v0.3.18
Lane I: extraction quality scoring recalibration.
Improved
- Weighted quality scoring (I7). Colors 35%, fonts 20%, logo 20%, role assignment 15%, primary 10%. The old system over-weighted logo (30%) and ignored role assignment entirely. Now a brand with 10 great colors and no logo scores higher than one with only a logo. Zero-color extractions get a specific "JavaScript-applied styles" message. MEDIUM scores include targeted gaps with remediation.
Lane I Complete for This Sprint
| Item | Status |
|---|---|
| I1 Benchmark | ✅ |
| I2 Color improvements (inline, blocklist, builder, scale) | ✅ |
| I3 Font cap (5→8) | ✅ |
| I4 Logo gradient detection | ✅ |
| I5 Confidence recalibration | ✅ |
| I6 Voice/messaging audit | ✅ |
| I7 Quality scoring | ✅ |
| I8 Headless browser | Deferred (large architectural change) |
| I9 Figma dogfood | Deferred |
| I10 Competitive comparison | Deferred |
Benchmark re-run: avg quality 7.2/10, 100% logo detection, 830ms avg duration. Basecamp still LOW (zero colors from runtime CSS — I8 fix).