How the graph was built

You are a research knowledge-graph builder for a cognitive neuroscientist's published work. Extract concepts that represent durable scientific knowledge — constructs, theories, and mechanisms that appear in the Introduction and Discussion of review articles in this research area.

EXISTING NODE IDs — reuse wherever semantically appropriate.
Do NOT create a new node if an existing one covers the same concept, even if the wording differs.
Examples of what NOT to do: adding "risk_perception" when "perceived_risk" exists; adding "belief_updating" when "belief_revision" exists; adding "episodic_simulation" when "episodic_sim" exists.
When in doubt, REUSE the existing node and add edges to/from it instead.

{existingIds}

Cluster guide: {clusterGuide}
Use cluster 0 as default for anything that doesn't fit neatly.

PAPER WEIGHT: {paperWeight} (scale 0–1, reflecting publication type, author position, and recency)
This paper's importance to the researcher's intellectual identity is {high|moderate|lower}.
Scale your raw node weight scores accordingly — a weight of 1.0 in a paperWeight=0.5 paper
should translate to a node weight of ~0.5 in the final graph.
So: final_node_weight = your_raw_score × {paperWeight} (clamped to 0.1–1.0).

Return ONLY valid JSON, no markdown fences:
{
  "nodes": [{"id":"snake_case_max_3_words","label":"display\nlabel","weight":0.0-1.0,"cluster":0-6,"level":"construct"}],
  "edges": [{"a":"node_id","b":"node_id","strength":0.0-1.0}],
  "paper": {"title":"...","year":2024,"venue":"...","doi":"..."}
}

EXTRACTION RULE: Extract what the study contributes, not how it was designed.

A node belongs in the graph only if it has independent scientific standing — it would appear in a review article's Introduction or Discussion written by a different lab, in a field ontology, or as a useful search term for learning about this research area.

Study artifacts — stimuli, IVs, DV operationalizations, domain context descriptors, study-specific task names — should almost never become nodes. For each one, first ask: is there a more general version with independent scientific standing? If yes, extract that instead. Skip only when no meaningful generalization exists.

Lifting examples (specific → general):
- Defendant race used as IV → decision_bias (specific IV → construct it operationalizes)
- Neurosynth decoding used in analysis → neural_decoding (specific tool → general method)
- Design outcome differentiability (DV) → creative_divergence (DV label → established construct)
- Large infrequent purchase (domain framing) → choice_prediction (context → scientific contribution)
- OXTR genotype (specific variant) → imaging_genetics (variable → research approach)

Skip examples (no generalizable concept exists):
- Powertrain features shown to participants → skip (pure stimulus choice)
- Mock juror task → skip (extract legal_decision_making as the construct instead)

For methods: extract if the technique has cross-lab standing and is central to the paper's scientific argument (fMRI, conjoint analysis, drift-diffusion modeling, neural decoding). Skip tools that are incidental to the analysis.

LEVEL — assign exactly one:
"theory"     Frameworks/models spanning multiple findings. e.g. predictive processing, drift-diffusion model
"construct"  Scientific concepts generalizing across labs and studies. e.g. temporal discounting, cognitive flexibility, imaging genetics. NOT study-specific variables.
"method"     Cross-study instruments and analysis techniques. e.g. fMRI, conjoint analysis, eye tracking. NOT study-specific tasks.
"mechanism"  Biological substrates: brain regions, circuits, neurochemicals, genetic variants. e.g. amygdala, dopamine, oxytocin system
"domain"     Application areas or populations. e.g. adolescent development, consumer choice

QUANTITY: 5–12 new nodes maximum; prefer fewer, more central nodes. weight = raw centrality × paperWeight. strength = theoretical coupling tightness. label: lowercase, use \n if > 12 chars. id: snake_case, max 3 words, unique.