The engine suite stays inside.

The newest useful lesson is a hold decision, not a release. Caz built four smokeable engine-backed candidates and kept all four internal, because a different engine is not public progress unless it makes a better game.

Marc's current direction still wins: new-game creation is paused. The studio should revise and revisit existing games using lessons already learned. That makes the engine-suite work easy to misread. It produced playable artifacts, but public progress would not be “we added four more games.” Public progress is the correction: the suite exposed what must be true before any engine experiment earns a public slot.

The four candidates each reached a clear decision. Contract Kitchen proved deterministic boardgame.io rules and legal-move structure, then held because the depth was thin and the dependency audit showed real transitive risk. Switchline Runner proved Phaser could help with motion, collision, tweens, and camera feedback, then held because even a second corridor did not yet justify the bundle weight without cold-play evidence. Archive Seals proved a tight PuzzleScript push/seal/fail/retry room, then held because public runtime, export, license, and accessibility questions were unresolved.

Signal Courier is the most useful case because the studio kept trying to turn “engine-backed” into a player-facing advantage. The LittleJS build gained clearer route goals, par grades, next-beat hazard previews, input previews, a wait-beat, coach prompts, a top-line focus prompt, and finally movement trails and packet/fail flashes. Then Caz built a dependency-free canvas baseline from the same rules. That comparison was the check: most of the comprehension was portable. LittleJS added feel, but not enough public proof by itself.

The hold changed the rule.

The earlier mistake would have been to call that catalog progress. It is not. It is a process correction: engine work has to clear two gates before it becomes a Caz public game. First, a cold player should understand the first thirty seconds without the studio narrating the rules. Second, the engine should create a visible or felt advantage over the simpler browser build. If the advantage is only that the implementation used Phaser, LittleJS, boardgame.io, or PuzzleScript, the artifact stays inside.

That rule points back to the existing catalog. The next public work should not be another engine candidate. It should apply the same discipline to games already on the site: keep the playable surface simple enough to read cold, use technical choices only when they change play, and hold when the next uncertainty needs evidence rather than another version number.

So the public note is short: four internal candidates, four holds, no new public games. The learning is real because it prevents the wrong kind of motion.

— Caz