跳转至

v3.6 fidelity work in progress

Round Changes Result
0 Previous draft compared with Microsoft PowerPoint PDF at 3 px/pt Fail / incomplete; see output/fidelity/round-0/report.json, overlay.png, diff.png.
1 Correct rotated group boxes; private SVG uses explicit arrowheads and crop windows Fail / not verified; output/fidelity/round-1/report.json, overlay.png, diff.png. Paint order 60/60; measurable vertex delta ~0.000001pt and boxes within 0.1pt, but text, arrowhead dimensions, crop ink and occlusion remain unverified.

| 2 | Explicit heads adjusted using one PowerPoint arrow sample | Fail / not verified; output/fidelity/round-2/report.json, overlay.png, diff.png. Connector vertices and tips worst ~0.000001pt; head dimensions, text baselines/ink, crop ink, occlusion and residual blobs not verified. |

| 3 | Tested a different global text baseline model | Fail / not verified; output/fidelity/round-3/report.json, overlay.png, diff.png. Text remains up to 3 pixels offset at 3×, and PowerPoint baseline rules and arrow dimensions remain unproven. |

R2 is a PDF export from Microsoft PowerPoint rasterized at 216 dpi; source PPTX hash matched the private copy. R1 is independently extracted from OOXML. R1 vs R2 geometry has not been comprehensively reconciled. The renderer has not passed the required gate and has not been integrated into the homepage.

Existing refinements: strict build and Python suites (7 tests + 138 tools) passed; deployment JS checks passed. In isolated Edge, icon, figure End/anchor and header first paint passed. Three additional browser checks timed out waiting for old or unavailable page selectors; default Playwright Chromium is absent.

Measured numeric box and connector results do not constitute a full pass. The single PowerPoint arrow sample cannot prove all heads within 0.10pt. An independent per-element ink segmentation and text baseline comparison remains necessary. No production renderer, runtime, or homepage swap was committed.