Add the payer contract of this burn tracker to this repository: a Foundry project under contracts/ (its own foundry.toml, dependencies inside contracts/), with every other file unchanged (dist/, refresh/, package.json). The contract is BurnTracker, plus BurnTrackerTest, which inherits BurnTracker and adds nothing, because the first deployment is a test. A later launch from this repository deploys BurnTrackerTest on Ethereum mainnet with these constructor arguments: intake 0xa43e6f75ee006411f79ac1c84120606c2330de82, asset 0xd34a99bc0f67ae1bbd63c660e6d0b0dd03e263b7, openAction "job.open@job-1" and continueAction "job.continue@job-1" as bytes32 strings, bodyPrefix and bodySuffix = the UTF-8 bytes of the continuation template before and after its {{PARENT}} placeholder, cooldown 86400, staleAfter 172800; the owner role goes to that launch's payer. No token, no pool, no other contracts.

WHAT IT DOES. Its owner calls open(bytes body) once: the contract orders job.open@job-1 through IMD Intake v2 (0xa43e6f75ee006411f79ac1c84120606c2330de82) with that body, paying the Intake's live price in IMD (0xd34a99bc0f67ae1bbd63c660e6d0b0dd03e263b7) taken from the owner. That makes the contract the hosted site's payer; the plane accepts job.continue only from the address that paid the project, so only this contract can order later versions. After every completed version anyone may call updateBurntracker(bytes resultJson, uint256 maxPrice) with the result document (result.json) of the last completed job as its proof: the contract accepts the proof only if its keccak256 equals the result hash the Intake handed to onResult for that completion, reads the parent job id from it, and orders job.continue@job-1 with its stored body template around that id, paid by the caller at the live price, never above maxPrice (a higher price reverts before any token moves). Callers supply only this proof and cannot change what the swarm is asked to do.

INTERFACE (exact; existing tools call it): open(bytes) returns (bytes32 requestId); updateBurntracker(bytes,uint256) returns (bytes32); recover(string parentJobId) returns (bytes32); setResultHash(bytes32); onResult(bytes32 requestId, bytes32 resultHash, string uri); onFailure(bytes32 requestId, uint8 status, bytes32 reason, uint16, uint16, bytes); parentJobIdOf(bytes) pure returns (bytes); views price() and nextUpdateAt(); status() returns (uint8 state, uint8 pendingKind, uint40 lastDone, uint40 orderedAt, uint40 staleAt, uint32 cooldown, uint32 staleAfter); public owner, intake, asset, openAction, continueAction, bodyPrefix, bodySuffix, resultHash, lastRequestId; owner-only setBody(bytes,bytes), setActions(bytes32,bytes32), setIntake(address,address), setTiming(uint32,uint32), withdraw(address token, address to, uint256 amount), two-step transferOwnership and acceptOwnership; events Ordered(bytes32 indexed requestId, uint8 kind, address indexed payer, uint256 amount, string parentJobId), Updated(bytes32 indexed requestId, bytes32 resultHash, string uri), Failed(bytes32 indexed requestId, uint8 status, bytes32 reason, uint8 newState). States UNOPENED 1, IDLE 2, PENDING 3, STUCK 4; kinds open 1, continue 2, recover 3.

RULES. One order at a time. updateBurntracker needs IDLE and the cooldown passed since the last completion. Callbacks count only from the Intake, only for the order awaited, only once. onResult stores the hash, returns to IDLE and restarts the cooldown. onFailure status 1 (refused): a refused update returns to IDLE and restarts the cooldown, a refused open returns to UNOPENED, a refused recover goes to STUCK; status 2 (failed) goes to STUCK. open() is allowed again after the hosting job failed (STUCK) or stayed pending past its stale time. Blocked jobs never call back, so once an order has been pending past staleAfter (fixed when it is placed), or in IDLE or STUCK, the owner may recover(jobId), a continuation of the job it names paid by the owner, or setResultHash(hash) to adopt a completion whose callback was lost. A callback for an order no longer awaited is refused. cooldown 1 hour to 30 days, staleAfter 1 hour to 30 days.

MEASURED FACTS (do not adapt them away): the Intake calls back with exactly 200,000 gas, and since Glamsterdam a storage slot written from zero costs about 111,000 of it, so every slot a callback writes must already be nonzero from the constructor on. result.json is canonical key-sorted JSON whose keccak256 is the hash onResult receives, 750 to 4,545 bytes, with the key "jobId" exactly once and a lowercase UUID as its value; the contract rejects a document without exactly one such key or with a malformed id. The launch's empty-chain check deploys only this contract, so the constructor must not require code at the Intake or the token.

DECIDE, DO NOT ASK: a question blocks the job and costs the payer; where this text is silent, choose and report it.