POLICY, NOT DATA. Two constants on this board are stated priors, solved to a target Erik named — they are not fitted and no holdout can set them. The TE multiplier (L4 · QB 1.000 · RB 1.233 · WR 1.000 · TE 0.596, re-solved on this 444-row distribution) encodes a superflex TE-premium league format and is solved to place the TE1 at the backside of the top ten. The curved freshman discount (D=0.405, γ=1.5) is solved to Erik's stated target — FR1 in early-second startup territory, FR2 inside startup relevance. Never present either as backtested or fitted. The QB convex lever (T=87, premium 0.30) is policy on the same footing.
Gate: +0.0362 ALL, worst-of-two-splits (n=3,306), on the shipped configuration. A bootstrap of 1,000 player-clustered draws put zero draws below the +0.005 bar, and the zero-tail threshold moved the gate by 0.0008 across its whole sweep. QB does NOT clear at the position level (19.9% of draws negative) — the QB half of this board is its weakest evidence. Predicted P(day 2+) is an OUTPUT of the model, never an input.
NO FRESHMAN PLACEMENT ON THIS BOARD IS BACKTESTED. The holdout contains historical pool player-seasons only, so for 85 of these 444 players the eye test is not a supplement to the gate — it is the only check that exists. Incoming freshmen are marked FR and have no college production yet: best PPG, college PPG and production score are shown as —, which means “not played” and never zero.
Tiers carry a backtested hit rate instead of a range. The hit rate is the share of historical-pool players in that percentile band who reached 5+ PPG in the NFL, against a base rate of 4.5%. It is a POOL statistic about the band, not a probability attached to this player — and it is not evidence about any freshman. Nothing on this board publishes a range, an interval or a floor/ceiling. ⚠️ The rates were measured on the holdout with n = 13 / 37 / 100 / 226 / 376 / 623 / 1,122 players per tier, top to bottom, and they have not been re-measured on v5.0's A'-reweighted comps — the percentile bands are the same, the comp construction underneath them is not.
Known limitation — the Elite Producer tier. 9 of its 13 members are incoming freshmen who have never played a college snap. Freshmen carry no own production, so the λ=0.30 own-production cap that trims produced vets never reaches them, which is how unplayed players reach a produced-vet tier. The tier name is descriptive; read it as a percentile band, not as a claim that a given freshman has produced. The real repair (a freshman own-production proxy) is NOT done.
Position reliability differs and this board says so. Share of all 10+ PPG producers our top decile actually catches: WR 72.4% · RB 69.0% · QB 43.7% · TE 41.2%. A 92-rated tight end and a 92-rated receiver are not the same bet. At QB, lead with P(day 2+) rather than the rating — round 1-2 QBs hit 5+ PPG 91.8% of the time and undrafted QBs 0.6%.
Alpha Rating is the currency: expected NFL outcome from 25 historical comparables, expressed against the top-10 blend of the best scores ever recorded at that position, then adjusted for league format. A ratio on a relative scale, not a points projection. Raw comp E[PPG] is the same expectation BEFORE the own-production cap and BEFORE format adjustment — it is the basis the tier is cut on, which is why tier and rating can disagree. For a freshman it is a pedigree prior with a buzz overlay, not a comp mean, and it is labelled as such per row. College PPG and NFL expectation are different scales by design — never read the gap between them as a regression.
Buzz is 0-5 and is never rescaled (5 = shocking / unexpected praise · 2.5 = about what you'd expect · 0 = silence when there shouldn't be). It is a labelled prior with no holdout — no historical buzz exists to backtest against. Only the 85 incoming freshmen carry a buzz overlay on this board.
Flag, don't rescue. SMALL SAMPLE keys on CAREER games (<10), not final-season games. OUT FOR SEASON forces a move; INJURY WATCH does not. PEDIGREE means the rank IS the recruiting profile and the player is not post-ready regardless of where he sits.
What v4.7 showed and this board does not. The named production comps are back, now recomputed on v5.0's A'-reweighted basis: each vet's list is his 25 nearest historical player-seasons in the best-season-weighted comp distance, the same neighbours whose mean reproduces his Raw comp E[PPG]. Freshmen and vets with no qualifying college season carry no production comps — there is no production to comp against. What remains absent is the v4.7 “where his 25 comps landed” distribution bar (the c_p10..c_max spread); that histogram describes the v4.6 comp construction and has not been re-derived on the v5.0 basis, so it is not shown beside a v5.0 rank.
Alpha Universe
v5.0 • SYSTEM ONLINE