Lab · Retrospective

The pivot that almost did not happen

The project could have shipped at "close enough." That would have been the wrong project.

Published

~2 min read · 506 words

On this page

Close enough was available

There is a version of this project that ships much earlier.

It interprets enough ADS and TTM bytecode on the PS1 to show Johnny walking, fishing, and waving. The palette is mostly right. The island is mostly right. The sounds play when something like the right thing happens. It runs in DuckStation. It has a release zip. People say “neat” and move on.

That version was available. It was also the wrong project.

The problem with approximate

Johnny Castaway lives in timing. The gag is not “Johnny goes fishing.” The gag is the pause before he reacts to the starfish. The small walk back. The way the ocean keeps moving while the foreground holds. The holiday decoration that is barely there unless you happen to notice it.

Approximate playback erases those edges. It makes the port impressive as a technical demo and weak as a preservation object.

Once that became obvious, the target changed: the PS1 would replay captured foreground packs, not reinterpret the whole Sierra runtime at scene time.

The host-and-replay move

The host build does what the desktop is good at. It runs the decoded Sierra engine, captures foreground frames, records sound events, and writes compact FG2 packs.

The PS1 does what the console is good at. It loads those packs, composites them over its own background, runs waves, stamps holidays, plays SPU samples, and keeps VBlank cadence.

That split is the project. It is the reason a 1992 screensaver fits into a 1994 console in a way that can be tested.

What the pivot cost

It cost tooling. A lot of tooling.

The moment you choose pixel-perfect replay, you need pack formats, manifests, dirty-region templates, reference captures, validation scripts, build scripts, scene ledgers, timing logs, and a way to compare the PS1 output against the host without lying to yourself.

It also cost pride. A straight interpreter port is cleaner to brag about. A hybrid pipeline is less romantic. It says: the console is not running the whole original engine. It is running the right artifact for the hardware.

I am fine with that trade. The disc plays.

What it bought

It bought a real acceptance bar.

FISHING 1 can be validated because there is a reference to validate against. The scene either matches or it does not. The sound either lands on the right frame or it does not. The runtime either survives night, tide, holiday, and raft variants or it does not.

The rest of the project inherited that bar. Sixty-one scenes remained at the time of the pivot; the 63-scene grind closed on 2026-05-05 with ACTIVITY 9 as the final row, and the post-validation work moved to the performance arc.