Elephant Hawk — Opportunity Seeker
A funnel, not an oracle. SEEKER generates opportunity candidates from real public signals (openFDA, ClinicalTrials.gov, PubMed, WHO burden), triages most away cheaply, runs the full SCORER engine on survivors, then ranks by composite and the independent need-supply gap so genuine white space isn't lost to rubric-gaming. Output = 3–4 theses to test with named experts — humans still make the call.
Stage 1 — Evidence-first candidates (decoupled from scoring)
Real signals per domain → AI clusters & names candidates. Every candidate carries provenance; none is allowed without a signal it ties to.
Stage 2 — Coarse triage (kill most, cheaply)
Four decisive screens: capability/focus, regulatory quick-read, white-space ratio, payer plausibility. We do not run all 14 layers on everything. Dropped candidates are shown — nothing is silently truncated.
Stage 3 — Full SCORER on survivors
The complete 14-layer engine (same gates → floor → bands as SCORER) runs on each survivor.
Stage 4 — Rank + heatmap (anti-Goodhart)
Ranked by composite (TOPSIS) and the independent need-supply gap. Favour candidates where the two agree; genuine high-need / low-supply white space is surfaced even when the rubric isn't maxed; rubric-gaming is flagged.
Stage 5 — Theses to test (the actual output)
For the top hotspots: a thesis to test, the key uncertainty, and which experts to talk to. Feeds the existing experiment-and-test loop. Export each as a standalone SCORER report.