ran. Pro-rata shares floor so payouts never exceed a budget. Budgets are frozen once a day starts, refunds need zero power, and recycling needs full accrual. Accrual is idempotent and cannot be blocked. Reentrancy, fee-on-transfer and missing-asset paths were already tested and roll back cleanly.

Observations, not defects

Sybil splitting of the daily budget is documented as a trust assumption. The brief's 3-wins cap is unreachable under the stricter 24h cooldown, which README discloses. Every rational player picks difficulty 2, so power is nearly flat per wallet. The hardcoded IMD address differs from the brief's older one; README says the chain table overrides it, and the task's example pool uses the same address, which I could not verify on chain.

Added failure-path, fuzz, and multi-actor invariant tests.

forge build passed. forge test: 83 passed, 0 failed, 0 skipped.

Reported one low-severity prize-asset address conflict, with a reproduced failing proof, in .imd-findings.json. Live-chain asset validation remains outstanding.

The review is complete and the findings file is written at the repository root. No source, test, doc or manifest file was changed. The only new file besides the findings record is a scratch reproduction under test/scratch/, which is discarded.

Outcome: one low finding kept, four specialist reports merged into it, two dropped.

• Kept (low, merged from math, permissions, flow and economics). Any non-zero epoch budget marks the day rewarded in rewardAvailable at src/Game.sol:266, and a rewarded win always burns the wallet's 24-hour cooldown. A 1-wei fund call, a settled epoch's recycled rounding dust, or one cosmetic purchase is enough. A win after 23:40 UTC on such a day locks the wallet out of the whole following day, which may hold the real budget. I reproduced three scenarios in a scratch test that passes on the current code: a 1-wei day forfeiting a 1000-token next day for a 1-wei payout, two winners on a 1-wei day both receiving zero while on cooldown, and natural dust making an unfunded day "rewarded" with no attacker at all. No funds are lost and solvency holds, so the severity stays low.

• Dropped: prize-token address conflict (write_foundry_tests). I checked both addresses on chain. The hardcoded 0x5F7B…7127 has code on Robinhood Chain 4663, reports symbol IMD and 18 decimals, and equals the launch pool's paired currency. The brief's 0xD34a…63B7 is IMD on Ethereum mainnet and 0x5F7B…7127 has no code there. The README's chain-table override is consistent with the launch, so this is not a defect in the source.

• Dropped: README owner address (permissions and economics, info). The manifest already writes "$owner", so nothing inherited the literal. The README's chain-table claim about IMD proved correct, and I cannot substantiate that the owner line comes from a different source, so it was not reported.

My own pass found nothing further. I traced funding freezes, accrual rounding, refund and recycle ordering, the rewarded-finish epoch guard, cooldown and win caps, reentrancy and exact-transfer checks, ownership scope, action decoding, and the deployment model. Game's runtime is 15,579 bytes and initcode 16,047 bytes, within limits. The committed suite passes with 83 tests, the protected token test passes, ABI exports match the compiled contracts, and launch.json has the right kind, argument order, lowercase even-length hex and a string notes field. The coverage record answers all 17 entry points plus four invariant and manifest rows.

Implemented the playable game, wallet/iframe claims, cover, source and static export.

Passed: build, typecheck, 9 unit tests, 13 browser scenarios and 16 design checks. No live transactions sent.

Validation evidence documents results and limitations.

Commit blocked: .git is read-only. Completed files remain in the permitted paths for publisher collection.

Saved six findings in .imd-findings.json. Website files remain unchanged.

1. High — bug: Wallet victory results clip “View rewards” at 320px.
2. Medium — bug: Recruitment erases saved control groups and movement orders.
3. Medium — bug: “Retry checks” does nothing after configuration loading fails.
4. Medium — bug: An outdated quote can prepare approval of a different token from the confirmation.
5. Medium — bug: Move orders let Pepes enter solid headquarters geometry.
6. Low — fidelity: Briefings omit the requested grain, scanlines and stencil treatment.

Each finding includes an exact source citation, proposed change and measured reproduction.

Build, typecheck, nine unit tests, twelve Chromium scenarios, WebKit mobile interaction, manifest/ABI verification and live read-only RPC checks passed. The rebuilt manifest matches the submitted export. Wallet transactions were mocked; none were broadcast.

The visual idea is a solo Pepe strategy game presented as a neon-lit military miniature battlefield.

“Needs requester” below identifies original proposals incompatible with the supplied deployment or later economy-v2 requirements. These are not frontend findings.

Reproduced and fixed all six findings. Rebuilt dist/ and verified its deployment manifest.

Passed: build/typecheck, 12 unit tests, 18 browser scenarios and 16 design checks. Candidate bundle: 3.34 MiB.

Responses are in .imd-responses.json; evidence is in docs/REVISION.md. No live transactions were sent. Files are ready for collection; read-only Git metadata prevented a local commit.

All six previous findings are fixed. No new reproducible defect was substantiated. .imd-findings.json contains {"findings":[]}. Site files remain unchanged.

Build, typecheck, all 12 unit tests, manifest verification and 18 distinct browser scenarios passed across the initial run and targeted reruns. One recruitment setup timed out initially; the unchanged scenario passed with a longer timeout. Chromium and WebKit gameplay were exercised. Wallet writes were mocked. The rebuilt export is byte-identical to the submitted export.

The inventory below counts duplicated brief text once. “Needs requester” identifies original proposals requiring changes to preserved contracts or external confirmation; these are disclosed departures, not reopened frontend defects. The approved economy-v2 specification also supersedes the original multiplayer and token-emission proposals.