How a study is made

What Blockstudy does with your chart, what it checks, what it can't check, and how it records what happened afterward.

Matching your chart to real prices

From a screenshot, Blockstudy reads the symbol, the timeframe and the time of the newest visible candle. If you pick a symbol instead, the study runs up to now, or up to the date you chose for a past study. Either way, it fetches 300 bars of historical prices for that symbol and timeframe from Twelve Data, ending at that moment. That is the study's cutoff: nothing after it is used, so a study of an older chart can't see what came next.

If the symbol, timeframe or time can't be matched with confidence, the study is not checked against prices. It says so, and every point in it is labelled as read from the chart alone. US stocks use regular trading hours unless you ask for extended hours.

Calculations come from prices, not pixels

Indicators are calculated by code from the matched bars: moving averages (EMA 20 and 50, SMA 200), RSI 14, ATR 14, MACD 12/26/9, average volume, swing highs and lows, Fibonacci retracements of the last swing, and chart patterns found from turning points. The model writing the study cites these values. It does not estimate them from the image.

How each study is checked

A draft study goes through checks written as code before anyone sees it. Levels must sit inside the matched price range, support must sit below resistance, a pattern must match what the code detected, a cited calculation must exist, and the study may not tell you what to buy or sell, or promise a result. A separate review step then reads the draft against the evidence it cites.

A draft that fails gets one chance to be corrected. If the corrected draft still fails, the study is not published, and it doesn't use your allowance. A chart that can't be read is rejected the same way.

Evidence coverage is not a probability

Each point, and the study as a whole, shows its evidence coverage: Full, Partial or Limited. It describes where the support for a point came from, not how likely it is to be right.

• Full: the point cites the matched price data or a calculation from it.
• Partial: the point cites only what is visible on the chart.
• Limited: no price data was matched, so nothing in the study was compared with prices.

A whole study shows Full only when every point is Full and nothing it lists as uncertain has a high impact.

When it can't verify something

Gaps are listed, not filled. A point the checks can't confirm stays labelled as read from the chart. What the study couldn't establish is listed under Uncertainties, with how much it matters. When the evidence points both ways, the study gives both scenarios and says what would strengthen or invalidate each one.

How past outcomes are recorded

For a study matched to real prices, Blockstudy takes its nearest level above the last close and its nearest level below it. It then follows the next 20 bars on the same timeframe and records which level price closed beyond first, on which bar, or that it stayed between them. If not enough bars arrive, the study is marked as not followed instead of being given a result.

A study of a past date, or of an old chart whose next bars had already closed, is never followed. Those bars were known before the study ran, so following them would only be hindsight.

These records are price events. A close beyond the upper level does not mean the bullish scenario was right, and price can cross one level and then the other. The app shows counts, never percentages. Below 10 followed studies, it says there are too few to read anything into.

Methods are versioned

The calculations, checks and outcome rules each carry a version number. When one changes, it changes under a new version. Every study keeps the versions it was made with, so improving a method never quietly reinterprets an older study.

What this does not do

Blockstudy studies charts for learning. It does not predict prices, tell you what to buy or sell, or place trades. Market data is provided by Twelve Data. Questions: support@blockstudy.io.