Behavioral Quotients

What It Solves#

Computational systems face one tension: retain exhaustive history and state grows without bound; discard it prematurely and you make commitments you cannot take back. Behavioral Quotients as a Design Principle for Computation formalizes state reduction as a quotient of histories by future indistinguishability — two histories are identified exactly when no future can tell them apart.

The paper is a unifying lens and a design criterion, not a new minimality theorem. Its contribution is to put one construction underneath several fields, and to state honestly which bridges are proven, which are analogies, and which are still conjectures.

Key Innovation#

One quotient, several names. The minimal sufficient state is the quotient, and it satisfies the universal factorization property of a coequalizer. In specific regimes this is:

Regime	What the quotient is
Finite deterministic	Myhill–Nerode congruence
Stochastic	Causal states (computational mechanics)
MDP	Exact state aggregation
Categorical	Coequalizer / final-coalgebra universal property

The generated geometry is the Gromov visual metric on a tree boundary and the Fisher–Rao geometry of a statistical manifold.

Evidence tiers, stated up front#

The paper marks every cross-regime bridge as a consequence, an analogy, or a conjecture — and says what would upgrade it:

Conjectural: prime weighting (§5.2) and the Riemann “asymptotic lock” (§5.3).
Externally verified: the solver line’s correctness results — witnesses checked against the original instances by separate verifiers.
Informal: the Laplace-horizon derivation.

This tiering is deliberate: the theory does not borrow the solver’s verification, and the solver does not borrow the theory’s novelty.

A reference implementation#

The solver line is a reference implementation of the design criterion. The same quotient principle appears in NitroSAT’s state reduction, in Geometry of Conditional Logic’s CRT repair, and at runtime in FUTCache.

Architecture#

Behavioral Quotient
├── Histories
│   └── H  ~  H'   iff no future distinguishes them
├── Quotient (minimal sufficient state)
│   └── universal factorization / coequalizer
├── Generated geometry
│   ├── Gromov visual metric (tree boundary)
│   └── Fisher–Rao (statistical manifold)
└── Instantiations
    ├── KV cache: answer-preserving compression
    ├── Continuous solvers
    └── Commitment / repair

Website#

Paper: sethuiyer.github.io/behavioral-quotients PDF: paper.pdf

Distill-style rendering of the source manuscript, with a continuous-verification pipeline that proves the page corresponds to the paper 1:1 — nothing added, dropped, reordered or reworded.

Connection to Core Vision#

The Arithmetic Manifold is a statement about the geometry of constraint systems. Behavioral Quotients is a statement about which distinctions that geometry is allowed to keep. It is the general form of the move the rest of the lab makes in specific settings: remove the distinctions that do not change the answer, then search the residual space.