Cognitive-science baseline review

Scientific record

Project: Project Fiction

Reviewed: 2026-10-06

Purpose: Constrain the migration handover's current hypotheses with an actual literature map before committing resources to an Invention Classifier pre-test.

Main assessment: Creative transformation drawing on memory is well grounded as a broad research framing. A discrete, independently identifiable innovation event with confound-resistant signals that transfer across domains is still a hypothesis. Classifier utility in runtime assistance and in reward optimization requires separate evidence.

Scope and provenance

This is a targeted, critical baseline, not a systematic review or a complete account of current cognitive science. Searches combined specific constructs, authors, and paper titles: controlled retrieval and creative ideation; conceptual combination; analogical transfer; restructuring and insight; episodic retrieval; divergent/convergent thinking; semantic scoring; source monitoring; speech under cognitive load; incubation; creativity prediction; and computational evaluation. Searches and source checks occurred on 2026-10-06. Foundational papers and recent work through 2026 were included where directly relevant. General searches for "creativity" alone did not define the map.

The map contains 26 source records. Sources are original research reports, authors' theoretical papers, meta-analyses, or scholarly reviews, accessed through publishers, PubMed, author/institutional manuscripts, and arXiv. Aggregator summaries were discovery leads only. Publication dates follow the source record, not search-engine crawl dates. No Hermes experiments were run.

Reading depth: A means the primary publication's abstract or publisher preview was inspected; M means selected full-text methods and results were also inspected. Indexed primary-source passages supplement some abstracts. Several direct PMC requests returned browser checks, and some publisher pages restricted full-text access; those were not represented as full-paper audits. Abstract-level records do not establish unreported exclusions, statistical robustness, or reproducibility. Limitations below are Hermes' critical assessments unless explicitly attributed to authors.

The handover is preserved locally (not included in this public selection: ../../../docs/handovers/2026-10-06-migration-handover.txt). Its report of older Fiction work is provenance for current positions, not independent experimental evidence. The older conversations and underlying research documents have not been imported.

Literature map by research tradition

Source extractions

Each record separates the source claim and observed finding from Hermes' interpretation and proposed consequences. No record below validates the Fiction thesis.

L01 — Memory in Creative Ideation (MemiC)

Benedek, Beaty, Schacter, and Kenett (2023), *The role of memory in creative ideation*, Nature Reviews Psychology 2, 246-257. Perspective; depth A. Publisher record.

• Claim: Creative ideation draws on controlled semantic and episodic memory processes.
• Method: Theoretical synthesis proposing stages of memory search, candidate construction, novelty evaluation, and effectiveness evaluation.
• Finding: An integrative framework, not a new experiment demonstrating a universal sequence.
• Limitation: Distinguishable stages need not be discrete, serial, or externally identifiable.
• Relevance: Very close precedent for Fiction's provisional process model.
• What changes: Attribute the precedent; stop treating search-to-transformation-to-evaluation as Fiction's distinctive mechanism. The event and signal claims must carry their own tests.

L02 — Memory-creativity meta-analysis

Gerver, Griffin, Dennis, and Beaty (2023), *Memory and creativity: A meta-analytic examination of the relationship between memory systems and creative cognition*, Psychonomic Bulletin & Review 30, 2116-2154. Meta-analysis; depth A. Publication abstract.

• Claim: Memory's relationship with creativity depends on memory type, task, and modality.
• Method: 525 correlations from 79 studies/unpublished datasets, covering 12,846 participants.
• Finding: Overall association was small, r = .19; semantic memory/verbal fluency was important, with differential working-memory and modality relationships.
• Limitation: Correlations and heterogeneous tasks do not establish a common event mechanism.
• Relevance: Supports memory involvement and warns against universal signatures.
• What changes: Separate domain from modality and avoid interpreting modest trait associations as event-level evidence.

L03 — Structured imagination

Ward (1994), *Structured Imagination: the Role of Category Structure in Exemplar Generation*, Cognitive Psychology 27, 1-40. Experiments; depth A. Publisher abstract.

• Claim: Generation in unfamiliar settings is structured by familiar category knowledge.
• Method: Drawings/descriptions of imagined extraterrestrial animals, with category and instruction manipulations; comparison with writers' creations.
• Finding: Familiar animal properties and retrieved exemplars influenced generated creatures; instructions altered reliance on knowledge frameworks.
• Limitation: Imagined-category outputs do not establish actionability or comprehensive knowledge control.
• Relevance: Existing representations both enable and constrain transformation.
• What changes: Record instruction and cue effects; fixed prompts do not imply identical information states. Preserve the distinction between Fiction and imagination tasks.

L04 — Cross-category conceptual combination

Rabini, Ubaldi, and Fairhall (2021), *Combining concepts across categorical domains: a linking role of the precuneus*, Neurobiology of Language 2, 354-371. fMRI experiment; depth A. Publication abstract.

• Claim: Conceptual combination can connect representations across semantic categories.
• Method: N = 24; single-category versus cross-category sentences across five semantic domains.
• Finding: Cross-category combinations increased precuneus activity while category-selective representations persisted.
• Limitation: Sentence comprehension is not invention; a brain contrast is not a universal event detector.
• Relevance: Cross-domain combination has a concrete cognitive precedent.
• What changes: Treat combination as a component process; require novelty and actionability evidence beyond the mere presence of combined concepts.

L05 — Analogical problem solving

Gick and Holyoak (1980), *Analogical problem solving*, Cognitive Psychology 12, 306-355. Five experiments; depth A. Publisher abstract.

• Claim: A relation from a distant source problem can guide a target solution.
• Method: Prior military-story analogy followed by a medical radiation problem; variations in hints and source relevance.
• Finding: Analogical solutions were more readily generated when participants were prompted to use the source.
• Limitation: Source presentation and hints supply information; puzzle transfer is not general-purpose invention.
• Relevance: Relational restructuring is a plausible route to actionable novelty.
• What changes: Log the acquisition and accessibility of source knowledge, and distinguish a new inference from new external input.

L06 — Constraint relaxation and chunk decomposition

Knoblich, Ohlsson, Haider, and Rhenius (1999), *Constraint Relaxation and Chunk Decomposition in Insight Problem Solving*, Journal of Experimental Psychology: Learning, Memory, and Cognition 25, 1534-1555. Experiments; depth M. Author/institutional paper.

• Claim: Impasse can be overcome by changing restrictive representations.
• Method: Matchstick arithmetic problems varying constraint scope and chunk decomposition; subsequent transfer tests.
• Finding: These manipulations affected difficulty and transfer in the tested problem family.
• Limitation: Repeated exposure changes participants' knowledge; a narrow puzzle family does not establish cross-domain innovation.
• Relevance: Restructuring is experimentally approachable without a retrieval-versus-construction ontology.
• What changes: Separate the need for restructuring from difficulty and prior solution exposure. Do not mistake transfer within a problem family for domain generalization.

L07 — New versus recalled ideas

Benedek et al. (2014), *To create or to recall? Neural mechanisms underlying the generation of creative new ideas*, NeuroImage 88, 125-133. fMRI experiment; depth A. Publication abstract and figure description.

• Claim: Newly generated and recalled object-use ideas can be compared while both involve memory processes.
• Method: Self-paced idea generation, vocal responses, and retrospective categorization of ideas as old or new.
• Finding: New ideas differed in parietal activation; the authors interpret generation as involving controlled semantic retrieval.
• Limitation: Retrospective origin labels and product originality can be confounded.
• Relevance: Close prior art for process comparison, not proof of a domain-general invention signal.
• What changes: Include original-but-recalled responses as a comparison, informed by L08.

L08 — Creating original versus recalling original ideas

Benedek et al. (2018), *To create or to recall original ideas: Brain processes associated with the imagination of novel object uses*, Cortex 99, 93-102. fMRI experiment; depth M. Author manuscript.

• Claim: Creation of new original uses can be separated from recall of original and common uses.
• Method: Initial N = 42; retrospective response-based analysis retained 38. Independent judges rated product creativity; participants judged origin.
• Finding: New and recalled original ideas shared activation patterns, with a localized difference for newly generated ideas.
• Limitation: Good agreement on creativity is not validation of origin judgments; original/common generation periods differed.
• Relevance: Direct precedent and a warning about process labels and timing.
• What changes: Separate origin, transformation, product novelty, usefulness, and timing. Match timing in comparisons rather than copying condition differences uncritically.

L09 — Executive functions and creative performance

Benedek, Jauk, Sommer, Arendasy, and Neubauer (2014), *Intelligence, creativity, and cognitive control*, Intelligence 46, 73-83. Latent-variable study; depth A. Publication abstract.

• Claim: Executive functions contribute differently to intelligence and divergent-thinking performance.
• Method: Updating, inhibition, shifting, intelligence, and creativity measures analyzed in a latent model.
• Finding: Updating and inhibition predicted creativity; shifting did not in this model.
• Limitation: Individual-difference associations do not identify innovation events or causal necessities.
• Relevance: Cognitive control is both a possible component and a competing explanation for signals.
• What changes: Add demanding, non-innovative retrieval/control conditions; avoid labeling executive effort as invention by default.

L10 — Divergent/convergent semantic-control dissociation

Krieger-Redwood et al. (2025), *Divergent and convergent creativity relate to different aspects of semantic control*, Imaging Neuroscience 3, imag_a_00502. Behavioral/resting-state study; depth A. Publication abstract.

• Claim: Distinct semantic-control processes relate to different creativity tasks.
• Method: Weak-association retrieval and feature selection compared with unusual uses, remote associates, and intrinsic connectivity.
• Finding: Unusual-use performance related to weak-association retrieval; remote-associate performance related to selection, with different connectivity associations.
• Limitation: Task and trait associations do not establish a single transferable state signature.
• Relevance: Challenges collapsing all creativity-related tasks into one positive class.
• What changes: Define the component tested and report task-stratified results; generalization across components remains separate from across domains.

L11 — Episodic retrieval manipulation

Madore, Addis, and Schacter (2015), *Creativity and Memory: Effects of an Episodic-Specificity Induction on Divergent Thinking*, Psychological Science 26, 1461-1468. Two experiments; depth A. Publication abstract.

• Claim: Episodic retrieval can contribute to divergent creative thought.
• Method: Brief detailed recollection induction versus control, followed by divergent, association, and convergent tasks.
• Finding: Divergent performance improved selectively under the tested induction conditions.
• Limitation: The induction changes retrieval strategy; it does not prove a universal invention process or a fixed internal state.
• Relevance: Retrieval is a substrate, and constructive memory links Fiction to simulation research.
• What changes: Document preparation manipulations; do not exclude episodic retrieval from innovation or merge the Hermean Projects on this basis.

L12 — Semantic richness constrains fluency and originality

Beaty, Kenett, Hass, and Schacter (2023; online 2022), *Semantic Memory and Creativity: The Costs and Benefits of Semantic Memory Structure in Generating Original Ideas*, Thinking & Reasoning 29, 305-339. Four studies; depth A. Publication abstract.

• Claim: Cue richness can affect quantity and originality differently.
• Method: Alternate-use cues with differing associative set sizes, alongside individual-difference measures.
• Finding: Sparse cues favored originality at a fluency cost; cognitive-control differences mattered.
• Limitation: Cue effects and originality ratings do not establish actionability or event identity.
• Relevance: Prompt structure can create both signal and confounding.
• What changes: Balance cue richness and response opportunities; keep fluency, length, and originality separate.

L13 — Semantic-distance assessment

Beaty and Johnson (2021; online 2020), *Automating creativity assessment with SemDis: An open platform for computing semantic distance*, Behavior Research Methods 53, 757-780. Five validation studies; depth A. Publication abstract.

• Claim: Semantic distance can predict human creativity/novelty ratings on verbal tasks.
• Method: Multiple embedding models assessed against alternate-use and word-association judgments.
• Finding: A combined semantic-distance factor predicted ratings and correlated with related measures.
• Limitation: Rating correspondence does not identify the process or establish usefulness; random distance can lack meaning.
• Relevance: Useful candidate output baseline, not ground truth for invention.
• What changes: Compare against semantic-distance-only prediction; never derive labels from the same semantic proxy that is then celebrated as a detector.

L14 — Verbal-fluency clustering and switching

Troyer, Moscovitch, and Winocur (1997), *Clustering and switching as two components of verbal fluency*, Neuropsychology 11, 138-146. Behavioral experiments; depth A. Publication abstract.

• Claim: Retrieval performance includes distinguishable clustering and switching components.
• Method: Younger/older adult fluency tasks and a divided-attention manipulation.
• Finding: Clustering and switching related differently to semantic/phonemic fluency; divided attention affected switching in phonemic fluency.
• Limitation: Fluency retrieves words; transitions are not inherently invention.
• Relevance: Associative search and effort can generate observable temporal structure.
• What changes: Use controlled retrieval/search as a negative comparison; do not label semantic switches or pauses as innovation events by definition.

L15 — Incubation meta-analysis

Sio and Ormerod (2009), *Does incubation enhance problem solving? A meta-analytic review*, Psychological Bulletin 135, 94-120. Meta-analysis; depth A. Publication abstract.

• Claim: Incubation benefits depend on problem type and interval conditions.
• Method: Synthesis examining preparation, interval duration, cues, and intervening demand.
• Finding: Positive average incubation effects, with divergent tasks benefiting more and high-demand intervals reducing effects.
• Limitation: Heterogeneous effects do not uniquely establish unconscious innovation or spontaneous-event identity.
• Relevance: Context for delayed transformations and the boundary with Project Daydream.
• What changes: Treat incubation as a manipulation to study, not a positive label; record intervening information and activity.

L16 — Mind wandering and incubation

Baird et al. (2012), *Inspired by distraction: mind wandering facilitates creative incubation*, Psychological Science 23, 1117-1122. Experiment; depth A. Publication abstract.

• Claim: An undemanding intervening task may support later creative performance.
• Method: Unusual-use problems with demanding, undemanding, rest, and no-break conditions.
• Finding: The undemanding condition improved performance on previously encountered problems and was associated with more mind wandering.
• Limitation: The design does not uniquely isolate mind wandering as the causal mechanism or prove all off-task cognition productive.
• Relevance: A documented overlap to investigate with Project Daydream.
• What changes: Preserve preparation and task-history boundaries; keep spontaneous cognition and measured innovative output separate.

L17 — False insights and metacognition

Danek and Wiley (2017), *What about False Insights?*, Frontiers in Psychology 7, 2077. Experiment; depth A. Publication abstract.

• Claim: Subjective Aha experiences are multidimensional and can accompany errors.
• Method: N = 70 solving magic-trick problems with ratings of Aha, certainty, suddenness, and related experiences.
• Finding: False insights occurred; correct/incorrect solutions differed in experience magnitude and quality.
• Limitation: Puzzle phenomenology is not identical to Fiction's broader innovation target.
• Relevance: Confidence or an Aha report cannot be an infallible positive label.
• What changes: Assess correctness/usefulness separately; retain ambiguous and falsely confident cases in label-validity work.

L18 — Source-monitoring framework

Johnson, Hashtroudi, and Lindsay (1993), *Source monitoring*, Psychological Bulletin 114, 3-28. Theoretical review; depth A. Publication abstract.

• Claim: Judgments about memory origins use flexible experiential criteria and can err.
• Method: Framework integrating empirical findings on attribution, familiarity, cryptomnesia, and memory disruption.
• Finding: A theory of origin judgments, not a validated Fiction labeling instrument.
• Limitation: General memory attribution does not quantify label error in the proposed tasks.
• Relevance: "I invented this" can misattribute prior knowledge; independent raters also lack direct access to origin.
• What changes: Triangulate available knowledge, task record, self-report, and product judgments; distinguish inter-rater reliability from validity.

L19 — Voice under cognitive load

MacPherson, Abur, and Stepp (2017), *Acoustic Measures of Voice and Physiologic Measures of Autonomic Arousal during Speech as a Function of Cognitive Load*, Journal of Voice 31, 504.e1-504.e9. Experiment; depth M. Publication abstract and author manuscript.

• Claim: Cognitive demand can affect physiology and voice quality.
• Method: N = 16; sentence-level congruent/incongruent Stroop conditions, acoustic/autonomic measures, regression including participant identity.
• Finding: Skin conductance and two voice-quality measures related to load condition.
• Limitation: Small, controlled sample; this is neither invention nor cross-domain classification.
• Relevance: Direct alternative explanation for acoustic signals.
• What changes: Include load-matched non-innovation conditions, participant-disjoint evaluation, and nuisance-feature baselines.

L20 — Recent conversational cognitive-load prediction

Chowdhury (2026), *Predicting Cognitive Load from Speech and Interaction Dynamics in Dyadic Conversations*, arXiv:2606.12971, submitted 11 June. Preprint; depth A. Author preprint.

• Claim: Speech/interaction features may predict perceived load in collaborative conversation.
• Method: Audio from 53 dyads on nine tasks; recurrent prediction model.
• Finding: The author reports predictive information for several perceived-load dimensions.
• Limitation: Preprint; abstract-level inspection; perceived load is not invention. Split robustness was not audited.
• Relevance: A current competing target for the same broad feature family.
• What changes: Treat effort and interaction dynamics as measurable nuisance candidates; this source supplies no accuracy threshold or validation of Fiction.

L21 — Creative brain-network dynamics

Beaty, Benedek, Silvia, and Schacter (2016), *Creative Cognition and Brain Network Dynamics*, Trends in Cognitive Sciences 20, 87-95. Review; depth A. Publication abstract.

• Claim: Creative cognition involves interactions among large-scale systems.
• Method: Synthesis across divergent thinking and artistic tasks.
• Finding: Reviewed studies suggest cooperation between default and executive-control networks.
• Limitation: A cross-study interpretation is not a single event detector or an invariant brain signature; tasks and analysis levels differ.
• Relevance: Supports interacting generative and evaluative processes.
• What changes: Keep neural mechanisms separate from behavioral/acoustic features; L10 warns that connectivity relationships depend on the component and measurement context.

L22 — Predicting creative ability from connectivity

Beaty et al. (2018), *Robust prediction of individual creative ability from brain functional connectivity*, PNAS 115, 1087-1092. Predictive study; depth A. Institutional author manuscript.

• Claim: Functional connectivity can predict individual creative-thinking ability.
• Method: Initial task-fMRI sample of 163; cross-validation and three independent validation datasets.
• Finding: Connectivity models predicted creative ability in the reported datasets.
• Limitation: Individual ability prediction is different from detecting when a particular innovation occurs; neural measures do not validate speech signals.
• Relevance: Demonstrates relevant prediction precedent without settling Fiction's target.
• What changes: Specify the unit of prediction and level of generalization. Do not claim Fiction is the first attempt to predict creativity from observable signals.

L23 — Reverse-inference discipline

Poldrack (2006), *Can cognitive processes be inferred from neuroimaging data?*, Trends in Cognitive Sciences 10, 59-63. Methodological analysis; depth A. Publication abstract.

• Claim: Activity associated with a process is not deductive proof that the process occurred.
• Method: Bayesian analysis of activation evidence and process selectivity.
• Finding: Reverse inference depends on how selectively a signal relates to the proposed process.
• Limitation: This does not establish that Fiction classification is impossible; it identifies an inference problem.
• Relevance: A good classifier can detect correlates or nuisance variables rather than innovation.
• What changes: Test alternative explanations, prevalence, calibration, and negative controls before giving predictions an event interpretation.

L24 — Counterfactual thinking and behavioral regulation

Epstude and Roese (2008), *The functional theory of counterfactual thinking*, Personality and Social Psychology Review 12, 168-192. Theoretical review; depth A. Publication abstract.

• Claim: Alternatives to past events can support regulation and performance improvement.
• Method: Synthesis distinguishing content-specific and content-neutral routes to behavioral effects.
• Finding: A framework linking counterfactuals to action, not a universal benefit or invention metric.
• Limitation: Generating a hypothetical does not by itself demonstrate novelty or effective action.
• Relevance: Related tradition connecting constructed possibilities to behavioral consequences.
• What changes: Separate a novel interpretation, a useful intention, and demonstrated action; preserve the distinct question of Project Imagination.

L25 — Computational creativity evaluation

Jordanous (2012), *A Standardised Procedure for Evaluating Creative Systems: Computational Creativity Evaluation Based on What it is to be Creative*, Cognitive Computation 4, 246-279. Methodological proposal; depth A with indexed primary-text sections. Publisher paper.

• Claim: Computational creativity needs explicit, system-appropriate evaluation standards.
• Method: SPECS: state the definition, derive standards, then test; discusses product/process evaluation.
• Finding: An evaluation procedure, not proof of human-like capability in a system.
• Limitation: Standards still require construct validity and suitable evidence.
• Relevance: Output quality and process reproduction are different claims.
• What changes: Define evaluation for the human phenomenon, detector, runtime intervention, and reward branch separately; never equate narrative style with capability transfer.

L26 — Current human/LLM semantic-divergence comparisons

Bellemare-Pepin et al. (2026), *Divergent creativity in humans and large language models*, Scientific Reports 16, 1279, published 21 January. Benchmark study; depth A with indexed primary-text passages. Publisher paper.

• Claim: LLM and human output divergence can be compared on shared verbal tasks.
• Method: Divergent Association Task and writing benchmarks, including a 100,000-human comparison dataset; prompt/temperature variations.
• Finding: Tested models could exceed average DAT performance but not higher-performing human aggregates under the reported scoring.
• Limitation: Semantic-divergence metrics do not establish actionable innovation, a human-equivalent process, or a controlled prior knowledge state.
• Relevance: Existing systems already exhibit relevant task behavior.
• What changes: Do not presume a missing capability; compare against existing systems on justified Fiction tasks before attributing progress to a new intervention.

Synthesis: what is supported and what remains open?

1. What parts of Fiction are already well grounded?

External evidence / current interpretation: Creative thought can depend on stored semantic and episodic knowledge, controlled access, combination, restructuring, and evaluation. The claim that retrieval participates in transformation is consistent with L01, L02, L07-L12; it is not a novel Fiction discovery. Established task traditions examine related processes, with distinctions between fluency, originality, task solution, and evaluative behavior (L03-L06, L10, L13-L14).

Supported methodological caution: Output originality does not guarantee new construction on that occasion; recalling an original idea is an important comparison (L08). Aha reports, origin judgments, and rater agreement cannot alone establish process truth (L17-L18). Load and control can produce observable signals (L09, L19-L20). Creativity-related prediction already exists at other units of analysis (L22).

This supports a research framing and specific controls. It does not validate the handover's entire process chain, a discrete event ontology, a new dataset label, or an Invention Classifier.

2. What parts are plausible extensions but not established?

• CURRENT HYPOTHESIS: A useful operational event/window may be identifiable under controlled external information availability. Whether innovation is categorical, dimensional, gradual, or multistage remains open.
• CURRENT HYPOTHESIS: Features may add predictive information beyond difficulty, retrieval effort, confidence, hesitation, prompt cues, length, participant identity, and semantic-distance baselines.
• CURRENT HYPOTHESIS: A signal may transfer to a genuinely held-out domain while modality stays constant. Related component/task effects do not imply such transfer.
• UNRESOLVED: Whether process labels can be sufficiently valid and reliable without exposing raters to detector features or relying exclusively on self-report.
• CURRENT HYPOTHESIS: A narrower detector could improve runtime behavior even without broad generalization. Benefits must be measured against matched prompting/intervention controls.
• UNRESOLVED: Reward optimization might improve an independent capability criterion; accuracy alone says nothing decisive about gaming, proxy optimization, or distribution shift.

3. What appears to be Fiction's distinctive scientific claim?

Candidate claim, not established priority: Under an explicitly controlled external information boundary, independently assessed innovative transformations may have observable, temporally localized signals that remain informative across different domains within the same modality after credible nuisance controls.

The distinctiveness lies in the conjunction of event-level target, label validity, information control, confound resistance, and held-out-domain transfer, not in memory recombination, creative task performance, brain correlates, or a classifier alone. This review has not established that no prior study tests that conjunction. A focused prior-art search on event-level cross-domain detectors is required before claiming scientific novelty.

If event validity fails, first revise the measurement hypothesis. If a controlled within-domain signal fails, revise the instrument/features or signal hypothesis without declaring that humans cannot innovate. If cross-domain transfer fails, narrow that claim; assess runtime value separately only if it has an independent rationale. None of these outcomes establishes reward suitability.

Consequences for the next research decision

NEW PROPOSAL: Start with label/construct feasibility and a cheap, falsifiable same-modality pre-test. Include recalled-original, effortful non-innovative retrieval, difficult non-innovative reasoning, and superficial novelty comparisons where valid. Keep origin, transformation, novelty, and usefulness as separate assessments before deciding whether a binary event label is justified.

Use participant/session-disjoint evaluation, hold the second domain out from fitting and threshold selection, and compare nuisance-only, semantic-only, and candidate-signal models. The proposed sequence and gates are recorded in the methodology audit, pilot redesign, and decision gates. These are research proposals, not executed experiments or confirmed results.

Limits and next depth checks

Before protocol freeze, inspect complete methods, supplements, exclusions, and analyses for the most relevant event/label studies, assess replication and measurement validity, and search specifically for prior cross-domain event detectors. Only three sources here received selected full-text methods/results inspection. Sample sizes are stated only where checked in primary material.

Deepen concept formation/learning, standalone speech planning and hesitation, actionability measurement, computational process models beyond evaluative frameworks, and the distinction between incidental concept combination and innovation. Expand neural-decoding and counterfactual/episodic-simulation coverage as needed without prematurely adopting neuroimaging or a multimodal architecture. v0.1 deliberately keeps those gaps visible rather than inferring their answers from adjacent findings.