Scoring settings

Every threshold the code uses is visible and editable. Models never see or change these.

Composite weights

Per deal type; renormalised over the metrics Qloo could measure.

Deal type	m1	m2	m3	m4	m5	m6
M&A bolt-on
Licensing / co-brand
VC thesis

Red-flag rules

AUDIENCE_ONE_SIDED when min(a,b)/max(a,b) <TASTE_CLASH when buyer top-20 tags missing >... and m2 <DEMOGRAPHIC_MISMATCH when m4 <FADING when momentum z <NO_GEO_OVERLAP (licensing) when m5 <

Verdict bands

ADVANCE when F ≥ (and no red flag)WATCH when F ≥No score below weighted coverage

Settings are read-only on this public deployment.