A calendar says a number is due at 13:30. A graded calendar says what an above-consensus print, an in-line print and a below-consensus print each read as, and which assets each band touches. That is a genuine improvement and it is still not a forecast.
An ordinary economic calendar tells you that US CPI is due at 13:30, that it is high impact, that consensus is 2.9 per cent and that last month printed 3.1. Four facts. Everything you actually want to know — what happens if it comes in at 3.2, which of the things you trade care, and whether this release has historically moved gold more than the dollar index — is absent, and you were supposed to supply it from memory at 13:28.
So: what can an AI-assisted calendar tell you that a plain one cannot? Three things, concretely. It can grade a release into bands before it happens, so you have read the above-consensus, in-line and below-consensus cases in advance. It can attach a reading and a volatility state to each asset you follow, with the headlines and releases that reading rests on. And it can record what this event has typically done, as a threshold and a direction rather than a story. What it cannot do is tell you what the number will be, what you should do, or whether your trade is a good one.
INFO
Quick answer. A graded calendar pre-computes the interpretation, not the outcome. For a high-impact release it produces three bands — above consensus, in line, below consensus — each with a one-sentence reading and the specific assets it touches. It is preparation moved earlier in time. It is not a prediction and it is not advice.
The design decision underneath all of this is worth stating plainly, because it is what separates a useful tool from a horoscope.
A model that predicts the CPI print is making a claim it cannot support. A model that says "if this prints above consensus, here is what that conventionally reads as, and here are the instruments where that reading usually shows up" is doing something entirely different: it is preparing the interpretation for each outcome that can occur. The uncertainty stays where it belongs, in which band happens.
So a graded release looks like this, illustratively:
Nothing in that table is a forecast. All of it is work you would otherwise be doing in the ninety seconds after the print, which is the worst possible moment to be doing it.
Each band carries a short summary and up to four named assets with a direction. That limit is deliberate — a band that names fourteen instruments has stopped being a reading and become a hedge.
Three other layers, each with a different epistemic status, which is the part most tools blur.
A per-currency line. One sentence of bias per currency in the feed, plus the commodities the platform prices — gold, silver, platinum, palladium and oil. This is a summary of that day's commentary, at the top of the calendar, so you do not start the week without having read anything.
A per-asset reading. For each tracked asset: a bias of bullish, bearish or neutral, a reason in plain words, and the specific headlines and releases that reason rests on. The sources matter more than the bias does. A reading you can trace to three named articles is something you can disagree with; a reading with no sources is something you can only believe or ignore.
A volatility state, which is not from the model at all. Low, normal, elevated or high — computed from recent daily ranges against that asset's own twenty-day average, arithmetic rather than judgement, and bumped one step when a high-impact release is due within twenty-four hours. This is the number we would keep if we had to delete everything else, because position sizing has a legitimate use for it and interpretation does not.
A historical reaction per event. Expressed as a threshold and a direction — what magnitude of surprise has historically been associated with what kind of move — rather than as a narrative about last time.
Five limits, stated as limits rather than as caveats in small print.
It cannot tell you the number. Nothing can. The entire structure above exists because the print is unknowable and the interpretation is not.
It cannot tell you what to do. A reading is a reading of commentary, not a recommendation, and we are not authorised to give investment advice and would not want to. There is no position size in the panel, no entry, no "consider buying". Where the tool touches your risk it does so through arithmetic you configured.
It cannot tell you that the conventional reading will hold today. Markets reprice against the textbook regularly — a hot inflation print that sends the dollar down because the market was positioned for hotter still. The band tells you what the outcome conventionally reads as. It does not promise the tape agrees.
It cannot see positioning, flows or the thing nobody has published yet. The inputs are public headlines and the public calendar. A move driven by an unreported liquidation is invisible to it, and will be explained afterwards by commentary that the next day's run will dutifully summarise.
It cannot be right about an event it did not receive. Calendars revise. Consensus figures move in the days before a release, times shift, and events are occasionally cancelled. A grading produced on Monday for a Thursday release was graded against Monday's consensus.
WARNING
The most dangerous use of any graded calendar is as permission. "The panel says bearish gold" is not a reason to be short gold. It is a summary of what commentary said, attached to sources you have not read yet.
A small feature that says more about the honesty of a tool than its features list does.
The fundamentals run can fail. The model call can time out, the headline fetch can come back thin, the calendar feed can be late, and the scenario grading can produce nothing usable. Earlier versions of ours responded to a failed desk call by writing every tracked asset with a blank bias, and the panel simply showed nothing — indistinguishable from a quiet day.
Now each stage keeps its own note on the run: why the per-asset desk call produced no bias, whether the event reactions were freshly generated, reused or carried forward, and why the banded grading produced nothing. Those notes surface rather than sitting in a log nobody reads.
This matters because the alternative is a screen that looks confident when it is empty, and that is a worse failure than an error message. A blank panel with no explanation trains you to assume nothing is happening. A blank panel that says why trains you to check the source yourself.
As a pre-read, once. Ten minutes on Sunday or first thing in the morning: the currency strip, the bands for anything high impact this week, the volatility states on the instruments you trade. Then close it. The value is that you have already thought about Thursday before Thursday.
As a sizing input, through the volatility state. Elevated or high volatility on an instrument is a fact about recent ranges. Whether that means you size down, widen stops or stand aside is your rule, written in advance, and the calendar's job is only to tell you which state you are in.
As a calendar-avoidance check. The most mechanical use and probably the most valuable. Is a high-impact release due on something I am about to trade, inside the window I care about. That is a yes-or-no question, it needs no artificial intelligence whatsoever, and getting it wrong is expensive.
No, and a tool that implies otherwise is misrepresenting what it does. What can be prepared in advance is the interpretation of each possible outcome — what an above-consensus, in-line or below-consensus print conventionally reads as, and which instruments that reading usually touches. The uncertainty remains in which band occurs, which is exactly where it belongs.
Three pre-graded scenarios for a high-impact release: above consensus, in line and below consensus. Each carries a one-sentence reading and up to four named assets with a direction. They are produced before the release so that the interpretation exists before the print does, rather than being improvised in the minute after it.
No. A bias in our fundamentals panel is a reading of that day's commentary, shown with the specific headlines and releases it rests on, and it carries no position size, entry or instruction. We are not a broker or an adviser and do not provide investment advice. The value is that you can check the sources and disagree, which is not possible with a bare signal.
From price data, not from a model. Recent daily ranges are compared with that asset's own twenty-day average to produce low, normal, elevated or high, and the state is bumped one step when a high-impact release is due within twenty-four hours. Keeping this arithmetic rather than judgement is deliberate, because it is the output a sizing rule can legitimately consume.
The panel says so, with the reason. Each stage of the run keeps a note explaining why it produced nothing — a failed desk call, thin headline coverage, a late calendar feed — and those notes are shown rather than logged. A blank panel that explains itself is safer than one that looks the same on a quiet day as on a broken one.
The calendar is part of the free tier, along with the SMC indicators and journal basics. The intention is that the tools you use to avoid a bad decision should not sit behind a payment, and the paid tiers are for the copier, the signal sources and the multi-account features. Current plan contents are listed on the pricing page.
For the trading side of this — how to build a routine around releases rather than react to them, and what an AI-assisted playbook looks like in practice — see /blog/trading-economic-calendar-ai. And because a graded calendar is a tool that invites you to hand over judgement it cannot hold, /blog/what-should-never-be-automated is a useful corrective to read alongside it.
