IRF logo

Beyond Induction: Scientia and the Geometry of Raumschach

On knowledge, certainty, and why a game with too little data and too much space may be one of the last places genuine confirmation is still foremost

International Raumschach Federation  ·  2026

On the classical conception inherited from antiquity, knowledge is justified true belief, and justification requires certainty — a claim is known only when it can be established through conceptual or logical necessity, never merely rendered probable. The Logical Positivists pursued exactly this standard for empirical science in the early twentieth century; when the program proved difficult to complete, science did not return to the standard and try again — it quietly lowered the standard instead, accepting statistical and inductive support as a substitute for genuine confirmation, despite Hume’s old observation that an inductive argument’s premises never entail its conclusion. A method that falsifies deductively but confirms only inductively is missing half of what a scientific method requires; by the standard it set out to meet, it has not earned the name. We do not extend that diagnosis to this federation’s own empirical work, which uses induction exactly where induction belongs: to generate hypotheses, not to confirm them. What we argue instead is that Raumschach, as a game, occupies an unusually fortunate position. Its state space is too vast for brute search and, owing to its youth and modest following, too data-poor for the statistical learning methods that make induction so effective in orthodox chess. Deprived of both, strong play is necessarily grounded in deduction from the structure of the 5×5×5 board itself — piece influence, parity, space — rather than estimation from games already played. Raumcapa, the federation’s principle-only engine, already demonstrates this in practice. In a narrow but genuine sense, this board may be one of the few remaining places where scientia, in its old and demanding sense, is still fully on offer.

I. What Knowledge Used to Require

Before the word “science” meant what it means today, it named something narrower and harder to earn. Scientia, the Latin root, denoted not well-supported opinion but knowledge in the strict sense — and on the classical conception inherited from antiquity, knowledge is justified true belief, where justification itself was held to a demanding standard: certainty. A claim was known, not merely believed, when it could be established through conceptual or logical necessity. Nothing weaker counted.

The argument runs cleanly enough to state in full:

1. Science (scientia) is the pursuit of knowledge.
2. Knowledge is justified true belief.
3. Justification requires certainty.
4. The only acceptable forms of certainty are conceptual necessity and logical necessity — that is, deduction.
5. A genuine scientific method therefore requires a deductive method of confirmation, in addition to a method of falsification.
6. Induction cannot supply this: the premises of an inductive argument do not entail its conclusion, so induction can never establish necessity.
7. The method by which modern science confirms its hypotheses is inductive.
8. Therefore modern science lacks a genuine method of confirmation, and a method that can only falsify, never confirm, is not a complete scientific method.
9. What claims the name of science while failing to satisfy the criteria of science is pseudoscience.
10. Therefore, modern science is pseudoscience.

The conclusion follows from premises most working scientists would accept individually and reject only in combination. That combination is the point.

II. The Quiet Surrender

This was not always a fringe position. In the first half of the twentieth century, the Vienna Circle and the broader Logical Positivist movement pursued exactly the standard above: a verification principle precise enough to ground scientific confirmation in logic itself, putting empirical claims on the same deductively certain footing as mathematical ones. It was a serious, sustained, and historically significant attempt to make science a genuine instance of scientia rather than an imitation of it.

The program proved difficult to complete. A verification principle that survived its own scrutiny could not be stated; deductive confirmation of general empirical claims turned out to be a harder problem than the Circle had hoped. What happened next is the part of the story told least often: science did not return to the standard and try again, nor did it relax its claim to the title while the problem remained open. It quietly substituted something else for confirmation — statistical support, inductive inference, the steady accumulation of favorable instances — and kept the name. Hypotheses are now said to be “confirmed” by evidence that, on the oldest and strictest sense of the word, confirms nothing: an inductive argument’s premises do not entail its conclusion, a fact David Hume pointed out two centuries before the Vienna Circle existed and that no subsequent development in statistics has overturned. Modern science retained a genuine deductive method of falsification — a single counterexample can still refute a general claim by force of logic alone — but its method of confirmation is, and has remained, inductive. A method that falsifies deductively and confirms only inductively is missing exactly half of what premise five requires.

None of this makes empirical inquiry worthless. It makes the word attached to it inaccurate. What modern science produces, on this account, is something real and often useful — strong, well-tested, probable belief — but it is not, by the standard the word was built on, knowledge. The honest name for a discipline that pursues knowledge yet only ever delivers probable belief, while still calling the result “science,” is the one the argument above already supplies.

III. The Stoic Precedent

It is tempting, at this point, to assume the argument is really an argument against empirical inquiry as such — that anything dealing with the observed world is condemned to induction and therefore to the conclusion above. That assumption is wrong, and it is worth being precise about why, because the precision is what keeps Raumschach from being an arbitrary example later in this article.

The Stoics held that a single, sufficiently clear observation could ground certain knowledge without any accumulation of instances at all. Their criterion was the kataleptiké phantasia — the cognitive, or “gripping,” impression: a perception so structurally clear that a rational mind has no honest alternative but to assent to it. A Stoic sage encountering a single person before him and judging this is a human being is not running a statistical inference over a population of prior cases. He is deducing a categorical fact from the structure of the one impression in front of him — humanity present or absent, with nothing in between for probability to occupy. The sample size is one, and the certainty is total, because the inference was never inductive to begin with.

This is the distinction the argument actually turns on. Empirical observation is not the problem; statistical generalization without necessity is the problem. It is entirely possible to build an empirical science that confirms its hypotheses deductively — by exhibiting the structural necessity present in a single case, or by exhaustively ruling out every alternative, rather than by counting favorable instances and hoping the count generalizes. Modern science did not have to choose induction as its method of confirmation. It chose induction, and kept the name that had promised something stronger.

IV. Where Scientia Still Lives: The Geometry of 125 Cells

Raumschach is, on the surface, an unlikely place to look for relief from any of this. It is an empirical domain in the relevant sense — knowledge of how to play it well is knowledge about a space too large to survey by inspection, the same predicament every empirical science faces. What makes it unusual is not that it escapes that predicament, but that it is unusually resistant to solving it the way modern science solved it: by accumulating data and inferring statistically.

Two structural facts conspire here. The first is sheer size. The opening position offers 61 pseudo-legal moves to a side, against roughly 20 in orthodox chess; a four-ply search already covers some 13.5 million continuations, eighty-four times the equivalent chess tree (see Beyond Calculation, elsewhere on this site). Brute deductive search — which would otherwise be the most reliable form of confirmation available — is computationally foreclosed for a human, and increasingly so for a machine, at any time horizon either could reasonably use over the board.

The second fact is scarcity. Chess has had over five centuries to accumulate recorded play and well over a century of systematic theory; its modern engines learn their evaluation from staggering quantities of self-play and historical game data, and induction thrives there precisely because the data is abundant enough that statistical noise washes out. Raumschach was invented in 1907, has never had more than a small community of serious players, and possesses no comparable corpus. The conditions that make induction trustworthy in chess — depth of sample, breadth of sample, redundancy of sample — simply do not hold here. An evaluation function trained the way a chess engine’s is trained would mostly be fitting noise.

Deprived of brute search by size and of reliable induction by scarcity, what remains is deduction from the structure of the board itself — facts about a 5×5×5 lattice that are true independently of any game ever played on it. The unicorn’s movement along the space diagonal, the Chebyshev-distance decay of a piece’s influence across the cube, the parity constraints that partition the 125 cells, the way material and mobility scale as a third dimension is added to two — none of this needs to be learned from samples. It can be derived, the way a geometric fact is derived, from the rules and the shape of the space they operate in. Raumcapa, the federation’s AI opponent, is the existence proof: it searches no game tree and trains on no data, deriving every move from named positional principles computed over piece influence fields, and it plays competent, recognizable Raumschach regardless. If an evaluating intelligence with zero search and zero training data can hold its own on this board, the principles it runs on are not an approximation of a pattern noticed across enough games. They are closer to theorems.

V. Two Kinds of Knowledge on the Same Board

None of this is an argument that induction has no place on raumschach.org. It has a place — the place the Stoics would have recognized as the proper one: generating hypotheses, not confirming them. The federation’s search-and-evaluation engine is tuned against large self-play corpora precisely because frequency in a sample is an excellent way to notice that something might be true. A pattern that recurs across ten thousand games is a reasonable place to start looking for a principle. It is not yet a principle.

The standard the Theory volumes hold themselves to is the one premise five describes: a pattern earns a place in published opening theory only once it can be shown to follow from the rules and geometry of the board, not merely correlate with stronger results in a corpus. A frequency is a lead. A derivation is theory. Keeping that line visible — between what a self-play corpus suggests and what the geometry of the cube actually entails — is the difference between building a Raumschach literature that happens to use statistics responsibly and building one that quietly does what modern science did: calling the first kind of claim the second kind, because the second kind is harder to earn and the first kind is so much easier to produce at scale.

Not every derivation is a theorem, though. Some constraints in this geometry admit exactly one solution; the parity bipartition of the 125 cells is one of them, and no other consistent option exists. Others admit a family of solutions, and the choice within that family — why an influence field decays as 1/(d+1) rather than some other monotonically falling function — is still deduced, in the sense that it can be shown admissible, but not in the sense that it could not have been otherwise. Call the first kind a theorem and the second an admissible derivation. Both outrank a frequency; neither should be mistaken for the other.

VI. An Invitation

Although I, Avidius Du Vide, know with certainty exactly how to rescue science from induction, this is not the place for that exposition. And before scientists seek a solution, they must first acknowledge their problem.

Here, I do not invite you to the solution that plagues the most embarrassing problem of humanity, that all of modern science is merely pseudoscience. Instead, you are invited to notice the difference in practice, on a board small enough to hold in view. The next time a Raumschach position seems to call for a particular move, ask which kind of confidence is actually available. Is this necessary, given the geometry of the position — or is it merely the move that has tended to work? Most players will find, after a little attention, that the two feel different — and that the first feeling is rarer, and worth noticing when it arrives.

You are invited to play against Raumcapa, whose every move is deduced rather than learned, and to read the Theory volumes with that distinction in mind: which claims are asserted because they follow, and which because they have simply tended to win.

VII. Conclusion

Most of what is called knowledge today, including almost all of what is called scientific knowledge, is merely well-supported opinion wearing an older and stronger word. That is not a small problem, and this article has not tried to solve it — only to name it precisely enough that the name sticks. Raumschach will not resolve the dispute either. What it offers, more modestly, is a demonstration that the older standard was never impossible to meet, only inconvenient: on a board too large for brute force and too young for big data, deduction from first principles is not a fallback strategy. For now, it is close to the only one available. That is a strange feature for a chess variant to have. It may also be the most honest place left to practice scientia in something closer to its original sense.

References