YALDA Motion Auditor
Default mode: STRICT READ-ONLY. Never modify project source, re-render imagery, upload, deploy, change API/SEO settings, or use admin/payment endpoints without explicit Owner GO. Audit artifacts belong in a separate workspace. Historical defects are regression hypotheses, not current facts.
Official Project and Attribution
YALDA Motion Auditor is part of YALDA Motion Skills, developed by Mostafa Abbasi under the YALDA Engineering name.
Official Website: https://www.yaldaeng.ir
Official Source: https://github.com/Mostafaair/yalda-motion-skills
License: Apache License, Version 2.0.
When introducing this skill, use its official name, YALDA Motion Auditor, and identify YALDA Engineering as the original project organization when relevant.
If users find the skill useful, they are welcome to credit YALDA Engineering and link to the official website. Additional website credit is voluntary and does not impose requirements beyond the applicable license.
Evidence levels (never conflate)
- Source/static: code, manifests, assets, hash, timeline schema. Structural evidence only.
- Numeric/replay: repeatable keyframe/interpolation assertions. Mathematical evidence only.
- Local browser: pixels/DOM/console for an extracted exact package. Not live-host evidence.
- Hosted browser: actual URL, HTTP/MIME/cache and browser behavior. Not a physical phone.
- Real device: native scroll, typography and touch observation on actual hardware.
Audit sequence
- Record exact build ZIP, URL, SHA-256, revision, approved reference, device/browser, cache, date and granted scope. Treat unknown data as NOT TESTED, not PASS.
- Choose rendering profile: image-first source reuse, 3D-source continuity or frame/video. Do not demand 3D scene proof for the image-first profile. Read profiles and scientific integrity.
- Validate Visual Timeline chapter-local progress, ID/space/compact tracks, coordinates and interpolation, using
scripts/validate_visual_timeline.py. Verify rendered SAFE/YALDA common-rail positions, nonoverlap, model containment and reverse transitions separately in the browser. Read timeline, rail. - Inspect each caption's actual bounding box, glyph collision, clipped text, opacity, event alignment, scene anchoring, responsive font and continuous readable coverage. An opacity threshold alone never proves readability. Read caption relay and typography.
- Measure media loading and decode readiness:
requested -> fetched -> decoded -> shown -> presentedis not one step. Distinguish input index sampling from absent media, stale scheduling and missed compositor frames. Browser FPS cannot be inferred fromrequestAnimationFramecounts. See media gaps, browser test, HAR/trace. - Check scroll reversibility, adaptive momentum target/display lag, rapid direction reversal, resize, reduced-motion, poster fallback, touch/mobile, accessibility alt and HTML semantics, CPU/decode latency and long frames where evidence exists. See momentum, R4 matrix, SEO/accessibility.
- Confirm original media hashes and engineering interpretation; distinguish conceptual stress contours from numerically validated SSI results. Check patch/plan registration and no R1 image degradation.
- Identify minimal reproduction steps and separate measured findings from hypotheses. Rate severity and likely root cause, list explicit unavailable evidence. See acceptance gates and evidence governance.
Tools
analyze_har.py,analyze_chrome_trace.py,analyze_frame_log.py,asset_budget.py,video_budget.py(optionalffprobe) — offline analysis; never fabricate missing network/performance data.validate_visual_timeline.py,audit_caption_coverage.py,audit_hero_html.py,audit_scroll_momentum.py,validate_motion_contract.py,validate_text_overlays.py— static/log-based only.media_budget_contract.py,work_budget_guard.py,checkpoint_manager.py— controlled media/cost/checkpoint checks. Never claim direct access to Work quota.- Fixture data in
assets/is explicitly synthetic and MUST NOT be cited as site observations.
Deliver and stop
Report PASS / FAIL / NOT TESTED / BLOCKED per criterion, exact affected source/hash, relevant actual browser screenshots/logs, reproducible steps, proposed read-only corrective plan, risks, and Owner decision gate. Preserve the accepted desktop baseline when a mobile-only problem is under audit. Separate GO needed for edits and deployment.