The box score is the last place a breakout shows up.
We surface players whose underlying skill has pulled ahead of — or fallen behind — their box-score results, and we say where that points in a stat you already use. For a starter: “his skill points to an ERA around 3.5, while his results sit at 4.0 — the skill is ahead.” For a hitter: “his skill points to an OPS around .840, ahead of his current .760.” (Relievers get a directional read, not a number — their signal is real but too noisy for a precise figure, and we won’t fake one.)
Over the coming weeks. Across four seasons of held-out history, players showing this pattern tended to see their results move toward their skill over roughly the next month. We never put a date on one player — we give you the direction and the window the pattern has actually played out over.
Advanced machine-learning and statistical modeling, validated on history the models never saw. The rigor is the point; the recipe stays in the vault — we publish what we found, never how we find it.
From the track record, never a promise. Across 3,278 held-out player-windows over four seasons, the players our signal flagged outperformed the rest by a meaningful margin (a +0.039 wOBA edge, p≈0). And every published call is graded in the open as games finish — you watch the record fill in real time. What we will never give you is a per-player percentage; the honest numbers are the aggregate edge and the public ledger.
Every swing carries information the box score hasn’t absorbed yet. Prophet 9 measures a player’s underlying performance signals over the windows where they become reliable, compares them with his surface line, and flags the gap. When the underlying leads, improvement follows. We proved that relationship on four seasons of held-out history before we published a single call — and our track record stays public so you can keep auditing it.
The picture to hold in your head is below — drawn from today’s live lead signal, not a sketch. A player’s underlying signals — the quality of what he’s actually doing on the field — move first. His surface line follows, late, because surface stats need sample to catch up. The shaded area between the two readings is what we measure, every night, for every player.
Before publishing a single call, we validated the signal on 3,278 held-out player-windows across four seasons. Players our signal flagged went on to outperform by +.039 wOBA versus the unflagged — a result with p ≈ 0. The methodology is ours; the result is public.
Reads the most recent form aggressively. Quickest to believe a genuine surge — and the first to be fooled by a hot streak, which is why it never publishes alone.
The blended view. Where the two readers agree, it speaks plainly; where they split, the spread is reported to subscribers as information in its own right.
Shrinks toward the player’s established level. Slowest to chase noise — and the last to recognize a real transformation, which is why it never publishes alone either.
Which signals we read, over which windows, with which thresholds. That’s the edge, and publishing it would erase it for everyone who pays for it. What we publish instead is everything you need to judge us: the validation, the intervals, and a permanent public record of every call.
Trust the audit, not the adjectives.
What we do tell you is whether his skill is leading or lagging his results — and where his stats are heading because of it. You won’t see which inputs we read, over which windows, or where the lines are drawn — that’s the part that stays in the vault.
Two readings of every player: what the inputs say, and what the box score says. The gap between them is the signal.
3,278 held-out player-windows, four seasons, p ≈ 0. The proof came before the product.
Every projection ships with an honest interval, and every call lands on a public ledger — hits and misses alike.