mirror of
https://github.com/garrytan/gstack.git
synced 2026-11-13 11:57:19 +01:00
Garry Tanandcapy-ai[bot]
c85dd0a799
* 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>
78 lines
2.5 KiB
JSON
78 lines
2.5 KiB
JSON
{
|
|
"schema_version": 1,
|
|
"note": "GSTA-152 plan item 8. Per-meter constants for /ship budget projections, fitted conservatively from metered runs: each constant is the largest observation times `headroom`, rounded up to `round_to` (lib/ship-budget.ts fitConstant). test/ship-budget-calibration.test.ts checks the constants equal the fit and that a leave-one-out fit on two runs is at or above the meter on the third. Savings claims count completed ships only.",
|
|
"headroom": 1.5,
|
|
"round_to": 10000,
|
|
"observations": {
|
|
"per_call": [
|
|
{
|
|
"run": "gstack-evals#19",
|
|
"value": 240000,
|
|
"source": "GSTA-152 plan v3 evidence: late parent calls charged 0.22M then 0.24M (solved from the 29.3M and 16.3M counter windows; derived)"
|
|
},
|
|
{
|
|
"run": "gstack-evals#26",
|
|
"value": 400000,
|
|
"source": "GSTA-152 plan v3 evidence: late turns cost about 0.4M each (root counter read after every step)"
|
|
},
|
|
{
|
|
"run": "gstack-evals#31",
|
|
"value": 500000,
|
|
"source": "GSTA-152 plan v3 evidence: the /ship child's context was about 0.5M per call at the handoff"
|
|
}
|
|
],
|
|
"per_dispatch": [
|
|
{
|
|
"run": "gstack-evals#19",
|
|
"value": 2400000,
|
|
"source": "GSTA-152 plan v3 evidence: about 2.4M per child, solved from the two counter windows (derived)"
|
|
},
|
|
{
|
|
"run": "gstack-evals#31",
|
|
"value": 4950000,
|
|
"source": "GSTA-152 plan v3 evidence: the /ship child and its 10 review dispatches account for about 49.5M (4-5M per dispatch)"
|
|
}
|
|
],
|
|
"stop_overhead": [
|
|
{
|
|
"run": "gstack-evals#26",
|
|
"value": 1650000,
|
|
"source": "GSTA-152 plan v3 evidence: the stop message and the parent's checks took the counter from 39.40M to 41.05M"
|
|
}
|
|
]
|
|
},
|
|
"constants": {
|
|
"per_call": 750000,
|
|
"per_dispatch": 7430000,
|
|
"stop_overhead": 2480000
|
|
},
|
|
"step_calls": {
|
|
"7": 12,
|
|
"8": 12,
|
|
"9": 12,
|
|
"default": 3
|
|
},
|
|
"step_dispatches": {
|
|
"7": 1,
|
|
"8": 1,
|
|
"14.5": 1
|
|
},
|
|
"segment_dispatches": {
|
|
"9": 4,
|
|
"11": 2
|
|
},
|
|
"units": {
|
|
"capy-tokens": {
|
|
"fit": "observations above (Capy root counter, children included)"
|
|
},
|
|
"claude-usage-tokens": {
|
|
"seeded_from": "capy-tokens",
|
|
"why": "no Claude Code-metered /ship yet; refit once one completes"
|
|
},
|
|
"estimate-tokens": {
|
|
"seeded_from": "capy-tokens",
|
|
"why": "the estimate meter projects with the Capy fit, which is at or above the meter on the metered runs"
|
|
}
|
|
}
|
|
}
|