Research note · provenance-first

The Correlation-Signal Archetype: borrowing direction from a second reading, until the link breaks

Direction comes from the relationship between two readings, not from either one alone; when that relationship comes apart, the model stands aside rather than guess.

Published Jun 16, 2026

Method archetype

Two flowing lines move together across the left of the frame, a blue leading line and an emerald following line, then drift apart toward the right. — Two readings that move together, until they do not. The model trades the link while it holds.

A correlation-signal model does not read one series and ask what it is doing. It reads two, and asks what they are doing relative to each other. The unit of evidence is a pair, not a point: the thing being measured is the relationship between two readings, not the level of either one on its own.

Neither reading carries the signal by itself. One can look unremarkable and the other can look late, yet together they can be informative, because what the model trades is their joint behaviour. The rest of this explainer follows that idea to its consequences: where the direction comes from, what makes it actionable, and the one event that switches the whole thing off.

Leader and follower: where direction is borrowed from

When the relationship includes a lead, one reading tends to move first and the other tends to follow. A move already visible in the leading reading is then an early indication of where the following reading is likely to go. The model acts on the follower and uses the leader as its reason. The direction is inherited, not generated by the traded reading crossing some level of its own.

The leading reading is the one whose turns tend to arrive first. Its move is treated as information about what has not happened yet in the other reading.

The following reading is the one that tends to repeat the move a little later. On its own it can look slow; paired with the leader, its likely next step is what the model leans on.

A blue leading line and an emerald line that traces the same shape a little later. — When one reading tends to move first, the other tends to echo it; the model reads the leader as an early indication for the follower.

Why the link, not the level, is the signal

This is what makes the family its own archetype rather than a variation on the others. The side the model takes comes from the relationship's implication. It is not one reading ranked against its own history, it is not a single reading reaching an extreme and being let through a gate, and it is not several readings pooled into one score. The second reading is a source of direction carried across a link, never just another term added into a sum.

Two panels side by side: one reading judged against its own past, beside two readings joined by a highlighted link. — The direction is carried across the link between two readings, not read off either reading's own level.

Hold that distinction firmly, because it is easy to lose. The moment the second reading is described as one more input that gets combined, the method has quietly turned into a different archetype. Here the link itself is both where the direction comes from and the thing the model watches.

The window the gap creates

If the leader moves and the follower tends to respond a little later, then the interval between the two is what makes the borrowed direction actionable at all. It is also the part most exposed to practical error, which is why it is necessary but never the whole of the edge.

If the relationship is measured in a way that lets later information leak in, the gap is an artefact and there was never a real window.
If the real gap is short, the cost and delay of acting can be larger than the response the model was trying to capture.
If the relationship was never durable, the window was a coincidence that will not repeat.

Two coupled lines with a deliberately thin shaded interval between a leading move and a following response. — The gap between the leader and the follower is the only window to act, and it is easily emptied by leakage, by cost, or by a link that was never durable.

Leaning by how firmly the two hold together

Because the direction is borrowed, the model is only as committed as the relationship behind it. How hard it leans follows the live state of the link rather than how dramatic either reading looks on its own: a firm link supports leaning into the implied direction, a loosening link supports only a light lean, and a broken link means standing aside.

Link state	What it means	What the model does
Holding	The two readings still move together	Lean into the implied direction
Loosening	The link is slackening	Lean only lightly, if at all
Broken	The readings have come apart	Stand flat

One honest caveat belongs here. These are three plain states, not a precise measured law. Whether a given model grades its conviction smoothly with the strength of the link or steps between a few settings is a design choice that varies from model to model, and this explainer does not claim a specific one.

Three state cards, holding, loosening and broken, each with a two line motif and a lean depth bar. — How hard the model leans follows the live state of the link, shown as three plain states rather than a precise measured curve.

The off-switch: when the two come apart

This is the part that most separates the archetype from a simple bet on one reading. When the two readings decouple, the model goes flat. The trigger is the relationship failing, not the traded reading doing anything in particular, and because the direction was borrowed from the link, losing the link removes the very source of the read.

Two lines coupled then splitting apart at a marked point, with a position lane below dropping to flat at the split. — When the two readings come apart, the source of the direction is gone, so the position falls to flat at exactly that point.

Strong is not the same as durable

The hardest assumption in the whole method is that the relationship is durable, not merely visible right now. Any two readings will appear linked over some stretch by chance, and a tight co-movement in one window says nothing about whether it will hold when conditions change. Strength and durability are different questions, and they need not coincide.

How durability differs from strength, in plain terms

Strength is how closely the two readings move together in the stretch you are looking at. Durability is whether that closeness survives a change in the backdrop. A momentarily tight but fragile link is something to distrust rather than trade, because a coincidence carries no information about direction. Magnitude is used only once the link has shown it holds; it is not a substitute for the link being real.

Two snapshots: one link that holds its shape across a marked change, and one equally tight link that dissolves at the change. — A durable link survives a change in conditions; a coincidence that looked just as tight comes apart when the backdrop moves.

This is also the cleanest line between this archetype and its neighbours. No model that reads a single series has an edge that rests on the stability of a relationship between two of them. That dependence is specific to reading a link.

One relationship recipe, many possible pairs

What makes this an archetype rather than a single strategy is that the recipe does not care which two readings it is given. Read the link, lean while it holds, stand flat when it breaks: point that at one masked pair and it is one model, point it at another and it is a different model with the same discipline. This is a statement about the method's portability, not evidence that any particular pair carries an edge.

A central relationship recipe connected to several anonymous candidate pair slots. — The recipe for reading a link is the reusable part; pointed at a new pair it becomes a new model with the same discipline.

What its output looks like: mostly silence

A usable position needs two things at the same time: a move in the leading reading, and a link that is currently intact and firm enough to trust. With either one missing there is no trade, so a model of this archetype is flat far more often than it is positioned. Long quiet stretches are the normal, expected output of a relationship-conditional signal, not a sign that something has gone wrong.

A long quiet activity baseline broken by a few short positioned segments. — The model is flat far more often than it is positioned; long quiet stretches are the expected output, not a fault.

Grading the evidence honestly

Because the link itself can strengthen, weaken, or come apart over time, a co-movement measured on studied history is only weak evidence that the relationship will still hold ahead. A backtest, a held-back walk-forward, and a single live stretch are three different kinds of evidence, and they are decreasingly certain in that order.

Three bands of decreasing certainty above a staged review path with one masked internal step. — Evidence gets less certain the closer it sits to live, and the model keeps its internal wiring out of view through to staged review.

What we can and cannot claim

This explainer describes the family mechanism, not the wiring of any one model. Which reading leads, how far ahead it tends to move, how the strength of the link is judged, and which pair is used are model internals, and they are not stated here because they are estimated and re-checked for each model rather than fixed constants the article is withholding. A lead and a lag is a statistical relationship, not proof that one reading causes the other.

A method is not a recommendation, and a model built this way is not a guarantee of future results. It is a disciplined way of trading a relationship while it lasts and stepping aside when it does not, and it earns its place only by surviving an idea, a backtest, a held-back test, and a staged review before it is relied on.

Sources

Method family: cross-series correlation and lead-lag signals. The family infers a likely direction for one reading from the move of a related, leading reading, and stands aside when the relationship between them breaks down.
Stonewell One documented model-family taxonomy, the catalogue of distinct model archetypes and how each one forms its directional read.
Stonewell One model lifecycle and staged review process: how a candidate signal moves from an idea, through a backtest and a held-back walk-forward, to a staged review before it is relied on.