Cinema knowledge graph: Fable 5 vs Opus 4.8 vs GPT-5.6 Sol

/goal Build me a visually stunning, explorable cinema knowledge graph in the browser — movies, actors, and directors as an interconnected web I can wander through, with a timeline.

Run this in three phases:

1. RESEARCH + DATA — first, figure out the best freely available real movie data source and wire it up. IMDb's official API is enterprise-only, so evaluate: TMDB's free API, the OMDb API, or IMDb's non-commercial TSV datasets. Pick one, pull a real dataset of at least ~300 notable films (roughly 1970-today) with their directors and principal cast, and store it as clean local JSON the app ships with. Real titles, real people, real years — nothing invented. Note the source and its attribution requirement in the app footer.
2. BUILD —
   (a) GRAPH VIEW: a force-directed graph (D3 or similar, WebGL if needed for performance) — films, actors, directors as distinct node types; edges for acted-in / directed. Click any node to focus its neighborhood; double-click to expand. Hover cards with poster-style info. Search that flies you to any node.
   (b) TIMELINE VIEW: the same data along a scrubbable timeline by year — see a director's career or an actor's run unfold.
   (c) A "six degrees" mode: pick two people and the app finds and animates the shortest connection path between them.
   Design it like an editorial data-viz piece — restrained palette, beautiful type, smooth physics — not a hairball.
3. QA — deploy live to Vercel, screenshot with Playwright, verify search, pathfinding, and both views perform smoothly with the full dataset.

DONE when: the deployed graph is fast, beautiful, and correct — real data, working six-degrees pathfinding, both views usable — and wandering it is genuinely fun. Keep iterating until the graph feels alive, or you've run 40 turns.