Why This Site Keeps a Record It Does Not Advertise

The How it works page carries one sentence that most sites in this business would never write: the record of this site's calls is being built, not advertised. It makes no claim about forecasting accuracy, and it will not make one until there are enough resolved calls for the answer to mean something. That is not modesty. It is a description of a machine, and the machine is the interesting part.

This piece is about the machine: what gets written down, when, by what rule it is graded, what it refuses to print, and why a record kept this way is worth keeping even while it says nothing about whether the calls were any good.

Written Before, Graded After

Twice each weekday, at fixed times on the UTC clock, the FX Monitor's directional bias is written to a ledger: one row per rated pair, plus one for the dollar index. Each row carries the direction the monitor was showing at that moment, in the same fixed vocabulary the page itself uses, and the time it was written.

The order matters more than the content. The row exists before anyone knows how the market moved. Its publish-time fields cannot be edited afterwards; the only thing that is ever added to a row is its outcome, and that is set once. Nothing is deleted. A call that turned out badly sits in the ledger with exactly the same standing as one that turned out well, because at the moment of writing nobody knew which it would be.

That is the whole reason to write things down in advance. A record assembled afterwards, from whichever calls the author remembers making, is a story. A record written before the outcome, by a process that does not know the outcome, is evidence about the process, whatever it eventually shows.

No Backfill, and Gaps Shown Rather Than Filled

The second rule follows from the first. If a publication window is missed, because the scheduled run fired late or not at all, that window is skipped. No row is created later to stand in for it, however easy it would be to reconstruct what the monitor would have said. A row written after its own window is a row written with hindsight available, and one such row poisons the standing of every other.

So the record has holes, and the record page lists them: which windows were never published, and why, rather than smoothing over them. A gap that is visible is information about the process. A gap that has been filled is a small lie about it.

One Fixed Rule, Set Before the Fact

Grading is the part people imagine is subjective, and it is the part that is least so. Each row is graded at its own horizon, not whenever someone gets round to looking, against a rule that was fixed and published before the row existed. The rule reads price and returns one of a small set of outcomes: the call was confirmed, it was invalidated, or the horizon passed without a decisive move either way and the row is marked neutral. Once an outcome is written it is never rewritten; running the grader again produces the same answer, which is the property that makes a grader trustworthy.

The site keeps a second record alongside this one: the calls made in its written reports, which name a trigger and an invalidation level and are graded on daily bars over a fixed window. Where a single day's bar touches both levels, the rule resolves the call against the author, because the order of events inside a day is not knowable from daily data and the record is not allowed to flatter itself. The dollar index, whose series carries daily closes only, is graded on closes, so a touch means a close through the level; the page says so in its own words.

The two records are kept apart and shown side by side. They are graded differently, so they are never blended into a single figure, and each one's rule is printed beside its own results.

What the Page Refuses to Print

Most of the record page's design is a list of things it will not do. It will not show a rate for any bucket of calls until that bucket holds a minimum number of graded outcomes; below that it prints a word, accumulating, in place of a percentage, because a rate from a handful of calls is noise wearing the costume of evidence. It will not fold neutral outcomes into the rate to make it look better or worse; they are counted and disclosed beside it. It will not quietly drop rows that cannot be graded, whether because no price was recorded at publication or because the levels were unusable; it counts them, says why, and keeps them out of the rate in the open.

It will not even state its own coverage as a number. The share of published calls the figures are drawn from is described in words, so that a reader cannot mistake a precise-looking fraction for a precise fact. Rows still inside their horizon are shown as open and count toward nothing.

None of those figures appear in this article, and that is deliberate. They change with every window, and the only honest place for a changing number is the page that recomputes it.

Not Audited, but Checkable

The site's disclaimer says plainly that it does not maintain a verified, audited track record, and nothing here contradicts that. What it maintains is a checkable one. The ledger is public data; the record page publishes a fingerprint of the current snapshot, so that anyone holding an earlier copy can tell whether a row has been altered since. That is a weaker guarantee than an audit and a much stronger one than a claim, and it is available to any reader rather than to a firm the site pays.

It is also why the record page is not advertised. It is not indexed by search engines, and no page on the site links to it, and neither is a matter of timing. Both follow from the sentence this piece opened with: the site publishes no performance claim, so it publishes no performance page, and a page that reports rates is a performance page whether or not the author means it as one. That holds for as long as the policy does, and changing it would be a decision, not a drift. The mechanism runs regardless. Rows are written at every window, graded at every horizon, and the day there are enough of them for a rate to mean something, the rate will have been earned by a process that was in place long before anyone looked.

Two Records on This Site, Not One

The E.A.G.L.E. Watch board keeps its own follow-through record, and it is easy to confuse the two. That one counts, for every directional read the sensor issues, whether price kept moving the sensor's way over the next few check-points, and grades itself against a printed chance baseline. It is a record of a sensor's reads. The record described here is a record of published calls, graded against fixed levels at fixed horizons. They share a principle, every read or call counted and none excluded, and share nothing else, which is why neither page quotes the other.

None of this says anything about the next call. A record kept this way cannot; that is the point of it. What it produces is a condition: a body of calls, written before their outcomes, graded by a rule that was fixed in advance, with its gaps and exclusions on display. Whether that body of evidence eventually says the monitor reads the market well, or badly, or no better than chance, is a question the page will answer when it has enough rows, and the reader will be able to check the answer against the file. That is the whole of what is being promised.