CORTEX
Measurement — projects the result from the data
Runs without sight of ORACLE's output. Agreement between the two is evidence only because it was not coordinated.
CORTEX is the measurement layer. It evaluates historical data, efficiency metrics, projections, probability gaps and expected value, and reports what the numbers support — not what the matchup sounds like.
Inputs
- MLB: starter FIP/xFIP, park factors, weather, bullpen workload, handedness splits
- WNBA: pace, rest days, home/away splits, four-factor efficiency
- Liga MX: standings strength, recent form (W/D/L + GF/GA), fixture density, group context
Signature move
Luck-corrected efficiency in basketball, FIP over ERA in baseball — the underlying numbers over the surface result.
Constraint
Every claim must trace to a figure in its inputs. No external ratings, no personnel detail it was not given.
Measurement starts the process. It does not finish it.
ORACLE
Context — prices what the box score misses
Runs without sight of CORTEX's output. A situational read that merely echoed the model would add nothing.
ORACLE is the context layer. It prices injuries, rotation changes, travel, rest, congestion, coaching matchups and market movement — the situational factors a pure model underprices.
Inputs
- Injury reports, rotation changes, bullpen fatigue
- Schedule context: rest, travel, getaway days, lookahead spots
- Coaching matchups, motivation, rivalry dynamics
- Market movement: line movement, sharp vs public splits
Signature move
Letdown and trap identification, getaway-day flagging — the spots a model prices last.
Constraint
A read with no number attached is a pass, not a lean. Negative-edge angles are never surfaced as plays.
Context is allowed to contradict the model. That is the point of running them apart.
ARBITER
Decision — rules, sizes, and caps
Confidence is arithmetic: the price's implied probability plus a capped, named edge. Never a free-form number.
ARBITER is the decision engine. It weighs the measured edge against the situational read and the posted price, assigns a confidence figure, sizes the stake within hard caps, and decides whether the play publishes at all.
Inputs
- CORTEX's ranked plays with named edges
- ORACLE's independent situational reads
- Original game slate, lines, market data
Signature move
The PASS. When the two reads disagree on a side, no play is forced — the disagreement is the output.
Constraint
3u maximum on any play, 10u maximum per day, one Pick of the Day across every league — or none.
The most important output is the PASS. No play is forced when the edge is not there.
At a glance
Three questions. One verdict.
| Model | Persona | Primary question | Output |
|---|
| CORTEX | Measurement | What do the numbers support? | Probability, named edge, expected value |
| ORACLE | Context | What does the situation change? | Independent situational read |
| ARBITER | Decision | Is the price wrong enough to stake? | Ruling, confidence figure, capped stake |
| NEMESIS | Dissent · in development | What does this number know that we do not? | A bear case on every staked play |