BTC$61,987−1.7%ETH$1,757−1.1%SOL$73.15−0.7%BNB$580.3−0.4%XRP$0.9604−0.8%DOGE$0.05823−0.9%ADA$0.1519+0.3%AVAX$6.99−0.6%LINK$9.01−0.9%TRX$0.3104+0.2%DOT$0.7140−0.7%LTC$42.33−3.9%BCH$206.4−1.1%NEAR$3.21−2.8%
Provyan
Home / Technology

The Provyan Proof Engine

Provyan is not a chart site and not a signal seller. It is a measuring instrument: a method, built for this site, that takes a trading signal through six stages and reports what it finds, including when the answer is “no better than chance”.

Studies run4,13244 signals on 94 coins since 2017
Random draws per study10,000corrected across 3,565 studies
Reach strong evidence1334% of those judged
Calls sealed before the outcomeEvery onehash-chained, public

The six stages

  1. Detect. Each of the 44 signals is an exact rule with no discretion, such as “14-day RSI closes below 30 after being at or above it”. A signal that repeats within 7 days of itself is counted once, so one event is never scored several times.
  2. Measure against the coin itself. The move after every occurrence is compared with what the same coin did over the same number of days from any day. A coin that rose for years makes every signal look good; subtracting its ordinary behaviour removes that illusion. This is the step most published backtests leave out.
  3. Test for luck, across everything at once. For each study the engine draws 10,000 random sets of days of the same size and counts how often they produce a gap at least as large. Because 3,565 studies are tested together, the results are then corrected jointly (Benjamini-Hochberg), so the handful that would pass by accident do not get through.
  4. Check that it held twice. The history is cut in half. A pattern that appears in only one half is marked down, however strong it looks overall.
  5. Score. Sample size, the corrected luck test, the stability check and the size of the gap are combined into one proof score from 0 to 100. Only 133 of 3,565 judged studies reach “strong evidence”, and a direction is called only at 55 or more.
  6. Seal. At every daily close the new signals are committed to a hash-chained ledger before anyone can see how they turn out, revealed a day later and graded after 7 days. No line can be changed afterwards without breaking the chain, so the live record cannot be tidied up.

What sets it apart

What it has shown so far

Of 3,565 studies with enough cases, 198 came through with strong or some evidence. Those are the setups worth a trader's attention, and each one is published with every case behind it: the ranking. The call rule built on them is sealed night by night in the public ledger, where its full replay is laid out as well.

The live layer

Around the engine the site runs collectors of its own, and none of them stops: orders are read as they trade, liquidations every 25 seconds, the Bitcoin and Ethereum networks block by block, and the two joins are rebuilt every minute.

Inputs

The engine works on daily candles: open, high, low, close, volume, trade count and taker volume for each coin, from the first day it traded. The current source is Binance spot, the deepest record available; the method itself does not depend on the venue and the same stages run on any series. Details and limits are in the methodology.

Full methodology · Public ledger · Guides · Open data