When a programme publishes a clear pattern of past winners, we score your profile against that pattern and show the evidence — which marks you hold, which you lack. When a listing is scraped without that pattern, we say fit is unknown and order by relevance to your goals, deadlines, and requirements already on file. Coach language never overrides those arithmetic results. You always choose what to apply to; Fill assists and never auto-submits.
Are you a measured fit?
Published-rules engine + winner cohort stats → coverage band
Coverage over discriminating (not near-universal prerequisite) cohort signals:
- likely ≥ 0.85
- match ≥ 0.60
- reach ≥ 0.35
- else far_reach
Missing a prerequisite (≥ 0.90 prevalence among winners) blocks banding — you are outside the counted pattern, not a soft “reach.”
Too few discriminating stats → unknown (honest: winners too alike / evidence thin).
Is this text relevant to what you wrote?
Keyword / catalog shortlist + optional suggested score
Relevance is labelled as relevance, not eligibility.
- Only open + competitive + fit unknown may get a suggested score.
- If off or fail, local total order: soonest deadline → lower/absent fee → more already-met published requirements → title.
- Models cannot invent, drop, or duplicate rows.
