Done. forge build is clean (no warnings) and forge test passes 27/27 — 19 new plus the 8 inherited starter tests, which still pass untouched.

What was built

src/FirstSwapTickRecorder.sol — a BaseHook with beforeSwap: true and every other permission false.

Sampling happens in beforeSwap, so the tick written for a block is the one the previous block left behind. That choice is the substance of the hook: a tick sampled after the first swap would be a tick the first swapper picked, so whoever lands first in a block could set it to whatever their capital reaches. Sampling before puts the value out of reach of everyone in the block it belongs to.

Storage is mapping(PoolId => TickSnapshot[]), append-only, in strictly increasing block order. A {uint48 blockNumber, int24 tick} struct fits one slot, so a recording swap costs one cold SSTORE. Reads:

• snapshotCount — 0 for a pool never swapped
• latestSnapshot → (bool recorded, uint256 blockNumber, int24 tick) — never reverts; the flag is what separates "no data" from a genuine tick 0
• tickAtBlock — binary search, never reverts for any block or pool; a quiet block reports false rather than a neighbour's value
• snapshots — the whole history
• snapshotAt — the one revert, SnapshotIndexOutOfBounds(poolId, index, length), because an index read has a defined domain

No fee (beforeSwapReturnDelta off, so v4 never reads a delta from it), no owner, no admin function, no delegatecall, no way to receive value.

One judgment call worth flagging: uint48 block numbers require a decision at the boundary. It refuses to write past block 281,474,976,710,655 rather than truncating, since a truncated block number would file a snapshot under a block that never happened and collide with real history. Unreachable in practice; both sides of it are tested.

On the tests

Since there's no separate reviewer here, I checked the suite rather than trusting it: I compiled four mutated copies of the hook in test/scratch/ — no same-block guard, overwrite-instead-of-ignore, one global history instead of per-pool, and record-in-afterSwap — and re-ran the identical suite against each. All four were killed (3, 2, 1 and 11 failures respectively). The per-pool mutant is caught by exactly one test, test_HistoriesAreIndependentPerPool, which builds a second launch pool on the same hook.

Beyond the four required cases, the suite covers: the recorded tick differing from the post-swap tick, quiet-block gaps, sells as well as buys, exactly one event per recording block, per-pool isolation, the swapper's delta being untouched and the hook holding neither balances nor ERC-6909 claims, the pool fee unchanged, beforeSwap rejected from a non-PoolManager caller, both ends of the block range, and a fuzz property over arbitrary block gaps and swaps-per-block.

Nothing outside the two granted paths changed; test/scratch/ is removed.