Inside realbacktesting

An AI analyst writes our daily market notes. Here is how

Every day this site publishes a market note and a piece on trading. They are written by a specialised AI analyst — kept honest by the same rules as our backtests.

Most trading content is written to sell you something. This is the piece where we tell you exactly who writes ours, how, and where the guardrails are — because on a site whose whole promise is proof over promises, the writing has to be held to the same standard as the backtests.

So here is the plain truth: the daily market note and most of the educational pieces on this site are written by a specialised AI analyst. Not a generic chatbot pointed at a keyword — a purpose-built agent with a narrow job, a strong leash, and a short list of things it is never allowed to do.

Why an analyst, and why disclose it

Markets move every session. A useful note has to land every day, cover indices, commodities, forex and crypto, and explain not just what moved but why — consistently, without a human burning an hour on it each morning. That is exactly the kind of work a well-scoped agent does well.

The reason we tell you is simple. An AI byline only earns trust if it comes with the constraints that make the output reliable. Hidden, it is a liability. Disclosed and bounded, it is just a faster, tireless desk.

The rules it cannot break

The analyst runs against a fixed set of prime directives. The important ones:

  • It never invents a number. Market levels and data come from sources it verified against at least two reputable outlets, and they are attributed. Any figure about our own systems is read straight from the file that drives the site — never recalled from memory, never rounded into something flattering.
  • It never gives financial advice. No predictions, no price targets, no "buy this". It describes the session and explains the drivers. A market note is context, not a signal.
  • It separates fact from interpretation. What happened is stated plainly with a source. What it might mean is marked as a read, not a certainty.
  • It will not repeat itself. Before writing, it checks everything already published so you are not served the same idea twice.
  • It cannot touch anything that matters. It publishes to this website through the same locked-down, reviewed pipeline as everything else, and it has no access to trading, money or accounts.

What it actually does each day

JobHow it works
Market noteGathers the prior session's news and data, finds the one driver that mattered, writes the note, publishes it, and sends the link. Fully automated.
Blog pieceProposes three topics each morning; a human picks one; the analyst writes and publishes it. The ideas are machine-generated; the editorial call is not.

That second row matters. The breadth and consistency are automated; the judgement about what is worth saying stays with a person. It is the same division of labour we trust in the systems themselves — the machine handles the relentless, repeatable work, and a human owns the decisions that need taste.

The stack, in the open

No content farm, no anonymous freelancer. Here is exactly what produces these pages:

  • A private VPS — the same server behind our trading research, not a rented content mill. The agent has no access to trading, money or accounts on it; it can only write to the website.
  • openclaw — the orchestration layer. It schedules the daily run, hands the analyst its tools (web search, the publish pipeline, the Telegram link back to a human), and enforces the leash: nothing ships except through the same reviewed, gated pipeline as the rest of the site.
  • OpenAI's GPT-5.5, at high reasoning effort — the model that reads the session, finds the driver and writes the note.
  • Anthropic's Claude — engineered the agent, its skill library and every guardrail described above.

That split is deliberate: the model that writes the words is not the model that built the rails it runs on. Two labs, two jobs — one to produce, one to construct the constraints it produces inside. It is the same instinct as the backtests, where the thing being measured should never be the only thing doing the measuring.

How this connects to the rest of the site

The thesis behind everything here is that a claim is only worth as much as your ability to check it. Our backtests are built to be redrawn in your own cTrader, on real costs, with a 30% out-of-sample hold-out — the full method is in the methodology. The writing follows the same instinct: sourced, bounded, and honest about what it is.

If a note ever states something you can't square with the tape, that is a bug, not a style. The whole point of putting the analyst on a leash — and telling you about the leash — is that you never have to take the words on faith.

You can read the market the machine can't predict it for you, and neither can we. What we can do is show up every day with an honest read and never pretend it is more than that.

Published Jun 14, 2026 · realbacktesting · Educational content and market commentary — not financial advice. Trading involves risk; past performance does not guarantee future results.