osmWorkflow/.trellis/tasks/archive/2026-08/08-03-experiment-glb-compression/prd.md

Experiment GLB compression

Goal

Create and validate a repeatable GLB compression experiment for nantaizi-lake-innovation-valley, focused on reducing file size without making the current uncompressed GLB output the wrong baseline.

Current measurement:

• nantaizi-lake-innovation-valley.glb: 20.50 MB
• embedded images: 17.72 MB
• non-image binary buffer: 2.45 MB
• conclusion: texture compression / resize should be tested before geometry compression.

Requirements

1. Add a standalone compression tool or script that consumes an existing GLB and writes a separate compressed GLB output.
2. Do not replace the default <area>.glb in this first experiment.
3. Prefer texture-first compression; geometry compression may be evaluated but should not be treated as the primary win.
4. Keep the experiment repeatable from the command line.
5. Validate only nantaizi-lake-innovation-valley.
6. After visual review accepts the WebP result, integrate it as an explicit optional compress stage that does not run in all and does not replace the default GLB.
7. Report:
   • source GLB size
   • compressed GLB size
   • image vs non-image size before compression
   • GLB extensions introduced
   • Cesium preview compatibility status or blocker

Acceptance Criteria

• Compression command/script runs from this repo.
• It produces a separate compressed GLB for nantaizi.
• Source GLB is left intact.
• scripts/glb-digest.js can inspect the compressed GLB or the blocker is documented if the digest needs extension support.
• Visual/Cesium compatibility next step is clear.
• Optional compress stage can generate compressed nantaizi artifacts from the existing Cesium outputs.
• compress is not included in all and cannot be enabled from area stages.
• No outputs/ files are committed.

Out Of Scope

• Making compressed GLB replace the default pipeline artifact.
• hanyang validation.
• CDN/server gzip/brotli configuration.
• Large frontend UI changes.