The pattern wizard: patterns, track records and significance

Start in one minute

1. Open any chart and pick a timeframe. A badge such as “5 patterns found on H1 · Show” tells you what the wizard sees.
2. Press it (or ✦ Wizard in the toolbar, then Scan). The panel lists every finding, best first.
3. Press Show on the chart to draw a finding, Keep as my drawing to keep it, or 🔔 Notify me to get its levels as alerts.
4. Before you act on one, read its track record: it is the part that tells you whether the pattern has worked.

Reading a pattern

The suggested trade

▲ Long 3,995.04 · at the market now · Stop 3,833.04 (4.05%) · T1 4,180.03 · R:R 1:1.14 · T2 …

• Entry: at the market, a limit order in a zone (waits for the price to come back), or a stop order on a breakout (waits for the price to break through).
• Stop: where the pattern is wrong, with its distance in %.
• T1, T2, T3: take-profit targets, each with its reward to risk: 1:1.14 means the target is 1.14 times as far as the stop.
• A pattern without a sensible trade says so (for example when no target pays back at least the risk). Levels are never below zero.

📐 Position size for this trade (under each trade, here and in the market scan): enter your balance, its currency and the risk (a % of the balance or an amount), and the wizard works out the size that loses exactly that much if the stop is hit, from the ticker's real lot size and the live rate to your currency: lots and units, the risk in money and pips, the value of one pip, the margin, and what each target pays. Sizes are rounded down to 0.01 lots, so the risk is never more than you chose. Your balance and risk are remembered in this browser and shared with the position size calculator. On the chart, Fill in a paper order puts the side, order type, price, size, stop loss and the take profit you pick into the Trade panel; check it and press Buy or Sell.

Candle confirmation

A pattern says where to enter; a candle at that place says whether the market agrees. Under each trade the wizard shows one of these:

Candles smaller than half an ATR are ignored: on any market a tiny candle says nothing. Definitions follow Nison and Bulkowski, as proportions of each candle's range, so they work on every market and timeframe.

The track record

Track record (all tickers tested): 279 trades, target 1 first 55% (needs 41%), +0.22R per trade

For Elliott waves the wizard goes further up: degree by degree, to the largest swings the chart shows, it checks whether the wave is part of a larger Elliott wave, and names the primary pattern: the largest Elliott structure found that contains it (for example “Part of a larger Elliott wave: wave 3 of an impulse up → wave C of an A-B-C correction down”). Zoom out (more candles) to see more degrees.

The record and its verdict are shown to every account; the intervals and significance need Pro.

Confidence intervals

95% interval: wins 49%–61%, per trade 0.05 … 0.41R (bootstrap); 0.03 … 0.40R (t)

A record of a few hundred trades is a sample, not the truth. The interval is the range the true value most likely lies in (95 times out of 100):

• Wins: a Wilson interval for the win rate, reliable even for small samples.
• Per trade, bootstrap: the trades resampled 4,000 times; it makes no assumption about how results are distributed (trading results rarely follow a bell curve).
• Per trade, t: the classic interval from the average and spread. When both agree, you can trust the range.

Read it like this: when the whole interval for the result per trade is above zero, even the cautious estimate makes money. When it includes zero (−0.11 … +0.28R), a small profit may just be luck.

Expected return

Expected return: +0.22R per trade (+0.31% of the price) · 95% interval +0.05 … +0.41R (bootstrap), +0.03 … +0.40R (t); +0.07 … +0.58% · ✓ 90% ✓ 95% ✗ 99%

The win rate alone can mislead: a strategy that wins often but makes little on each win and loses a full risk on each loss can lose money. The expected return is what one trade earned on average, in multiples of the risk (R) and in percent of the price (each trade's result times its stop distance). It has its own 90/95/99% confidence intervals (bootstrap and t) and its own test: ✓ at a level means both the t-test and the bootstrap show it is above zero at that level.

Statistical significance

✓ 90% ✓ 95% ✓ 99% means the edge is unlikely to be luck at that confidence. The wizard checks it three ways, and all must agree:

1. Win rate against its break-even: an exact binomial test: how likely would 55% wins out of 279 be if the strategy really won only the 41% it needs?
2. Expected return against zero, without assumptions: the bootstrap: how often do the resampled averages fall to zero or below?
3. Expected return against zero, classic: a one-sided t-test of the average result per trade.

The levels match the intervals: significant at 95% means the whole 95% interval of the win rate is above its break-even and the whole 95% interval of the expected return is above zero (each test is one-sided at 2.5%).

Many strategies are tested on every ticker, so a few would look significant by chance alone. The Benjamini-Hochberg correction adjusts for that: a strategy stays significant only when it holds up after counting how many were tested together. Fewer than 10 trades are never called significant.

Significant does not mean certain. It means past prices show an edge that is unlikely to be chance. Markets change; an edge can fade.

What each verdict means

A worked example

A scan of a gold chart listed six findings pointing different ways. Their records:

Figures from one scan, as an illustration; your chart shows today's.

How the back-test works

1. The ticker's candles of one timeframe are read (up to 8000).
2. Every few candles, the wizard scans the 600 candles that ended there: exactly what a customer would have seen on the chart at that moment. No look-ahead: it never sees a later candle.
3. Each suggested trade then meets the candles that followed. A market entry fills at the next candle's open; a limit or stop entry must be reached first. Target 1 before the stop is a win; the stop first is a loss; both in the same candle counts as a loss (the cautious reading); neither within the horizon is valued at the last close.
4. The same setup seen at several moments counts once.

Automatic: the most used tickers are back-tested on M15, H1, H4 and D1 in the background, on every new candle, so their records are always there and only grow. Your own back-test (the Back-test button in the panel; Pro) runs it on the ticker and timeframe on screen; its results join the shared record.

Not included: spread, commission and slippage. They matter little on daily candles and more on minutes; subtract your own costs from the result per trade.

Scanning a past moment

Under Scan, choose As of a past date (or start Replay and pick a candle): the wizard scans the chart as it was then and shows what happened next for each trade: target reached, stop hit, still open, or entry never reached, with the result in R. It is the quickest way to see the wizard at work on a market you know.

Patterns that already reached their target or failed are listed apart under Finished recently, never as valid ones.

Asking an AI for a second opinion

Keep the patterns you want an opinion on with Keep as my drawing, then press ✦ Ask AI on the chart and tick them. The assistant you choose (ChatGPT, Claude, Perplexity, Gemini or DeepSeek) gets each kept pattern's analysis as text: its quality and larger-wave context, the suggested trade with entry, stop, targets and R:R, the rules met and missed, and the track record with its win rate against break-even, result per trade, 95% intervals and significance. Patterns you did not keep are left out, and you tick which indicators to include. Open What is shared to read it before sending. Nothing is sent by us; it opens in your own account there, in the assistant's app when it is installed on your phone.

Good questions: “Which of these setups has the best record, and is its edge proven?”, “Do the larger waves support the trade?”, “Where would you put the stop instead, and why?”. An AI can explain and challenge the numbers, but it cannot see more history than the wizard has tested.

What it finds

Harmonic patterns (Gartley, Bat, Alternate Bat, Butterfly, Crab, Deep Crab, Cypher, Shark, 5-0, three drives, AB=CD), Elliott impulses, diagonals, corrections and triangles, head and shoulders, double and triple tops and bottoms, triangles and wedges, rectangles (trading ranges), bull and bear flags and pennants, cups with handles (and inverted), reversal candles at support and resistance, Fibonacci pullbacks and expansions, Andrews' pitchforks, Gann fans, channels, support and resistance and trend lines; and the indicators that suit the market's state. Breakouts are checked against volume where the market reports it (1.2× the 20-candle average or more confirms). The rules of each are on the documentation page.

Market scan

Market scan (Pro+) runs the wizard over every ticker of the markets you choose (forex, stocks, indices, metals, energies, commodities, crypto, ETFs) on one timeframe, and keeps only the setups that are:

1. at their entry now: confirmed by a candle (or, if you allow unconfirmed ones, within half an ATR of the entry with nothing pointing the other way),
2. significant at your level (90, 95 or 99%): a win rate above break-even and an expected return above zero, after the multiple-testing correction,
3. in the direction you choose.

More filters narrow the list further: the kinds of pattern (harmonics, Elliott waves, chart patterns, Fibonacci and Gann, reversal candles), the pattern size, the entry (at the market, a limit order on a pullback, a stop order on a breakout), a minimum quality, a minimum reward to risk at target 1, a minimum number of trades in the track record, a minimum expected return per trade, only candles that have closed, and only the tickers you list (in any market; a quick pass). Your choices are remembered on this device.

The best come first: confirmed ones, then the highest significance, then the highest lower bound of the expected return. Each pattern delivered uses one pattern-wizard scan of your allowance: ask for 10 and find 4, and 4 are used; find none, and none are. When more than you asked for pass every filter, the results end with Load 10 more and Reveal all: they show the rest, best first, one credit each, for three hours after the scan (then scan again for fresh setups). Opening a delivered chart shows its scan without using another one for six hours. The scan of each ticker is shared, so a pass over all markets takes a few minutes and the panel shows its progress. One market scan runs at a time: reopening the page shows the one still running (and your last results), and you can follow it on the dashboard or cancel it under Tasks once it is 30 minutes old or stuck. In the phone app it is under Tools.

What each plan includes

Some drawing tools behind patterns (for example Gann fans and pitchforks) belong to a plan: their findings are named, with the plan that shows their levels. Compare the plans.

Insights and the API

Strategy insights ranks the strategies with the best record per ticker and timeframe. Developers get the same statistics with GET /v1/patterns/stats?symbol=EURUSD&timeframe=H1 (see the API documentation).

Limits: read this

• The wizard is automatic analysis of past prices. It is not investment advice and it cannot know the news.
• A past edge is evidence, not a promise. Records mix calm and volatile years; the all-tickers record mixes markets.
• Several findings can point opposite ways at once. Prefer the ones with a significant record, with the larger trend, and with a sensible stop.
• Always use a stop loss and size positions with the position size calculator; practise first with paper trading.