Deutsch
Cartography · August 29, 2026 · Audit log

The
Feasibility­Space

Anyone generating a random instance is always subject to a limitation imposed by the format. Rules can never enrich randomness, only constrain it. The question is how much freedom remains after that — and whether hardness is hiding in what remains.

3-SAT, n = 60, at the threshold · 5304 bits drawn · 1 cell ≈ 40 bits
44.6 % format coercion
discarded by the format — 2364 bits mere renaming — 332 bits genuinely free — 2608 bits
Finding 1 · holds

The bit budget

A naive generator draws a fixed number of bits for every instance. A large share of that produces no valid instance at all — the format throws it away. A second share does produce valid but indistinguishable instances: mere renaming of variables and signs. What remains after that is the space in which the randomness actually moves.

ClassRaw bitsvalidsymmetrygenuinely freediscard

The discard rate lies between 44 and 58 percent across all four classes. Roughly half of the random draw is not freedom but format coercion. That is the limitation, and it is a measurable quantity.

Test bench

Five candidates for a structure measure

What was sought was a measure that distinguishes between randomly generated problem classes and carries the hardness. Every candidate was pitted against a control that could topple it.

MEASURE 01

Symmetry share

rejected
Claim: the share of the valid space that is mere renaming separates the classes. Measured at the respective threshold: 3-SAT 11.3 %, 3-XORSAT 51.9 %.

The two classes were compared at different m — 3-SAT at 256 clauses, XORSAT at 55. At equal m the contrast vanishes: 11.3 against 13.7 percent. The symmetry share measures constraint density, not the class.

Symmetry share against clause count, n = 60. The curves run parallel; the apparent class difference was the offset on the x-axis.
MEASURE 02

Annealed gap

rejected
Claim: the distance between the first moment and the true threshold measures how strongly the solution space clumps — and therefore the hardness. Density-independent, so not confounded like 01.

Computed across 13 classes. The gap does not separate polynomially solvable from NP-hard: 3-XORSAT sits at 1.089, right between 4-SAT and NAE-3-SAT. 4-XORSAT sits at 1.024, below 5-SAT. The measure describes clumping, not computability.

First row: polynomially solvable. Second: NP-hard. The ranges overlap completely.
MEASURE 03

Redundancy ρ

calibrated
ρ = (effective independent assignments) / |S|. One counts what share of all possible constraints is compatible with the solution set, and computes back how many purely random assignments that would correspond to. ρ = 1 means: the solution set is indistinguishable from noise.

The measure is normalized by |S| and by the constraint's selectivity, and needs no clustering algorithm — it reads the structure straight off the compatibility geometry.

Calibration point. NAE-3-SAT has an exact complement symmetry: with x, ¬x is always also a solution, and both kill the same constraints. The prediction is therefore ρ = 0.500 exactly. Measured: 0.500.

ρ against solution count, log-log, n = 16. Every class falls as a straight line — the decay is a power law.
MEASURE 04

Decay exponent β

reducible
ρ ∝ |S|⁻ᵝ. The ordering of the classes by β is identical at n = 14, 16, and 18; the confidence intervals of neighboring classes do not overlap.

It is reproducible. It is not independent: β is explained to 90 percent by the selectivity of a single constraint (R² = 0.904). The measure is largely another way of writing p_kill.

And it does not separate: 1-in-3-SAT is NP-hard and sits at 0.794, above 2-SAT (polynomial, 0.740). The same class has already toppled three other separations before.

β at n = 18 with bootstrap 95 % intervals. The ordering is sharp; the complexity classes lie interleaved.
MEASURE 05

Residual after subtracting p_kill

refuted
Claim: what remains after subtracting selectivity carries the hardness. Across seven classes, all three polynomial ones sit above all four NP-hard ones — p ≈ 0.03 in the exact enumeration test.

Seven data points, and the shape of the fit was chosen after the fact. So it was pre-registered and tested on a cleanly matched pair: Horn-3-SAT has the same arity and the same selectivity as 3-SAT, but is polynomially solvable. If the residual carries tractability, β(Horn) must be > β(3-SAT).

Pre-registered test at p_kill = 1/8, n = 16. Horn-3-SAT sits at 0.369, 3-SAT at 0.574 — intervals disjoint, direction reversed.

The prediction did not fail narrowly — it inverted. The residual separation was a 1-in-35 fluke.

Instrument

The solution space, exactly

Near the threshold the solution set is small enough that a DPLL can output it completely as disjoint sub-cubes. From that one can draw exactly uniformly — no solver, no bias. Verified against brute force, χ²/df = 1.11.

What one can then see: the real clusters, defined as connected components under one-bit jumps within the solution set. Below, the two spaces, fully enumerated, projected onto three principal components.

drag to rotate
MEASURE 06

Majority closure, drawn uniformly

explained
Solution triples are far more often majority-closed than the independence benchmark (1 − 6/343)^m would lead one to expect — a factor of 20 to 290. No sub-cube artifact: same-cube share 0.000, and the effect survives when all three must come from different cubes.

Yet it is explained. A model that merely scatters three points around a shared center — fit to the measured Hamming distance, with no solution property at all — matches the rate. Exact in two of six rows, and it even overestimates the rest. Real solution sets are, if anything, less majority-closed than uncorrelated points of the same spread.

Each line one data point. The distance to the independence benchmark is large; the distance to the correlation model vanishes.

Geometrically this looks like: three uniformly drawn solutions and their majority point, in the plane of the first two principal components.

Gray is the solution cloud. Green where the majority point is itself a solution, red where not.
MEASURE 07

Solver bias in the backbone

own claim refuted
The claim was: probSAT overestimates the share of frozen variables because it gets stuck in the largest cluster. Evidence: backbone 0.017 exact against 0.117 from probSAT.

The comparison was unequal — exact backbone over all solutions against probSAT over twelve. At twelve samples, every variable with a skewed marginal distribution looks frozen, regardless of the sampler. At equal sample size, uniform drawing overestimates more than probSAT in all five rows.

At twelve solutions each. probSAT spreads wider (0.345 against 0.308 Hamming) and therefore sits closer to the truth.
REMAINDER

How many solutions one actually needs

useful

What remains of the refuted claim is a calibration. The frozen estimator is massively biased at small sample sizes, and how strongly depends almost entirely on how large the backbone actually is.

Dashed is the true backbone of the respective curve. Small backbones still have not converged even at 1024 solutions.

A measured frozen value of 0.2 from twelve solutions is just as compatible with a true backbone of 0.04 as with 0.15. In that range the estimator carries almost no information.

Balance

What remains

HOLDS

The limitation is a quantity

Roughly half of the drawn bits are format coercion, not freedom. Between 44 and 58 percent across four classes. The space in which randomness is allowed to move is precisely quantified.

HOLDS

ρ measures structure without clustering

The effective number of independent solutions falls straight out of the compatibility geometry, without needing to search the solution space. Calibrated on the known case NAE-3-SAT to three decimal places. That is an instrument, independent of the hypothesis about it having failed.

HOLDS

The structure is generator-independent

Mersenne Twister, a congruential generator, and a SHA stream deliver identical ensemble figures. What is being measured belongs to the format, not the randomness source.

HOLDS

A verified uniform sampler

DPLL cube decomposition, verified exactly against brute force, χ²/df = 1.11. Runs up to n ≈ 100 near the threshold and delivers the complete solution set including real clusters — no solver, no bias. That is an instrument, independent of the fact that two hypotheses died on it.

CAUTION

Frozen estimators from few solutions are unusable

At a true value of 0.042, eight solutions overestimate it ninefold, and even 1024 still misses by 20 %. The error is a sampling effect and hits every sampler, not just local search.

FALLS

Structure does not carry hardness

Four candidates, four times no separation between polynomial and NP-hard. What the encoding forces is selectivity — how many assignments a single constraint kills. Computability is not contained in it.

OPEN

Non-uniform variable choice

The only surviving thread from the previous day: markedly harder at unchanged α, 0 of 12 within budget. The threshold confound is untested — the same control that toppled the sign titration is still outstanding there.