Press Raven · The Mechanics Catalog · Coverage Audit
Yes on motivators — the second brainstorm closed the Building gap outright. But the families are only one axis, and the other one has two holes with nothing built in them at all.
Family coverage is now comprehensive, and Building is no longer the gap. It went from the thinnest family — two entries, both unbuilt — to the second-largest.
But roughly forty of the ideas in that batch collapsed into mechanics the catalog already had. The genuinely new count is seventeen, and where they landed is more interesting than how many there were.
| Family | Was | Now | Distribution |
|---|---|---|---|
| optimizing | 6 | 8 | +2 |
| building | 2 | 7 | +5 |
| finding | 3 | 6 | +3 |
| dodging | 3 | 6 | +3 |
| surviving | 3 | 5 | +2 |
| discovering | — | 1 | new — proposed |
| unplaced | 1 | 2 | layers |
Building was filled by Detective Board, Evidence Chain, the Balatro synergy idea, the bomb-disarm formula grab, and Restaurant Rush — five entries that all ask the player to construct something rather than pick it. That was the exact hole flagged after the first seed, and it closed without anyone aiming at it.
Eight separate ideas across both brainstorms are one loop wearing different clothes. They are now all filed as instances of stream-go-no-go:
This is the catalog earning its existence in one move. Eight ideas that read as eight games are one mechanic and three tuning dials — decay speed, false-positive cost, and whether the criterion changes mid-round. Wildlife Photography isn't a new game; it's Contraband Scanner with the false-positive cost raised until restraint becomes the skill. That's now written down in the entry.
Similar, smaller collapses: Confidence Betting and Confidence Duel into confidence-wager; Prediction Games, Explain Before Reveal and Prediction Freeze into predict-before-reveal; Mini Metro, Resource Allocation and Resource Puzzle into shared-clock-triage; Information Economy into minimum-evidence; Shortcut Race into call-then-execute.
The Police Checkpoint was filed separately as Stamp or Reject, against the obvious merge into the stream. Two things differ and both look load-bearing: the criteria escalate in complexity across a session rather than staying fixed, and the fiction gives the player standing rather than a score. The newspaper-copy editor and the construction-permit inspector are this entry, not the stream. If escalation turns out to be the only real difference, it merges back — the seed note says so.
This is the finding I'd most want to hand back. The five motivators organise by why the player cares. They are completely blind to what the player physically does in response — and two mechanics can share a family while asking for utterly different cognitive work. Tapping where something feels wrong and writing the rule that covers ten examples are both "Finding" adjacent, and they are not remotely the same act.
Sorting all 35 entries on that second axis produces the real coverage picture. Cell format is total(built):
| Family | gate | locate | select | allocate | assemble | produce | layer | total |
|---|---|---|---|---|---|---|---|---|
| finding | 1(1) | 4(0) | 1(1) | · | · | · | · | 6 |
| dodging | 2(1) | · | 3(1) | · | · | 1(1) | · | 6 |
| optimizing | · | · | 4(2) | 4(3) | · | · | · | 8 |
| building | · | · | 1(0) | · | 4(0) | 2(0) | · | 7 |
| surviving | · | · | 2(1) | 2(0) | · | 1(0) | · | 5 |
| discovering | · | · | · | · | · | 1(0) | · | 1 |
| unplaced | · | · | · | · | · | · | 2(1) | 2 |
| total | 3(2) | 4(0) | 11(5) | 6(3) | 4(0) | 5(1) | 2(1) | 35(12) |
Nothing built asks the player to locate, and nothing built asks them to assemble. Of twelve built mechanics, five are select and two are gate. Exactly one asks the player to produce anything — Go Undercover, where you voice the con's reasoning.
Both games are, structurally, tap-and-choose games. Every mechanic that would have the player point at a place, build a structure, or generate an answer from nothing is currently a seed.
This matters more than the family count because response mode is what bounds how deep the learning goes. Recognition is real and worth training — that was the founding thesis, and the whole Recognition Under Pressure section of the second batch defends it well. But a catalog that is 16 of 33 select-or-gate is a catalog that can only ever ask "do you know it when you see it." The Building column and the locate column are where "can you do it" lives, and neither has been tested in a real build yet.
The cheapest test is Fault Sweep — hidden-object mechanics are well understood, the content is paragraphs you already have, and it is the only locate entry with no unsolved asset problem. Cluttered Search, the other obvious candidate, is still blocked on scene art that doesn't exist.
Discovering — "What's the pattern?"
Rule Discovery doesn't fit the five. Nothing is located, nothing is assembled, nothing is being avoided or optimised or survived. The player's felt question is what the unstated principle is, and the payoff is the aha — a motivational hook the other five don't name.
If Discovering survives review it probably also claims Where Experts Look (currently Finding) and Compress It (currently Building), which would make it a family of three rather than a family of one. That's the test I'd apply: a sixth family that only ever holds one entry is a mis-shelved entry, not a family.
Two others strain their shelves and are worth a second look — Inspection Budget sits in Optimizing because it spends a resource, but what it actually trains is suspicion, which is Dodging's territory. Bank or Push is filed Surviving but is probably a layer.
Deferred, as agreed — but the batch produced hard evidence, so it's filed rather than lost. The Container now sits unplaced next to The Boss Gauntlet: the Escape Room, the Safehouse, and the Courtroom are not mechanics, they're shells that give three or four unrelated loops one reason to sit together. And Bank or Push and Roster Attrition both work as wrappers over any mechanic that produces a running score.
So there are now at least two distinguishable non-mechanic kinds — layers that wrap one loop, and containers that house several — plus three entries that suspect they belong in one of them. That's a real schema question with real evidence behind it whenever you want to open it.
High on the family answer and on the collapse — both are countable and both hold however you argue the edge cases. High on the response-mode holes, which survive any plausible reshuffling of individual entries.
Lower on the seventeen individual family assignments, which were made from prose descriptions rather than from playable builds, and lower on response mode as a taxonomy — it's a useful lens that has not yet had to survive contact with a real design decision.