[Flotapponnier]

What

film/ renders a brand film from the live API. 43 seconds, 1920x1080. Every figure on screen is a number the site publishes, fetched by a script and committed, so a render is reproducible and every claim carries its date.

cd film && npm install && npm run data && npm run render

Why it is a project and not an export

The first cut ran the same bar chart for every benchmark and became a list: by the third table the eye has learned the shape and stops reading. Each act is now drawn from what its bench actually measures, and the shape is declared in the data so the cut never has to know what a bench is about.

Act	Shape	Why
Aggregators	arrivals on one clock	A provider's marker lands at the moment its measured lag says it should, so the film's own timing is the measurement
Trading	accumulation into columns	Throughput is mass, not speed, so it piles up instead of racing
Bridges	subtraction from $10,000	Best to worst is about eighteen dollars: legible in dollars, invisible as a percentage bar
Blockchains	1:1 playback, one tick per block	Robinhood Chain fires forty ticks in four seconds, Blast fires two. Cannot be sped up without lying, so it is the longest act
RPCs	a grid of every chain probed	The claim there is breadth, not speed

Each act carries the category's own description, read from src/lib/categories.ts rather than written again, so the film cannot describe a category differently from its category page.

Data handling

scripts/fetch-data.ts reads /api/citable and /api/stat/<slug>. Three things it gets right that are easy to get wrong, each commented where it happens:

• Latency benches that declare s store milliseconds. The same conversion the site has in format.ts. Skipping it prints a 0.26 s head lag as "260 s".
• The full ranked field comes from /api/stat, not the two leaders in /api/citable. A film that never shows the losers never shows that several products were tested on the same thing.
• Only referenced logos are copied into film/public, so the directory stays reviewable.

Site bug fixed in passing

Bench launchpad-wars publishes its leader as pumpfun while the logo file, the product page and every other reference spell it pump-fun. The winning row of a live bench was rendering initials instead of the brand mark. One alias in logo-manifest.ts.

Honest gaps, written into the README

No audio (needs a licensed track, not an improvised one). 16:9 only, because the type sizes are tuned for it and a 9:16 cut needs its own layout pass rather than a second <Composition> that would look broken. And Mobula appears in two acts as a genuine leader, which is worth a conscious decision rather than a default, since OpenChainBench is a Mobula project.

Weight

node_modules and out/ ignored. Committed: source, data.ts, three fonts and the 33 referenced logos, about 540 KB, so a clean clone renders without the network.