Skip to Ozmium documentation
Ozmium Docs

FAQ

How Ozmium generates trade signals, how to report a problem with them, how to get a token listed, and how to send the newsroom a tip.

Updated August 9, 2026

How does Ozmium generate trade signals and analysis?

Three independent voices look at each market, and two of them have to call the same side before anything is published. Each may abstain, and an abstention never softens the requirement - with one voice silent, the other two must both agree.

1. Technical indicators

Every reading is a standard, published formula computed on closed bars. Nothing here is a house variant, so any quant running the same definitions on the same candles gets the same numbers:

Read Definition Period
MACD histogram EMA(12) - EMA(26), minus its own EMA(9) signal line, divided by ATR so one threshold means the same thing on every asset 26
Head and shoulders The classical formation across recent closes -
VWAP position Price against cumulative sum(typical price x volume) / sum(volume) session
RSI 14 Wilder's relative strength index, Wilder RMA smoothing 14
Bollinger %B (close - lower band) / (upper band - lower band), bands = SMA(20) +/- 2 standard deviations 20
Cumulative volume delta Buying minus selling volume, accumulated and ranked against its own window 20

Each read votes bullish or bearish in the direction it was measured to work, not in the direction convention assumes. That distinction matters: fed raw, RSI treats an oversold market as bearish, and measured across 33,918 observations that reading loses money - while reading oversold as bullish measures the largest single edge in the set. Where a reading has no opinion it abstains rather than voting a number nobody tested. The share of reads calling each way is the percentage published as X% Bullish Technicals.

2. Trader positioning

Open interest across the venues we can read, long against short in USD. The majority side is the read. Its strength is weighted by how large that market's book is relative to every other market - a 99% long reading on a market holding a thousandth of the ecosystem's money is noise, not evidence, and below the median market by size the voice abstains entirely.

3. Experimental signals

A proprietary approach to bundling the same data into markers for chapters of importance in macro patterns. These are experimental and exist primarily for the agent system's own use rather than for human trading - which is precisely why they are not explained further and are not financial advice. People like seeing them plotted on the charts, so we share them.

Entry, stop and target

Stated arithmetic, in order:

  1. Entry is the close of the last closed bar. No forming bar is used, so a published entry cannot change afterwards.
  2. ATR(14) on that bar gives the market's ordinary range.
  3. Stop sits beyond the nearest structural level on the risk side, clamped to 2.0-3.5 x ATR, defaulting to 2.5 x ATR when no level qualifies.
  4. Risk = |entry - stop|.
  5. Target must clear 3 x risk, snapped to the nearest structural level at or beyond that distance.

Those multiples are measured rather than chosen. Swept across 548 markers as real bracket trades - entry at the next bar's open, exit at whichever level the tape touches first, a bar spanning both counted as the stop, net of real fees - every geometry inside 2 x ATR underperformed: a tighter stop is not protecting the trade, it is handing it to the noise before the move develops.

Check it yourself

Every signal's derivation is published, free and unmetered, at /api/signal-proof:

https://ozmium.org/api/signal-proof?sym=BTC&cls=Crypto&gran=86400&side=buy

It returns the exact bars used (count, first and last timestamp, and a fingerprint of the closes so you can confirm you hold the same tape), every indicator with its formula and period, and the level arithmetic step by step. Recompute it and you should land on the same numbers to the digit. If you do not, the response tells you exactly which step to compare.

Signals are commentary and infotainment, not financial advice. They are published because they are interesting and auditable, not because they are certain.

How can I report potential bugs in analysis content?

Email [email protected] with the details - screenshots, logs, the market and timeframe, and anything that helps reproduce it. The more detailed and reproducible, the better.

We use AI systems in the pipeline, so sometimes things go askew. We are always working to improve these living, organic data systems, and a specific report is the fastest way something gets fixed.

How can I get a token listed in your trade signals?

Right now these are handled by hand: email [email protected].

You can also email [email protected] and he might consider it. That is an agentic inbox - human reviewed and whitelist gated - so a reply may not always be forthcoming, depending on Ozwald's schedule and filtering configuration. Much the same as emailing any broadcast studio.

How can I submit a tip or story for the newsroom?

Email [email protected]. Same caveats as above: agentic inbox, human reviewed, whitelist gated, and a reply is never guaranteed.

How can I support your work?

Contents

On This Page