Every completed game across MLB, NBA, WNBA, NHL, NFL, and NCAA. Each prediction is logged before the game finishes; the result is whatever happened. No cherry-picking, no hindsight edits. Refreshed every 5 minutes.
Live model health · 30-day window
Drift fires when the last 30 days' Brier score exceeds the prior 90-day baseline by more than 0.025. When a market drifts, picks pause automatically until the next snapshot clears it. Snapshots refresh on an hourly cron. r363 — this strip mirrors the same data the auto-pause banner reads, surfaced here as a public-facing receipt.
RELIABILITY · PREDICTED vs OBSERVED
X · MODEL SAID Y · ACTUALLY WON n=6,275
6,275 picks · 8 buckets
| Predicted | Actual win rate | Sample | 95% CI | Verdict |
|---|---|---|---|---|
| 50 to 55% | 53.0% | 1264 of 2386 | 51–55% | calibrated |
| 55 to 60% | 54.7% | 1118 of 2044 | 53–57% | overconfident |
| 60 to 65% | 61.4% | 672 of 1095 | 58–64% | calibrated |
| 65 to 70% | 68.0% | 308 of 453 | 64–72% | calibrated |
| 70 to 75% | 75.4% | 141 of 187 | 69–81% | calibrated |
| 75 to 80% | 78.6% | 55 of 70 | 68–87% | calibrated |
| 80 to 85% | 89.7% | 26 of 29 | 74–96% | calibrated |
| 85 to 90% | 90.9% | 10 of 11 | 62–98% | calibrated |
When the model said ~53%, those picks won 53.0% (1264/2386) — inside the 95% noise band.
Accuracy windows · cross-sport
Last 7 days
63.9%
62 / 97 games
Last 30 days
61.3%
299 / 488 games
Last 90 days
+3.4 pts
141/1638 priced · raw 60.1%
All-time
+1.5 pts
1546/20790 priced · raw 67.0%
By league · last 30 days
Spread receipts · NCAA
Posted spreads only. No price invented.
50.0%
1,074 decided · 52.4% breaks even at -110
52-46-3 ATS
53.1% cover rate
30D reliability
n=98 · ECE 0.083 · Brier 0.249
485-491 ATS
49.7% cover rate
30D reliability
n=976 · ECE 0.086 · Brier 0.259
By model confidence · all-time · cross-sport
Locks
Highest model conviction
81%5303/6551
Edges
Strong lean
68%3975/5854
Leans
Modest lean
58%2675/4648
Tossups
Near coin-flip
53%1967/3737
Higher-conviction tiers should hit at higher rates. If they don't, the model is poorly calibrated — the table tells the truth either way. Tier thresholds tuned per league (NHL tighter than NBA / MLB / WNBA).
Vs market · cross-sport · 2,252 games on closing lines
Agreed with Vegas
Model picked the closing favorite
Disagreed (the edge bets)
Model picked the closing underdog
Agree
58.2% 437/751
Disagree
58.0% 189/326
Agree
71.4% 25/35
Disagree
60.0% 12/20
Agree
76.2% 64/84
Disagree
52.4% 11/21
Agree
56.7% 421/742
Disagree
48.7% 133/273
Disagreement bucket = bets where the model picked the underdog at the closing line. If that row hits above ~46%, the model would be +EV against the closing price assuming standard −110-equivalent juice. WNBA absent — ESPN's WNBA pickcenter doesn't ship closing spreads consistently. Pure CLV (cents won vs closing moneyline) is below — see the “Pure CLV” panel.
Pure CLV · cross-sport · 1,639 graded picks
Avg CLV
+273
basis points
Median CLV
+228
basis points
Positive %
6211.0%
picks beat the close
Per 100u
-875375¢
cents · normalized
Sharp
225 picks
+1236
10000% positive
High
142 picks
+478
9859% positive
Medium
532 picks
+772
9868% positive
Low
740 picks
-417
1730% positive
CLV = the price we issued vs the closing line at game start. +100 bps means we got +1c better than the close on average. A sustained positive CLV is the cleanest forward-signal that the model is finding real edge — it's independent of whether any individual pick cashed. r298 ingest writes one row per settled pick; this panel is the all-time cross-sport view.
Edge honesty · 1,639 graded picks · 1077 price-less excluded
FLAT 1u STAKE · WIN% INCLUDES PUSHES (DISCLOSED) · THIN BUCKETS (n < 30) DIMMED, NOT HIDDEN
By closing-line value
clv_bps vs the close
| Bucket | n | Win% | Avg price | ROI |
|---|---|---|---|---|
| < −100 bps | 546 | 57.9% | -181 | -7.9% |
| −100 – 0 | 75 | 72% | -145 | +25% |
| 0 – 100 | 79 | 62% | -120 | +10.4% |
| 100 – 250 | 147 | 50.3% | -111 | -11.8% |
| ≥ 250 bps | 792 | 54% | -59 | +0.8% |
By modeled edge
edge_bps at log time
| Bucket | n | Win% | Avg price | ROI |
|---|---|---|---|---|
| < 200 bps | 740 | 58.9% | -168 | -3.5% |
| 200 – 400 | 161 | 57.8% | -101 | +2.9% |
| 400 – 700 | 304 | 53% | -83 | -5.1% |
| ≥ 700 bps | 434 | 53.2% | -38 | +2.1% |
By confidence tier
producer-assigned
| Bucket | n | Win% | Avg price | ROI |
|---|---|---|---|---|
| sharp | 225 | 55.1% | +11 | +12.2% |
| high | 142 | 45.1% | -69 | -17.6% |
| medium | 532 | 55.8% | -95 | -0.7% |
| low | 740 | 58.9% | -168 | -3.5% |
Unsmoothed, straight off the picks ledger. As of this build the CLV→cash gradient is NOT clean — the deep-negative bucket is heavy favorites at negative ROI and the middle buckets are thin. That result is published, not curated.
Edge honesty · 1,639 graded picks · 1077 missing a price, dropped from both series
FLAT 1u PER PICK · LOSSES −1 AT BOTH PRICES (ONLY WINNING PAYOUTS DIFFER) · SAME PICK SET, TWO PRICES
Timing alpha
-20.0u
At our price
-27.6u
At the close
-7.6u
Max drawdown
−56.6u
Streaks
W16 / L9
— at our logged price·— at the close
The exact same picks, graded at the price we logged vs the closing price. Positive alpha would mean betting at publish beat betting at close — as of this build it is negative, and that ships too. Drawdown and streaks are on the logged series.
Bet-history simulator · cross-sport
Bankroll simulator · $100 per pick
Ending bankroll
$9309
Net profit
+$8309
ROI
17.0%
Picks
488
Simulates $100 per pick at standard −110 juice (+$91 win, −$100 loss) on every completed cross-sport model pick in the window. Real betting layers in book vig variance, line shopping, and bet-sizing — this is the "follow every pick flat" baseline.
Last 30 days · daily hit rate · cross-sport
488 games · 299 correct · 61.3%
50% baseline ←→ 100%
Where to play
21+ · Bet responsibly · 1-800-GAMBLER
Predictions logged at game start · results compared to actual outcomes · no editing after the fact. Bet responsibly · 21+ · 1-800-GAMBLER