Glossary

Canonical vocabulary for the zettelkasten spine/matrix layer. These are the preferred terms; where a term replaced an older one, the deprecated alias is noted so existing code and stored data keep resolving.

Structure

Spine : A materialized organization whose row structure is written into the graph as persistent nodes (an apex plus per-dimension nodes). Because the structure is first-class, the rows become reusable entities you can build on — as opposed to a lens, which is a throwaway re-projection.

Apex (structural top node) : The single top node of a spine's structure tree (depth 0). It frames the whole spine (its synthesis title) and carries no members of its own; its rolled-up children are the substance. "Apex resolution" is the machinery that finds this node (resolve_apex, find_apex_id, APEX_TAGS, APEX_NODE_ID) — distinct from the column form below.

Structure node (structure_node column form) : The column form for a column that maps to a spine's structural node (the apex or a dimension node) rather than to a plain field/tag. This is the canonical name for what was historically written as the apex column form; readers accept both structure_node and the legacy apex value, and writers emit structure_node.

Sub-spine / component-of : A spine that is a component of a larger one. The relationship is recorded as a child-apex --component-of--> parent-apex edge, letting spines nest.

Nested spine : A spine composed of stacked ancestor tiers via component-of edges, so a document can be framed by its full root-first tier path rather than a single apex.

Identity & alignment

column_key_fingerprint (deprecated alias: spine_schema_fingerprint) : A spine org's schema identity — its sorted set of column keys. Two spines belong to the same spine-schema iff they share this fingerprint, regardless of per-spine labels or synthesis-graph names. It is the alignment key that folds independent spines into one comparison matrix. The old spine_schema_fingerprint name remains as a back-compat alias.

Spine-schema : The shared rubric a spine-group is built from: the set of column keys captured by column_key_fingerprint. A fingerprint shared by ≥ 2 spines forms a group (a matrix); a singleton stays an individual spine. Not to be confused with the grounded-extraction schema (the extraction rubric a source is read against).

Projection

Projection / tree-structured-by / ancestor_uids : A spine is a projection of one semantic base into a tree. "Tree-structured-by" names the relation that induces the parent/child structure; ancestor_uids is the ordered list of a node's ancestors (root-first) used to reconstruct that tree and to resolve nested-spine tiers.

Lens vs. materialized spine : A lens is an ephemeral, non-persisted re-projection of the corpus — it materializes nothing and routing degrades to tag + semantic matching. A materialized spine has written its structure (apex + dimension nodes and spine-member edges) into the graph, so its columns carry durable structure_node references and members route through real edges.

Grounding & evidence

Grounding modality : The channel a claim's evidence arrives through. Three exist: quote (a verbatim substring verified against the source text), equation (a transcribed LaTeX formula, optionally with a page snapshot), and data (a re-computable statistic verified by re-running the computation). A claim carries at most one grounding.method.

Data-grounding : Backing a claim with a re-computable statistic over a stored dataset rather than a verbatim quote. The claim carries a grounding.method: data citation the auditor verifies by recompute. Angelo verifies and stores this evidence; it never composes beliefs from it.

Data-citation : The method: data grounding block itself: the dataset_note + its pinned dataset_hash, a deterministic selection, the derivation_id (+ version / params), the asserted value, a tolerance, and opaque pass-through fields (confidence, validation, method_meta). The tuple (dataset_hash, derivation_id, derivation_version, selection, params) is the reproducibility pin — re-running it must yield value within tolerance.

Derivation : The deterministic function that turns a dataset into a number. Two forms: an expr (a whitelisted expression such as mean:col / quantile:col:0.9 / sum:col, evaluated by a built-in reducer with no host code, so it re-verifies anywhere), or a registered derivation (host code contributed via register_derivation, keyed by (id, version), same trust boundary as a custom agent). To angelo a derivation is opaque — it can only invoke and compare it.

Epistemic status : An optional, first-class claim field (grounded | measured | inferred) declaring what kind of backing a claim has, independent of the grounding method. The coverage matrix rolls it up as a per-dimension epistemic mix.

Grounded / measured / inferred : The three epistemic_status values. grounded — asserted verbatim from a source (a verified quote or equation). measured — a directly-observed statistic over data (a mean, a quantile; the default for a descriptive expr derivation). inferred — a fitted quantity (an inverse-optimization or model fit; a derivation declaring fit: true).

Hard violation vs soft unavailability : The two tiers a data-grounding verification can fail into. A hard violation (angelo could re-run and it disagreed — dataset_hash_mismatch, value_not_reproduced, nondeterministic_unpinned, validation_malformed) blocks coverage like an unverified quote and surfaces as unreproduced_data_claims. A soft unavailability (angelo cannot re-run here — derivation_unavailable, dataset_unavailable) is informational and retains the claim's prior verified / verified_in stamp, so a once-verified claim is not read as broken in a repo where its derivation isn't installed.