Experimental capability ยท bounded eight-case review

Experimental Research Ideas

Evidence-backed hypotheses generated from selected disclosure changes and reviewed against the full prior filing record.

Designed for selectivity

Reject unsupported stories.

The engine is designed to reject unsupported stories rather than generate an idea for every disclosure.

Experimental feature: These are AI-generated research hypotheses grounded in cited filing evidence. They are not facts, personalized investment advice, or validated trading signals. Analyst review is required.

Independent evidence review and automated grounding checks completed. Formal blinded human ratings remain pending.

Bounded result

Safe refusal is part of the product.

Two cases produced publishable research hypotheses. Six were safely held or rejected. None produced an actionable trade view, and broader generation remains blocked.

Published from frozen packets

Publishable hypotheses

Each published packet has full prior-filing review, 1.00 evidence coverage, a falsifiable research question, and a skeptic result of PASS_HYPOTHESIS_ONLY.

Correct rejection

When more context overturns the story

EFX remains the permanent negative fixture: an incomplete prior excerpt made an existing obligation appear new and led to unsupported causal claims.

Transparent failures

Other safe failures

Scale boundary

Broader generation remains blocked.

The 60-case phase is not authorized. Formal blinded human ratings and broader evidence-quality validation remain pending.

How research ideas are grounded