Results and research directions from Orientational Connective Logic and the geometries behind Quester GX-1. Each one states the mathematics, shows it working in a panel that computes live, and carries a label saying exactly what it is.
Each entry separates three things: what the mathematics establishes, what is specific to the Faciens constructions, and what comes next. The panels compute everything they display, and a self-check runs every time the page loads.
Let a reference orientation be a unit vector a. An orientation x is admissible when it lies within an angle α of a. The admissible set is a spherical cap, a region rather than a point, so admissibility changes continuously instead of flipping at the slightest deviation.
When several caps are in play, being admissible pair by pair does not make them admissible together. Three caps can share a direction with each of the others and still have no direction common to all three. Admissibility is a property of the whole configuration. That is the geometric heart of co-admissibility (CAA) in Quester GX-1, and the panel builds the case exactly.
Three equal caps at equal spacing. Change their size and how far they sit from the pole.
A population of orientations s₁ … sₙ with weights wᵢ has a first moment, its mean direction, and a second moment, its orientation tensor.
The mean gives the average direction. The tensor keeps the spread and its shape. Two populations can agree exactly on the first and differ on the second. In the Faciens programme the tensor is the descriptor of amendment A-3.1, where it predicts Quester design cost on held-out workloads better than the mean direction does.
Mixture A is the opposite pair ±x. Mixture B is an opposite pair you can turn anywhere on the sphere.
Add directions to mixture C, with equal weights. Goal: the same mean as A, a different tensor.
Take an orientation round a closed loop of states and it comes back pointing the same way, yet carrying a phase fixed by the loop. With the convention used across the Faciens pages, the phase is half the signed solid angle Ω the loop encloses.
This holonomy is classical geometry, and in optics it is the Pancharatnam–Berry phase. In OCL it is the first instance of a central idea: an orientation carries a record of its path. The panel computes the phase from the states themselves and the solid angle from the vertices, two independent routes that must agree.
A triangle from horizontal to diagonal to a third state you move.
A normalised complex pair (u, v) lies on the three-sphere S³. The Hopf map sends it to a point on S²:
Every point on one circle, the fibre, lands on the same point below. The coordinate the projection drops is the fibre phase.
Each Quester character is a sign pattern (a, b, c, d) with entries ±½, read as the state (a + ib, c + id). Under the Hopf map the sixteen characters land on exactly four directions, D, A, R and L, four on each. The table computes it. The grouping comes from the characters themselves: every entry has the same size, which places every image on the equator at quarter-turn spacing, and all three ways of pairing the entries give four groups of four.
| Image | Characters (a b c d) |
|---|
Let O be an observation map. Two structures are equivalent to that observer when O gives them the same reading. This is an equivalence relation, so the observer partitions the world into classes.
Adding observers can only split classes, never merge them. The Faciens application is OGD-NC1: the six orders of three non-commuting optical elements are the structures, and the probe beams are the observers. The panel shows how many orders each choice of probes can separate. Orders that act identically on every input stay together under any probe, which is the true limit of what can be read back.
Elements: quarter-wave plates at 0° and 90°, and a half-wave plate at 45°. Choose which probe states you send in.
Compare a ring of n nodes with a ring of n + 1. Around any node, the window of radius r is a straight chain until it wraps round. While 2r + 1 < n, every window in either ring looks the same, so no local observer can tell the rings apart. Once 2r + 1 ≥ n, the window closes on itself in the smaller ring and the difference shows.
For architecture this is a design rule: how far a unit must see before a global property of the system becomes visible to it. It bears on how much routing context a chiplet in Quester needs.
Rotations about different axes do not commute: Rₓ(α) Ry(β) ≠ Ry(β) Rₓ(α). Applied to one starting orientation, the two orders end in different places. In the Faciens geometry this is why order is part of an orientation, and why OGD-NC1 predicts that the order of a history can be read back from final measurements under realistic noise.
The commutator measures how much order matters. It is zero only when the axes coincide or an angle vanishes, and it grows with the area the two rotations sweep.
OCL gives every proposition an orientation as well as a truth value, and defines its relations geometrically. One of them, R, holds between two propositions when their orientations point away from each other.
Collect the inner products of a set of propositions in their Gram matrix Gᵢⱼ = ⟨uᵢ, uⱼ⟩. Two models can agree on every truth value and still differ in their relations, so OCL distinguishes situations that classical truth assignment merges. This is OCL’s expressiveness result in miniature: orientation adds distinctions beyond what bisimulation-based logics capture.
Four propositions in the plane. A proposition is true when its orientation points to the right. Model 1 is fixed; turn the propositions of model 2.
These are proved in the OCL papers by Arcades Cinza. Where a result can be exhibited by computation, the page does so below.
The most propositions that can pairwise satisfy R in n dimensions is n + 1, attained by a regular simplex. The page builds the simplex in each dimension and checks every pair.
| Dimension n | Propositions | Every pair satisfies R | Inner product |
|---|
Each has a precise target and a test that could settle it.