4.6 KiB
Midnight Grand Tour — visual review
Implemented on codex/gpt6-portfolio-evolution, based on main 642f5b5, in .worktrees/gpt6-portfolio-evolution. Local preview: http://localhost:41812. Ready for visual review; not merged or deployed.
The visible changes
The opening now combines an open typographic introduction with a longer-lens night drive. A larger coral horizon, broad cyan mountain waves and cooler metallic reflections establish one visual language. Camera compositions ease between chapters without changing native scrolling or the canonical journey/audio state.
The Blender-authored car retains the original wedge body, now with a swept floating wing, turbine-style wheel faces, rear-window louvers, segmented coral lamps and cyan rear guides. The editable source and provenance live in assets/blender/evolution/. It exports into eight material groups and 18,564 triangles; reflections are baked once into a small procedural environment, with no per-frame reflection pass.
The professional material has more space: a broad career timeline, an architectural infrastructure plate, larger real project screenshots, an evidence-led Footy Tracker layout, and clearer toolkit groups. Project images open at full size. Controls keep their existing behavior, with refined spacing, focus and touch states. Static and reduced-motion visitors receive desktop or phone artwork captured from the real world.
Before and after
Desktop before:
Desktop after:
Phone before · Phone after · Tablet opening
Project screenshots · Infrastructure · Toolkit · Contact
Blender detail render · Recorded drive, scroll reversal and navigation
Preservation and practical checks
- Normalized rendered main text is identical to the freshly built approved baseline: 12,333 characters. Original content links match. The eight section IDs and their order are unchanged.
- Content, approved metadata, audio source, canonical journey source and test files have no changes. All compiled approved project media and audio files match the baseline byte for byte.
- Final typecheck, client build, server build, validated prerender, artifact validation and transfer-budget check passed. No budget limits were changed. The initial baseline build also ran the required existing unit suite: 662 checks passed; it was not repeatedly rerun during visual iteration. No tests were added or edited.
- Inspected all chapters at desktop 1440×900, tablet 768×1024, phone 390×844, narrow 320×568 and short 1024×600. No horizontal overflow was observed. Exercised menu navigation, scroll, fast jumps, reversal, resizing, keyboard scrolling and focus states.
- Soundtrack reached the running state; mute/unmute, volume at 25%, pause/resume motion and graphics selection worked. Static mode, system reduced motion and a no-JavaScript browser retained content and appropriate poster artwork. Full-size project-image navigation worked.
- The final direct preview produced no JavaScript errors or console warnings, including after graphics-quality changes. Resetting texture-upload flags between renderer generations fixed the observed WebGL warning.
- Final compressed-transfer totals: initial application 90,134 bytes; first live scene including the car 279,010 bytes; both fallback posters together 189,323 bytes. All transfer checks passed. Detailed totals are in
evidence/gpt6-evolution/transfer-budgets.json.
Practical limits
The inspection browser reports ANGLE/SwiftShader software rendering. A short direct Low-quality sample averaged about 12 fps at 1440×900 and 40 fps at 390×844; an instrumented High-quality sample was slower. These are software-renderer measurements, not a hardware-GPU performance certification. Existing adaptive quality and Static controls remain available. Vite still reports the existing large deferred Three.js chunk warning; the measured compressed scene transfer remains within its budget.
The recording shows actual browser output and is silent. Audio controls and preservation were checked separately; no claim is made of a complete auditory mix review or physical-device testing.