Skip to content

The 9 roles

The role is the rhetorical function a claim serves in the paper’s argument. It is distinct from the claim-type (the epistemic character of the proposition itself). A claim’s role determines how the dependency graph treats it, what edges it can carry, and how downstream synthesis pipelines reconstruct the argument from the graph alone.

The methodology defines 9 roles. This page is the canonical reference for each — what it marks, signal phrases that distinguish it, typical claim-type pairing, the edges it carries, and an example from the corpus.

For the methodology’s own description, see docs/method.md § 4.2 in the corpus repo. This page elaborates with operational guidance.

hypothesis

Section titled “hypothesis”

What it marks. The paper’s organising bet — the proposition the rest of the paper will defend or test.

Signal phrases in the prose.

  • “we hypothesize”
  • “we propose that”
  • “we asked whether”
  • “we sought to test”
  • “we tested the hypothesis that”
  • “the central question is whether”

Typical claim-type. hypothesis (when the schema allows) or assessment. The methodology specifies claim-type: hypothesis for these claims; see methodology/schema for the schema’s official type vocabulary.

Edges carried. A hypothesis carries entails: edges to its predictions. It does not itself carry empirical content; it does not carry requires: to empirical claims (predictions do that).

Panel. Typically null — hypothesis claims are paper-level, not panel-grounded.

Headley example. hypothesis-distinct-compartmental-roles: “Perisomatic and distal dendritic inhibition serve distinct computational roles in regulating neuronal output.” Carries entails: to four prediction claims.

prediction

Section titled “prediction”

What it marks. The deductive consequence of a hypothesis under specific conditions. What we expect to observe if the hypothesis holds.

Signal phrases.

  • “if X, then we should observe Y”
  • “this predicts that”
  • “we therefore predict”
  • “the model predicts”
  • “should be maximally effective at”
  • “is predicted to”

Typical claim-type. prediction (extended schema) or interpretive.

Edges carried. A prediction carries derived-from: back to its hypothesis (reciprocal of entails). It is tests:-targeted by empirical claims (the empirical claim that tests the prediction carries tests: to it).

Panel. Typically null — predictions are deductive, not measured.

Headley example. prediction-beta-optimal-distal: “If the optimal frequency of rhythmic inhibition at a compartment is set by matching the rhythm period to the local spike timescale, then distal inhibition should be maximally effective at beta (~20 Hz).” Carries derived-from: hypothesis-frequency-compartment-matching.

empirical

Section titled “empirical”

What it marks. A measured or computed result, panel-grounded. The largest role bucket — 139 of 310 claims in the curated corpus.

Signal phrases.

  • “we found that”
  • “we observed”
  • “we measured”
  • “we computed”
  • “the result was”
  • “Figure N shows”
  • “doubling X reduced Y from A to B”

Typical claim-type. empirical.

Edges carried. Many. Empirical claims tests: predictions, dissociates-with: complementary empirical claims, requires: methodological and scope claims, supports: synthesis claims, rules-out: alternatives.

Panel. Required. Empirical claims are panel-grounded by definition.

Headley example. distal-inhib-drops-firing-02hz: “Doubling distal dendritic inhibition reduces somatic firing rate from approximately 5.5 Hz to approximately 0.2 Hz, primarily by suppressing dendritic Ca²⁺ and NMDA spikes.” Carries tests:, dissociates-with:, requires:, and supports: edges — typical density for a load-bearing empirical claim.

control

Section titled “control”

What it marks. An empirical result whose primary work is to rule out an alternative explanation. Functional, not contentful — the same proposition could be empirical in a different paper context.

Signal phrases.

  • “rules out”
  • “excludes”
  • “shows that the effect is not due to”
  • “fails to support”
  • “no significant effect of [potential confound]”
  • “control condition”
  • “control experiment”
  • “consistent with the null for [alternative]”

Typical claim-type. empirical.

Edges carried. rules-out: (the eliminative move) and scopes: (when the control bounds the empirical claims it verifies).

Panel. Usually a specific control panel.

Meijer R1 example. cortical-layers-show-no-differential-modulation: “Splitting cortical recordings by layer reveals no differences in modulation fraction, sign, or latency. This rules out a layer-specific cortical mechanism.”

scope

Section titled “scope”

What it marks. A boundary condition on the empirical claims. Often global — scopes: ["*"] qualifies every empirical claim in the paper.

Signal phrases.

  • “all results come from”
  • “results are restricted to”
  • “is not a physiological pattern”
  • “single-cell scope”
  • “the model assumes”

Typical claim-type. assessment (or scope in extended schema).

Edges carried. scopes: to specific claims, or ["*"] for global.

Panel. Typically null.

Headley example. l5-model-single-cell-scope: “All results come from a single-cell compartmental model — no network dynamics, no recurrent excitation, no population effects.” Carries scopes: ["*"].

methodological

Section titled “methodological”

What it marks. A procedural or analytical capability that warrants a downstream interpretation. Not a claim about the world; a claim about the apparatus.

Signal phrases.

  • “manifold extraction is on a pooled super-session”
  • “spike-sorting uses Kilosort 2.5”
  • “rhythmic inhibition is modeled as sinusoidal rate modulation”
  • “the analysis uses [method]”

Typical claim-type. assessment (or methodological in extended schema).

Edges carried. enables-method: to the empirical claims that depend on the methodology, scopes: to the same.

Panel. Usually a methodological figure panel; sometimes null.

Meijer R1 example. manifold-from-pooled-super-session: “Manifold analysis is on a pooled super-session… nulls are block-aware shuffles.” Carries enables-method: and scopes: edges to the manifold-derived empirical claims.

synthesis

Section titled “synthesis”

What it marks. A higher-order proposition integrating across multiple empirical claims into one statement. Stays inside the paper’s own evidence.

Signal phrases.

  • “taken together”
  • “in summary”
  • “these results show”
  • “this dissociation establishes”
  • (typically appears at section breaks or paragraph ends)

Typical claim-type. synthesis (or interpretive).

Edges carried. derived-from: (when the synthesis is a deductive consequence of a hypothesis) or supports: (when synthesis claims integrate empirical results).

Panel. Typically null — synthesis claims integrate across panels.

Meijer R1 example. orthogonality-derived-from-additivity: “Under a linear readout, additive modulation entails orthogonality of the stim and choice axes.” Carries derived-from: hypothesis-additive-modulation — a deductive synthesis that demotes orthogonality from independent finding to geometric corollary.

interpretation

Section titled “interpretation”

What it marks. A reframing of empirical results through a theoretical lens. An act of mapping to broader theory — distinct from synthesis, which stays inside the paper’s own evidence.

Signal phrases.

  • “may provide a functional interpretation”
  • “suggests a role for”
  • “points to a mechanism whereby”
  • “may explain”
  • (typically at the end of results or transitioning to discussion)

Typical claim-type. interpretive.

Edges carried. interprets: to the empirical claims being reframed.

Panel. Typically null.

Headley example. pv-gamma-sst-beta-correspondence: “Layer 5 inhibitory streams are functionally matched to interneuron type.” Carries interprets: edges to the four empirical claims that ground the mapping.

literature-context

Section titled “literature-context”

What it marks. A cited prior claim treated as a first-class node — the paper depends on this prior result, but the result wasn’t produced by this paper.

Signal phrases.

  • “as shown by Author (Year)”
  • “the reported association of X with Y”
  • “previous work has established that X”
  • “Author (Year) found that”
  • Slug pattern: interprets-<author>-<year>-<topic> or interprets-<topic>-associations

Typical claim-type. interpretive.

Edges carried. Targeted by interprets: edges from the paper’s interpretation claims (the paper’s interpretation depends on the literature claim).

Panel. Typically null. Paper context (the paper that asserts the literature-context claim) is in assertions[0].panel if the citing paper has a specific panel referencing the prior work.

DOI. Top-level doi: field is the cited paper’s DOI. This is the schema’s exception to the doi: ~ placeholder rule for non-literature-context claims. The CLI’s verify-refs reads from this field; CrossRef confirms the DOI resolves to a real paper (anti-hallucination check).

Headley example. interprets-pv-gamma-sst-beta-associations: “Empirical work across the cortical-interneuron literature has established two correlated associations: PV+ interneurons target perisomatic compartments and entrain cortical gamma rhythms; SST+ interneurons target distal dendrites and associate with beta rhythms.” Top-level doi: 10.1038/nature08002 (Cardin et al. 2009 Nature).

How the system assigns roles

Section titled “How the system assigns roles”

The three extraction agents each assign a provisional role from the prose-level evidence they see. The reconciler may revise based on cross-agent disagreements (with guidance to preserve role distinctions when agents surface different roles for related claims). The external reviewer (Step 4.5) further revises to recover roles the prose extraction systematically misses (hypothesis, prediction, control per the bias rules).

Final role assignment is the analyst’s call at Step 5. The methodology says: “Step 5 is where the schema’s role labels are first assigned definitively, because role-assignment requires the analyst’s judgment about what kind of work each claim is doing in the paper’s argument.”

Implementation references

Section titled “Implementation references”
  • prompts/external-reviewer.md — role recognition rules in the reviewer prompt (Biases 5, 6, 7)
  • elife_extract/schema.py — Role Literal type with the 9 values
  • docs/method.md § 4.2 in the corpus repo — methodology authority

Next steps

Section titled “Next steps”