Activation Is Not Querying

Querying and activation are different. A context window, vector index, or knowledge graph tells you where data lives and how to search it. They do not tell you which past experiences should influence the system right now.

Querying explicitly searches stored records. Activation models availability: which memories should exert influence based on goals, history, permissions, evidence, and current state.

The Retrieval Question Is Underspecified

Standard retrieval-augmented generation starts with a query $q$, stored items $M = \{m_1, \ldots, m_n\}$, and a similarity score:

$$s_q(m_i) = \operatorname{sim}(f(q), f(m_i)).$$

The system returns the top-$k$ items. Graph RAG expands this from passages to nodes, subgraphs, or summaries.

These methods improve search, but the mathematical object remains mostly query-conditioned relevance. A record can rank high whether it was verified yesterday or superseded last year, whether it represents a direct fact or a summary, whether it has repeatedly supported successful action or has merely been retrieved many times, and whether it aligns with current goals or merely shares vocabulary.

Query similarity misses a key variable: memory availability based on system history and state.

We call that variable activation.

Activation versus querying

Querying and activation are complementary operations over different conditionals.

Querying estimates:

$$P(m_i \text{ is relevant} \mid q).$$

Activation estimates:

$$P(m_i \text{ should influence inference now} \mid q, g, H_t, G, \pi, \sigma_i),$$

where $g$ is the active goal, $H_t$ is access history, $G$ is graph structure, $\pi$ is policy, and $\sigma_i$ is epistemic state (source, confidence, verification, and decay).

Dimension	Query-Conditioned Retrieval	Activation-Conditioned Memory
Core question	What matches this request?	What past experience should be active now?
State	Request-local	History-, goal-, and task-dependent
Unit	Passage, row, or node	Entity, episode, procedure, edge, or path
Temporal behavior	Static between indexing runs	Strengthens, decays, and updates with use
Structural behavior	Ranks candidates independently or by neighborhood	Allocates traversal budget across connected evidence
Epistemic behavior	Often delegated to downstream generation	Carries provenance, contradiction, and verification into selection
Output	Ranked context list	Bounded evidence graph

Vector scores are inputs to a memory policy, not the policy itself. Querying finds candidates. Activation decides which candidates shape action.

Stateful Activation

Our architecture maps experience as a graph of goals, episodes, chunks, facts, decisions, source documents, and agent actions. Edges capture temporal order, causality, containment, contradiction, update, provenance, derivation, similarity, assignment, and dependency.

Activation combines query relevance with recency, qualified successful use, salience, goal alignment, source authority, contradiction status, and policy constraints. This is an engineering use of cognitive-memory ideas such as ACT-R base-level and spreading activation, not a claim that an emergent memory system faithfully simulates human cognition.

Hard permission checks run before content scoring to prevent data leaks. The useful research property is auditability: when a memory enters context, the system should be able to explain whether it arrived because of semantic match, temporal state, repeated successful use, a causal relation, an active goal, or source authority.

Activation Guides Graph Search

Activation controls where retrieval goes next.

Starting from query cues, activation allocates a bounded search frontier across the graph. Traversal should account for both evidence value and cost, including the tendency of broad, high-degree cues to consume retrieval budget. Nodes such as "customer", "roadmap", "Q3", or "incident" may connect to thousands of memories. Their raw connectivity makes them easy to retrieve and weak for discrimination unless graph fan-out is controlled.

Retrieved nodes generate new cues, letting the system find supporting evidence whose text does not match the original prompt.

The output is an evidence graph preserving nodes, relations, provenance, and competing branches before inference begins.

Example: Decision Lineage

Consider: Why was Project X canceled?

Embedding search might pull up the cancellation notice, a meeting transcript containing "Project X", and status reports semantically close to "canceled." Those text snippets might support a fluent answer while missing the root cause.

Activation search starts at "Project X" and "canceled," then follows decision lineage across hops:

$$\text{account loss} \rightarrow \text{forecast revision} \rightarrow \text{budget cut} \rightarrow \text{cancellation}.$$

The LLM receives this causal chain with source links. Reconstruction happens during retrieval, so the model does not have to guess cause from loose text fragments.

This example illustrates the key research claim: for questions requiring temporal, causal, procedural, or decision lineage, the relevant unit is often a path under a goal-conditioned memory policy, not a passage under a similarity metric.

What Follows Analytically

Three design-level conclusions follow from the formulation.

1. Activation can bound a graph frontier

Naive breadth expansion over a graph grows quickly with branching factor and depth. Activation does not abolish graph-search complexity; it supplies a stateful pruning policy. A fair implementation must therefore report traversal work directly, including graph operations, latency, cache behavior, and the size of the retained frontier.

2. Evidence density matters more than token count

An evidence graph is not guaranteed to be shorter than a flat list of retrieved chunks. Its proposed advantage is higher evidence density: more causal, temporal, and epistemic structure per token, with less duplicated passage text. A fair evaluation must report both token cost and support coverage.

3. Reinforcement introduces a feedback-control problem

Reinforcement can be inexpensive computationally and dangerous statistically. If retrieval itself increments future availability, an early false positive can become self-reinforcing.

Reinforcement should therefore depend on qualified outcomes: citation, user confirmation, successful action, or later consistency. Negative evidence and supersession must reduce authority without destroying lineage. In this architecture, forgetting is not deletion; it is controlled loss of retrieval authority.

Testable Hypotheses

The architecture should earn its complexity empirically. We propose comparing five systems under matched generation models and context budgets:

1. lexical retrieval;
2. dense or hybrid top-$k$ retrieval;
3. graph retrieval without activation;
4. activation reranking over a fixed candidate pool;
5. activation-guided traversal with recursive cue expansion and evidence reconstruction.

The strongest evaluation tasks are those where flat similarity is plausibly insufficient: organizational "why" questions, decision-history reconstruction, incident postmortems, temporal supersession, customer-commitment recall, procedural next-action selection, and contradiction detection.

Primary metrics should include answer accuracy, evidence-node recall, causal-edge F1, temporal-order accuracy, contradiction recall, provenance completeness, stale-memory false-positive rate, and calibration of answer confidence against evidence support. Systems should also report retrieval latency, graph operations, serialized context tokens, and human audit time.

The critical ablations should test activation state, graph traversal policy, evidence reconstruction, authorization, and reinforcement. The claim is falsified or narrowed if a simpler graph reranker matches full traversal on causal and temporal recovery at equal cost.

Failure Modes And Boundary Conditions

Activation is not intrinsically safer or more accurate than querying. It creates state, and state creates path dependence.

• Popularity lock-in: frequently retrieved memories can crowd out rare but decisive evidence.
• Stale authority: recency and repetition can preserve an obsolete decision unless supersession edges carry enough weight.
• Reflection laundering: model-generated summaries can acquire the authority of direct observations if lineage is lost.
• Fan overcorrection: strong attenuation can suppress broad concepts that are genuinely diagnostic for a task.
• Goal capture: an incorrect active goal can steer traversal away from relevant evidence.
• Permission leakage: scoring an inaccessible node before authorization may reveal that it exists, even if its content is withheld.
• Audit overload: a large evidence graph may be more inspectable in principle but less usable in practice.

These are reasons to expose activation contributions, provenance, and path history as first-class outputs. A memory system that cannot explain why an experience became available is difficult to calibrate and unsafe to reinforce.

Relation To Adjacent Work

This proposal does not claim to invent activation, graph retrieval, hierarchical memory, reflection, or retrieval-augmented generation. ACT-R provides a mature account of base-level and spreading activation. RAG established retrieval from non-parametric memory for generation. GraphRAG demonstrated graph-based, query-focused summarization. RAPTOR developed recursive tree-organized retrieval. Generative Agents combined memory streams, reflection, and planning. MemGPT framed LLM memory management through an operating-system analogy, and Zep developed temporal knowledge-graph memory for agents.

The proposed contribution is the ordered runtime combination: goal- and history-conditioned activation; fan-aware traversal of a hierarchical organizational memory graph; recursive cue generation; evidence-graph reconstruction before language-model inference; and auditable reinforcement after use. The patent context is relevant because it fixes the claimed engineering boundary. The research question remains open because the value of that combination must be established through controlled comparison.

Summary

Querying asks which stored items match a prompt. Activation decides which past experiences should shape current behavior.

For short-lived question answering, that distinction may add little. For agents and organizations expected to accumulate experience, revise beliefs, preserve decision lineage, and operate under policy over months or years, it may be the difference between a searchable archive and a memory system.

---

Publication

Marvin Danig, Activation-Guided Graph Retrieval for Cognitive Memory Reconstruction in Language Model Systems (2026). The work is associated with U.S. provisional patent application 64/084,217, Hierarchical Activation-Guided Graph Retrieval for Cognitive Memory Reconstruction in Large Language Model Systems, filed June 6, 2026.

Copyright 2026 Achiral. All rights reserved. No patent license is granted by publication of this article.