From Raw Signal to Conviction Score: The Full Calculation

Most trading signals are a single number wearing a costume. Somewhere behind the "78% bullish" headline is one indicator, maybe two, dressed up to look like a system. The percentage feels precise. The math behind it usually isn't.

A conviction score works differently, and the difference is the entire point. It is not a confidence level pulled from one chart pattern. It is the output of five distinct measurements, weighted by how much each one actually predicts outcomes, run through a filter that can override everything else, and capped by a discipline that most frameworks don't have: the willingness to produce no score at all.

InDecision runs on this calculation for every call. The framework has posted 82.5% directional accuracy across its history, but that number is a average of averages — it masks a much more interesting structure underneath. Understanding how the raw signal becomes the final score explains why that structure exists, and why the number itself is only half the story.

The Five Inputs Before the Override

Every conviction score starts as a weighted sum of five factors, each scored independently on its own evidence.

Daily Pattern Analysis carries 30% of the weight — the largest single input. It measures whether current price action matches historical setups that resolved a specific direction with statistical consistency, not whether the chart "looks like" something a trader has seen before.

Volume Analysis carries 25%. This is not "volume is up." It's a specific comparison against a rolling baseline, with 4.2x average volume acting as the threshold where a move stops being noise and starts being a signal worth weighting heavily. Below that multiple, volume contributes little regardless of price action.

Timeframe Alignment carries 20%. A setup that agrees across the 4-hour, daily, and weekly chart scores high here. A setup that only exists on one timeframe — the classic trap of a trader anchored to a 15-minute chart — scores low, even if it looks compelling in isolation.

Technical Confluence carries 15%. This is where multiple independent technical signals — not variations of the same signal — either stack in the same direction or don't. Three momentum indicators agreeing is not confluence. A momentum signal, a structure signal, and a volume signal agreeing is.

Market Timing carries the remaining 10%. This factor accounts for the 8-hour funding reset cycle and other cyclical mechanics that shift the probability of follow-through independent of the chart itself. It's the smallest weight because timing amplifies or dampens a setup — it rarely creates one from nothing.

Multiply each factor's independent score by its weight, sum the five products, and the result is a raw conviction number between 0 and 100. This is the number before the framework asks the only question that matters more than the math: is it safe to act on it?

Risk Context Is Not a Sixth Factor — It's a Ceiling

Every InDecision article eventually has to correct a natural assumption: that Risk Context is the sixth factor in a six-factor system, sitting alongside the other five with its own percentage weight. It isn't, and treating it that way is the fastest way to misread the framework.

Risk Context is an override layer, not an input. The five weighted factors can produce a raw score of 85 — comfortably in high-conviction territory — and Risk Context can still cap the usable score below the threshold where a signal gets published at all. Thin liquidity, an imminent macro catalyst, extreme funding skew, or a market structure the framework hasn't seen enough of historically can all trigger this cap.

The reason this matters mechanically: a weighted average of five good inputs can still describe a bad trade. Averaging doesn't know about tail risk. A setup can score well on pattern, volume, and timeframe alignment while sitting directly in front of a liquidity air pocket that turns a clean signal into a disaster the moment size gets applied. Risk Context exists specifically to catch what the weighted sum cannot.

This is also why two setups with identical raw scores can produce different published conviction levels. The calculation is deterministic up to the five factors. The override is conditional on context the five factors don't measure. Both matter. Only one of them is a percentage in the weighting table.

Why the Bands Matter More Than the Score

A single number — "conviction: 74" — invites a false precision. Is 74 meaningfully different from 71? In most cases, no. This is why InDecision doesn't act on the raw score directly. It acts on the band the score falls into, and the bands carry very different performance profiles.

High conviction, 80% or above, has hit 91.2% directional accuracy historically. Medium conviction, the 60-79% range, has hit 78.4%. Below 60%, the framework does not publish a directional call at all. It abstains.

That gap between 91.2% and 78.4% is not decoration — it's the entire justification for banding instead of scoring. A framework that treated a 61 and a 79 as roughly equivalent "medium confidence" signals would be right about as often as a coin flip weighted slightly in its favor. A framework that treats 80 as a hard line, below which position sizing and conviction language both change, is making a falsifiable claim about where the edge actually lives. The 91.2% figure is that claim tested against outcomes.

The abstain threshold is the least glamorous part of the system and the most important. Below 60% raw conviction, after the risk override is applied, InDecision does not shade toward a soft lean or a "slightly bullish" hedge. It says nothing. This is a deliberate design constraint: a system that always has an opinion is a system whose opinions carry less information. Removing the noisy middle is what keeps the high and medium bands statistically honest.

What This Means When You're Actually Positioned

The practical consequence of this calculation shows up in position sizing before it shows up in anything else. A high-conviction call, north of 80 after the override, is a different trade than a medium-conviction call in the low 60s — not just in how confident the language sounds, but in how much capital the setup justifies. Treating a 65 and an 87 the same because both are "the framework is bullish" throws away the entire point of running the weighted calculation in the first place.

It also reframes what a losing trade means. A high-conviction signal that fails 8.8% of the time (the inverse of 91.2%) is not evidence the framework is broken — it's the expected failure rate of a system operating exactly as designed. The five-factor weighting, the risk override, and the abstain discipline together produce a specific, testable error rate per band. A loss inside that expected rate is the system working. A loss pattern that consistently exceeds it is the signal to re-examine the weights, not the outcome.

The raw signal was never the story. Six months of price data and a volume spike can be found on any chart, for any asset, on any given day. The calculation — five weighted factors, one override layer that can cap everything, and a threshold below which the system says nothing — is what turns that raw signal into a number worth acting on, and just as often, into a number worth ignoring entirely.

Weekly InDecision signals include the full five-factor breakdown and conviction band for every call. Subscribe to see exactly how the framework reads the market each week.