Fifteen terms, each defined in the PCT sense — not the engineering sense where the two differ — with a link to the page or section where the term does its work. Written and maintained by Łukasz Diener.
This page is a dictionary, not an argument. Each entry states the term in the sense it carries on this site, and points to the page or section where it is developed. The point is that a reader — human or automated — can lift a single definition out and use it, without needing the rest of the page.
Where a term has both a PCT sense and an engineering sense — and several do — the PCT sense is given, and the collision is named. Terms marked (Diener) are extensions of the framework, not part of Powers' original work; they are attributed accordingly.
A theory of behaviour according to which organisms act to control their own perceptual input against disturbance, rather than to produce responses to stimuli. The controlled quantity is a perception, not an output; the means of controlling it — whatever actions happen to work — is what an outside observer sees as behaviour. Developed by William T. Powers (1926–2013) across Behavior: The Control of Perception (Aldine, 1973) and a series of later works.
See also Theory · Powers & history · Controlled variable
The value at which a perceptual signal is to be held. In PCT the reference signal is internal to the system and autonomous: it is generated by a higher-order loop within the same hierarchy or, at the top of the hierarchy, is intrinsic. Nothing outside the system is a reference signal. External artifacts — files, records, source documents — are not references; they are part of the environmental basis of perception.
See also The loop in four terms · Substrate formulation · Perceptual signal
The system's internal representation of a controlled variable, produced by an input function operating on whatever the system can sense. It is not the variable itself. A control system never has access to the world; it has access to \( p \), and the loop acts on the difference between \( p \) and the reference.
See also The loop in four terms · Reference signal · Error signal
The output of the comparator: the difference between the reference signal and the perceptual signal. In PCT it is the entire output of the comparator and it drives the output function. There is no additional processing between error and action, and no separate representation of what caused the error.
See also The loop in four terms · PCT vs LQR · Comparator
The element computing \( e = r - p \), the difference between the reference signal and the perceptual signal. In a control hierarchy the comparator sits at every level. In a verification loop with no oracle — the architecture specified in No Argument for the World — the comparator sits outside the effector and inside the system boundary, not in the weights and not in the gradient.
See also The loop in four terms · Comparator architecture · Reference Signal Engineering
A quantity that a system acts to hold at a particular value despite disturbances. It is identified not by inspecting the system's internals but by disturbing candidate variables and observing which ones are restored. A system's controlled variables are frequently not the ones its designers intended, and rarely the ones its outputs advertise.
See also Core principles · Test for the Controlled Variable · PCT in AI
An experimental procedure to determine which variable an agent is actually controlling. Three conditions are required, and all three are necessary: resistance (a disturbance to the candidate is opposed), specificity (disturbances outside the candidate are not compensated), and invariance (compensation is invariant across physical channels). Resistance alone is insufficient, because a reflex arc resists a disturbance without controlling anything.
See also TCV protocol · Controlled variable · Audit kit
PCT's account of learning: a random-walk restructuring of the control hierarchy driven by sustained intrinsic error, not by external reward. When persistent error cannot be eliminated through normal control action, the system randomly varies its own parameters until the error drops. The specific neural mechanisms remain an active research question and are not fully specified.
See also Core principles · The 11 levels
The open-loop gain of a PCT control loop, equal to the product of the input gain, output gain, and feedback gain: \( G = K_i \cdot K_o \cdot K_f \). As \( G \to \infty \), the loop drives the controlled perception to the reference and rejects additive disturbance; the closed-loop transfer takes the form \( p = \frac{G}{1+G}\,r + \frac{K_i K_d}{1+G}\,D \). High gain is the engineering reason a controller with no plant model can outperform one that has it.
See also PCT vs LQR — derivation · Substrate formulation
A generalisation of PCT's classical linear form to an environment that is shared, stateful, and finite-capacity: \( p_i = F_i(q_1, \dots, q_n, x) \), where \( x = [C, \rho, \tau]^{\mathsf{T}} \) encodes capacity, load, and transport delay. In the limit \( \rho \ll C \), \( F_i \) is locally affine in \( q_i \) and the classical form \( p = M \cdot q + d \) is recovered as a first-order Taylor expansion. Diener's extension, developed in the theory section.
See also The loop as an equation · Differential gain collapse · Clean-display regime
The uniform vanishing of the marginal perceptual return on output effort, \( \sup_{q_i} |\partial F_i / \partial q_i| \to 0 \), as a shared substrate approaches saturation \( \rho \to C \). It is mathematically distinct from additive disturbance: under disturbance the loop still closes because \( \partial p / \partial q > 0 \); under collapse the loop is decoupled because the derivative vanishes. The controller cannot compensate by increasing \( q \), because increasing \( q \) no longer moves \( p \). Diener's extension.
See also The collapse condition · Three empirical calibrations · Substrate formulation
The regime in which the classical PCT form \( p = M \cdot q + d \) is exact: stateless, effectively infinite-capacity environment with no interaction between agents, so that \( M \) is constant, positive, and bounded away from zero on the operating range. Powers' tracking experiments were run in this regime. The name locates the boundary of Powers' original mathematics and states, without criticism, where it stops applying.
See also The loop as an equation · Substrate-limited PCT · Differential gain collapse
The name for the whole extension of PCT when it must be referred to with one term: the classical theory plus the substrate formulation and the differential gain collapse analysis. It is not a replacement for PCT; it is the same theory in a regime where the environment has finite capacity and can run out.
See also The loop as an equation · Gain collapse · Calibrations
The architecture for closing the verification loop around a language model without an oracle, specified in No Argument for the World (§8). The system does not control for the claim is true — a perception unavailable to it — but for the claim and an independently obtained record agree. The construction is that of double-entry bookkeeping: neither record is privileged, and control is exercised over the perception of their agreement. Claims for which no independent channel exists are marked as unchecked rather than passed through.
See also No Argument for the World · Comparator architecture · Reward hacking
A verification structure in which no channel is privileged as "the truth". Pairwise residuals \( \rho_{ij} = d(c_i, c_j) \) are computed between channels, and the claim is accepted when \( \max \rho_{ij} \le \tau \) for a declared tolerance \( \tau \). It is the operational form of Reference Signal Engineering, developed at scale in The Metric Was Green.
See also The Metric Was Green · Reference Signal Engineering · Gain collapse
Perceptual Control Theory was created by William T. Powers (1926–2013) and developed further by the PCT research community (Marken, Mansell, the IAPCT). Terms marked (Diener) are Łukasz Diener's own extensions of the framework — the substrate formulation, the differential gain collapse, Reference Signal Engineering, and the comparator architecture. The rest of the glossary defines terms in Powers' sense. His full work is set out in six open-access audits with permanent DOIs.