AQ Mentor — reverse engineering
The other modules search for strategies. AQ Mentor does the opposite: it learns from you. You mark on the chart the entries that look ideal to you and it discovers what they have in common — not to imitate your clicks, but to answer «why did you choose those market situations?».
Menu Strategies → AQ Mentor (reverse engineering). Like a master teaching an apprentice: you point, it deduces the rules.
You mark your ideal entries on the chart. You choose neither indicators nor rules — you just point at situations.
AniQuant computes everything it knows on every marked bar and discovers which conditions are characteristic of your entries.
Every finding is tested over the whole history (In/Out‑of‑Sample) with the Lab's real engine.
Phase 1 — Mark your trading
With the toolbar tools you mark, with one click on the bar, your ideal decisions inside the In-Sample stretch (the learning zone):
An ideal buy.
An ideal sell. In an always‑in‑the‑market system, your exits are short entries.
An ideal close.
Removes a mark.
The In‑Sample / Out‑of‑Sample cut is a slider (the amber line): you mark in the IS, and the OOS is reserved for validation. Your marks are auto‑saved per symbol+TF — you can build them up over weeks and come back.
Phase 2 — The micro‑edges
Press 🔬 Analyze and AniQuant compares your marks against the rest of the market. For every objective condition (e.g. RSI(14) < 34) it computes how much it characterizes them, and ranks them by significance.
% of your entries where the condition holds. If out of 150 marks it holds on 28 → 19%.
% of the rest of the market (outside your marks) where it also holds. To rule out the «always‑true» conditions.
Recall ÷ Rest. 19% ÷ 4% = 4.8×: it shows up almost 5 times more often in your entries than by chance.
The lift corrected for sample size (a two-proportion test). z > 2 significant, z > 4 strong. This is the queen column.
Group A vs Group B. The A conditions are an indicator's threshold (RSI < 30); the B ones are relations or crosses between two (SMA4 < SMA8). A badge tells them apart. Note: percentile thresholds tend to have a low z (they fire on a fixed % of bars); relations capture specific events and stand out more.
A panel with the whole catalog grouped (oscillators, moving averages and bands, candlestick patterns, calendar) and a configurable range of periods. Tick whatever you want analyzed; everything, by default.
It shows live the number of combinations you will sweep (pairs grow as the square). The more you select, the more exhaustive — and the longer it takes.
Diversity: it collapses micro‑edges that fire on almost the same marks → the top shows different patterns, not the same one with another operand.
Tolerance ±N: a mark counts as a hit if the condition holds within ±N bars (marking one bar late is not penalized). It applies equally to marks and baseline → fair lift/z.
📋 Ranking — the full appraisal
The analysis shows a curated Top‑40. The Ranking shows all the conditions (up to 2,000) in a sortable table, with three blocks of real backtest:
Trades · Profit · Loss · Net $ in the learning stretch.
The same out of sample — the acid test.
The whole history (IS + OOS).
The $ comes from the Lab engine (it enters on the next bar's open, exits at Horizon bars): it reproduces the Lab to the cent when you seed it. Sort by Total Net to see the ones that really make money. The descriptive figures (recall/rest/lift/z) and the real $ live side by side: one measures your style, the other whether it pays.
It measures a micro‑edge's forward edge over the whole history, split IS / OOS / Total, with a verdict: ✓ it holds (profitable and significant OOS), ~ weak or ✗ no.
Validate all highlights the survivors.
It paints the micro‑edge's trades on the chart (green/red) and a cumulative profit curve (in $ or %), with a tooltip box on hover.
To see at a glance how that edge behaves.
By default no overlap (sequential, one position at a time, realistic). Ticking it, every signal opens a trade even if you are already in the market (pyramiding).
It only changes the result on conditions that fire densely (thresholds); on crosses, it makes no difference.
🧬 From the finding to the strategy
Tick the micro‑edges you want (the survivors auto‑tick) and press 🧬 Seed the Lab. They are sent to the Lab combined as an entry AND, with the exit BarsSinceEntry = Horizon already set — and there the engine combines and optimizes them. Also from the Ranking (double-click = one row; ☑ + button = several).
Marks → Analyze (micro‑edges) → Validate (IS/OOS)
→ Seed the Lab (combine & optimize) → a real strategy
That a condition characterizes your marks (high recall) does not guarantee it predicts the future. Which is why AQ Mentor discovers hypotheses (guided by your marks, cheaply) and the rest of the ecosystem validates them over thousands of occurrences of the full history — that is where overfitting falls apart. A condition on its own is weak; the value is in combining the survivors. AQ Mentor is the human-guided front of the whole discovery chain.
The chance bar: why you will see fewer stars
There is a statistical trap that affects the modules which compare many things at once —AQ Mentor and AQ Contrast—, and that is very easy to fall into.
If you measure one indicator and get a |t| of 3, that is rare: it happens by chance less than once in three hundred. But if you measure eight hundred and keep the best, a |t| of 3 stops being rare — it is what you should expect. Keeping the best of many is not the same as looking at one.
The chance bar is the highest value you should expect out of pure luck after making N comparisons.
It grows very slowly: multiplying the comparisons by ten raises it by only half a point.
The stars do not come from fixed thresholds, but from the margin by which each indicator clears the bar. Zero stars means something precise: «it does not reach what chance gives». With a fixed threshold, a |t| of 2 would earn two stars — and with 800 indicators compared, that 2 is below the bar.
The ranking is unchanged, but now you know where the line is: the rows that do not clear it come out dimmed, and a note tells you how many do. If none clears it, the module says so explicitly — and that is a legitimate result, not a failure of the module.
You will see few stars. The ones you do see mean something.
AniQuant can be tried free for 30 days, with every module and no card.