Skip to main content

Conviction Scaling

Even when opposing signals are detected and a direction winner is chosen, the mere existence of a dissenting model is a reason to trade smaller. Conviction scaling is the mechanism that translates the degree of confidence advantage into a position sizing multiplier.

The Core Idea​

Imagine two scenarios:

  • BUY at confidence 0.85 vs. SELL at confidence 0.30 — a gap of 0.55. Strong conviction on the long side with a weak dissenter.
  • BUY at confidence 0.55 vs. SELL at confidence 0.50 — a gap of 0.05. Barely any edge in either direction.

Both scenarios have a "winning" direction, but they represent very different levels of certainty. Sizing both trades the same would be irrational. Conviction scaling ensures the second scenario produces a dramatically smaller position (or none at all) than the first.

The Sizing Ramp​

Conviction scaling maps the normalized confidence gap to a sizing multiplier using a continuous ramp function. The ramp has two key parameters: a floor multiplier (minimum size when conviction is weakest) and a full-conviction threshold (the gap level at which full standard sizing is reached).

  • At minimum gap (near-tie): sizing is at the floor multiplier — the trade is either very small or not taken.
  • At or above the full-conviction threshold: sizing reaches 1.0 — full standard size.
  • In between: sizing scales proportionally with the gap.

The specific formula, floor values, and full-conviction thresholds are proprietary and calibrated per profile.

Profile-Tuned Parameters​

The floor multiplier and full-conviction threshold are tuned to each strategy profile:

  • Conservative — highest floor (even minimum-conviction signals trade at a meaningful fraction of standard size); largest gap required before reaching full size.
  • Balanced — intermediate floor and threshold.
  • Aggressive — lowest floor (minimum-conviction signals may be near-zero); smallest gap required before reaching full size.

Specific parameter values are proprietary and not disclosed.

Interaction With the Composite Score Gate​

Conviction scaling produces a sizing multiplier that feeds into the composite score gate's tier evaluation. A high-quality signal that passes the confluence gate but was generated alongside a conflicting signal will be tier-assessed on its composite score — and then have the conviction scalar applied on top.

This means a borderline composite score combined with low conviction can produce a very small final position size, even if the signal nominally qualified to trade. This layering of sizing adjustments is intentional: each dimension of uncertainty compounds the sizing reduction rather than any single gate being a binary pass/fail.

What This Looks Like in Practice​

A trader reviewing the audit trail for a given trade will see the conviction gap and sizing multiplier recorded alongside the composite score and tier. If a position was notably small, the conviction scalar is one of the first places to look — it is one of the most common reasons a technically valid signal trades smaller than the profile's standard size.

Conviction scaling is never a surprise: the lineage record captures it explicitly, and the dashboard surfaces the key sizing factors in the trade detail view.