The 14 edge types

Edges are propositions about logical structure between claim entities — not citations. A claim file declares each edge as a top-level YAML key whose value is a list of target claim slugs. Reciprocal edges (predicts / confirms) are populated symmetrically at build time.

The methodology defines 14 edge types. This page is the canonical reference. For the methodology’s authoritative description, see docs/method.md § 4.3-4.4 in the corpus repo.

The edge inventory

Edge	Reasoning form	Meaning	Count in 12-paper curated corpus
requires	dependency	A would be invalid if B were false (mechanistic / hierarchical dependency)	148
supports	abduction (induction)	A provides evidence for B; multiple supports drive the abductive loop	126
entails	deduction	A (typically a hypothesis) deductively implies B (typically a prediction)	65
derived-from	deduction	A is the deductive consequence of B; reciprocal of entails	61
tests	deduction → empirical loop	Empirical claim A tests prediction B (closes the hypothesis-prediction-test loop)	73
refutes	abduction (negative)	A’s evidence is incompatible with B (B is the prediction or hypothesis being refuted)	8
rules-out	elimination	A’s evidence eliminates an alternative explanation B	15
dissociates-with	dissociation (symmetric)	A and B jointly establish a dissociation between two empirical claims	65
validates	disconfirmation control	A is a control or sign-flip whose specific result strengthens the warrant for B	54
predicts	predictive validation	A predicts B (typically model-to-experiment)	3
confirms	predictive validation	Reciprocal of predicts; populated at build	74
interprets	reframing	A reframes empirical B through theoretical lens (an act of mapping, not a derivation)	42
enables-method	methodological warrant	A is the methodological capability that warrants B’s interpretability	79
scopes	scope qualification	A is a boundary condition on B (or, if ["*"], on every empirical claim in the paper)	197

Edges by reasoning form

The methodology operationalizes six argumentative moves through these 14 edges. For the full treatment with examples from the corpus, see docs/method.md § 4.4. Summarized:

Deduction

entails and derived-from carry hypothesis-to-prediction deduction. The Headley paper’s hypothesis-distinct-compartmental-roles entails: four predictions; each prediction derived-from: the same hypothesis.

The Meijer R1 paper’s hypothesis-additive-modulation entails: prediction-near-zero-choice-stim-interaction and (notably) entails: orthogonality-derived-from-additivity — a synthesis claim that is itself a deductive consequence of the hypothesis, demoting the empirical orthogonality finding from independent evidence to geometric corollary.

Induction (hierarchical support)

requires and supports carry mechanistic dependency and inductive support. Standalone empirical claims that aren’t themselves predictions tested in a hypothesis loop nonetheless carry supports: edges to higher-order claims via inductive accumulation.

The Headley ca-spikes-couple-20ms-before-ap supports beta-bidirectional-dendritic-control and beta-gates-distal-apical-inputs — the timescale measurement is the inductive ground for the period-matching argument.

Abduction

supports and refutes from empirical claims back to hypotheses close the abductive loop. The Meijer R1 near-zero-choice-by-stim-interaction supports: hypothesis-additive-modulation and refutes: prediction-multiplicative-gain-yields-significant-interaction — abduction to additivity by elimination of the alternative.

Elimination

rules-out carries the eliminative move: A’s evidence eliminates an explicit alternative B. The corpus carries 15 rules-out edges, scattered across papers; they are diagnostically interesting because they are scrubbed by abstracts (the Section 7 synthesis comparator pipeline shows this).

Dissociation

dissociates-with is symmetric — between two empirical claims that jointly establish a contrast. Neither claim alone establishes the dissociation; the contrast does.

The Headley distal-inhib-drops-firing-02hz dissociates-with perisomatic-inhib-drops-firing-07hz — neither alone establishes the compartmental distinction; the paired comparison does.

Scope qualification

scopes carries the boundary condition. A scope claim with scopes: ["*"] qualifies every empirical claim in the paper.

The Headley paper’s two global-scope claims (l5-model-single-cell-scope, naturalistic-drive-parameterization) qualify all empirical results — no network dynamics, no sensitivity analysis over synaptic parameters.

Where edges live in the claim file

Two locations in the YAML frontmatter:

Top-level keys — the named edges are top-level YAML keys whose value is a list of target slugs:

tests:
  - prediction-distal-dendritic-spike-mechanism

dissociates-with:
  - perisomatic-inhib-drops-firing-07hz

belongings: block — a generic typed-edge mechanism for edges that don’t have a dedicated key, or for legacy / ad-hoc edges:

belongings:
  - relation: requires
    target: l5-model-single-cell-scope
  - relation: supports
    target: hypothesis-distinct-compartmental-roles

Both are equivalent for graph purposes; the top-level keys are the preferred form when the edge has a name.

What the system emits

The CLI emits claim files with empty belongings: [] and no top-level edge keys. Step 6 (dependency mapping) is intentionally analyst work in the methodology; the CLI scaffolds for it but doesn’t infer edges automatically.

A future LLM-suggestion pass for edge inference is plausible but unimplemented. The empirical question — whether an LLM can produce analyst-quality edge maps — hasn’t been answered. For now, the analyst at Step 5 maps edges based on their reading of the paper’s argument structure.

Implementation references

• docs/method.md § 4.3-4.4 in the corpus repo — methodology authority
• elife_extract/write.py — claim file emission (currently emits empty edge sections)

Next steps

• The 9 roles — what roles claims with which edges
• The claim schema — where edges fit in the full claim file
• Design decisions — why Step 6 is scaffolded rather than auto-inferred