3–6 analysts
Enough for observed workflow variation without pretending to establish broad adoption.
Proposed next experiment · 2–4 weeks
Test workflow value before making a commercial claim.
Can 3–6 analysts identify and explain the most important disclosure change faster without reducing evidence quality?
Pilot design
A bounded side-by-side workflow comparison
The analyst remains responsible for the final conclusion.
Review 8-K and 10-K disclosure changes
Use a bounded issuer and filing set with known source availability.
Normal vs assisted workflow
Compare current process with Pure News Intelligence-assisted review.
2–4 weeks
Long enough to observe repeat use, corrections, and research-record value.
Example pilot scorecard
Proposed measurement—not actual performance
Baseline and assisted values remain blank until a controlled pilot occurs.
| Measure | Role | Normal workflow | Assisted workflow | Evidence captured |
|---|---|---|---|---|
| Time to identify and explain the most important change | Primary | To measure | To measure | Task timestamps + analyst explanation |
| Material-event recall | Secondary | To measure | To measure | Reference-set comparison |
| Category and materiality agreement | Secondary | To measure | To measure | Analyst labels vs system labels |
| Excerpt correction rate | Guardrail | Not applicable | To measure | Corrections per reviewed record |
| Novelty-label acceptance | Trust | Not applicable | To measure | Accept / revise + reason |
| Repeat use and exports | Adoption signal | To measure | To measure | Sessions + evidence packets |
| Stored-context usefulness and preference | Qualitative | To measure | To measure | Post-task interview |
Predefined decision rules
Go, revise, or stop
Threshold values should be finalized before the pilot begins; these are directional criteria.
- Meaningful time savings
- No reduction in evidence quality or recall
- Strong repeat use and analyst trust
- Time savings but weak label acceptance
- Useful comparisons but excessive correction burden
- Value concentrated in only part of the workflow
- No meaningful time savings
- Lower material-event recall
- Low trust or workflow friction exceeds benefit
Help test the next version
Choose the smallest useful contribution.
No personal information is collected on this site.
Proposed idea-engine evaluation
Measure research-agenda quality before market performance
No results are reported here.
Workflow outcomes
- Analyst acceptance rate
- Accept-with-edit rate
- Rejection rate
- Time to create a research agenda
- Edit burden
- Analyst preference
Grounding quality
- Grounding score
- Novelty accuracy
- Mechanism clarity
- Falsifiability
- Usefulness and non-obviousness
- Affected-entity accuracy
- Safe-rejection quality
Comparison groups
- Disclosure change only
- Generic AI summary
- Grounded AI hypothesis
- Grounded hypothesis plus skeptic
- Analyst-edited hypothesis
- Human-created research agenda
Sequencing rule: Only after hypotheses are frozen prospectively should a separate study evaluate IC, Sharpe ratio, hit rate, signal decay, or realized-versus-implied volatility.