§ 001 / THE OUTPUT
Eight hours became fourteen merged changes.
This is the evidence ledger for one bmdpat Codex run. The first merge landed at 17:45 UTC on August 12. The last landed at 02:00 UTC on August 13. The observed merge window was 8 hours 14 minutes.
14
merged pull requests
Not prompts. Not draft branches. Changes merged into the public product.
§ 002 / THE TRACE
Open every receipt.
The animated line is the one visual summary. Every marker is a real merged GitHub pull request.
Live merge trace
17:45 UTC Aug 12 → 02:00 UTC Aug 13
01 / 14
Contain receipt card rendering
Below-fold receipt cards no longer compete with the first render.
Category: performance. Select any marker to inspect the result.
§ 003 / WHAT CHANGED
Output grouped by consequence.
Performance
7
Less work before first paint.
Removed speculative route fetches, deferred a global font preload, contained below-fold rendering, kept blog metadata warm, and moved known topic hubs to static delivery.
Measurement
5
Metrics stopped lying by accident.
Fixed inverted legacy-event matching, replaced an impossible raw ratio with a bounded visitor-level metric, and added missing Run Doctor and newsletter-window signals.
Verification
2
Claims ended in receipts.
Updated generated architecture truth and recorded production limitations instead of turning green CI into a larger claim than it could support.
§ 004 / 39-SECOND CUT
Watch the proof reel.
1080p HDR10 with an SDR fallback · silent and caption-first
§ 005 / FULL LEDGER
Fourteen changes. No hidden denominator.
+672 / −69 / 68 file touches
- #1387GitHub ↗
Contain receipt card rendering
Below-fold receipt cards no longer compete with the first render.
- #1388GitHub ↗
Match legacy events in either order
A 73 ms event-order inversion stopped hiding real handoff-path views.
- #1389GitHub ↗
Disable speculative tracked-link prefetch
Tracked calls to action stopped issuing unsolicited route fetches on first paint.
- #1391GitHub ↗
Stop receipt and blog link prefetch
Receipt and blog indexes stopped preloading every linked detail page.
- #1390GitHub ↗
Sessionize funnel capture readout
The decision metric became a bounded 33.3%, replacing a misleading raw 218.2% ratio.
- #1392GitHub ↗
Refresh architecture test count
Generated inventory and architecture records returned to the same source of truth.
- #1393GitHub ↗
Clarify raw funnel ratios
Raw beacon volume was separated from the visitor-level conversion metric.
- #1396GitHub ↗
Stop desk secondary-link prefetch
The desk stopped fetching receipts and desk markdown before a visitor asked for them.
- #1394GitHub ↗
Measure Run Doctor plan exposure
A new PII-free event now distinguishes no plan from a visible plan that was not copied.
- #1398GitHub ↗
Record merged performance verification
Production verification and the rolling Speed Insights limit were recorded instead of implied.
- #1397GitHub ↗
Measure newsletter signups in-window
All-time subscribers and signups inside the experiment window became separate facts.
- #1399GitHub ↗
Defer mono font preload
A below-fold font stopped blocking first paint on every route.
- #1400GitHub ↗
Keep blog metadata cache warm
A 24-hour tagged cache removed recurring cold database reads while preserving publish invalidation.
- #1401GitHub ↗
Statically deliver topic hubs
Known topic pages moved from first-visitor server work to build-time static delivery.
§ 006 / CONFIDENCE
High on shipped code. Still scoped.
All fourteen PRs cleared the required check, browser E2E, and risk gate before merge. That is 42 required green checks. It proves those guarded paths passed. It does not prove every performance change produced a durable user-level lift. Rolling field metrics still own that claim.