PatternX AI: the context layer explained

PatternX AI is an evaluation layer, not a detector. It runs on Pro and Elite accounts switched to AI mode, and only after the deterministic engine has detected a setup, passed it through the universal filters and given it a PatternX Score of at least 50. Its job is narrow: rate the market context around that setup, then keep it or drop it.

Where it sits in the pipeline

The order never changes: detection, the universal filters and the PatternX Score come first, then the context layer, then your own settings, the risk engine and execution. Because it runs after detection, the context layer cannot add a pattern the detectors didn't find. Its inputs are the candles the engine already used, PatternX's own closed-trade record on that market (or a fixed per-pattern prior when there are too few trades) and, for perpetual futures, the funding rate. It makes no external calls.

The 8 factors

Each factor is scored from 0 to 100. The weights add up to 9.1, and trend strength and confluence carry the most: 1.5 each, or about 16.5% of the total apiece.

The context score and the veto

The context score is the weighted mean of the eight factor scores: each score times its weight, summed, then divided by 9.1. The veto rule then drops the setup unless both of these are true:

1. The context score is at least 48.
2. It is not the case that market structure is below 35 and momentum is below 40 at the same time.

The second condition catches one specific contradiction: a breakout while the swing structure works against the trade and momentum is weak. Either weakness on its own is tolerated; together they veto the setup even when the overall context score passes.

A worked example

For example, take a hypothetical long setup with these factor scores: trend strength 70, market structure 25, volatility regime 80, volume quality 60, momentum 35, confluence 65, historical similarity 62 and funding sentiment 55.

• Weighted sum: 70 × 1.5 + 25 × 1.3 + 80 × 0.8 + 60 × 1.2 + 35 × 1.0 + 65 × 1.5 + 62 × 1.2 + 55 × 0.6 = 513.4.
• Context score: 513.4 ÷ 9.1 ≈ 56.4, well above 48.
• Veto check: structure (25) is below 35 and momentum (35) is below 40, so the setup is vetoed despite its passing score.

With momentum at 45 instead, the context score would be about 57.5. Structure would still be weak but momentum would not, so the setup would be kept and its confidence re-weighted.

How confidence is re-weighted

A setup that survives the veto keeps its pattern, entry, stop and target. Only its confidence changes: the layer starts from the engine's confidence, which is the PatternX Score, and re-weights it using the context score. The re-weighted number is the confidence you see, and it is the number your minimum confidence is checked against. Because the adjustment is part of a fixed formula, the same inputs always give the same result.

AI mode starts from every setup the engine accepts at 50 or more, so its output can differ from Classic in both directions. A setup Classic would have sent you can be vetoed, and a setup the engine scored below your minimum can still reach you if the context layer keeps it. Each kept setup also gets a short reasoning note, generated from a template, that summarises its context.

What the context layer does not do

• It does not detect, find or invent patterns. It evaluates only setups the 20 detectors have already produced.
• It does not touch your trade. The take-profit and stop-loss levels, the position size, the risk checks and the route to execution are exactly what Classic would use.
• It does not read news, social media, the order book or correlations. Its only sentiment input is the funding rate.
• It is not a learned model. It is neither machine learning nor a language model but a deterministic formula. No model is trained and the formula does not adapt: the weights are fixed, and only the inputs move, including PatternX's live win rate on a market as more trades close.
• It does not manage open trades. Once a position is open, its stop, break-even and trailing behaviour work exactly as they do in Classic.

If the AI evaluation fails with an error, the setup falls back to Classic evaluation and carries on through the normal pipeline.

Plans and availability

PatternX AI is available on Pro and Elite, while PatternX Classic stays on every plan and is the default mode. On Pro, AI mode allows up to 5 trades a day, compared with 100 in Classic; on Elite, AI mode has no daily plan limit. Pro and Elite can also run backtests in AI mode. See PatternX AI and pricing for plan details.

Where large language models are used

Language models are part of PatternX as writers, not deciders. When configured, PatternX uses models from OpenAI or Anthropic for:

• the plain-language explanation attached to a signal,
• the assistant and the support assistant,
• daily, weekly and monthly reports,
• education lessons and optional backtest write-ups.

None of them sits on the detection, risk or execution path. They do not detect patterns, compute the context score, apply the veto, size positions or place orders, and they are instructed not to give buy or sell advice. A language model can also be wrong or leave things out, so read its text as commentary on a signal, never as part of the signal itself.

Limitations

The context layer reads the same recent candles as the engine, so it cannot see what those candles don't show, such as news or scheduled events. Its thresholds (48 for the context score, 35 for structure and 40 for momentum) are fixed and apply equally to every market and timeframe. Until a market has 10 closed trades, historical similarity rests on a fixed prior, not a measured statistic, and funding sentiment says nothing on stocks or FX. A veto is not a prediction: some vetoed setups would have worked, and some kept ones will fail. For the wider picture, compare Classic vs AI and read what PatternX can't do.