Methodology

A claim graph is a decomposition of a scientific paper into typed propositions and the typed logical relations between them. Each proposition is one declarative sentence in active voice, recorded with provenance (which paper asserts it, in which panel, using which analysis and dataset), an epistemic marker, and a reproduction status. Relations are not citations; they are statements about logical structure — A requires B means A's validity depends on B's, so invalidity propagates through the graph when a claim fails to reproduce.

The unit of work is the claim, not the figure. A figure panel records where a paper instantiates a claim; the claim is the stable entity. The same proposition could be asserted by different papers in different panels — the UUID identifies the claim while one or more assertions blocks attach it to specific paper-panel-analysis tuples.

The method has two distinct phases. Claim induction is the process of reading a paper and extracting its claim structure — the propositions, their roles, their logical dependencies, and their provenance (which figure panel, which analysis, which dataset). Claim induction is a reading act: it requires comprehension of the paper's argument and judgment about what constitutes a claim. It produces a claim graph. Claim verification is the process of running code against deposited data to check whether an induced claim reproduces. Verification is an execution act: it requires data, code, and compute. It produces a pass/fail/warn verdict per claim, optionally with reproduced figures that can be compared to the published originals.

The two phases are separable. A claim graph is valuable before any verification runs — it makes the paper's argument structure explicit and navigable. Verification adds an empirical layer: did the computation actually produce the reported result? But induction comes first, and induction is where provenance should be captured — figure URIs, panel assignments, dataset links — because the agent performing induction has the paper's structured source in front of it — JATS XML, whose figure elements carry the publisher's own ids (what the readers are actually given). Deferring provenance capture to a later build step (e.g. guessing figure filenames from panel labels) is fragile and loses information that was available at induction time.

The corpus is a 10-paper prototype assembled to test whether the schema is expressive enough to capture the argumentative structure of recent neuroscience papers, whether the verification step can re-enact published analyses against deposited data and code, and whether downstream pipelines (paper summaries, synthesis from the claim graph alone, comparison against the published abstract) yield findings that would not be visible from the prose alone. It is reverse-engineered from finished papers; forward construction by authors at submission would look different.

What this document is not: a specification of how the schema should evolve at scale, a proposal for editorial workflow, or a comparison with related schemes (Wikidata, Semantic Web claim representations, micropublications). Those discussions belong elsewhere.

↑ Contents

---

The published corpus is 10 papers from eLife, spanning the journal's neuroscience subfield mix. Two further papers — the two bioRxiv versions of one preprint — are carried as method examples rather than corpus members: they are the only pair here that shows how a claim tree changes between a preprint and its revision, and they are counted separately everywhere (66 claims across 2 versions), never folded into a corpus total.

Total: 289 claims and 4 parts across 10 papers, carrying 916 typed relations — 17 of the 21 relation types the vocabulary defines, and all 9 roles. A claim tree has two grains: the count leads with the wholes a reader of the argument follows, and reports the parts — components of those wholes, one comparison or measure each — separately.

Distribution by role:

Role	Count
empirical	139
prediction	46
hypothesis	31
control	27
scope	23
methodological	16
literature-context	12
synthesis	9
interpretation	7
Total	289

Per-paper claim counts: artiushin 17, bouyeure 30, ejdrup 25, gadeke 27, headley 26, kammer 23, kolb 20, meijer-orthogonal 24, meijer-additive-r1 41, rozak 23, scheller 23, wengert 31.

The R1 revision of the Meijer paper carries 41 claims against 24 in the v1 preprint — the inflation reflects new hypotheses (additivity, orthogonality-as-derivation), new empirical claims (per-neuron GLM coefficients, receptor-expression GLM), and new literature-context nodes added to anchor the receptor-pharmacological reframing.

↑ Contents

---

No human checks this corpus

Every claim in this repository was produced by a language model, and no person has verified any of it. That is true of the whole pipeline, not one step of it:

• The claims are found by three model calls reading the paper independently, and reconciled by a fourth.
• The relations between claims — the deductive spine — are inferred by a model call.
• The method requires an analyst's review gate. There is no such gate. What stood in for one was a flag — --review-mode external, which is another model revising the draft, or --review-mode auto-approve, which wrote the files unread — and both are gone. The model's revision is now a declared layer, external-review, so changing the prompt behind it makes every claim tree built on it stale. Approval is what it should always have been: an operation on a version that already exists, recorded in runs/<paper>/approvals.jsonl by scripts/pipeline.py approve. That file is empty for every cell in this corpus — the same fact as before, now stated where it can be checked rather than inferred from a default.
• The agent: field on a verification record names the agent identity that ran the script, not a person who checked the result. All 225 records read agent: mainen-z; none of them means Zach Mainen read that claim.
• The coverage verdicts in mappings/ — covered, gap, no-assertion — are model judgements, made from the paper and the claim text.

Read the corpus as a draft annotation layer, not as adjudicated output. Where this document says "curated", it means "assembled by this pipeline", not "checked by a person".

A human-review step is compatible with the format — claim files carry provenance fields, and the tooling can already write anchored comments and claim assignments into the paper for a person to disposition. It is not implemented, and it has not been run. Adding it would be a change to the method, and would be recorded here as one.

Claim induction is the translation of a paper from argument format into claim-graph format. It demands reading comprehension, domain judgment, and decisions about what constitutes a claim — and in this version a model makes all of them.

Induction is not one act but a sequence of them, and each is a declared layer with its own question, its own inputs and its own record of having run. What each does, and how, is documented beside its declaration in pipeline/layers/ and gathered in the Layers section. What follows here is what holds across all of them.

What the readers are actually given

Before any reading happens, the paper has to be turned into text, and the format that text comes in bounds everything downstream. This step is the prepare layer, and it is stated here because a method that says "three agents read the paper" without saying what they read has skipped the decision that constrains the rest.

For an eLife DOI the source is JATS XML, not the PDF and not the web page. JATS — the Journal Article Tag Suite, the archival XML that eLife and most journals typeset from — is fetched from the article's CDN URL:

https://cdn.elifesciences.org/articles/<article-id>/elife-<article-id>-v1.xml

The choice is not incidental. JATS carries the structure a claim tree needs and a PDF has already discarded:

	JATS XML	PDF
Section boundaries	tagged (<abstract>, <sec>)	inferred by regex against heading text
Figure captions	tagged, one element per figure, with its own id	recovered from body text by position
Figure identity	eLife's own <fig id> — fig2, fig3s1	guessed from a label like "Figure 2"
Reading order	explicit	a column-order guess from pdfplumber
Tables	<table-wrap> with a caption	text that may or may not still be a table

The panel-level grounding the whole format rests on — every empirical claim naming the panel it came from — is only as good as the figure structure at intake. Read from a PDF, a claim's panel field is an inference about a label in running text; read from JATS, it is the publisher's own element id.

The PDF path still exists and is still used, but only deliberately: --input-format pdf, or --pdf-path for a paper that is not on the eLife CDN. There is no silent fallback — with a DOI and the default --input-format auto, the intake is JATS or it fails. A method that quietly degraded to PDF when a fetch failed would make every claim below it weaker without saying so.

What prepare writes is runs/<paper>/prepared.json: the four slices the readers are given — abstract, results, figure captions, methods — the typed figure captions, and extraction_path recording which route was taken, with the source URL. That file is the verbatim text the model saw, which is what makes an extraction re-examinable rather than merely re-runnable. For Gädeke it records:

extraction_path  jats
source           https://cdn.elifesciences.org/articles/105391/elife-105391-v1.xml
abstract          1,931 characters
results          33,886 characters
captions          8,496 characters   (7 typed figure captions)
methods          32,690 characters

The fetch is cached under ~/.cache/elife-extract/, so every later layer on the same paper runs offline and costs nothing. The practical consequence for anyone debugging a bad extraction: check the slice sizes first. A short results means section detection failed, and everything downstream read the wrong text — which is a fault in intake presenting as a fault in reading.

Reading the abstract first

Read the abstract and identify two to four top-level claims — the paper's main bets. For each, write a candidate slug (3–5 words, lowercase, hyphenated, verb-phrase form). These will be the synthesis or interpretation nodes at the top of the dependency graph. They typically have no single figure of their own — they are the synthesis of the figures below them. The Headley paper, for example, surfaces pv-gamma-sst-beta-correspondence as a single synthesis node interpreting the simulation results in light of prior interneuron-rhythm associations; this node's panel is "fig10 (synthesis / discussion)" rather than a single quantitative panel.

Why three readers

The three agents are deliberately partitioned along the axes along which extractions most often disagree: framing versus literal numerics versus computational structure. A claim that all three surface independently is high-confidence; a claim that only one surfaces is single-source and may be either real-but-buried or an artefact of the reading strategy. The reconcile layer records both cases distinctly.

The partition is by section, and each reader sees its sections and nothing else. The results reader is given the abstract, the Introduction, the Results and the Discussion, preceded by a panel inventory; the Introduction and Discussion were added on 2026-09-11 (#56), after the first runs showed that cutting them lost the organising hypothesis and every literature-context premise. The caption reader is given the figure and table captions with the inventory. The structure reader is given the Methods, the appendices and the supplementary material. No reader sees the whole paper.

A single reading of the whole paper is kept as a comparison, not as a layer. It runs through the same reader code on the raw paper (reader_from_raw) and is scored by evaluate as the fourth-reading row beside the three-reader chain; on Gädeke it recovered slightly more of the curated tree at higher precision and missed every alternative the paper rules out. Whether the partition earns its place is a scheme question (#85), to be settled against adjudicated trees, and the comparison row exists so that it can be. Until it is settled, the three-reader chain is the accepted scheme and every recorded run uses it.

Which model reads is a profile, not a property of the readers. The standard profile puts the three readers on Sonnet and the reconciler, reviewer and edge inference on Opus; frontier puts everything on Opus with leaner tasks. A run answered by Claude Code subagents rather than an API records, in the ledger's by, the model that actually answered each prompt, which need not match the profile it stands in for; Gädeke's v3 readers, for instance, were answered by Opus under the subagent profile.

Where short-form fields fit

Three fields are populated during authoring but are not the primary claim sentence:

• displayClaim (one to two sentences) is the form rendered when the claim is presented in body text or in a card view. It softens the formal claim sentence into something readable in a paragraph; it preserves the proposition's content but allows shorter constructions, parenthetical units, and contractions where the formal claim field cannot. Authored with the claim files; can be revised without changing the underlying claim.

• shortClaim (single short clause, ≤90 characters typical) is the headline form: what fits in a tooltip, a hover preview, or a graph-node label. Required for synthesis, interpretation, and literature-context nodes that must be readable at a glance in the synthesis layer; optional but recommended for hypotheses, predictions, and high-traffic empirical claims. Authored with the claim files.

• number and numberParts are not authored manually. They are computed by the build pipeline from the claim's role and its position in the dependency graph (hypotheses get H#, predictions hang off the hypothesis they derived-from as H#.P#, empirical claims testing those predictions hang off as H#.P#.E#, scope claims become Sc#, methodological become M#, controls become C#, literature-context becomes L#, synthesis becomes S#, interpretation becomes I#, and standalone empirical not under any hypothesis loop become E#). The numbering is regenerated on every build; do not paste numbers into source files.

role is assigned at the review gate and is the most consequential single field in the schema after claim, because it governs how the synthesis layer groups the claim and how the claim is numbered in the build.

↑ Contents

---

One markdown file per claim, stored at claims/<paper-slug>/<claim-slug>.md. The file has YAML frontmatter and an optional markdown body for prose elaboration. The body is for caveats, alternative interpretations, pointers to contradicting claims in other papers, and reasoning that does not compress into frontmatter; it is rendered as prose where the claim is displayed in detail.

4.1 Required frontmatter

Field	Value
uuid	UUID4, generated once at creation, immutable
slug	filename slug, lowercase, hyphenated, 3–6 words, verb phrase
doi	placeholder ~ for now (claims are not yet citable units)
claim	one declarative sentence, active voice, quantitative where the result is quantitative
claim-type	empirical / interpretive / existence / synthesis / assessment / hypothesis / prediction
role	one of nine values (Section 4.2)
concepts	controlled list of domain terms
priority	date the claim was first registered
epistemic	strong / moderate / weak / contested (analyst's overall assessment of support across all assertions)
assertions	list of blocks linking the claim to specific paper-panel-analysis tuples
reproductions	list of blocks recording verification attempts

claim-type and role are orthogonal axes: claim-type is the epistemic character of the proposition (is it observed, inferred, asserted to exist, synthesised, or methodological?); role is the rhetorical function the claim serves in this paper's argument. A claim-type: empirical claim can carry role: empirical, role: control, or role: scope depending on whether it is the primary observation, a check that rules out an alternative, or a boundary condition.

A role: hypothesis claim, and a rejected alternative (an alt- claim asserted with stance: rejects), also carry a top-level addresses field naming the research question they answer, as a q<N> id. A question is not a claim — it is a declarative-sentence axis a question does not lie on — so questions themselves are not claim nodes; they are listed in the asserting paper's index.md frontmatter under questions: ([{id: q1, text: "…"}]), and addresses links a hypothesis or an alternative to the one it answers. The hypothesis is the paper's committed answer; the alternatives it rules out are the other answers to the same question. Every other role omits addresses.

4.2 Roles — the role inventory

Role is the rhetorical function the claim plays in the paper's argument. The synthesis pipeline reads this field directly to organise reconstruction.

Role	What it marks	Typical claim-type
hypothesis	The paper's organising hypothesis or framing question. Anchors deductive chains via entails: to predictions.	hypothesis
prediction	A specific empirical prediction derivable from a hypothesis. Carries derived-from: back to its hypothesis and is tests:-targeted by empirical claims.	prediction
empirical	A measured or computed result, panel-grounded. The largest role bucket (125 of 289).	empirical
control	A check ruling out an artefactual or alternative explanation. Carries scopes: or rules-out: edges.	empirical
scope	A boundary condition that qualifies a set of claims (single-cell scope, dataset boundary, optogenetic-vs-physiological scope). Often global (scopes: ["*"]).	assessment
methodological	A procedural or analytical capability that warrants a downstream interpretation (manifold-from-pooled-super-session, particular sorting pipeline). Carries enables-method:. A contrast run only to define a region or a set of trials for a later analysis — a localizer — is methodological, not empirical, because the paper does not argue from it.	assessment
synthesis	A claim integrating across multiple empirical claims into a higher-order proposition (the dissociation, the receptor-reconciliation). Top of the within-paper graph.	synthesis / interpretive
interpretation	A reframing of an empirical result through theoretical lens, marked separately from synthesis. Carries interprets: edges.	interpretive
literature-context	A cited prior claim treated as a first-class node. Section 5.	interpretive

4.3 Edges — the edge inventory

Edges are propositions about logical structure between claim entities, not citations. Each is a top-level YAML key whose value is a list of target slugs. The deductive pair entails / derived-from is populated symmetrically at build (the reader emits entails; derived-from is written as its reciprocal). confirms and refutes are not reciprocals — they are the two outcomes of a test, stated directly from the result to the prediction it settled (§ 4.3, issue #28).

Edge	Reasoning form	Meaning	Count in corpus
requires	dependency	A would be invalid if B were false (mechanistic / hierarchical dependency).	104
supports	abduction (induction)	A provides evidence for B; multiple supports drive the abductive loop.	148
entails	deduction	A (typically a hypothesis) deductively implies B (typically a prediction).	54
derived-from	deduction	A is the deductive consequence of B; reciprocal of entails.	48
tests	deduction → empirical loop	Empirical claim A tests prediction B (closes the hypothesis-prediction-test loop).	75
refutes	test outcome (negative)	A's result came out against prediction B — the negative outcome of a test, aimed at a prediction (parallel to confirms; distinct from contradicts/rules-out, which assert B is false).	1
rules-out	elimination	A's evidence eliminates an alternative explanation B.	22
dissociates-with	dissociation	A and B jointly establish a dissociation (symmetric edge between two claims whose difference is itself the finding; neither bears on the other's truth). Neutral — no MIRA predicate.	62
in-tension-with	tension	A and B are both asserted and both stand, yet pull a shared implication in opposite directions or cannot be jointly explained without a further claim (symmetric). Exported under mira:opposes, which overstates it.	1
validates	disconfirmation control	A is a control or sign-flip whose specific result strengthens the warrant for B.	44
predicts	predictive validation	A predicts B (typically model-to-experiment).	0
confirms	test outcome (positive)	A's result came out as prediction B said it would — the positive outcome of a test, aimed at a prediction (parallel to refutes).	70
interprets	reframing	A reframes empirical B through theoretical lens (this is an act of mapping, not a derivation).	36
enables-method	methodological warrant	A is the methodological capability that warrants B's interpretability.	76
scopes	scope qualification	A is a boundary condition on B (or, if ["*"], on every empirical claim in the paper).	144
part-of	composition	A is a component of B — one comparison, condition, measure or study of a proposition B states whole. B is weakened but not falsified by A alone; the page shows B with its parts folded beneath it.	4

4.4 Edge-to-reasoning-form mapping

The edge inventory operationalises seven argumentative moves:

1. Deduction. entails and its reciprocal 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.

2. Induction (hierarchical support). requires and supports carry mechanistic dependency and inductive support. Standalone empirical claims that are not 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.

3. Abduction. supports from an empirical claim back to a hypothesis, and refutes from a result to the prediction it came out against, 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. confirms and refutes are the two outcomes a tested prediction can carry, and aim at predictions only; a result bearing on a hypothesis takes supports (issue #28).

4. Elimination. rules-out carries the eliminative move: A's evidence eliminates an explicit alternative B. The Meijer R1 paper's rules-out-multiplicative-gain-control synthesis claim explicitly aggregates this move at the discussion level. The corpus carries 15 rules-out edges, scattered across papers, and the synthesis layer shows they are diagnostically interesting because they are scrubbed by abstracts.

5. Dissociation. dissociates-with is a symmetric edge between two claims whose difference across a condition, region, population or measure is itself the finding; neither bears on the other's truth. The Headley distal-inhib-drops-firing-02hz dissociates-with perisomatic-inhib-drops-firing-07hz — neither claim alone establishes the compartmental dissociation; the contrast does. It is a neutral contrast marked by "whereas", "in contrast", "selectively", and has no MIRA predicate.

6. Tension. in-tension-with is the symmetric edge the #125 ruling split from dissociates-with: two claims the paper asserts, both of which stand, that pull a shared implication in opposite directions or cannot be jointly explained without a further claim — marked by "although", "however", "despite", "no correlation with", "at the cost of". The Gädeke insula guilt response matches the published Yu/Koban signature at the group level (dot-products-between-individual-neural), while individual signature scores do not track individual behavioural guilt (individual-grbs-dot-product-values-not): the replication is real and the null bounds what it can mean. A tension is not a contradiction (both claims hold), not a qualification (neither narrows the other), and not a rules-out (nothing is eliminated). The paper usually resolves it in the discussion with a claim that interprets both.

7. 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. The Meijer R1 paper's optogenetic-activation-not-physiological-pattern scopes the brain-wide additivity claim to optogenetic stimulation, leaving open whether endogenous, mixed-selectivity DRN release would yield the same signature.

4.5 Auxiliary fields

displayClaim — one to two sentences, used in body text and card views; preserves the proposition while allowing readable phrasing.

shortClaim — single short clause for graph nodes and tooltips; required for synthesis / interpretation / literature-context nodes, recommended for high-traffic claims.

number / numberParts — computed at build from role + graph position; do not author manually. The numbering convention is hierarchical: H1.P2.E1 reads as "first hypothesis, second prediction, first empirical test." Standalone empirical (no hypothesis loop) become E#; controls C#; scope Sc#; methodological M#; literature-context L#; synthesis S#; interpretation I#. 242 of the 245 published claims carry computed numbers.

reproductions: — list of blocks. Each block records a verification attempt:

reproductions:
  - agent: mainen-z
    date: 2026-03-30
    status: verified
    script: verification/<paper-slug>/verify.py
    original_figure: verification/originals/<paper-slug>/fig4.jpg
    figure: verification/<paper-slug>/fig4a-firing-rates.png
    original_script: <URL to deposited notebook>
    script_execution: unmodified | patched | from-notes
    script_execution_note: short string describing any patch
    time_fast: "~2 min"
    time_full: "~6 hrs (NEURON + 1.88 GB Dryad)"
    notes: |
      Free prose recording reproduced numerics, comparison with paper,
      and any caveats. The notes field is the primary place where the
      reproduction's evidentiary basis is recorded.

status vocabulary (per reproductions[].status) — see the verification layer for criteria.

Status	Meaning
verified	Ran the analysis end-to-end (or read the deposited intermediate); output matches the assertion within tolerance
verified:partial	Ran a defined subset; matched portion documented in notes
unverified	Not yet attempted (default; reason genuinely unknown)
unverified:no-data	Data deposit not accessible
unverified:no-code	Code not accessible
unverified:code-error	Code runs but errors before producing output; record the exact error and any known fix
unverified:compute-infeasible	Code runs but would require compute beyond available resources; record estimated runtime and any deposit-first workaround
failed:mismatch	Ran; output does not match — discrepancy logged in notes

unknown appears in 95 claims of the rendered claims.json; this reflects either the absence of a reproductions: block on the source claim file (default unknown) or a non-empty block whose status was never set. It should be read as "not yet adjudicated" rather than as a verification outcome.

4.6 Examples per role and edge

A single example per role, drawn from the corpus:

• hypothesis. hypothesis-distinct-compartmental-roles (Headley): "Perisomatic and distal dendritic inhibition serve distinct computational roles… entails: four predictions." A hypothesis carries entails edges to its predictions; it does not itself carry empirical content.

• prediction. prediction-beta-optimal-distal (Headley): "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. distal-inhib-drops-firing-02hz (Headley): "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 — the typical density for a load-bearing empirical claim.

• control. cortical-layers-show-no-differential-modulation (Meijer R1): "Splitting cortical recordings by layer reveals no differences in modulation fraction, sign, or latency. This rules out a layer-specific cortical mechanism." Empirically computed but functions to eliminate an alternative.

• scope. l5-model-single-cell-scope (Headley): "All results come from a single-cell compartmental model… no network dynamics, no recurrent excitation, no population effects." Carries scopes: ["*"], qualifying every empirical claim.

• methodological. manifold-from-pooled-super-session (Meijer R1): "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. orthogonality-derived-from-additivity (Meijer R1): "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. pv-gamma-sst-beta-correspondence (Headley): "Layer 5 inhibitory streams are functionally matched to interneuron type." Carries interprets: edges to the four empirical claims that ground the mapping.

• literature-context. interprets-pv-gamma-sst-beta-associations (Headley): the inherited PV/gamma–SST/beta correspondence from prior literature. Section 5 develops this role.

↑ Contents

---

literature-context is the ninth role, added in iteration 4 of the schema. It treats cited prior work as a first-class claim node — not a citation in a bibliography, but a proposition with the same schema as any other claim, that the present paper's argument inherits.

5.1 Distribution

Twelve literature-context claims appear across eight papers in the present corpus:

Paper	Count	Examples
meijer-2025-serotonin-additive-r1	5	interprets-gain-control-default-framework, interprets-5ht2a-gain-control-visual-cortex, interprets-lottem-2016-additive-piriform, interprets-cohen-li-matias-phasic-5ht-responses, interprets-paquelet-correlated-ensembles
ejdrup-2026-dopamine	2	interprets-cragg-rice-vmax-ratio, interprets-may-wightman-1989-fscv
gadeke-2026-guilt-insula	1	(single literature-context anchor)
headley-2026-inhibitory-rhythms	1	interprets-pv-gamma-sst-beta-associations
kammer-2026-foveal-feedback	1	(single literature-context anchor)
meijer-2025-serotonin-orthogonal	1	(single literature-context anchor)
scheller-2026-self-prioritization	1	(single literature-context anchor)
Total	12

The Meijer R1 paper is the densest case because its central reframing (additivity rather than gain control) requires explicit engagement with the prior-literature gain-control framework. Without literature-context nodes, the rules-out: multiplicative-gain-control synthesis claim would have no explicit referent for "multiplicative gain control" — the move would be eliminative against an unnamed alternative. The literature-context node interprets-gain-control-default-framework makes the Servan-Schreiber lineage explicit, so that the eliminative move has something specific to engage.

5.2 Structural function

A literature-context claim is structurally distinct from an interpretation claim in two respects.

First, its empirical content is not the present paper's evidence — it is content from a cited prior paper (or several), inherited as a load-bearing premise. The claim file's assertions: block records this: method: literature interpretation; cited as the receptor-specific instantiation of the gain-control framework for serotonin. The confidence is bounded by the strength of the prior literature, not by anything the present paper does.

Second, its role in the graph is to give downstream synthesis or scope claims an explicit referent. The Headley interprets-pv-gamma-sst-beta-associations claim explicitly notes: "The literature-context registration matters because the correspondence claim is specifically not a prediction the paper tests — the paper's simulation uses generic inhibitory inputs parameterized by location and frequency, without simulating PV+ or SST+ neurons directly. The biological correspondence is an inherited literature premise that connects the mechanistic result to observed interneuron-type behavior. Without an explicit node for the PV/gamma and SST/beta associations, the synthesis claim's interpretive weight would lean on an unnamed referent."

The role makes inherited premises auditable. Where a paper's interpretation depends on a literature claim that is itself contested, the literature-context node is the place that contest is recorded; downstream claims that requires: or interprets: the literature-context node inherit the contest.

5.3 Cross-paper deduplication

The schema is designed so that a single literature-context node — say interprets-servan-schreiber-1990-gain-control — could be referenced by multiple papers' claims. In the present corpus this is not exploited; each literature-context claim lives in the asserting paper's directory with one assertion block. But the UUID-based identity is constructed so that, at scale, such a claim could migrate to a flat claims/ namespace, accumulate assertion blocks from each paper that cites Servan-Schreiber 1990 in this role, and become a corpus-level node with a single graph identity.

The implication is that citation, in this schema, is not a flat list at the end of a paper. It is a graph: papers connect to prior work through typed edges that name the role the prior work is being asked to play (interprets, requires, validates). The literature-context primitive is what makes citation queryable as graph structure — which papers in the corpus engage the gain-control framework, which engage Lottem 2016, which inherit the Cragg-Rice DAT Vmax ratio. None of these queries is currently realised; the primitive is in place, the deduplication is not.

The forward construction case (claim graphs assembled by authors at submission) is where literature-context would scale. An author with a graph in hand can declare which existing literature-context nodes their paper inherits rather than re-create each one. The deduplication then becomes the corpus's cross-paper primitive.

↑ Contents

---

The methodology described above is the disciplined process the prototype would adopt at scale. The prototype's actual workflow falls short of this discipline in several respects, and the document is honest about the gap.

Authoring discipline not strictly enforced

The procedure above — three independent extractions and a mandatory review gate — describes a workflow the prototype did not strictly enforce. In practice, authoring was prompt-guided LLM extraction with intermittent rather than systematic human review. The 289 claim files (and 4 parts) should be read as a draft annotation layer, not as adjudicated output. A scaled-out version — the version this document is the methodology for — would enforce the three-extraction reconciliation and the review gate as actual procedural checkpoints. The corpus is the prototype's draft; the methodology is the discipline the draft should be brought up to.

Verification coverage is shallow

Of the 10 papers, 7 carry a verification script. Across the corpus, every claim a re-run could settle — 133 of 151 empirical and control claims — carries a reproduction record. Coverage was never the problem. Provenance was.

Until 2026-09-10 those records were narrated: written afterwards by an agent describing what it believed had run. Instrumenting one script to record what it actually opened showed what that costs. A record named Code/csv/fMRI - Choices_singleTrialData.csv while the code opened Behav - Choices_singleTrialData.csv. And one claim — the convergent-validity argument for the paper's central result — carried status: verified and a reproduced value while the function it cited raised shapes (4,4) and (5,5) not aligned and returned nothing at all. The bug was ours, the status was written from what the analysis was expected to show, and nobody could have caught it from the record alone.

The fix is structural: a verification script now emits its own provenance — every file it opens, with size and hash, and every value beside the paper's — and the claim record carries that emitted value rather than a description of it. Gädeke has been through this; the other 6 papers with scripts have not, so their verified statuses should be read as unaudited until they are.

The ~150 claim status labels in the corpus that are not backed by either live execution or from-notes records are agentic extraction judgments — the LLM authoring agent's assessment of whether a claim is observationally direct, requires re-execution, or is methodological. These are draft annotations.

Two documented mismatches

The two mismatches preserved through verification rather than papered over are worth naming.

• Bouyeure prior-threat (anatomical mismatch). Reproduction finds 36 significant voxels with peak at MNI [-9.0, -92.5, -6.0] (occipital pole). The paper localises the prior-threat effect to the fear network. The claim file carries failed:mismatch; the verify log records PASS on the meta-claim that the discrepancy itself is reproduced.

• Wengert maximal firing (quantitative mismatch). Direction reproduced (WT > KI); magnitude and significance off (paper: WT ≈ 201, KI ≈ 126, p < 0.001; reproduction: WT = 207.8, KI = 175.8, p = 0.166). The claim is verified:with-nuance rather than plain verified. The discrepancy is in the n recruited per group and statistical power; the underlying biology direction is correct.

These two cases are the prototype's evidentiary weight. They are what the verification step is for.

Reverse-engineered, not forward-constructed

The corpus was reverse-engineered from finished papers. Forward construction — claim graphs assembled by authors at submission, against a schema they author into rather than retrofit — would look different. The richest contrast is what literature-context becomes at scale (Section 5.3); a smaller contrast is that authors would correct quantitative-hallucination errors in real time rather than on review, and would surface negative results their own discussion glosses past.

Synthesis-pipeline magnitude is not robust

The comparator finding that abstracts scrub rules-out and refutes edges and absorb validates controls — a finding of the synthesis layer — is a structural property of the comparison and is robust to LLM stylistic variation. The magnitude — how much abstract prose is devoted to each move type, how much the synthesis expands beyond what the abstract carries — is sensitive to the agent's stylistic recovery: an LLM that pads the synthesis with connectives the schema does not encode will inflate the apparent gap. The diagnostic findings (which edges are scrubbed) are robust; the magnitude is not. Specific volume comparisons should be treated as illustrative rather than as measurements.

↑ Contents