# Pixel craft evidence · 0.1.0

Evidence distinguishes researched, implemented, exercised, transferred and independently reviewed capability. It is not a categorical mastery certificate.

## Demonstrated work

| Capability | Actual evidence |
|---|---|
| Editorial hybrid craft | Original workshop generation, critique and targeted revision; changed-context orbital and synthwave studies. Parent and independent reviewer inspected source images and focal crops. The workshop depicts cartridge positioning/repair, not unambiguously completed insertion. |
| Native-grid light/material control | Authored 384 × 216 procedural scene: warm, cool and neutral presets, distinct ceramic/steel/emissive treatments, exact palette reservation, failure/correction pair and integer nearest exports. All suite PNGs decoded and compared to fresh renders. |
| Native behavior | **13 passing Python tests** after a reproduced/fixed extreme-vector normalization issue. Covers sRGB roundtrip, linear blending, changed light response, material contrast, accent reservation, invalid input and every integer-expanded pixel. |
| Contained effects | **7 passing pure-buffer tests** and **29 passing Chromium browser checks** for separate Off/CRT/VHS/digital-glitch approximations. Actual RGBA bytes, exported PNG blocks, real download and different source images were exercised. |
| Lifecycle and fallback | Live reduced preference both directions, explicit pause, real offscreen observation, actual bfcache restoration, stale decode success/failure guards, missing-image recovery, forced Canvas/readback/export failures and repeated teardown. Hidden-tab handling uses a labeled fixture, not a physical background-tab observation. |
| New-subject transfer | A fresh agent read a copied method, froze a different brief, made a 96 × 72 right-lit heron with 12 colors, critiqued/revised the catch and verified its exact 5× enlargement and four reserved accent pixels. Parent inspected and reran verification. See [transfer source](assets/transfer/README.md). |
| ProDyn integration | The separate host guide passes 1440/700/390 layout/axe checks, actual custom-font glyph inspection, keyboard skip, no-JS clean fallback, embedded treatment/export and stable/Mk.ii switching without changing artwork bytes. Source/host screenshots were visually inspected. |
| Package safeguards | Five real ZIP-builder fixture tests pass: explicit inventory/readback, unlisted-file exclusion, missing dependency rejection before destination creation, unsafe/symlink/duplicate rejection and existing-output preservation. The real 60-file source allowlist resolves required entries and local Markdown links. Final archive extraction is an artifact-specific delivery check, not inferred from these fixture tests. |

## Verification environment and repository gates

Observed local environment: macOS, Python 3.9.6, Pillow 11.3.0, Node 24.13.0, Chromium 153.0.8010.12 via Playwright 1.63.0. Browser screenshots/layout checks used DPR 1. The skill's platform metadata describes compatible code paths, not proof of execution on every OS.

Parent execution: complete repository `npm test` passed, including 30 root Python checks, the existing motion/integration/palette browser harnesses, and the new native/effects/lab/host checks. `npm run test:export` separately passed all 31 existing exporter tests. New effects remain independent of existing M1–M18 contracts.

The source repository retains independent craft/native and runtime reviews plus parent resolutions under `dsm/pixel-craft/evidence/`. Neither reviewer approved scopes they did not inspect. Medium native numeric and attribution issues were corrected; the runtime review's earlier absent-allowlist finding was resolved during parent integration. The source package is a reviewed allowlist, not a broad copy of the private repository.

## Research grounding

The craft and retro ledgers preserve exact URLs, short retrieved quotations, creator/rights notes, static visual observations and source limitations. Citation IDs are local to their own ledger. Text evidence checks and direct visual critique are distinct: a text-only quotation gate cannot certify an image-based observation. Independent source spot checks found no substantive contradiction in the inspected claims; not every historical attribution or image was independently re-audited.

## Not established

No manual Aseprite/editor proficiency, production character-animation pipeline, calibrated display/print color matching, spectral or energy-conserving light transport, real VHS/CRT hardware emulation, historical-console compatibility, physical mobile/Safari/Firefox behavior, screen-reader certification, sustained performance benchmark or human approval of a commercial application is claimed. Canvas export comparisons used opaque art; browser alpha/profile limitations remain documented. The heron's tiny catch remains symbolic at native size.

Use the method, practice on an unfamiliar brief, critique the actual result, and add only demonstrated lessons. Repository availability does not install the skill in other agents or authorize restyling other projects.