Inside the glass box
See the reasoning, not just the number.
Most products hand you a pick and ask for trust. We show the machinery: thousands of independent reads, converging on a signal — and every one graded and recorded so it can be checked.
One system, many rooms
The intelligence system.
Every surface below is the same engine, read through a different lens. Step into any room — the tab bar carries you between them.
Many reads, one signal
Edge lives in the disagreement.
Independent referees each price the game on their own. When they converge, that's confidence. When they agree the market is wrong in the same direction — that's the signal worth surfacing, with the reasoning attached.
From many independent reads to one auditable signal.
How the engine works
Independent referees
Several estimates score each game without looking at the sportsbook's price — a sharp exchange, a structured model, the wider market. Different lenses, judged on their own.
A signal is a case, not a badge
Every signal is prosecuted before it's published.
No lonely confidence number. The engine argues the case against itself — evidence, counter-evidence, and the falsifier that would break it — then returns a verdict, including the honest verdict of no-bet.
Illustrative matchup · Home −3.5
IllustrativeHome −3.5 carries a measurable edge over the closing price.
Confidence · Moderate
Real edge, but a falsifier is still in play.
Prosecution
Why the edge exists
Two independent estimators land short of the market line and agree on direction.
Consensus engineRest and travel sit in the home side's favour this window.
Schedule modelThe number drifted toward the thesis without a public surge behind it.
Line-movement read
Defense
Why it might be wrong
Public exposure on the favourite is heavy — some of the value may already be priced.
Public-pressure readA single questionable injury status sits upstream of the edge.
Roster-shock agentClosing-line history on this matchup type is thin.
Calibration check
Judge
What would flip it
- The questionable status is downgraded to OUT before lock.
- The number moves past −4.5 on public action, not new information.
- Book consensus breaks against the thesis in the final hour.
Confirm the injury status and the edge upgrades to PLAY; downgrade it and the case collapses to NO-BET.
Inputs current · re-checked 11s ago
Not one AI — a council
Watch the verdict change — and see which agent moved it.
Behind every read is a council of specialist agents — each with one job and an escalation threshold. When the recommendation changes, you can trace exactly which agent escalated and why. No opaque black box.
The council
IllustrativeIllustrative read · Home −3.5
Council opens. Independent estimators diverge from the price and agree on direction. The council opens the case at PLAY.
Line Movement
How the price moves vs. information
Sharp Pressure
Money moving against the public
Public Bias
Where the crowd is piling in
Injury Freshness
Roster status and how stale it is
Matchup
The structural edge in the game
Model Disagreement
When independent estimators split
Narrative Signal
The story around the athletes
Responsible Decision
Guards against over-exposure
After the whistle
We grade the thinking, not the scoreboard.
A win is not proof and a loss is not failure. Every settled signal is graded on the square it actually lands in — so a lucky win gets flagged and a correct read that lost gets respected.
Good outcome
Bad outcome
Good process
Earned
Right read, right result. The standard — not the highlight reel.
Respected
Right read, wrong bounce. We keep the process; variance owes us nothing.
Bad process
Lucky
Wrong read, right result. The most dangerous square — we flag it, never frame it.
Corrected
Wrong read, wrong result. Logged, learned, fed back into calibration.
Verdicts most sites won't publish
Transparency is the product
We detect. You decide.
Numbers about our track record stay hidden until there's enough settled, calibrated history to publish them honestly. Until then, this is the part we can show you in full: the method.