How the numbers are made - and what they don’t mean

The data

Chart Echo indexes years of historical price data (OHLCV - open, high, low, close, volume) for stocks and crypto across multiple timeframes. Crypto history is sourced and gap-filled from market data providers; equities from established market-data APIs. Data is de-duplicated and time-ordered before indexing, and the daily timeframe is available free while intraday timeframes are a Premium feature.

Building the index

Every candidate historical window is normalised to remove absolute price and scale, then encoded as a vector that captures its shape. Those vectors are stored in a similarity index (FAISS) so that, at search time, the nearest neighbours to your pattern can be retrieved in milliseconds rather than by scanning the whole dataset. The index spans on the order of 450 million windows.

Scoring a match

A match’s similarity is the closeness of its vector to your pattern’s vector, reported as a 0–100 match score. Optional time-warping allows a historical setup that unfolded faster or slower than yours to still register as similar in shape. A self-match (the pattern against itself) scores 100% and is labelled as such.

Outcome statistics

For each match we measure its continuation window (typically the next ~60 candles) and aggregate across all matches:

• Directional split - the share of matches that rose, fell, or finished roughly flat.
• Average & median return - central tendency of the continuation move (median resists outliers).
• Average maximum gain & drawdown - the typical best-case and worst-case excursion along the way.
• Projection cone - the median continuation path with quantile bands (e.g. 25–75% and 10–90%).

The confidence score

Confidence is a 0–100 blend of four things: how many matches were found, how strongly they agree on direction, how tight their outcomes are, and the quality (similarity) of the matches. Lots of close matches that resolved the same way → high confidence. A handful of loose matches pulling in different directions → low. Confidence describes the strength of the historical evidence, not the probability that you’ll make money.