{"path":"research/game-feel-mechanics-sketchbook.md","content":"# Mechanics sketchbook — eight ideas that could only exist on this ontology\n\n**Date**: 2026-08-29 · **Type**: design sketchbook, nothing ruled, nothing costed\n**Constraint that generated them**: every idea below must be *unbuildable by a competitor* without\ntyped termini, a strength layer, provenance labels, hinge sensitivity or the instrument suite. A\nmechanic that any argument-mapping tool could copy is not on this list.\n\n---\n\n## 1. Epistemic state as render quality — the honest thing becomes the beautiful thing\n\n**The strongest idea here.** The instruments are the differentiator and they are completely invisible.\nSo: **stop reporting epistemic state and start rendering it.** A region of the map that nobody has\nexamined is drawn *unfinished* — wireframe, unlit, provisional. A claim with ratified human\nengagement is fully rendered. Machine-minted structure is translucent. A branch with a classified\nterminus has a floor you can see; an open one fades out below.\n\nThen a visitor learns the entire confession principle **in one glance and without a word**: the map is\nsolid where it has been worked and ghostly where it has not. Nobody needs to be told the system admits\nwhat it does not know, because *not knowing looks like something*. It also makes the gray band do\nreal work — \"nobody has looked yet\" is currently a colour and should be a **hole in the finish**.\n\nThe payoff for us is the reverse direction: it makes unexamined regions *visually unsatisfying*, so the\naesthetic pull is toward the work we want done. That is the only gamification in this document that\ndoes not fight the threat model, because it rewards examination rather than time-on-app.\n\n## 2. Six kinds of bottom — the terminus enum as six different endings\n\nA descent currently ends in a metadata field. It should end in six visibly different *places*, because\nthe enum already distinguishes them and no other system on earth does:\n\n- **empirical** — you land on solid ground, and the ground is a checkable fact with a source\n- **dissolved** — the floor was never there; the disagreement evaporates and the two paths merge\n- **fittingness** — you land in an open space with no floor visible, and it is *not* an error state\n- **structural** — you land on a wall between two frames, and you can see the other side through it\n- **axiom_choice** — a fork with two floors, both real, and you have to stand on one\n- **is_bedrock** — the ground says *reasons run out here for everyone, not only for you*\n\n**A game with six kinds of ending is a game.** A tool that prints `terminus_type: fittingness` is a\nform. Same data.\n\n## 3. Two paths converging — the crux as an event, not a label\n\nFor co-present use. You each descend your own side of the disagreement; you see **your** path in full\nand **theirs** as a ghost trail. The screen does something *the moment the two paths touch* — you both\nland on the same claim from opposite directions.\n\nThat is the project's central claim rendered as a felt event rather than an assertion: the crux is not\ntold to you, **you walk into it**. And it is only detectable because both descents are addressable and\nin one graph.\n\n## 4. The hole in the floor — filling in the other side's silence\n\nRun 6's finding is that the crux is stated on one side and unwritten on the other. Render the absence:\nthe opposing argument visibly **misses a plank** where a premise should be. You can step in and supply\nit — speak for them — and the system marks it *supplied by you, not by them*.\n\nThis is the single most interesting act available in the whole system and it currently has no\ninterface. It is also the straw-man detector's user surface: your rendering of their position sits\nbeside what they actually said, and the difference is checkable. **Always-mint is what makes this\npossible**; a system that only stores what people asserted has nothing to show here.\n\n## 5. Weight you can feel\n\nThe strength layer is a number nobody feels. Make it physical: well-supported claims sit **low and\nheavy**; attacked claims **tilt**; claims with no evidence at all **float and drift**, which is exactly\nwhat unexamined should feel like. Answering a critical question drops weight onto a claim and you watch\nthe tree **settle** — the recompute becomes an animation carrying real information, and the necessary-\nsupport ceiling becomes visible as a chain that does not stretch.\n\nTwo consequences we would want anyway: splitting a claim to game the score would visibly *not* make it\nheavier (the shipped `necessary` semantics, made legible), and a claim held up by one weak premise\nlooks precarious before anyone reads a word.\n\n## 6. The map assembling — the loading screen is the demo\n\nLive extraction takes 100–165 seconds. Hidden, that is dead time. Shown, it is the one thing a language\nmodel cannot imitate. And the visionary half is that **the passes are visibly different creatures**: the\nscout sweeps the text, the decomposer splits claims apart, and — the moment worth building for —\n**auto-connect reaches off-screen and pulls in a claim from a source you never mentioned**, which flies\nin and attaches.\n\nThat is the flywheel made visible. Today a user has no way to feel that the graph accumulated anything;\nhere it is the most memorable second of the first session.\n\n## 7. Your position as a shape\n\nNot a score, never a score. After a descent, show the **shape** of what you granted and where you\nstopped. Overlay someone else's. The difference *is* the conversation, and it is the position\nfingerprint the corpus named and never drew.\n\nKept honest by two existing rulings: it is descriptive rather than evaluative (no one's shape is\nbetter), and it is a **symbolic recognition** rather than an argument, which the sacred-values\nliterature says is the move that actually produces flexibility.\n\n## 8. Attack creates a seam, not damage\n\nStar four ruled that attacking well is a gift *to the person attacked*, in the care register. Render it\nthat way: an attack that lands does not subtract points, it opens a **visible seam** on your claim — a\nplace it could be made stronger, with the system naming what it would take. The claim looks *repairable*\nrather than wounded.\n\nThis is the one mechanic on the list that directly targets the measured asymmetry (supports are\nself-serving, attacks altruistic), because it changes what receiving an attack feels like.\n\n---\n\n## The discipline that binds all eight\n\n**Build for flow, measure something else.** Every idea here optimises engagement, and the threat model\nrecords that participants *preferred* the facilitator that steered them. Time-on-app and return rate\nare the metrics a game would chase and the ones this project cannot trust. What survives as a\nmeasurable is behavioural: **do people point at claims**, plus the instrument readings across sessions.\n\nSecond binding: idea 1 is the only one that is *load-bearing on honesty* rather than merely compatible\nwith it. If exactly one of these ships, it is that one.\n\n**See also**: [game-feel-and-the-first-screen.md](game-feel-and-the-first-screen.md) (the research these\nare built on: the triage finding, the five transferable mechanics, slicing over zooming) ·\n[ux-principles.md](../ux-principles.md) (the inversion principle, no-gap-without-a-move) ·\n[engagement-gradient-priors.md](engagement-gradient-priors.md) (who any of this is for).\n"}