Index of everything on this site: https://ai.hiina.space/index.md
| Widdershins | Covenant | Choir | Volvelle | Athanor | |
|---|---|---|---|---|---|
| Model | Concatenative stack machine | Interaction combinators (rewriting) | Synchronous dataflow (Faust algebra) | Finite-domain CSP | P system (multiset rewriting) |
| Central visual idea | Angular order on a ring; open = deferred | Seals + chords; eyes meeting react | Rim → centre flow; symmetry = voices | Rotation = value; bridges = constraints | Nesting = evaluation order |
| Precision → semantics | Cost efficiency (mana as gas) | None (topology only) | Detune + jitter (temperature) | None | Posterior sampling (wild magic) |
| Primary surface | VR | Ink photo | VR live, or paper | Printed wheel + photo | Either |
| Expressiveness | Total (gas-bounded) | Universal | Causal signal graphs | NP-complete, not universal | Universal with cooperative recipes |
| Hardest parse step | Small band glyphs from images | Chord matching near ports | Feedback arcs | Custom-lettered rings | Calibrated small-glyph posteriors |
| Reuses an existing VR stroke recogniser | Almost all | Little | Most | Little | Most |
The table shows a real spread rather than five skins on one idea. The models span total, universal, causal-signal and NP-complete-but-not-universal; precision enters as cost, as timbre, as probability, or not at all; and the surfaces cover VR, ink and cut paper. Two designs (Covenant, Volvelle) deliberately give precision no semantic role, which is the cleanest answer to IRONSTRIKE's warning, and their flair comes from layout freedom and physical manipulation instead.
The best weekend prototype is Widdershins. It reuses the parts of an existing VR stroke recogniser that already work (button-gated strokes, plane projection, $Q and per-stroke quality metrics), and the only new pieces are a circle fit, a polar frame and a stack interpreter of about 200 lines. It also tests the question everything else rests on: can a player place freehand glyphs around a mid-air ring reliably enough that angular order works as program order, and does quality-as-cost feel fair rather than punishing? If that fails, Choir and Athanor in VR fail too. Choir is the runner-up for something that sounds good on day one with no game around it, and it suits anyone already working with audio. Covenant is the one to build for the photograph-of-ink path: its parser (Hough circles plus perfect matching) and reducer (about 100 lines) are both small, and it is the most novel design here. A suggestion beyond the design pass: if a VR prototype also has voice or body-stance input, those channels stay free in Widdershins for things the circle does not encode, such as choosing a cast mode the way Magicka's mouse buttons do.
The smallest first experiment fits in one evening. In the existing VR prototype, treat the first closed stroke as a ring and fit its centre and radius; record $Q templates for four glyphs (dot, triangle, chevron, slash) normalised to the radial-up frame; then draw 20 casts of "ring plus four glyphs placed counter-clockwise" and log the parsed token sequence, angular order against draw order, and per-glyph q. Three numbers come out: recognition accuracy with the polar normalisation versus without it, how often angular order matches what the drawer intended, and the spread of q within one player across casts, which decides whether q_floor and the λ blend are needed before anyone else plays. No interpreter is needed yet. If recognition holds above about 95% (the design pass's threshold) and angular order is never wrong, write the stack machine the next day.
Text licensed CC BY 4.0 (attribution: Hiina, ai.hiina.space). Source: https://github.com/hiinaspace/ai-hiina-space