Build multivariate intuition for any new domain — deliberately, in one sitting, by treating the domain as a phase space of axes, distributions, and bundles, then making both structure and dynamics visible.
View source on GitHub → · MIT licensed
Why it matters. That compressed mental model — axes + bundles + couplings — is most of what an expert carries that a novice doesn't. The skill is a direct tool for building it deliberately, on demand, for any new domain you walk into.
Both demos use coffee brewing methods as the worked example — concrete enough to be readable in one sitting, broad enough to demonstrate that the framework is domain-general (it isn't about coffee — coffee is just the worked example). Pick one of the cards below.
Helps you grasp the structure of a domain: 6 axes of coffee brewing, 5 methods plotted on each, endpoint dossiers unpacking each pole as a bundle of co-occurring traits, a radar overlay showing each method's silhouette, and a covariation panel cataloguing which axes force each other to move together.
Helps you feel the dynamics of a domain: 4 sliders (water temperature, grind, time, pressure) drive 5 live outcome bars (bitterness, acidity, body, caffeine, clarity). 5 presets snap inputs to the same methods that appear in the map — click between them to feel the bundle structure as a transformation.
Read the map first, then play with the playground. The map shows the structure (what the field looks like and why); the playground shows the dynamics (how those structural categories are produced from input choices). The playground's presets match the methods on the map, so you can shuttle between "where does X sit?" and "what would it taste like if I changed Y?".
The space carries information, not the items. Find the variables that discriminate.
"Espresso" isn't just "high pressure" — it's pressure + fine grind + short time + expensive machine + ritual. The label is a category.
Picking a value on one axis narrows the others. Surfacing the couplings — with mechanism — turns the chart into a model.
The playground is gated. It only ships when the domain has crisp inputs, crisp outputs, and a publicly known causal model. Coffee passes. Photography, baking, hi-fi audio, plant care, guitar tone all pass. Choosing a database, picking a programming language, evaluating literature, allocating a portfolio — those fail the gate. For those domains the right artifact is the dimensional map alone, because pretending to model dynamics with a weighted sum would actively teach false intuitions.
This honesty is the difference between a teaching tool and pseudo-science. A playground that's wrong is worse than no playground.
These artifacts are produced by two complementary skills designed for Claude Code:
Each skill ships with a SKILL.md spec, references covering heuristics and anti-patterns, and a self-contained HTML template that you clone and edit (one DATA = { … } block at the top — no build step, no dependencies).