↩ ExploRational · A Catalogue of Divergent Rationality
explorational · a divergent move
When you decompose a problem into parameters and walk every combination — including the absurd ones — the empty cells are the discoveries. But the space explodes: a handful of parameters yields hundreds of configurations, six yields thousands. How do you enumerate exhaustively without drowning?
Fritz Zwicky, the astronomer who catalogued supernovae and named dark matter, had a method for inventing things nobody had thought to invent. He called it the morphological box: name the independent parameters of a problem, list the possible values of each, and the parameters-times-values grid enumerates every configuration the problem admits. Zwicky ran his box over jet and rocket propulsion in the 1940s and produced a table of concepts, many of which no one had built, some of which no one had even considered absurd enough to reject. The point of the empty cells is precisely that they are empty: no one had yet checked whether the combination they name was impossible or merely unimagined. Enumeration converts "I can't think of another option" into "here are the 141 options I haven't examined."
The absurd cells are the discipline. A single vivid design fills the imagination and crowds out its neighbours; the box refuses that monopoly by generating the neighbours mechanically, so the combination you would never have proposed still gets its cell. But the same multiplication that makes the box exhaustive is the trap the catalogue names for this move — endless deferral. Four parameters with a few values each already run past a hundred configurations; add a fifth and a sixth and you are staring at thousands, more than any afternoon can walk. The move opens options by multiplying, and multiplication is exactly what runs away from you. The instrument below is an honest morphological box: real combinatorics, and the one discipline — cross-consistency pruning — that keeps exhaustion possible.
the signature instrument
Each row is a parameter; each chip a value. The total is the product of the value counts — it updates the instant you add a value or a row. Click chips to walk a configuration and mark it explored, novel, or ruled-impossible. Switch to link mode to mark two values incompatible, and watch the viable count collapse below the raw product. The displayed values are computed live.
The transport box
4 parameters · a way to move people across a city
Total combinations · Π of the value counts
—
every configuration the box admits, most never built
Viable after cross-consistency pruning
— ruled out by 0 incompatibilities
the empty cells are the discoveries
A design problem does not arrive as a grid; it arrives as a few salient options and a strong feeling that those are the options. The morphological box breaks that feeling by refusing to reason about whole designs at all. It reasons about dimensions: what, independently, can vary? A way to move people across a city varies in its medium (road, rail, air, water), its power (human, engine, gravity, magnetic), its guideway (fixed, free, tracked), its occupancy (individual, shared, mass). Four parameters, and already the product is 4 × 4 × 3 × 3 = 144 configurations. Most of them have never been built. A few of them are things you use every day. And a handful are sitting in empty cells because no one ever generated them to look.
The move therefore opens options by multiplying rather than by listing. A list is bounded by what you can call to mind, and what you can call to mind is dominated by what already exists. The product is bounded only by the parameters themselves, so it reaches configurations no memory would have surfaced: gravity-powered, tracked, mass transit is a funicular; magnetic, tracked, mass transit is a maglev; the cell for magnetic, free, mass transit is empty — and the interesting question is whether it is empty because it is impossible or empty because no one has tried. Zwicky's claim was that the second category is far larger than anyone expects, and the only way to tell the two apart is to enumerate the cells and walk them one at a time.
what to try
On the transport box, add one value to any parameter — say a fifth medium, "underground." Watch the total jump by a whole factor: 144 becomes 180, because you multiplied a dimension of five against everything else. Adding one option does not add one configuration; it adds a slab of them.
Switch to link mode and forbid pairs — "air" with "tracked," "water" with "fixed," "magnetic" with "free." Each rule deletes every configuration containing that pair, and the viable count drops well below the raw product. Load sparse after pruning to see 144 collapse to a walkable 27. Pruning, not generating, is what makes exhaustion possible.
Load explosion — six parameters, 2,304 configurations, no constraints. Now try to walk it. You can't; that is the point. Add a seventh parameter and the total triples again. Without the discipline of constraints, exhaustive enumeration stops being a method and becomes a way to never decide.
pruning makes exhaustion possible
Zwicky knew the box was useless if it only multiplied. His own method carries a second step he called cross-consistency assessment: go through the values pairwise and mark the pairs that cannot coexist — not the whole configurations, which are far too many to judge, but the pairs, of which there are few. A configuration is then viable only if it contains no incompatible pair, and the enormous majority of the raw product is eliminated by a small number of pairwise rules. In the box above, four honest incompatibilities on the transport parameters cut 144 down to 102; nineteen of them cut it to 27. The cross-consistency step is the reason a morphological box scales at all: you judge O(pairs) constraints and get O(product) eliminations for free.
The instrument computes this the honest way. Marking a pair incompatible recomputes from scratch how many complete configurations survive when every configuration containing that pair is struck out — including the overlap where a single configuration would have been killed by two different rules at once, which must not be double-counted. That last subtlety is why the viable count sometimes drops less than you expect when you add a rule: the configurations it forbids were partly forbidden already. What the box gives you at the end is the exact set of internally consistent designs — small enough, if you have pruned honestly, to walk cell by cell and ask of each empty one: impossible, or merely unbuilt?
model
The total combination count is the plain product of the value counts, Πᵢ |valuesᵢ| — four parameters of sizes 4, 4, 3, 3 give 4·4·3·3 = 144, and adding one value to a size-3 parameter multiplies the total by 4/3, from 144 to 192. The viable count is the number of complete configurations — one value chosen per parameter — that contain no pair you have marked incompatible. The instrument obtains it by inclusion–exclusion over the forbidden pairs: it sums, over every subset of the rules, a signed count of the configurations forced by that subset, so overlapping eliminations cancel exactly rather than being counted twice.
That inclusion–exclusion count has been checked against a brute-force enumeration of every single configuration: on the transport box (144 raw, 4 rules → 102 viable), on the sparse box (144 raw, 19 rules → 27 viable), and on the explosion box (2,304 raw, 0 rules → 2,304 viable), the two methods agree to the integer. The walk tally counts the configurations you have personally marked and reports how many viable ones remain unwalked. And the curse-of-dimensionality meter is just the product read as it grows: each parameter you add multiplies the total by its number of values, which is why the total climbs geometrically and why, without pruning, "walk every cell" is a promise no one can keep.
the move ↔ the box
| The move | The box |
|---|---|
| a parameter | an independent dimension of the design — a thing that can vary on its own. |
| a value | one option on that dimension; a chip in the parameter's row. |
| the product | the size of the possibility space — every configuration the parameters admit. |
| a configuration | one complete option: exactly one value chosen per parameter. |
| an incompatible pair | a real-world constraint — two values that cannot coexist in any workable design. |
| the viable count | what is left worth walking once the impossible configurations are struck out. |
how this opening fails
risk · endless deferral
The catalogue's named risk for this move, and the box makes it literal. Each parameter you add multiplies the total, so six parameters already put thousands of configurations on the table and a seventh triples that again. An enumeration that cannot be walked is not rigour; it is a way never to choose, dressed as thoroughness. The discipline that saves the move is cross-consistency pruning — judging the pairwise constraints so the viable set shrinks to something an afternoon can actually cover. A box without constraints is not exhaustive. It is just large.
risk · the parameters are not really independent
The product is exact only if the parameters vary independently, and the pruning catches only incompatibilities you can name as a pair. But real design dimensions interact in ways no pairwise rule sees: three values can be individually compatible and jointly incoherent, or a "value" on one axis can silently change what the values on another axis even mean. So a configuration the box certifies as viable can still be nonsense the arithmetic cannot detect. Read the viable set as the configurations not yet ruled out — a sharpened agenda for judgement, never a catalogue of things that will work.
Lay out the parameters, take the product, and walk the cells no one thought to check.
can you use it?
RECOGNITION — Which is enumerating the combinations? A: listing ten ideas as they come. B: naming the parameters (say, channel × format × audience), then walking every cell of the grid — including the absurd ones. C: picking the best two ideas to combine.
B. It is systematic: parameters times values, every cell visited, the empty ones being the discoveries. A is unstructured; C prunes before enumerating.
THE NEAREST NEIGHBOR — Quantity before judgment also piles up options. What separates enumerating the combinations from it?
Structure. Quantity before judgment sets a free-form quota; enumerating the combinations imposes a parameter grid and forces every cell, so the gaps are visible and the absurd combinations get their day in court.
PRODUCTION — You are naming a product. Pick two parameters (e.g. tone × root-word source), give each three values, and read three cells of the nine — including one you'd normally skip.
A version: tone {playful, technical, mythic} × source {Latin, invented, everyday}. Playful+invented, technical+Latin, mythic+everyday ('Hearth'). Yours works if you generated a cell your taste would have deleted unexamined.