Codebase Improvement Loop

The measured GLCO loop used to replace a WebGL hero while protecting navigation, layout, accessibility, build health, and contact behavior.

Industrial test bench with signal display, measurement tools, checklist, and staged result trays by Lake Erie.

This rebuild removes a large WebGL package while keeping the site's visual identity. The change stays only when the build, browser, access, and contact checks pass.

The start state was easy to name

The homepage loaded Pixi to make a ripple effect over the main image. That added a large graphics package. The page's real job is to explain the offer and help a visitor start a pilot.

The desired state protects the real job

The new hero must show the GLCO setting. Its static image must fit each screen and reserve space before it loads. The main action must stay visible. Reduced-motion settings and all public tasks must still work.

The editable area stays narrow

The change removes the Pixi component and package. It adds a picture component, an image build script, and a new hero layout. Contact storage, email delivery, and Cap stay outside this test.

Tests decide whether the result stays

The acceptance gate checks:

  • production build success;
  • no Pixi package or Pixi client chunk;
  • no browser console or first-party network errors;
  • visible desktop and mobile navigation;
  • keyboard access and automated accessibility scan;
  • working article, pilot, and contact journeys;
  • contact payload and response compatibility;
  • known image dimensions and low layout shift.

What was tested

The repo records unit tests, browser tasks, content checks, and a production build. The live check also reads every sitemap URL and tests the contact path.

This is one site test. It does not prove that static media fits every product. A product may need an interactive canvas for direct control, a model, or live data.

Read Autoresearch for Codebases for the general pattern or Agentic Software Improvement for the service.

AI assisted with code changes and tests. GLCO defined the acceptance gate, reviewed failures, and owns the release decision.

Sources

Next action

Name one code measure that affects a real user task. Record the current result and required regression checks before changing the code.

Apply this method to one business process.

Bring one repeat process. We will map it, build a tested prototype, and show what changed.

Start a Workflow Pilot