AI Confidence

How much evidence a read rests on, shown as High, Medium or Low. Short or vague entries, or few trades, give Low. Treat Low as a hint, not a finding. The word is set by how much of your own data a read rests on, not by how fluent the AI sounds.

How it's calculated#

Limbix doesn't ask the model how sure it feels. It counts the evidence. Two surfaces show the word, each counting its own evidence:

$$\text{Confidence} = \begin{cases} \text{High} & n \ge n_{\text{high}} \\ \text{Medium} & n \ge n_{\text{medium}} \\ \text{Low} & \text{otherwise} \end{cases}$$

n

The evidence behind the read: analyzed trades for the Behavioral Profile, paired sessions for a sentiment comparison.

n_high, n_medium

The cut-offs for that surface, in the table below.

On the profile, the ? beside the chip prints the count behind it, for example "45 analyzed trades · 60+ required for High, 30+ for Medium".

Each bias on the profile also carries its own confidence, from the evidence for that bias alone. A 400-trade profile can still hold a bias read resting on four post-loss trades, and that bias is marked low confidence.

When NEX was offline. If a read came from the fallback path instead of the live model, the confidence chip is replaced by Estimated · NEX offline. An automated estimate should never wear a confidence word.

The model's own number. Each notebook entry's sentiment read also stores a 0 to 1 confidence from the model. Entries that read as pure analysis have it capped. Limbix does not print that number on any screen today.

Worked example#

Your latest profile was built from 45 analyzed trades. 45 is at least 30 but under 60, so the chip reads Medium confidence.

On the Sentiment page, the NEX READS line compares your Positive band (9 paired sessions) with your Negative band (5). The thinner band has 5 sessions: at least 4, under 8. The line reads Medium confidence.

When it shows "—"#

What's a good AI confidence level?#

High, but only because High means more of your record is behind the read. Medium is a real finding with a thinner base. Low is a direction to watch, not a conclusion. Whatever the word, it never makes a read about the market. Every NEX read is about your own trades and your own writing.

How to improve it#

1. Log every trade. The profile's confidence grows with the trades it reads. See add trades.
2. Link trades to your notebook entries. Sentiment comparisons only count sessions with linked trades. See link trades to an entry.
3. Refresh the profile after a busy month. See Behavioral Profile.

Where you'll see it#

• Behavioral Profile (/ai/profile): the confidence chip next to Trader archetype, and low confidence on thin bias rows under Cognitive bias scores. See NEX behavioral profile.
• Sentiment Analysis (/ai/sentiment): the confidence word on the NEX READS line of Tone → outcome. See NEX sentiment.

Frequently asked questions#

How reliable is NEX's analysis of my trading?#

As reliable as the evidence behind each read, and Limbix tells you how much that is. Every profile and sentiment comparison carries a High, Medium or Low word set by the count of trades or sessions behind it. When NEX writes your profile, it is told to copy every figure from your data and never estimate one. Low means "watch this", not "act on this".

What does "low confidence" mean on my behavioral profile?#

It means that particular read rests on thin evidence. For a bias, that is the count of trades that show it, not your total trade count. A profile built from many trades can still hold one bias seen only a handful of times. Read it as a pattern to watch over the next few weeks, and check whether it holds after your next refresh.

Is AI confidence the same as the model's own confidence score?#

No. Each sentiment read stores a 0 to 1 confidence from the model, but Limbix doesn't display it. The word you see is computed from your data: how many analyzed trades or paired sessions sit behind the read. A model can sound certain about four trades. Counting the evidence is harder to fool.