Cryptolean

Crypto risk, without hindsight

Every historical value on this site could have been computed on the day it happened. Expanding-window normalisation, daily close, and every definition versioned in public — including the studies that failed.

// 393 charts4,832 days of Bitcoin132,154 markets trackedNo login to read any of it
Bitcoin · historical risk level
Risk, 0 to 1
0.2148
Bitcoin$64,054.44
As of5 Aug 2026
Observations4,832
0.0 — accumulation0.51.0 — distribution

Risk tops when separation from the 365-day mean peaks — not when price peaks. The two are weeks or months apart.

risk-v1
Daily close
Not real-time
// The model

Most risk charts quietly use tomorrow's data.

A risk metric is normally scaled between its own minimum and maximum. Take those from the whole history and every point in 2015 is drawn using the range of 2021 — a number nobody could have computed at the time, on a chart that exists to tell you what to do today. It renders perfectly. It is still a lie about the past.

Ours expands: each day is scaled only against the days before it. Backtests come out worse and the chart is harder to sell. It is the only version that means anything.

risk-v1  ·  separation from the 365-day mean, expanding-window normalised  ·  first 1,000 days dropped  ·  peaked at 0.917 in Feb 2021, and 0.809 at the April price high
// 01 · The library

Every chart carries its own definition.

Title, the date it was computed, the version of the metric that produced it, and one plain sentence saying what it measures — on the chart, not in a help article behind a login.

Risk, valuation and cycles

Risk bands, logarithmic regression, dominance, drawdown, ROI by cycle and the technical set — across 14 assets, on daily close.

Versioned, so history cannot be rewritten

Retuning a metric silently rewrites every historical chart drawn with it. Each definition carries a version, and a change to the definition changes the version.

Bitcoin risk
risk-v1 · 4,832 days
Drawdown from ATH
5,832 days
Mayer multiple
price / 200-day mean
Bitcoin dominance
share of total cap
393 charts  ·  49 families  ·  14 assets  ·  8 categories  ·  all readable without an account
// 02 · Settlement trust

A forecast market is only as good as its referee.

We read the rules text of every market on both major venues, score how ambiguously it is written, and record every contract that has already been disputed — always with the clause that raised the flag, never the score on its own.

Where disputes concentrate

Disputed volume over time, the contracts contested more than once, and dispute rate by category — from settlement records already held, not estimates.

Ambiguity, before it is contested

Five flags: subjective wording, no named source, no edge cases, no timezone, prior dispute. Three of the five fire on what the rules text does not say.

132,154 markets  ·  758 disputed at least once  ·  340 twice or more  ·  $5.4bn of volume on disputed contracts  ·  worst case contested at $242m
// 03 · Market-implied

What the odds say, beside what the model says.

Prediction markets price the macro events that move crypto — rate decisions, inflation prints, recession calls. Those probabilities sit on the same axis as the risk metric, so a forward-looking number and a backward-looking one can be read together.

Implied probabilities, filtered

Illiquid days are left as visible holes rather than interpolated, and the filter that dropped them is stated on the chart. A smooth line drawn through no trading is an invention.

Priced, not predicted

These are risk-neutral prices carrying a risk premium, not forecasts. It says so on the chart, because the difference is the whole thing.

Kalshi and Polymarket  ·  public endpoints, no key  ·  the Fed ladder loses 35% of its days to the liquidity filter, and we say so on the chart
// 04 · The honesty layer

We tested the paper our own pitch cites. It failed.

A 2026 study reports that repricing in Kalshi macro contracts forecasts crypto volatility. It was going to be the flagship of this product. We replicated it on our own data, it did not hold — so the feature is not being built, and the study is published instead.

The paper
t = 3.63Reported over Jan 2023 – Mar 2026
Our data
t = −0.16313 days of overlap, p = 0.88
Out of sample
1.039Above 1: the signal made the forecast worse
Regressions run
66All reported, including the eight that looked real
A null we publish is worth more than a feature we cannot defend.

The eight results that survived multiple-testing correction all vanished once a time trend was added: macro repricing rose steadily across the sample while crypto volatility fell, and a regression reads two crossing trends as a relationship. That is in the study too.