From 265502a125d3fa26d570d59c44028792dfdea3be Mon Sep 17 00:00:00 2001 From: Paul Bakaus Date: Mon, 13 Jul 2026 11:48:05 -0700 Subject: [PATCH] Update Live Lab checkpoint evidence --- site/data/live-codex-worker-result.ts | 64 ++++++++++++++++++++------- site/pages/live-lab/index.astro | 33 +++++++++----- 2 files changed, 70 insertions(+), 27 deletions(-) diff --git a/site/data/live-codex-worker-result.ts b/site/data/live-codex-worker-result.ts index 819ab733c..17e56068e 100644 --- a/site/data/live-codex-worker-result.ts +++ b/site/data/live-codex-worker-result.ts @@ -1,5 +1,5 @@ export const liveCodexWorkerResult = { - generatedAt: '2026-07-13T03:57:44.263Z', + generatedAt: '2026-07-13T18:39:45.449Z', codexVersion: '0.144.0-alpha.4', transport: 'stdio JSONL', model: 'GPT-5.6-Sol', @@ -49,29 +49,61 @@ export const liveCodexWorkerResult = { warmRepeatCacheRate: 0.823, }, productionBrowser: { + runs: 1, + selectedTarget: 'article.offer-card', + firstVariantMs: 28823.47, + firstVariantP95Ms: 28823.47, + secondVariantMs: 72698.01, + allVariantsMs: 145234.51, + allVariantsP95Ms: 145234.51, + firstToSecondMs: 43874.54, + secondToAllMs: 72536.5, + serverPickupMs: 56.31, + sourcePreflightMs: 43, + validationMs: 5.67, + acceptRuns: 1, + acceptedVariant: 2, + acceptToResetMs: 231.1, + acceptToNextGoDispatchMs: 833.68, + acceptToNextFirstVariantMs: 52612.73, + blockedTailBaselineMs: 68141, + }, + 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, - deliveryGapMs: 33308, - serverPickupMs: 72.81, - sourcePreflightMs: 44, - validationMs: 5, - acceptRuns: 1, - acceptToResetMs: 196.5, - acceptToNextGoDispatchMs: 790.94, - acceptToNextFirstVariantMs: 23820, - blockedTailBaselineMs: 68141, + variants: 9, + calibratedPassed: 3, }, renderedQuality: { + runs: 1, + variants: 3, + passed: 1, + commandFidelity: 6, + brandFidelity: 7.67, + renderQuality: 6, + taskCompletion: 6, + }, + calibratedHeadingQuality: { runs: 3, variants: 9, - passed: 1, - commandFidelity: 5.67, - brandFidelity: 5, - renderQuality: 5.44, - taskCompletion: 5.11, + passed: 3, + commandFidelity: 6.11, + brandFidelity: 6.44, + renderQuality: 5.78, + taskCompletion: 5.56, }, atomicExperiment: { runs: 1, @@ -91,5 +123,5 @@ export const liveCodexWorkerResult = { }, 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. Accept interrupts unfinished variants; cleanup stays on the foreground control lane while the next generation can start independently.', + architecture: 'A dedicated persistent Live app-server connection and worker thread with native Impeccable skill, sub-command, project context, and annotation-image inputs. Variants 1 and 2 publish as independent reviewable checkpoints; variant 3 and parameters finish last. Accept interrupts unfinished variants; cleanup stays on the foreground control lane while the next generation starts independently.', } as const; diff --git a/site/pages/live-lab/index.astro b/site/pages/live-lab/index.astro index 4a70facf9..793a80132 100644 --- a/site/pages/live-lab/index.astro +++ b/site/pages/live-lab/index.astro @@ -406,33 +406,43 @@ const providerColumns = [
Dedicated Codex worker

- Five real Pick → /bolder → Go runs produced variant 1 in {displayMs(liveCodexWorkerResult.productionBrowser.firstVariantMs)} median and all three in {displayMs(liveCodexWorkerResult.productionBrowser.allVariantsMs)}. Accept returned the page in {displayMs(liveCodexWorkerResult.productionBrowser.acceptToResetMs)}; rotating the canceled producer cut Accept → next first variant from {displayMs(liveCodexWorkerResult.productionBrowser.blockedTailBaselineMs)} to {displayMs(liveCodexWorkerResult.productionBrowser.acceptToNextFirstVariantMs)}. In matched full tasks, warm app-server reached usable output in {displayMs(liveCodexWorkerResult.architectureComparison.warmAppServerFirstUsableMs)} versus {displayMs(liveCodexWorkerResult.architectureComparison.directFirstUsableMs)} direct; both passed 4/4. + With the target proven as {liveCodexWorkerResult.productionBrowser.selectedTarget}, variant 1 arrived in {displayMs(liveCodexWorkerResult.productionBrowser.firstVariantMs)}, variant 2 in {displayMs(liveCodexWorkerResult.productionBrowser.secondVariantMs)}, and variant 3 in {displayMs(liveCodexWorkerResult.productionBrowser.allVariantsMs)}. Accepting variant 2 while variant 3 was pending returned the page in {displayMs(liveCodexWorkerResult.productionBrowser.acceptToResetMs)} and dispatched the next Go in {displayMs(liveCodexWorkerResult.productionBrowser.acceptToNextGoDispatchMs)}. In matched full tasks, warm app-server reached usable output in {displayMs(liveCodexWorkerResult.architectureComparison.warmAppServerFirstUsableMs)} versus {displayMs(liveCodexWorkerResult.architectureComparison.directFirstUsableMs)} direct; both passed 4/4.

- + - + + - + - - - - - + + + + + + + + + + + + + +
Production Codex delivery candidates and quality evidence
CandidateFirst reviewFull setRendered gateDecision
CandidateVariant 1Variant 2Full setRendered gateDecision
Progressive{liveCodexWorkerResult.productionBrowser.runs} Sol/medium runsRoot · 1 → 2 → 3{liveCodexWorkerResult.productionBrowser.runs} Sol/medium run {displayMs(liveCodexWorkerResult.productionBrowser.firstVariantMs)}p95 {displayMs(liveCodexWorkerResult.productionBrowser.firstVariantP95Ms)}{displayMs(liveCodexWorkerResult.productionBrowser.secondVariantMs)}independently acceptable {displayMs(liveCodexWorkerResult.productionBrowser.allVariantsMs)}p95 {displayMs(liveCodexWorkerResult.productionBrowser.allVariantsP95Ms)} {liveCodexWorkerResult.renderedQuality.passed}/{liveCodexWorkerResult.renderedQuality.variants}production /bolder variantsDefault · quality work nextDefault · replicate quality
Atomic{liveCodexWorkerResult.atomicExperiment.runs} Sol/medium run{displayMs(liveCodexWorkerResult.atomicExperiment.firstVariantMs)}waits for trio{displayMs(liveCodexWorkerResult.atomicExperiment.allVariantsMs)}single turn{liveCodexWorkerResult.atomicExperiment.passed}/{liveCodexWorkerResult.atomicExperiment.variants}production /bolder variantsOpt-in · replicateRoot · 1 → 3{liveCodexWorkerResult.twoCheckpointControl.runs} Sol/medium control{displayMs(liveCodexWorkerResult.twoCheckpointControl.firstVariantMs)}full-card task{displayMs(liveCodexWorkerResult.twoCheckpointControl.secondVariantMs)}arrives with variant 3{displayMs(liveCodexWorkerResult.twoCheckpointControl.allVariantsMs)}two turns{liveCodexWorkerResult.twoCheckpointControl.passed}/{liveCodexWorkerResult.twoCheckpointControl.variants}production /bolder variantsControl · slower review
Heading transport{liveCodexWorkerResult.transportBaseline.runs} Sol/medium runs{displayMs(liveCodexWorkerResult.transportBaseline.firstVariantMs)}p95 {displayMs(liveCodexWorkerResult.transportBaseline.firstVariantP95Ms)}{displayMs(liveCodexWorkerResult.transportBaseline.allVariantsMs)}legacy 1 → 3 path{displayMs(liveCodexWorkerResult.transportBaseline.allVariantsMs)}title-only target{liveCodexWorkerResult.transportBaseline.calibratedPassed}/{liveCodexWorkerResult.transportBaseline.variants}recalibrated screenshotsTransport evidence only
-

Timing and rendered-quality samples are reported separately. The current blocker is output quality, not worker pickup: server pickup is {displayMs(liveCodexWorkerResult.productionBrowser.serverPickupMs)} median.

+

Root-task samples prove target identity in the posted Generate payload. The earlier five-run heading sample remains useful only for transport latency. Next gates: replicate root-level rendered quality and reduce the {displayMs(liveCodexWorkerResult.productionBrowser.acceptToNextFirstVariantMs)} cold next-result wait; server pickup is already {displayMs(liveCodexWorkerResult.productionBrowser.serverPickupMs)}.

@@ -495,6 +505,7 @@ const providerColumns = [

Dedicated Codex worker

First review
{displayMs(liveCodexWorkerResult.productionBrowser.firstVariantMs)}
+
Second review
{displayMs(liveCodexWorkerResult.productionBrowser.secondVariantMs)}
All variants
{displayMs(liveCodexWorkerResult.productionBrowser.allVariantsMs)}
Accept → page ready
{displayMs(liveCodexWorkerResult.productionBrowser.acceptToResetMs)}
Accept → next Go
{displayMs(liveCodexWorkerResult.productionBrowser.acceptToNextGoDispatchMs)}
@@ -502,7 +513,7 @@ const providerColumns = [
Worker ready
{displayMs(liveCodexWorkerResult.coldWorkerReadyMs)}
Rendered gate
{liveCodexWorkerResult.renderedQuality.passed}/{liveCodexWorkerResult.renderedQuality.variants}
-

Progressive remains the Codex default. Atomic is measured but opt-in; rendered quality is the next gate.

+

Variant 2 is now independently reviewable and acceptable. The next gate is cold post-Accept generation, followed by replicated root-level rendered quality.