A temporal reading

Memory. Context. Understanding.

State is what a running system holds. Read in time, the things state holds are three: what was, what is, what will be. Cognitive science has names for each, and the names are precise. The state stack’s nine axes turn out to be where those names live.

past + present + future = memory · context · understanding

The temporal reading · three functions of state, all three from nine axes

The reading

Three functions, sixty years of formalism.

Each temporal function has a canonical decomposition in the cognitive science and AI literature, and each decomposition lands on an intersection of axes the stack already names. The reading does not add a dimension; it makes a structural pattern across the existing ones visible.

01 · PAST

Memory

what the system retains

Memory is state about state — the system's record of where it has been, retained into the present so the present has something to be relative to. The canonical analysis (Atkinson and Shiffrin 1968; Tulving 1972; Squire 1992; Baddeley 1974, 2000) decomposes this into four distinct stores.

Canonical types

Working memoryBaddeley & Hitch 1974 · central executive + buffers

The active buffer of what is currently being held and operated on. Bounded, fast, dissolved at the end of the task.

Episodic memoryTulving 1972

Specific events located in time and place. The system’s record of what happened, when, and where.

Semantic memoryTulving 1972

Decontextualized concepts and facts. The system’s store of what is, independent of when it learned it.

Procedural memorySquire 1992

How-to. Skills and action patterns. Often held implicitly, exercised through performance rather than recall.

Where it lives in the stack

Lifetime. Episodic and semantic memory require persistence beyond the run — versioned or durable values. Working memory lives at the ephemeral end.

Composition. History pseudostates in hierarchical state machines (Harel 1987) are the formal compositional mechanism for retained past: a super-state remembers which sub-state was last active when re-entered.

Isolation. Whose past is readable. Encapsulated state hides its history; message-passing state publishes only what it chooses.

Formalism. Each memory type has a substrate: event logs and episodic vectors for episodic; knowledge graphs and ontologies for semantic; rule bases and behavior trees for procedural; working buffers for working.

Memory without context is replay. The record exists, but nothing in the present can act on it because the surround it belonged to has been lost.

02 · PRESENT

Context

what frames the step

Context is the present taken seriously — the active state plus its surround: the parallel concerns running alongside, the ambient scope, the events the system is admissible to. The formal tradition runs from Minsky’s frames (1974) and Schank’s scripts to McCarthy and Hayes’s frame problem (1969), and on to Clark and Chalmers’s extended mind (1998) and Hutchins’s distributed cognition (1995).

Canonical types

FramesMinsky 1974

A chunk of structured knowledge that captures a stereotypical situation. Slots, default fillers, expectations. The present is read as an instance of a known kind.

ScriptsSchank & Abelson 1977

Frames for sequences. A canonical ordering of expected events. The script is the context that says what comes next.

The frame problemMcCarthy & Hayes 1969

The foundational logical problem of representing what stays the same when something happens. The unspoken rest-of-context is the hard part.

Extended cognitionClark & Chalmers 1998; Hutchins 1995

Context is not bounded by the system’s shell. Notebooks, tools, artifacts, and other agents participate in the present the system is responding to.

Where it lives in the stack

Composition. AND-parallel composition is the formal mechanism for holding several contexts at once — the orthogonal region that runs alongside the main concern is the present’s width.

Isolation. The boundary decides what is in scope. Shared memory dissolves context; message-passing makes context an explicit handoff.

Reactivity. What the present admits as a stimulus. A closed-loop system has a narrow context; a reactive one is wide; an autonomic one selects its own.

Formalism. Working memory in the Common Model of Cognition (Laird, Lebiere & Rosenbloom 2017) is the canonical realization. Modern context windows and retrieval-augmented stacks are the contemporary version.

Context without memory is myopia. The system responds to what is in front of it with no sense of whether this is the first time, the tenth time, or a recovery.

03 · FUTURE

Understanding

what the system reasons about

Understanding is state about the trajectory — the system’s own model of what it will become and what would be invariant across that becoming. The predictive-processing tradition (Friston 2010; Clark 2013, 2015) treats this as the central operation: maintain a generative model, predict forward, attend to error.

Canonical types

Cognitive mapTolman 1948

The organism’s internal map of its environment. The original pre-formal name for the model that lets a system imagine reachable states.

Generative modelpredictive processing tradition

A model that produces predictions which can be checked against incoming evidence. Perception is top-down inference; learning is error-driven update.

Active inferenceFriston (free energy principle)

The system acts to bring the world into alignment with its predictions, not only updates its predictions to fit the world. Action and inference are one mechanism.

Procedural understandingCommon Model of Cognition (Laird et al. 2017)

Rules with utility metadata governing what to do next. The engineering version of "knowing what comes." Embodied in ACT-R, Soar, CLARION, Sigma, LIDA.

Where it lives in the stack

Self-reference. The machine holds a model of itself as a value. Reflective state is the prerequisite for reasoning about what next.

Reactivity. Autonomic reactivity is the machine choosing its own next move from that model. Below autonomic, the future is what the environment hands the system.

Verifiability. Model-checked invariants over reachable states make understanding rigorous — the future the machine reasons about is one it can prove things over.

Formalism. Cognitive architectures (ACT-R, Soar, CLARION, Sigma, LIDA) and Hierarchical Temporal Memory (Hawkins) are the substrates the field has built for the future-reasoning function.

Understanding without memory drifts. The system makes claims about its trajectory with no record of whether earlier claims held.

Where all three hold

The temporal cell.

Memory and context are widely held in production systems — databases retain the past, working buffers frame the present. Understanding is rarer. The cell where all three are first-class — where the past is retained, the present is held with structure, and the future is a runtime artifact rather than a guess — is the cell the predictive face already names. The convergent characterization in the literature is the Common Model of Cognition (Laird, Lebiere & Rosenbloom 2017), the abstract architecture under ACT-R, Soar, CLARION, Sigma, and LIDA.

The reflective hierarchical controller

OR-hierarchical composition holds the present; reflective self-reference reasons over what next; checkpointed lifetime retains the past. All three at L₀.

on the predictive face →

The TLA⁺-runtime-coupled machine

The specification is part of the runtime state — understanding the future is literally a runtime artifact, refused-into rather than guessed-into.

on the predictive face →

P10

The verified autonomic state machine

Dynamic-structure composition (memory of every prior shape), autonomic reactivity (understanding what to be next), model-checked verifiability (proven across the becoming).

on the predictive face →

The structural reading

The cell the genealogy is walking toward is the cell where state holds its past, structures its present, and reasons about its future as one running object. Each era opened an axis the prior one had not yet named; the temporal cell is the conjunction the axes already make available, occupied only in part by what has been built so far.

Across the family

The same reading recurs.

The temporal functions are not state-specific. They have counterparts on the sibling stacks, and the shared axes carry the correspondence without translation.

Memory in state ↔ the temporal axis (τ) of the data stack — how records hold history.
Context in state ↔ the workload zone (W) and verifiability (V) of the data stack — what surround a record is read against.
Understanding in state ↔ the encoding regime (R) of the synthesis stack — what a system’s substrate actually encodes about its trajectory.

Because the shared axes agree, the temporal reading travels. The cell that is whole in state has counterparts in the other stacks, and a primitive that lights up all three functions there will light up the same coordinates here.

Continue

From the reading to the cells.

The predictive face References & citations The research note