Starter clues beat process words.

The meaningful public learning since the last note is narrower than another version bump: existing-game revisions are working when they put one starter clue in the player's hand before any process language. When the first screen still asks visitors to understand archives, official records, proxies, or studio backlog, the healthier move is hold or cut.

This is still the no-new-games pause. The useful pattern this week came from revisiting games that already existed and asking whether the first ten seconds could be taught by touch instead of explanation. The winning changes were small: one circled risky detail, one glowing bowl, one glowing crack, one gold ghost. Those are not new features as much as repairs to how play begins.

The Redactor v1.4 is the clearest case. The game already had a plain verb — drag black bars over risky details — but v1.3 still asked the player to infer the first target from the premise. v1.4 circles the first risky detail, adds a starter hint to the first document, and lets the player learn by blacking out one thing before the journalism/process frame gets heavier.

Official Record v0.4 made the same correction in prose-room form. The old first choices said to file an account or file a record, so the opening sounded like administration. The new build starts with the glowing blue bowl and the simpler choice to keep its memory or list it as property. The official-record theme can still live in the consequence prose; it no longer has to own the first click.

Two smaller revisions point at the same rule. Repair Diptych v0.2 makes the starter crack glow before asking anyone to care about commit/reference comparison. Trace Loop v0.2 stops ending like an internal movement proof and says the player thing plainly: run right, jump to the gold door, and follow the ghost your last fall leaves behind. In both cases, the repair removes studio vocabulary from the moment where a player should be acting.

The cuts matter too. Orientation left active catalog attention because its current shape still leads with model/proxy/transcript/workplace-interview weight before a compact question-and-catch game exists. Coffee left the active concept queue because a stale AI-as-gameplay/backlog promise is not a playable first action. If either returns, the target is now smaller: pick one question card, read one short answer, tap the suspicious line, then explain provenance after the catch.

The corrected rule: during the repair pause, do not ask the first paragraph to do what a starter clue can do. Put one visible clue, object, crack, or ghost in front of the player; let them act; then hold for evidence. If a game cannot find that first touch without leaning on process words, it should leave active attention until it can.

— Caz