Files
pbakaus_impeccable/site/data/live-codex-worker-result.ts
T
Paul Bakaus 50fc88ec85 Update Live Lab performance evidence
AI-assisted implementation under maintainer direction.
2026-07-13 13:09:29 -07:00

161 lines
4.5 KiB
TypeScript

export const liveCodexWorkerResult = {
generatedAt: '2026-07-13T20:01:08.415Z',
codexVersion: '0.144.0-alpha.4',
transport: 'stdio JSONL',
model: 'GPT-5.6-Sol',
effort: 'medium',
profile: 'quality',
defaultInCodex: true,
prewarmReturnMs: 87.26,
coldWorkerReadyMs: 656.82,
timings: {
coldHandshakeMs: 36.83,
coldThreadStartMs: 400.19,
turnStartResponseMs: 0.45,
wakeToTurnStartedMs: 1.47,
trivialTurnCompletedMs: 2060.18,
},
protocolProbe: 'GPT-5.3-Codex-Spark · low',
quality: {
tasks: ['bolder', 'polish'],
runs: 4,
passed: 4,
medianGenerationMs: 34339,
p95GenerationMs: 42780.1,
averageScores: {
commandFidelity: 8,
brandAndSystemFidelity: 8.5,
frontendQuality: 7.5,
taskCompletion: 7.75,
},
sparkControlPassed: 1,
sparkControlRuns: 2,
sparkMedianGenerationMs: 16470.5,
},
architectureComparison: {
directFirstUsableMs: 64649.73,
coldAppServerFirstUsableMs: 35591.71,
warmAppServerFirstUsableMs: 34931.08,
directQualityPassed: 4,
directQualityRuns: 4,
warmQualityPassed: 4,
warmQualityRuns: 4,
annotatedQualityPassed: 3,
annotatedQualityRuns: 3,
coldInputTokens: 41066,
coldOutputTokens: 2148,
warmRepeatInputTokens: 46973,
warmRepeatCachedInputTokens: 38656,
warmRepeatCacheRate: 0.823,
},
productionBrowser: {
runs: 1,
selectedTarget: 'article.offer-card',
firstVariantMs: 29682.58,
firstVariantP95Ms: 29682.58,
secondVariantMs: 54597.81,
allVariantsMs: 102182.9,
allVariantsP95Ms: 102182.9,
firstToSecondMs: 24915.23,
secondToAllMs: 47585.09,
serverPickupMs: 76.82,
sourcePreflightMs: 45,
validationMs: 9.33,
},
fullSourceControl: {
runs: 1,
selectedTarget: 'article.offer-card',
firstVariantMs: 28823.47,
secondVariantMs: 72698.01,
allVariantsMs: 145234.51,
firstToSecondMs: 43874.54,
},
sourceDeltaExperiment: {
runs: 5,
fullSourceFirstVariantMs: 34177.39,
deltaFirstVariantMs: 28074.92,
firstVariantImprovement: 0.1786,
baselineFirstToSecondMs: 43874.54,
deltaFirstToSecondMs: 24624.41,
secondVariantImprovement: 0.4388,
finalDeltaAllVariantsMs: 102182.9,
status: 'Enabled for all Codex source-mode phases',
},
earlyAcceptVerified: {
runs: 1,
acceptedVariant: 2,
sourceCommitted: true,
followupCleanupPassed: true,
acceptToResetMs: 238.18,
acceptToNextGoDispatchMs: 831.36,
acceptToNextFirstVariantMs: 16938.05,
controlNextFirstVariantMs: 42969.34,
nextFirstImprovement: 0.6058,
},
standbyExperiment: {
controlFirstVariantMs: 33283.06,
prewarmedFirstVariantMs: 35049.12,
controlNextFirstVariantMs: 42969.34,
prewarmedNextFirstVariantMs: 50920.73,
status: 'Rejected: slower under app-server contention',
},
twoCheckpointControl: {
runs: 1,
selectedTarget: 'article.offer-card',
firstVariantMs: 48437.46,
secondVariantMs: 139587.21,
allVariantsMs: 139587.21,
variants: 3,
passed: 2,
},
transportBaseline: {
runs: 5,
selectedTarget: 'h2.offer-card__title',
firstVariantMs: 19283,
firstVariantP95Ms: 20412,
allVariantsMs: 52795,
allVariantsP95Ms: 55115,
variants: 9,
calibratedPassed: 3,
},
renderedQuality: {
runs: 1,
variants: 3,
passed: 2,
commandFidelity: 6.67,
brandFidelity: 7.67,
renderQuality: 6.67,
taskCompletion: 6.33,
variant1Passed: true,
variant3Passed: true,
},
calibratedHeadingQuality: {
runs: 3,
variants: 9,
passed: 3,
commandFidelity: 6.11,
brandFidelity: 6.44,
renderQuality: 5.78,
taskCompletion: 5.56,
},
atomicExperiment: {
runs: 1,
firstVariantMs: 40225,
allVariantsMs: 40227,
variants: 3,
passed: 1,
status: 'Opt-in experiment',
},
annotatedBrowser: {
runs: 2,
firstVariantMs: 14316.21,
allVariantsMs: 45424.24,
comments: 1,
strokes: 1,
screenshotAttached: true,
},
result: '4/4 full-task quality gate',
passed: true,
architecture: 'A dedicated persistent Live app-server connection and worker thread with native Impeccable skill, sub-command, project context, and annotation-image inputs. Every source-mode variant publishes through an independently validated fenced delta; the final delta adds deferred parameter manifests without rewriting prior output. Accept interrupts unfinished variants; cleanup stays on the foreground control lane while the next generation starts independently.',
} as const;