Think of systems that observe and then decide how to act - instruments take measurements, models estimate state from measurements, policies map state to action and so on. But the world does not expose variables. It produces reactions. A thermometer distinguishes conditions by expansion. A test particle reveals a field by moving through it. Immune cells identify pathogens through reactions triggered by molecular binding. The latent thing is known through its action on something else.

Response coordinates

Suppose a system maintains a repertoire of behaviors, B = {b1, b2, ..., bn}. Place it in some condition z, and let each behavior produce a response Ri(z). A response can include activation, stability, error, energy use, return, confidence, failure, or time to recover.

Together, those reactions give the condition a coordinate: Phi_B(z) = (R1(z), R2(z), ..., Rn(z)).

If two conditions produce different signatures, the repertoire can distinguish them.

If every available behavior responds in the same way, the conditions are indistinguishable from the system’s present point of view, however different they may appear to an outside observer.

Its effective state space is the partition induced by its possible reactions.

What the reactions reveal

A repertoire can become more than a menu of actions. It can operate as a distributed instrument. The behaviors need not know what they are measuring. In other words, Its members need not estimate the latent state, or even agree on what the state is. Their differential reactions make distinctions available upstream.

This differs from asking several models to estimate the same target and averaging their answers. The informative object is the pattern of agreement, disagreement, activation, failure, and persistence across the repertoire.

One behavior may respond to a fast disturbance. Another may ignore it until the disturbance persists. A third may remain stable under both and fail when a slower constraint changes. Their differences create observability.

The repertoire acts like an eye whose receptors are competencies rather than pixels. Some members are effective actuators. Some are sensitive probes. Some are both. A behavior that is currently a poor action may still reveal that the environment has changed.

Diversity therefore has a practical meaning. A smaller population whose members disagree under consequential conditions may see much more than a bigger one. It is not useful because disagreement is inherently good. It is useful when different behaviors separate conditions that matter. A behavior can be a poor action and still be a good sensor.

Changing the map

When the repertoire changes, a new behavior adds both a possible action and a new response coordinate. Adding a behavior can add a distinction:

Search therefore changes both the policy space and the perceptual geometry.

Simply put, search changes what the system can do, but also what differences it can notice.

This also changes the role of behavioral niches. A niche may preserve a detector for a condition that is rare, transient, or poorly understood. Removing it because it loses on average can erase an axis of perception. A behavior need not be globally competitive to be informationally valuable. Losing a behavioral niche can mean losing a sensor.

Reactions through time

State is not only encoded in which behaviors react, but in when they react, how long they persist, and how they recover. Two conditions indistinguishable at one timescale may separate cleanly at another. Resolution belongs to the observer, not solely to the world.

This connects the map to multi-clock intelligence. Fast behaviors expose local change. Slower processes observe patterns across their reactions. Still slower processes alter the repertoire itself.

A market example

Markets are partially observed and continually changing. Instead of first asking whether the market is trending, crowded, liquid, or mean-reverting, ask what the repertoire can currently get away with.

Passive liquidity may work until informed flow begins to run it over. Continuation may strengthen while exhaustion stops paying. Directional policies may remain profitable even as execution deteriorates. Fast behaviors may reverse while slower ones remain stable.

These reactions form a market-response signature. The system does not need to name the regime first. It can infer useful distinctions from which behaviors activate, fail, persist, or recover.

The market is partly known by what it permits different strategies to do.

An allocator can use this signature to decide where control belongs. A slower process can notice when the repertoire stops separating conditions that matter and search for a behavior that responds differently.

A new strategy can therefore add more than a source of return. It can add a new way of seeing the market.

Trading also exposes an important ambiguity. A behavior can fail because the market changed, because the behavior is inadequate, or because execution destroyed an otherwise useful response. The record must preserve enough lineage to tell these apart.

The repertoire becomes an indirect market model expressed through consequences available to the system itself.

The substrate

A fixed policy searches for the right action. A repertoire can do something more: its reactions construct the distinctions on which later action depends.

Evolution changes both at once. A new behavior may solve a problem. It may also split one previously indistinguishable condition into two.

That is the deeper object here. Not a better regime classifier, and not merely a population of competing strategies, but a substrate in which perception, action, and the map between them can co-evolve.

Markets are the forge because they are difficult to bullshit. Conditions move, other agents adapt, execution intervenes, and every distinction eventually meets consequence.