Pure News Intelligence · one-page research-results summary

Negative and null results were preserved rather than tuned away.

1 · Combined multisource model

Question: Did 10-K information improve an existing 8-K and inexpensive-news model?

Result: No incremental improvement was supported.

Decision: Do not claim incremental alpha.

2 · Expected-disclosure model

Question: Did characteristics, 8-Ks, and news predict the next filing representation better than the prior 10-K alone?

Result: The prior 10-K persistence baseline performed better.

Decision: Stop added modeling complexity at this gate.

3 · Disclosure-change demonstration

Question: Can the system identify, evidence, and check meaningful changes?

Result: The workflow was operational; the bounded risk test was inconclusive.

Decision: Advance to an analyst workflow pilot—not trading deployment.

Claim boundary

Supported: document processing, historical comparison, point-in-time integrity, evidence extraction, reproducible classifications.

Not supported: alpha, causal prediction, improved expected-language reconstruction, adoption, willingness to pay.

Not tested: time savings, recall, trust, repeat use, institutional-memory value.