On this page

The finite list

The port has 63 routed scenes in the current ledger.

That number is a gift. It means the project is finite. Not easy. Not short. Finite.

Each scene has an ADS family and tag. Each scene has a foreground pack. Some have tide variants. Some have raft variants. Some need holiday overlays to avoid stomping pixels. Some are mostly standing animation and some are little movies. They all eventually need the same thing: watch the PS1 output, compare against reference, fix what is wrong, mark the row.

Why the first five mattered

At the time this essay was first drafted, five scenes were validated: FISHING 1, FISHING 2, FISHING 3, FISHING 4, and FISHING 6. The live ledger has since moved through the rest; all 63 routed scenes are validated as of v0.7.0-ps1 and the headless-perf battle card is now its own post-validation arc.

Five out of sixty-three could sound small if you treat the scene count like a progress bar. It is not the right read. The first validated scene built the pipeline. The second proved the pipeline was not a one-off accident. The next three proved the loop could survive placement bugs, residual cleanup bugs, and real variant checks. The remaining fifty-eight became work, but they were work inside a known frame.

The difference between “nothing is known” and “the loop is known” is enormous. The same difference shows up later in performance work: once the first row of the matrix has a number, the rest are numbers in context, not unknowns.

The daily loop

The loop is:

  1. Pick a scene.
  2. Capture or refresh the host reference.
  3. Generate or inspect the FG2 pack.
  4. Wire the runtime if this scene needs a new path.
  5. Build the disc.
  6. Boot DuckStation.
  7. Watch the scene.
  8. Compare screenshots, logs, and timing.
  9. Fix the obvious thing.
  10. Repeat until the human review feels boring.

“Feels boring” is important. A validated scene should not feel lucky. It should feel like you have watched it enough times that the next playthrough is uneventful.

The grind is the project

There was no grand final algorithm that validated the other fifty-eight. There were tricks. There were pack improvements. There was more codegen. But the work stayed one scene at a time, and the last cluster (the foreground-only multi-view scenes) closed on 2026-05-05 with ACTIVITY 9 as the final row to flip green.

That is not disappointing. It is the nature of preservation work. The only way to know a gag landed is to watch the gag — sixty-three times.

Why the website mirrors it

The site has 63 scene pages for the same reason the code has 63 ledger rows. The shape of the project is the shape of the source material. If a scene has a status, it deserves a page. If a regtest reference exists, it deserves a page. If a holiday has an emblem, it deserves a page.

The grind becomes navigable when every unit has a shelf.