Every term you'll see on the desk, in one sentence each. No finance degree required.
P(up)
The brain's estimated probability a name rises over the next ~5 trading days. 50% is a coin flip; 65% is a meaningful lean up; 35% is a lean down.
Conviction
How far P(up) sits from 50% — i.e. how strong the lean is. A conviction of 0.12 means P(up) is 62% (up) or 38% (down). Higher conviction = the brain is more sure.
Horizon
The holding window the brain predicts over. Here it's 5 trading days (about a week) — a swing-trade horizon, not day-trading.
Walk-forward accuracy
How often the brain was right when tested only on the future it had never seen during training. This is the honest test — far harder to fake than testing on data it already studied.
Drift vs Skill
Stocks tend to drift up over time, so part of being "right" is just riding that drift (that's beta). Skill is accuracy above that baseline — the only part that's a real edge (alpha). The desk separates the two so a drift-powered number can't masquerade as skill.
Base rate
How often a name rises over the horizon regardless of any prediction — the bar any real edge has to clear.
Brier skill (BSS)
A score for how well-calibrated the probabilities are versus random guessing. Positive = better than random; 0 = no better than a coin.
Calibration (Platt)
A correction applied so that when the brain says "70%," it really happens about 70% of the time — not 55% or 85%.
Confluence
When two independent signals — the brain and insider buying — agree on direction. The bet is that agreement is stronger than either signal alone.
Champion / Challenger
The brain always trains a rival model in the background. The challenger only takes over if it genuinely beats the current champion on unseen data — so the desk never quietly downgrades itself.