Scientific record
Current research question
How can innovative transformation of already-available representations, without materially new external information, be operationalized in humans; can its events or dimensions be assessed independently; and do observable signals generalize across domains within one modality after alternative explanations are controlled?
The later question is whether the capability can be reproduced or encouraged in AI models, with runtime and reward branches tested independently. Retrieval is a substrate of transformation. The contrast is mere reuse/reproduction versus innovative transformation, not construction versus retrieval.
Highest-priority register
All entries remain open. Links to hypotheses identify dependencies, not validated answers.
| ID | Open question | Why it matters · next evidence | Related claim |
|---|---|---|---|
| FQ-01 | What exactly counts as an innovation event? Is innovation categorical, dimensional, gradual, or multistage? | Define observable criteria and a temporal unit before labeling or detection. | FH-01, FH-02 |
| FQ-02 | What is fixed in a fixed information state? How do prior knowledge, internal retrieval, derived information, tools, hints, and feedback enter the boundary? | Specify external access and document preparation; do not claim that cognition freezes internally. | FH-01 |
| FQ-03 | Novel relative to what, and what makes a representation actionable? | Separate personal/task/reference-set novelty, meaningfulness, usefulness, and demonstrated action. | FH-01 |
| FQ-04 | How do we distinguish genuine innovation from random variation, paraphrase, novelty of wording, or superficial recombination? | Test matched negative cases and product/process assessments. | FH-01, FH-03 |
| FQ-05 | How do difficulty, uncertainty, hesitation, controlled retrieval effort, and cognitive load differ from the target? | Include demanding non-innovation comparisons and nuisance-only prediction. | FH-03 |
| FQ-06 | Can independent raters reliably identify transformation or events, and are their judgments valid? | Reliability is insufficient; test origin, transformation, novelty, and utility judgments separately. | FH-02 |
| FQ-07 | What evidence should raters see? Can output-only judgments support a process label? | Predefine evidence access; keep detector features and condition cues out where possible. | FH-02 |
| FQ-08 | What controls show a signal tracks innovation rather than difficulty or expression novelty? | Compare recalled-original, effortful retrieval, load, length, prompt, and participant baselines. | FH-03 |
| FQ-09 | What counts as true domain generalization while modality remains constant? | Define domain difference, lock the model, prevent participant/item/session leakage, and protect the test domain. | FH-04 |
| FQ-10 | Which cognitive-science findings constrain the hypothesis; what is known science versus a distinctive claim? | Use the baseline and focused event-detector prior-art searches; priority claims remain unestablished. | FH-01 to FH-04 |
| FQ-11 | How much directional evidence warrants a confirmatory study? | Fix continuation/failure rules before data; power a later study for a meaningful effect with participant clustering. | FH-02 to FH-04 |
| FQ-12 | Would a narrower classifier improve runtime decisions and model outcomes? | Independent intervention comparison; human-to-model transfer is separate. | FH-05, FH-07 |
| FQ-13 | Can a detector-derived reward improve the target without gaming, noise, proxy optimization, or shift? | Separate reward tests with an independent outcome measure; no accuracy-to-RL inference. | FH-06 |
Resolution discipline
No event definition, label rubric, actionability criterion, numeric gate, sample size, modality, domain pair, or complete protocol is adopted. The baseline review constrains the inquiry; the pilot redesign is a NEW PROPOSAL. Record any resolution, source, alternatives, and consequences in the hypothesis history and research log.
Complementary existing-AI capability register
NEW PROPOSAL: Fiction-CWR asks whether a scoped valid-synthesis deficit remains when supplied primitive knowledge is demonstrated and evidence for composition familiarity is controlled. This narrow behavioral question complements the current human question; it does not resolve FQ-01/FQ-06/FQ-07.
| ID | Open question | Required next evidence | Related claim |
|---|---|---|---|
| FQ-14 | Do pinned frontier models show a substantive withheld-versus-familiar composition loss despite primitive competence? | Frozen P/F/C protocol; all-model and qualified estimates; independent confirmation | FH-08, FH-07 |
| FQ-15 | Can every accepted solution's necessary withheld motif be certified, and F/C evidence be matched? | Independent enumeration/certificates, copy audits, exposure and difficulty manifests | FH-08, FQ-02/FQ-03 |
| FQ-16 | Is an apparent loss explained by state load, output burden, prompt/representation or budget? | P/R/E, higher-budget and representation arms; family sensitivity | FH-08, FQ-05 |
| FQ-17 | Do synthetic compositions predict useful constrained workflows or human-relative behavior? | Separate domain/practical/human comparison studies; no inference from solver success | FH-01, FH-07 |
| FQ-18 | What independent task count, mastery rule and effect threshold support confirmation? | Prospective simulation/precision design and frozen criterion rationale | FH-08, FQ-11 |
All remain unresolved. Candidate δ=0.20/τ=0.90 belong to EXP-0002's proposed design, not adopted human gates.
| ID | Open question | Required next evidence | Related claim |
|---|---|---|---|
| FQ-19 | Could workflow organization partly explain an apparent withheld-composition loss at comparable cumulative resources? | Certified F/C × SD/SW on one pinned model; equal iteration/history/evidence/permissions; actual usage; positive withheld gain as well as gap shrink; separate cost interpretation | FH-09, FH-08, FQ-16 |
EXP-0003 prepares a finite directional diagnostic for FQ-19. It records no model result and resolves no human construct or label question. Numeric continuation/qualification/resource tolerances and execution remain unresolved.
2026-10-23 — Construct-validity questions retained
The current research question above is unchanged. The source integration adds explicit questions to FQ-01–FQ-04/FQ-17:
- Can outcome-based innovation criteria be validated independently of search, analogy or other producing mechanisms?
- Can valuable transformation of a representation, objective or opportunity be assessed alongside fixed-goal solutions without rewarding easier goals or arbitrary substitution?
- What independent evidence distinguishes usefulness from finite-world feasibility, including benefits not immediately measurable under the initial goal?
- How can external information, goal, representation and legal environmental operations be varied independently while controlling search freedom and difficulty?
All remain UNRESOLVED. The missing Cycle 2/3 report bodies and CV-001 materials limit reconciliation; their absence is not evidence that the work never occurred.