Canonical terms

The Calyx glossary.

Every term grounds to a node in the concept graph and links to its full concept page. Use it as a quick reference, or jump from a term into the explainer behind it.

A

Anchor

A real, stored outcome a claim is measured against to be grounded. Grounding compares a claim’s constellation to the anchor’s; close enough and the claim is supported.

Assay

The Differentiate verb realized: KSG mutual information answers how many bits a lens adds, enforcing a redundancy contract (≥0.05 bits, ≤0.6 correlation).

C

Calculus of Association

The unifying theory: intelligence framed as the measured calculus of association, with the four verbs as first-class operations. The conceptual spine the whole system instantiates.

Constellation

One input seen through many lenses at once — the native Calyx record. The shape across its slots is the meaning, which is why the lenses are kept rather than averaged into one vector.

Cosine distance

A similarity measure between two embeddings based on the angle between them: near 1.0 means aligned (similar), 0 means a right angle (unrelated). The atomic operation behind every later verb.

E

Embedding

A representation of meaning as a vector — a point in a high-dimensional space — so that inputs meaning similar things land in similar directions.

F

Fail-closed doctrine

When something is unknown or uncalibrated, Calyx returns a structured error (from a 38-code catalog) rather than a silent wrong answer. Failing closed is preferred to bluffing.

Forge

The math runtime that executes Calyx’s vector operations: CPU SIMD with opt-in CUDA, bit-parity across backends, and seven quantization levels. The engine cosine and the verbs run on.

G

Grounded retrieval pattern

The end-to-end pattern for trustworthy retrieval: ingest → measure → search → guard_generate, then cite sources and abstain when the guard says so. RAG that will not bluff.

Grounding

Calyx’s mandatory discipline of measuring a claim against an anchored outcome rather than asserting it — an answer is provisional until it is grounded.

Grounding kernel

The minimal (~1%) set of records whose meaning explains an entire corpus, found by the Compose verb (Lodestar). It is usable as both an index and an answer path.

L

Ledger

Calyx’s append-only provenance chain — BLAKE3 hash links, Merkle checkpoints, ed25519 signing — that makes every result verifiable and reproducible. The receipts behind an answer.

Lens

A frozen, content-addressed embedder — one fixed way of seeing an input. Freezing it makes a lens a reproducible contract identified by a LensId.

Loom

The Count verb realized: it computes agreement (cosine), delta, and interaction (Hadamard) between slots to build the agreement graph of which ideas relate.

N

No-flatten doctrine

The rule that a constellation’s slots stay separate from end to end — never concatenated or averaged into one blob — because per-lens structure is what lets Count and Differentiate work on real signal.

O

Oracle

Calyx’s consequence predictor: it builds a butterfly consequence tree and walks it in reverse, gated by an honesty rule (panel_bits ≥ anchor_entropy) so it only speaks when it has the signal to.

P

Panel

The chosen set of lenses measuring a vault — the declared measurement configuration that fixes which perspectives every record in that vault is seen through.

S

Sextant

Calyx’s multi-signal search over a vault: ANN indexes (HNSW/DiskANN/SPANN) plus BM25, fused with reciprocal rank fusion. It turns the stored panel into retrieval.

Slot

One lens’s typed output for an input, held as a SlotVector that may be Dense, Sparse, Multi, or Absent. Slots are the typed cells a constellation is built from.

Symmetry of Knowing

The thesis of Q↔A epistemic symmetry: when asking becomes as cheap as answering, good questions and judgment become the scarce thing. It reframes value around the question side of grounding.

T

The four verbs

Calyx’s grammar of meaning — Measure, Count, Differentiate, Compose — the hub through which every mechanism subsystem hangs.

U

UTML

The learning theory the platform’s pedagogy engine is built on: it models learning as the factors ΔP/ΔK/ΔΩ/ΔΞ — a move from entropy toward coherence.

W

Ward

The fail-closed guard at query time: per-slot cosine against a threshold τ with conformal calibration, emitting verdicts accept / new-region / quarantine / refuse.