V1 was designed to prove the method. The current run tests whether the same evidence-engineering approach can operate against a research universe two orders of magnitude larger, using triage gates to focus local LLM compute where it adds the most value.
Research triage is deliberately separated from deep extraction so compute can be concentrated on the highest-priority evidence.
This page is a snapshot, not a fake “live” feed. The website is intentionally separated from the Ollama pipeline so the research run can continue independently and the published dashboard can be updated safely in batches.
| Stage | Count | Status |
|---|---|---|
| Indexed papers | 94,826 | complete snapshot |
| Triaged records | 94,591 | complete snapshot |
| Broad Ollama candidates | 70,198 | candidate pool |
| Strict deep candidates | 32,437 | Tier A + B |
| Deep extracted | 50 | first batch |
Acquire a broad literature corpus and preserve source metadata and identifiers.
Identify compound-level and class-level signals before spending deep-model compute.
Rank research into priority tiers, preserving lower tiers as searchable evidence rather than deleting them.
Run high-priority papers through Ollama to create structured fields, evidence scores, confidence indicators and later comparative analytics.