From 00aba4e13cb6bb8f98c1475469fab356c9960a43 Mon Sep 17 00:00:00 2001 From: garrytan Date: Thu, 29 Oct 2026 16:42:40 +0000 Subject: [PATCH] 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> --- lib/ship-budget.ts | 12 ++++---- lib/ship-handoff.ts | 4 +-- test/ship-budget-projection.test.ts | 44 ++++++++++++++++++++++++++++- 3 files changed, 52 insertions(+), 8 deletions(-) diff --git a/lib/ship-budget.ts b/lib/ship-budget.ts index 6768a01ad..ad93c3291 100644 --- a/lib/ship-budget.ts +++ b/lib/ship-budget.ts @@ -51,7 +51,7 @@ export const METER_CONTRACTS: Record = { export interface Reservation { id: string; kind: 'step' | 'dispatch' | 'child'; step: string; next: string; roles?: string[]; tokens: number; at: string; state: 'open' | 'consumed' | 'settled' | 'released'; token?: string; started_at?: string; settled_at?: string; - prior?: number; reading_at?: number | null; epoch?: number; actual?: number; + prior?: number; reading_at?: number | null; epoch?: number; actual?: number; unseen?: boolean; } export interface Sample { id: string; kind: 'step' | 'dispatch'; epoch: number; prior: number; projected: number; actual: number } export interface Handoff { @@ -172,12 +172,14 @@ export function projector(l: Ledger, cal: Calibration = loadCalibration(l.calibr /** * Measure settled reservations against the current (fresh) reading: a step * from its reading to now; settled dispatches as one group from the earliest - * of their readings, split by role. Returns one DRIFT line per measurement. + * of their readings, split by role. A charge the meter cannot see (settled + * --unknown, or a Claude-metered outside voice) is never measured. Returns one + * DRIFT line per measurement. */ export function measureSettled(l: Ledger): string[] { const last = l.meter.last; if (last === null) return []; - const pending = l.reservations.filter(r => r.state === 'settled' && r.kind !== 'child' && r.actual === undefined + const pending = l.reservations.filter(r => r.state === 'settled' && r.kind !== 'child' && r.actual === undefined && !r.unseen && typeof r.reading_at === 'number' && r.epoch === l.meter.epoch && last >= r.reading_at); const dispatches = pending.filter(r => r.kind === 'dispatch'); const from = Math.min(...dispatches.map(r => r.reading_at!)); @@ -277,10 +279,10 @@ export function settleIn(l: Ledger, r: Reservation, opts: { unknown?: boolean; c if (r.state === 'settled' || r.state === 'released') return; r.state = 'settled'; r.settled_at = new Date().toISOString(); - if (opts.unknown) { l.unknown += r.tokens; return; } + if (opts.unknown) { l.unknown += r.tokens; r.unseen = true; return; } if (l.meter.contract.meter === 'estimate') { l.estimate_used += r.tokens; return; } const unseen = l.meter.contract.meter === 'claude' ? (r.roles ?? []).filter(x => x === 'outside-voice').length : 0; - if (unseen && r.roles?.length) l.unknown += Math.round((r.tokens * unseen) / r.roles.length); + if (unseen && r.roles?.length) { l.unknown += Math.round((r.tokens * unseen) / r.roles.length); r.unseen = true; } } /** Validate and store one meter reading against the run's contract. */ diff --git a/lib/ship-handoff.ts b/lib/ship-handoff.ts index 48e2ac7a5..3ff8dd4cc 100644 --- a/lib/ship-handoff.ts +++ b/lib/ship-handoff.ts @@ -20,7 +20,7 @@ import { wtreeFingerprint } from './script-invocation'; import { resultLine } from './result-codes'; import { SHIP_STEPS, budgetError, budgetLine, ledgerFileFor, loadCalibration, mintRunId, mutateLedger, parseStep, pointMarker, - readLedger, reserveIn, runDirFor, segmentEstimate, tally, usageError, writeLedger, type Handoff, type Ledger, + projector, readLedger, reserveIn, runDirFor, segmentEstimate, tally, usageError, writeLedger, type Handoff, type Ledger, } from './ship-budget'; import { loadMeterState, saveMeterState } from './claude-usage'; import { parseArgs, readMeter, stopLines, str, tokensOpt, writeReturnFile, type Ctx, type Outcome } from './ship-budget-cli'; @@ -92,7 +92,7 @@ export function cmdHandoff(argv: string[], ctx: Ctx): Outcome { const call = readMeter(l, o, true); const cal = loadCalibration(l.calibration); const t = tally(l); - const room = l.cap === null ? Infinity : l.cap - t.used - t.reserved - t.unknown - l.tail - l.stop_overhead; + const room = l.cap === null ? Infinity : l.cap - t.used - t.reserved - t.unknown - projector(l, cal).reserve; const subCap = tokensOpt(o, 'sub-cap') ?? Math.min(room, segmentEstimate(cal, l.meter.contract.unit, from, to)); if (subCap <= 0 || subCap > room) return { stdout: stopLines(l, { call, next: from, step: from, projected: subCap, reason: 'budget_exhausted' }), exit: 3 }; const pins = pinsNow(ctx.cwd, base, ctx.env); diff --git a/test/ship-budget-projection.test.ts b/test/ship-budget-projection.test.ts index 940ae169e..21df72684 100644 --- a/test/ship-budget-projection.test.ts +++ b/test/ship-budget-projection.test.ts @@ -8,7 +8,7 @@ * ship-budget-projection-GSTA-152-prb.md in the GStack Minder drive). */ import { afterEach, describe, expect, test } from 'bun:test'; -import { loadCalibration, SHIP_STEPS, unitConstants } from '../lib/ship-budget'; +import { loadCalibration, measureSettled, settleIn, SHIP_STEPS, unitConstants } from '../lib/ship-budget'; import { budgetHarness, type Harness } from './helpers/ship-budget-harness'; const cal = loadCalibration(); @@ -197,3 +197,45 @@ describe('dispatch reservations settle when the reviewer returns (GST-101)', () expect(close.out).toMatch(/^BUDGET_DRIFT: steps=2 step_projected=\d+ step_actual=700000 dispatches=0 dispatch_projected=0 dispatch_actual=0$/m); }); }); + +describe('the gate fixes from /ship review (GST-101)', () => { + test('a dispatch settled --unknown is never measured, so the dispatch factor stays at the prior', () => { + const h = fresh(); + const s = started(h, '40M'); + const id = h.field(h.sb(['check', '--run', s.run, '--step', '11', '--next', 'dispatch:outside-voice', '--reading', '0']).out, 'RESERVATION', 'id'); + expect(h.reviewLog(['--start', 'adversarial-review', '--reservation', id]).code).toBe(0); + const settled = h.sb(['settle', '--run', s.run, '--reservation', id, '--unknown', '--reading', '100000']); + expect(settled.code, settled.out + settled.err).toBe(0); + expect(settled.out).not.toContain('DRIFT:'); + const next = h.sb(['check', '--run', s.run, '--step', '11', '--next', 'dispatch:outside-voice', '--reading', '200000']); + expect(h.field(next.out, 'BUDGET', 'calibration')).toEndWith('dispatch:prior'); + expect(Number(h.field(next.out, 'BUDGET', 'projected'))).toBe(K.per_dispatch); + expect(Number(h.field(next.out, 'BUDGET', 'unknown'))).toBe(K.per_dispatch); + }); + + test('a Claude-metered outside voice is booked unknown and never measured', () => { + const r = { id: 'r1', kind: 'dispatch', step: '11', next: 'dispatch:outside-voice', roles: ['outside-voice'], tokens: K.per_dispatch, at: '', state: 'consumed', reading_at: 0, epoch: 1 }; + const l: any = { meter: { contract: { meter: 'claude' }, epoch: 1, last: 100_000 }, reservations: [r], unknown: 0, samples: [] }; + settleIn(l, r as any); + expect(l.unknown).toBe(K.per_dispatch); + expect(measureSettled(l)).toEqual([]); + expect(l.samples).toEqual([]); + }); + + test('a calibrated delegate verdict is one the handoff admits: both hold back the same reserve', () => { + const h = fresh(); + const cap = 12_000_000; + const s = started(h, '12M'); + const sb = (args: string[]) => { const c = h.sb(['check', '--run', s.run, ...args]); expect(c.code, c.out + c.err).toBe(0); return c; }; + for (const [i, step] of ['1', '2', '3'].entries()) sb(['--step', step, '--next', String(i + 2), '--reading', String(300_000 * (i + 1)), '--prior-reading', String(300_000 * i + 200_000)]); + const d = h.field(sb(['--step', '4', '--next', 'dispatch:specialist', '--reading', '1200000', '--prior-reading', '1100000']).out, 'RESERVATION', 'id'); + expect(h.reviewLog(['--start', 'review', '--reservation', d]).code).toBe(0); + const c = sb(['--step', '4', '--next', '12', '--reading', '8000000', '--returned', d]); + expect(h.field(c.out, 'BUDGET', 'verdict')).toBe('delegate'); + const tail = Number(h.field(c.out, 'BUDGET', 'tail')); + expect(cap - 8_000_000).toBeLessThan(2_000_000 + K.stop_overhead); + const ho = h.sb(['handoff', '--run', s.run, '--from', '12', '--to', '22', '--base', 'origin/main', '--reading', '8000000']); + expect(ho.code, ho.out + ho.err).toBe(0); + expect(Number(ho.out.match(/^HANDOFF: \S+ sha256=\S+ sub_cap=(\d+)$/m)![1])).toBe(cap - 8_000_000 - tail); + }); +});