How a reading is made and checked
Every reading published by Horizon AI is written by Meridian Engine and states, before the outcome is known, the levels that matter and the conditions that would invalidate it. Below: the data it uses, what it leaves out and how you can check the outcome yourself.
Updated 2026-10-21
Principles
The six rules Horizon AI applies to every Meridian Engine analysis. Trust you can verify.
- Before, not after. Direction, levels and invalidation are written before the move, with the time of issue.
- Every outcome stays. Analyses stay in the record with their outcome, losses included: nothing deleted, nothing re-dated.
- The method shows. Every statement points to the price bar that produced it, and the outcome is checked against public price series.
- The same for everyone. No personalisation and no position size: the decision and the risk stay with the reader.
- Measure, don’t promise. No figure without sample and period, no promised result.
- Quality before quantity. Every page is checked before it goes out: data, sources and layout.
Structure of an analysis
Bias, entry, stop, targets, invalidation, rationale and outcome, recorded before the move.
Bias
Long, short or neutral. Neutral means no directional edge and is published with the reference band and the two closes that would invalidate it.
Entry
The level from which risk, targets and result are measured.
Stop
The price that invalidates the analysis. The distance between entry and stop defines 1R, shown in price on the card.
Targets
T1 and T2, in price and in multiples of R. A T2 at +1.08R means a move slightly larger than the stated risk.
Invalidation
The closing conditions, on the analysis timeframe, that void it before the stop: for example an H1 bar closing above 4,318.9.
Rationale and opposite scenario
The evidence, each item with bar and timestamp, and the strongest opposite scenario the engine assessed and rejected, with the reason.
Outcome
The first event within the horizon (target, stop or invalidation), with MFE and MAE, checked on public price series independent of the engine’s feed.
No field is optional: an analysis without a stop, an invalidation or an issue time is not published. No field changes after issue: every card is recorded with a UTC timestamp and the run ID before the move, and the public record carries its outcome.
Method
Structure breaks graded by quality, sweeps measured in ATR, every piece of evidence tied to its bar.
- Structure
- Higher or lower highs and lows (HH, HL, LH, LL); structure breaks (BOS and CHoCH) graded by quality: displacement (a break with momentum) or drift (a slow break). A drift break alone is not sufficient for a decision.
- Liquidity
- Clusters of equal highs and lows, where orders concentrate. Sweeps, that is price pushing past those levels, measured in multiples of the average bar range (ATR).
- Zones
- Supply and demand zones, and the price gaps left by momentum bars (FVG and BPR).
- Context
- Market session, ATR percentile (volatility against its average) and the macro calendar when available. A missing or stale input is stated on the card.
- Decision
- Two agents: one reads the chart structure, the other decides whether there is a trade. Every statement names the bar it comes from: in the BTCUSD reading of 2026-10-01, the H1 sweep of 30/09 at 13:00 UTC, 1.28 times the ATR deep. Without confluence the outcome is “no trade”, with the band and both scenarios.
The full report lists the evidence item by item: the BTCUSD report of 2026-10-01 lists ten, each with timeframe, bar and timestamp, together with the rejected opposite scenario and its reason.
What the engine reads
Meridian Engine reads price structure, macro data, positioning and scheduled events, and writes the analysis out in full in about 60 seconds. When an input is stale or incomplete it says so and gives it less weight: the reading reports this in its "Data and limits" block.
In the BTCUSD reading of 9 August 2026 the macro context was flagged as stale and positioning data was five days old. The gold reading of 24 September 2026 had no macro data available, and says so.
What a reading contains
- what the market did, with the figures and the chart they come from;
- which events moved it, and which data were missing;
- two readings of the same facts, including the opposite case;
- where the levels are, each with the reason it matters;
- what would invalidate the reading, written before the outcome;
- what happened next, added once the outcome is known, with the time of the check.
What it leaves out
No position size, no personalisation, no invitation to act at a given moment. The engine's internal parameters that are not published stay in the internal register. The reading is the same for everyone and does not take the reader's circumstances into account. It is not investment advice.
Times and market data
All times are in UTC and refer to the close of the hourly bar. Each reading states the feed it was produced from: the gold reading of 24 September, for example, a Capital.com demo feed, bid side, which is not a production feed.
How the outcome is checked
The outcome is checked against public price series, separate from the engine's feed. The series quote on their own scales: figures are brought onto the reading's scale, and the page says how.
The levels of every analysis are expressed in the price of the engine's series. To check them on a public series, for example COMEX gold futures (GC=F), add the basis to each level: the difference between the two series, measured on the last bar before issue and stated on the card of every analysis.
| Instrument | Check series |
|---|---|
| XAUUSD | COMEX futures (GC=F) |
| BTCUSD | Binance BTCUSDT |
If two series disagree, or a level was missed by less than the tolerated difference, the outcome stays "to be checked" and is read by hand.
The four checks anyone can do on any analysis are in the guide How to check a market analysis.
The public record
From 10/10/2026 every analysis enters the out-of-sample record at its issue time, before the outcome. Engine version frozen from 10/10/2026: rules and parameters do not change. No analysis is removed, invalidated and losing ones included. Analyses published earlier, from 09/08/2026 to 07/10/2026, sit in the calibration archive: they include analyses published with the outcome already known and do not count in the record. Aggregate figures are always published with the number of analyses and the period next to them, never a percentage on its own.
Publication and corrections
The analyses are produced by Meridian Engine and published by Horizon AI. Who is behind it is on the authors page; how an error in text or data is fixed is in the corrections policy. Every trading decision stays with the reader.
Comparison
Seven checkable criteria: UTC issue time, cited evidence, published losses, fixed exit rule, public check series.
| Meridian Engine | Paid signal channels (typical) | AI analysis tools (typical) | |
|---|---|---|---|
| Direction stated before the move | Yes, with a UTC timestamp | Yes, often without a checkable timestamp | Partly, as summary indicators |
| Rationale with cited evidence | Yes, bar by bar | No | Partly |
| Losses published | Yes, all of them | Rarely, self-reported | Partly, often backtests only |
| Exit rule fixed in advance | Yes, version 1 of 2026-09-26 | No | No |
| Checked on public series | Yes, up to two series | No | No |
| Forex, metals, indices, crypto | All four classes | Yes | Zero to three classes out of four |
| Reader’s decisions | Position size and risk | Often not even these | All |
Source: internal review of 19 operators, updated on 2026-09-19. Every Meridian Engine row can be checked in the public record.
Technical questions
Timeframe, price feed, the 1R risk unit, conditional analyses and the exit rule.
Which timeframe does the engine use?
Several: the timeframe of each analysis is written on its card, together with the horizon of validity. The first analyses in the calibration archive were on H1; since 03/10/2026 every analysis reads D1, H4 and H1. Horizons and exit rule.
Which feed do the prices come from?
The engine reads the bid of the feed named on the card. The outcome is checked on independent public series, for gold COMEX futures (GC=F), with a stated tolerance. Check series on the gold card.
What is 1R and why is it used?
The distance between entry and stop. Measuring in R makes instruments with different stop sizes comparable. On the gold card 1R equals $32.57. Gold card.
What is a conditional analysis?
An analysis that applies only if a trigger fires, for example an H1 close below a level. Without the trigger the outcome is “not activated, 0R”, and the row stays in the record. BTCUSD, 2026-08-09.
Why “no trade” in some cases?
Without confluence there is no directional edge. The analysis publishes the band and the two closes that would invalidate it, and that outcome also enters the record. BTCUSD, 2026-10-01.
What determines the outcome?
The first event within the horizon: target, stop or a closing invalidation condition. If stop and target fall in the same bar, the stop prevails. Exit rule of 2026-09-26.
Does the analysis say how many lots to open?
No. The analyses are the same for everyone and do not take the reader’s account into account: position size and risk stay with the reader. Meaning and risks.
Are the analyses personalised?
No. No information on capital, experience or goals is collected, which is why the analyses are not investment advice. Meaning and risks.
For a question not listed here: contact@usemeridianai.com. The answer is added to this list.
Meaning and risks
What long and short mean, how long an analysis lasts, when it fails and what it is not: Meaning and risks.
Trading leveraged products such as CFDs and forex carries a high risk of losing money rapidly: you can lose all the capital you deposit.
Every Meridian Engine analysis records direction, entry, stop, targets and invalidation before the move; the outcome is checked afterwards on the public price series named on each card, losses included. Waitlist members receive one or two analyses a week by email, free while on the waitlist.
Get the code for your first analysis →Four minutes on how you read the market: the questionnaire decides what we build first. Two minutes: why Meridian Engine is of interest and what you would change.
General analysis, identical for everyone: position size, risk and decisions remain with the reader. Not investment advice.