the-budget-check-and-the-rate-limit-checkpoint

Summary

The rest of itd-2609201925079472 after spc-2609202134341288 closed: pieces 4 and 5 of that spec, which wait on a runner that reports its quota and its rate-limit responses (itd-2609201916056194).

• The budget check (criterion 7, from itd-29): where the runner reports remaining quota, an estimate from the spec's size is compared before the run starts, and a run it exceeds is refused naming both numbers with no state written; a runner that reports none is named and the check is skipped out loud.
• The rate-limit checkpoint (criterion 8, from itd-29): a runner's rate-limit response ends the window early for the whole run, since every lane spends the same budget; every lane with work in flight is checkpointed to its own branch, next_eligible_at is written once, and the record names the lane the response came from (spc-2609202134341288, "The pace and the fix rounds, per lane").

The pace, the window clock and the ceiling with its parallel lanes (criteria 1 to 6 and 9) are built; spc-2609202134341288's design sections remain the record for them, as amended below.

Amendments to spc-2609202134341288

The closed spec is not reopened; these rulings amend its design and its criterion C4 from 2026-10-02, and the loop is built to them.

• A step's worktree is made only when a helper is free (ruling DR6d-1). The product thinker, verbatim: "preparing parallel steps' worktrees: ONLY WHEN A HELPER IS FREE — a step's worktree is made just before an agent takes it; at most the agent ceiling's worth on disk." It reverses the eager opening "The count" and C4 carried ("A stage the binary performs itself proceeds at the ceiling", read as a ready step's worktree and brief made whatever the ceiling). A lane opens for a ready spec step only while a slot is left for its implementer beside every lane opened whose implementer is not out yet, and fewer step worktrees than the ceiling are on disk (a lane holds one from its worktree stage until it lands or is discarded); the opening call makes its worktree, and its brief and implementer follow. A step waiting for a helper has no worktree and waits under waiting as step <n>. C4 reads: a step made ready at the ceiling waits for a helper with no worktree, and its lane opens at the next free slot (as C6 already says of - needs: none). The stages the binary performs on a lane already open (its brief, a round's close, a landing step, a sync, a hold) still take no slot and are never held by the ceiling.
• A missing preflight receipt holds only its lane (ruling DR6d-2). The product thinker, verbatim: "missing preflight receipt for one lane: CARRY ON, WAIT SHOWN — other lanes continue; the lane waits for its receipt like the merge wait (contend, not refuse), and the status shows 'waiting for its full check (since HH:MM)'." The landing's push without a receipt naming the lane's head is a contention, as the forge's merge is: the call moves another lane and names the wait under blocked, and gives the wait (exit 3) only when nothing else moves. The time the wait began is written once, on the lane's landing as check_wait_since (a version-9 field), with a record entry, kept across calls and cleared by the push; the status shows it in UTC. The remedy still names how the receipt is minted. Landing stays one lane at a time: a landing waiting for its full check is a landing under way, so a sibling's landing waits behind it as it waits behind a merge. Who starts the check (ruling DR6d-2b) is a separate intent and is not built here.

Footprint

• packages: internal/core/implement/loop, the runner adapter itd-2609201916056194 delivers
• tests: a fake runner reporting quota over and under the estimate, one reporting none, and a rate-limit response mid-lane with two lanes in flight

Steps

No steps listed: the spec is built as one step. To split it, list the steps in order as 1. <title>, each with - packages: and - tests: indented beneath it; - landed: <pull request or commit> marks a step that has landed.