# Gitdex source-filter frontend capture — review candidate

This set records the **real, unmodified Gitdex Solid frontend** at source `1bb5a9b`, running on loopback against a **synthetic fixture API**. It does not show Encore, PostgreSQL, GitHub discovery, a successful provider sync, or a valid GitHub token. The browser held only the literal localStorage marker `fixture-only-not-a-github-token`. Fixture owners `fictional-person` and `fictional-lab`, their repositories, descriptions and `.example.test` URLs are invented for this study. No private repository name or customer data was used.

The fixture API returned two configured sources, four fake repositories, stats, a health configuration and an empty optional stack aggregate. The actual frontend rendered Dashboard, Inventory, and its source dropdown. The paired `inventory-all-*` pixels show the default fictional-person owner filter; `inventory-source-*` shows the source dropdown changed to `organization: fictional-lab` and the two organization rows. This is evidence of **frontend filter behavior against fixture responses**, not evidence of real GitHub coverage or a persistent provider identity fix. The score and staleness numbers are fixture values, not calculated live results.

`evidence/fixture-api.mjs` is the exact disposable loopback responder. `capture-fixture.mjs` and `capture-motion.mjs` use Chrome CDP to seed the marker/theme and capture actual 1365×900 PNG browser pixels. `requests-{light,dark}.json` records network requests; neither capture contacted `api.github.com`. The 30 original PNG frames per theme are archived under `evidence/motion/`, with per-frame UTC, bytes, hash, theme and table rows in provenance JSON. The source-select change is a browser DOM change event. The paired 5-second, 6 fps VP8 WebMs are **edited silent frame sequences** made from those captures, not uninterrupted real-time interaction or narrated Remotion release films. Capture used only 127.0.0.1:3001 (Vite) and 127.0.0.1:4000 (fixture API); both and headless Chromium were stopped afterward.

`make-poster.py` builds paired 1600×900 poster studies from authentic filtered Inventory crops with surrounding editorial labels; `poster-thumb-{light,dark}.png` is the 384×216 legibility check. The poster studies are review candidates, not an approved release cover. `asset-manifest.json` lists exact bytes, SHA-256, MIME and media geometry for every review asset.

This branch is review-only. Paired Cartesia narration, Remotion animation, timed captions, final film, and exceptional release cover remain open. Do not link the assets publicly until independent visual/truth review, media-main merge, hosted byte/hash verification and accurate changelog copy pass.