Small bounded sample
The Experimental Research Idea Engine reviewed eight frozen cases. Two produced publishable hypotheses and six failed safely; the 60-case phase remains blocked and unrun.
Known limitations
The strongest claims are deliberately narrow.
The demonstration shows an inspectable workflow. It does not establish investment performance, analyst adoption, or commercial value.
Human review pending
Independent evidence review and automated grounding checks are complete, but formal blinded human ratings have not been collected.
No workflow pilot
No live institutional workflow pilot has measured analyst time, recall, trust, edit burden, adoption, or willingness to pay.
No performance claim
No Sharpe-ratio, information-coefficient, hit-rate, signal-decay, return-prediction, or causal improvement has been demonstrated.
No commercial validation
The project has no validated customer-adoption or commercial-success claim. Interviews informed the design but did not establish product-market fit.
Public-safe evidence limits
The public repository uses permitted, frozen evidence and cannot reproduce the breadth of licensed institutional datasets.
Static public packets
GitHub Pages serves frozen packets rather than live inference. This improves auditability but does not demonstrate a production generation service.
Generalization risk
Results from the bounded issuers, filings, periods, and disclosure categories may not generalize to other companies, markets, document types, or regimes.
Analyst accountability
A hypothesis can be grounded and still be unhelpful or wrong. Analysts remain responsible for interpretation, additional diligence, and every decision.
Release boundary: formal blinded human review is the next evidence gate. Broader generation is not authorized.