Evidence (SSI fixture matrix, c1 lane, 2026-07-04)
Run: SSI main 8421204 / transformer ebc3997 / wp-codebox fa9f02d7. All 7 complexity-1 fixtures pass the hard gate; visual parity (capture-only, 1280x1600 viewport, external requests blocked) tells the real story:
| Fixture |
Native rate |
Visual mismatch |
| 12-portfolio |
97% |
88.9% |
| 21-studio-web-seafood-popup |
100% |
75.9% |
| 2-onepager-coffee |
91% |
61.5% |
| 15-saas |
97% |
52.6% |
| 35-hyper-minimal-blog |
100% |
2.3% |
| 31-personal-cv-academic |
100% |
2.6% |
Native rate and visual parity are decoupled — block conversion is largely solved at this tier; the imported WordPress render diverges for styling reasons. Prior measured causes (SSI docs, 15-saas/38-medical): oversized inline SVG icons losing CSS sizing, and a frontend stylesheet not applied to the rendered page. Two fixtures at ~2.5% prove the render path can converge.
Task
Root-cause the mismatch for 12-portfolio and 21-studio-web-seafood-popup via the matrix visual-compare artifacts (source.png/candidate.png/diff.png + captured DOM), fix the owning cause in the transformer/emission layer, and prove before/after mismatch ratios plus no regression across the c1 lane. If the root cause is SSI-side (import-time stylesheet enqueue), report rather than papering over in blocks-engine.
Evidence (SSI fixture matrix, c1 lane, 2026-07-04)
Run: SSI main
8421204/ transformerebc3997/ wp-codeboxfa9f02d7. All 7 complexity-1 fixtures pass the hard gate; visual parity (capture-only, 1280x1600 viewport, external requests blocked) tells the real story:Native rate and visual parity are decoupled — block conversion is largely solved at this tier; the imported WordPress render diverges for styling reasons. Prior measured causes (SSI docs, 15-saas/38-medical): oversized inline SVG icons losing CSS sizing, and a frontend stylesheet not applied to the rendered page. Two fixtures at ~2.5% prove the render path can converge.
Task
Root-cause the mismatch for 12-portfolio and 21-studio-web-seafood-popup via the matrix visual-compare artifacts (source.png/candidate.png/diff.png + captured DOM), fix the owning cause in the transformer/emission layer, and prove before/after mismatch ratios plus no regression across the c1 lane. If the root cause is SSI-side (import-time stylesheet enqueue), report rather than papering over in blocks-engine.