gstack-evals #36 round 3 projected Step 8 at 16.43M (12 calls x the calibrated
750k + one 7.43M dispatch) and measured 0.34M (ledger
ship-20261030T020847-c940ef); that inflation blocks capped runs. check now
records each Capy per-call charge (--reading minus a fresh --prior-reading) and,
once three are measured, prices a step at step_calls x their median over the
last five, unscaled; the calibrated per_call applies until then. The audit
dispatches of Steps 7, 8 and 14.5 are reserved by their own check, so they move
from step_dispatches to segment_dispatches. step_calls are refit from the
round-3 replay ledgers' step samples at 330k per call (replay3-34.md): Step 8
13, Step 14 5; two per_call observations join the fit (constant unchanged).
Records the ship size ledger for this branch.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
gstack-evals #36 round 3: a child misparsed a reservation id out of the
RESERVATION: line, which gave UNMEASURED and then BUDGET_CHECK_SKIPPED at close.
check keeps RESERVATION: and adds RESERVATION_ID: <id>; the pass program, Steps
9, 11 and 14.5 and the Capy skill pass that line's value to --reservation.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
thread_id "this" is the root thread, so a lane subagent's budget counted every
sibling tree (#3633: root +37.3M while the lane's own counter moved +22.9M). A
jam's own counter includes its own subagents (measured 2026-10-30 on
jam_01M658JBRMTMQ1WQWYT6WFD22W: +10.92M across a six-reader dispatch). The capy
contract now reads the agent's own jam id (scope thread, descendants), the
handoff child baselines at its own verify-handoff reading, and reconcile charges
the child's returned own-counter used=.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* feat(ship): reviewers read one packet; the parent reads only a compact findings record (GSTA-167)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* fix(GSTA-167): keep the pinned CODEX_BOUNDARY literal in the resolver; a test holds the packet boundary equal to it
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* fix(GSTA-167): a Step 7/8 audit is one packet, never sharded
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* fix(GSTA-167): packets survive the replay: base as given, budget roles, read= ranges, data summaries, no stale partial manifest
- --base origin/main no longer renders origin/origin/main (resolveBaseRef shared by build and audits)
- PACKET lines carry budget_role and read= ranges under 30,000 bytes; build prints BUDGET_NEXT for gstack-ship-budget --next
- each packet ends with END OF PACKET; the brief reads it with the read tool in parallel ranges, about 3 calls
- .json/.jsonl/.ndjson/.csv/.tsv additions past 400 changed lines reach reviewers as a DATA SUMMARY
- a partial build keeps other reviewers only at the same head, base and diff content
- collect: a matrix counts only from a state-machines reviewer; a CRITICAL line failing validation is invalid
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* fix(GSTA-167): pass-2 review fixes: data summaries read only the diff, a stale Red Team build refuses, any-case CRITICAL lines
- topLevelKeys takes keys and columns from the diff's own lines, never the working tree, which a symlink can point at a private file (outside voice CRITICAL)
- a --findings build whose head, diff base or content differs from the manifest refuses with STALE_MANIFEST instead of dropping the specialists' coverage (adversarial + Red Team)
- collect: a CRITICAL line in any case, or one that fails to parse, makes its reviewer invalid (Red Team)
- docs and the misplaced Value header in the review-packet tests
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* docs(GSTA-167): CHANGELOG entry for reviewer packets; ship-budget names BUDGET_NEXT as the --next value
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* fix(ship-budget): calibrate projections from the run's own meter, settle dispatch reservations when reviewers return (GST-101)
The /ship budget projected every step and dispatch from fixed constants
(7.43M per dispatch, 2.25M per default step) plus a fixed 4.48M reserve, so
real caps refused work that cost 3-20x less (GSTA-152 PR B, GSTA-2). Each
settled reservation is now measured on the next reading, and the run projects
with its own factor (max actual/prior of its last three measurements x 1.5),
the prior only before the first measurement. The default tail reserve scales
with the step factor; the stop overhead keeps its fit.
Reservations of reviewers admitted one at a time no longer pile up: check
--returned settles named dispatches, --finish settles the pass's dispatches of
that step, and a step change settles consumed dispatches. DRIFT and
BUDGET_DRIFT lines print projected against actual.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* fix: pre-landing review fixes (GST-101)
Unseen charges (settle --unknown, a Claude-metered outside voice) are booked
at their reservation and never measured, so they cannot shrink the dispatch
factor. The handoff holds back the same calibrated reserve as check, so a
delegate verdict is one the handoff admits.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* docs(ship-budget): handoff sub-cap subtracts unknown charges and the calibrated reserve
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* chore: changelog (Unreleased) for GST-101 ship-budget calibration
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* test(ship-budget): give the GSTA-152 PR B replay an explicit 120s timeout
It spawns about 100 CLI processes; on windows-latest it ran past the
shard runner's 30s per-test timeout (windows-free-shard 5, run
40917695070), and on 2 cores it sat at bare bun test's 5s default.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* fix(ship-budget): thresholds above the host call floor; a child hands off instead of stopping; parallel batches admit a prefix (GSTA-167)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* fix(ship-budget): a child returns incomplete while a grandchild is unreconciled or an ADMIT remainder was never re-checked; join a child's meter only after its handoff succeeds (GSTA-167)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* docs(troubleshooting): BUDGET_CAP_REACHED admits a fitting prefix of a reviewer batch (GSTA-167)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* chore: changelog (Unreleased) for GSTA-167 handoff floor
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* chore: move GSTA-167 changelog entry into [Unreleased]
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* v1.91.325.0 stamp
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* fix(ship-budget): calibrate projections from the run's own meter, settle dispatch reservations when reviewers return (GST-101)
The /ship budget projected every step and dispatch from fixed constants
(7.43M per dispatch, 2.25M per default step) plus a fixed 4.48M reserve, so
real caps refused work that cost 3-20x less (GSTA-152 PR B, GSTA-2). Each
settled reservation is now measured on the next reading, and the run projects
with its own factor (max actual/prior of its last three measurements x 1.5),
the prior only before the first measurement. The default tail reserve scales
with the step factor; the stop overhead keeps its fit.
Reservations of reviewers admitted one at a time no longer pile up: check
--returned settles named dispatches, --finish settles the pass's dispatches of
that step, and a step change settles consumed dispatches. DRIFT and
BUDGET_DRIFT lines print projected against actual.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* fix: pre-landing review fixes (GST-101)
Unseen charges (settle --unknown, a Claude-metered outside voice) are booked
at their reservation and never measured, so they cannot shrink the dispatch
factor. The handoff holds back the same calibrated reserve as check, so a
delegate verdict is one the handoff admits.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* docs(ship-budget): handoff sub-cap subtracts unknown charges and the calibrated reserve
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* chore: changelog (Unreleased) for GST-101 ship-budget calibration
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* test(ship-budget): give the GSTA-152 PR B replay an explicit 120s timeout
It spawns about 100 CLI processes; on windows-latest it ran past the
shard runner's 30s per-test timeout (windows-free-shard 5, run
40917695070), and on 2 cores it sat at bare bun test's 5s default.
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* ship: token cap with real meters and fresh-context delegation (GSTA-152 PR A)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* fix: pre-landing review fixes (stopped or lost capped run no longer blocks uncapped review starts; short child return reconciles incomplete)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* fix: carry estimate spend across a meter epoch change; calibration docs match their evidence
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* test: load the committed calibration file by its repo path so pregate maps it to a lane
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* ship: fix fence placeholders and register claude-usage fixture (GSTA-152 PR A)
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
* test: run ship-budget and review-log bins through their interpreters so Windows can launch them
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
---------
Co-authored-by: capy-ai[bot] <230910855+capy-ai[bot]@users.noreply.github.com>
