Graded in public
Every prediction Scout makes is logged before the outcome is known, then graded against what actually happened. Nothing gets deleted, including our misses.
We grade ourselves in public.
Most recruiting tools tell you what to do and never check whether they were right. Prospecta logs every Scout prediction, grades it against the outcome, and shows the tally here. If we're wrong, you'll see it.
First cohort
Ledger opens with our first cohort. Every prediction is logged before the outcome is known — including our misses. Public counts unlock once the ledger crosses our threshold (100 predictions logged).
How this will fill in
Predictions logged
Immediately — every Scout brief is parsed into discrete, gradable claims and timestamped before the outcome is known.
Graded
As outcomes land — a nightly job compares predictions against verified replies and commits. Hits and misses both count.
Verified commits
As athletes upload NLIs or signing announcements, each commit gets a verified badge and the school name unlocks.
How we grade Scout
Prediction ledger
Every weekly brief Scout writes is parsed into discrete, gradable claims (reply likelihood, coach fit, roster gap, offer timing). Each one is timestamped before any outcome is known.
Outcome grading
A nightly job matches predictions against verified commitments and replies. Hits and misses are both counted — no quietly burying the misses. Athletes can also mark predictions correct or missed manually.
Drift detector
Reply-likelihood scores are snapshotted weekly. When the score moves outside the expected band — coaching change, timing window, etc. — you get a drift alert with the before/after.
See where Prospecta families actually landed.
The outcomes feed is fully anonymized — schools, divisions, states only. No athlete names, no scores, no marketing varnish.
Methodology: Scout writes weekly briefs that include numeric and categorical claims. Each claim is stored as a row before the outcome is known. A nightly job compares stored claims against verified outcomes (commit reports, dated reply receipts, NLI screenshots) and updates hit/miss counts. Reply-likelihood scores are recomputed weekly from deterministic signals (sport, staff size, contact timing, recent activity) and snapshotted so drift is detectable. No predictions are ever deleted to improve the score. This is calibration scoring in the Metaculus tradition — we're checking whether predictions we label "high confidence" actually come true more often than the ones we label "moderate" or "low," not just tracking a raw hit rate.