Public FAQ
Direct answers about what the project does—and does not prove.
The public demo is a frozen, evidence-linked research workflow. It is not a live stock-picking service.
Frequently asked questions
1. Does this generate stock picks?
No. It identifies evidence-linked disclosure changes and, for selected cases, proposes testable research hypotheses. It produced zero actionable trade views.
2. Did the original research beat the market?
No. The locked return-prediction tests did not support the original pure-news alpha thesis, and no Sharpe-ratio or information-coefficient improvement was demonstrated.
3. Why did the project pivot?
The null return-prediction result and early discovery work pointed to a more defensible problem: helping analysts prioritize disclosure changes, verify evidence, and preserve research context.
4. Why were only eight Research Idea Engine cases reviewed?
V2 was a bounded grounding repair. The same eight frozen cases were rerun to test whether the system could correct known defects before any broader evaluation.
5. Why is the 60-case phase blocked?
Formal blinded human review is still pending. Broader generation is not authorized until reviewers assess factual support, novelty, mechanism quality, falsifiability, and usefulness.
6. What does 1.00 evidence coverage mean?
Every factual proposition in a published packet was marked supported by the packet’s cited evidence. It does not mean the hypothesis is true, useful, or predictive.
7. What makes this different from a generic AI summary?
The workflow compares prior and current filings, searches earlier context, decomposes claims, normalizes financial quantities, retrieves counterevidence, requires falsification conditions, and permits safe rejection.
8. Why is EFX considered a successful outcome?
Full-prior evidence showed that a supposedly new term had already been disclosed and that compared amounts had different economic roles. Unsupported mechanisms were removed, and no trade view survived.
9. Does the public site run live AI models?
No. GitHub Pages serves public-safe frozen JSON packets and static interface code. No model inference occurs in the public browser workflow.
10. Can analysts edit or approve hypotheses?
The detail pages provide browser-only disposition controls and local notes. They do not alter frozen packet JSON, and no analyst approval is claimed.
11. What evidence is still missing?
Formal blinded human ratings, a live institutional workflow pilot, prospective hypothesis freezing, broader case evaluation, and only then investment-performance analysis.
12. What would be tested in a real pilot?
Review time, factual recall, novelty accuracy, mechanism clarity, falsifiability, usefulness, edit burden, analyst preference, and safe-rejection quality.
13. Is this investment advice?
No. It provides experimental research decision support, not personalized investment advice, position sizing, price targets, or trade instructions.
14. How were future outcomes kept out of the packets?
Evidence eligibility is point-in-time. Frozen source receipts, timing audits, and tests prevent post-filing outcomes from entering the generation and grounding record.
15. Which parts are reproducible from the public repository?
The static demo, public-safe packets, schemas, prompts, deterministic indexing and audit scripts, and automated tests are inspectable. Licensed or sensitive source material is not redistributed.
Audit the boundary