# PRD-108 Prior-Art Comparison

## Purpose

This memo frames the closest adjacent systems so PRD-108 can be positioned credibly for investors and advisors.

The goal is not to prove absolute novelty. The goal is to show that PRD-108 occupies a specific, differentiated design point.

## Positioning Summary

PRD-108 should be described as:

* adjacent to blackboard systems
* adjacent to semantic memory systems
* adjacent to field-based memory research
* different in its **specific combination** of:
  * mission-scoped shared field
  * agent-callable tool access
  * query-time resonance dynamics
  * lifecycle integration into orchestration
  * swappable A/B baseline via a common context interface

## Comparator 1: Blackboard Architectures

### Overlap

* shared workspace
* multiple contributors
* independent readers

### Difference

Classical blackboard systems are symbolic and rule-driven. PRD-108 instead uses:

* semantic vector retrieval
* query-time ranking by resonance
* temporal decay
* access reinforcement
* co-access strengthening

### Investor-safe takeaway

PRD-108 is not a brand-new answer to the idea of a shared workspace. It is a modern, semantic, LLM-oriented reinterpretation of that pattern with explicit retrieval and dynamics.

## Comparator 2: CrewAI Shared / Crew-Level Memory

### Overlap

CrewAI documentation describes:

* shared crew memory
* semantic retrieval
* recency-aware ranking
* automatic context recall around tasks

Reference point used for this memo: public CrewAI memory documentation describing crew memory, semantic retrieval, and recency-aware recall behavior.

### Difference

PRD-108 differs most clearly in these areas:

1. **Mission-bounded lifecycle**
   * create, seed, inject, destroy, garbage collect
2. **Common backend abstraction**
   * same calling code can run against Redis baseline or vector-field backend
3. **Explicit resonance formulation**
   * `cosine^2 * decayed_strength`
4. **Co-access strengthening**
   * not just retrieval scoring, but associative strengthening among co-retrieved patterns
5. **Coordinator-level orchestration integration**
   * field creation and cleanup are part of mission control, not just memory lookup

### Investor-safe takeaway

Do not claim that PRD-108 is the first example of shared semantic memory across agents. Claim instead that it is a differentiated orchestration architecture with specific retrieval dynamics and baseline comparability.

## Comparator 3: LangGraph / LangMem-Style State and Semantic Memory

### Overlap

Adjacent systems provide:

* graph or thread state
* semantic search over stored memories
* persistent store abstractions

Reference point used for this memo: public LangGraph / LangChain memory and semantic-search documentation describing shared state, store-backed memory, and semantic retrieval.

### Difference

PRD-108 is more specific about:

* mission-scoped field isolation
* lifecycle ownership by the coordinator
* reinforcement and decay dynamics as part of retrieval behavior
* direct experimental comparison against a message-passing baseline

### Investor-safe takeaway

PRD-108 should be positioned as a **coordination substrate**, not merely a memory store.

## Comparator 4: Mitra, "Field-Theoretic Memory for AI Agents" (arXiv:2602.21220)

### Overlap

This is the closest conceptual prior art:

* field-based memory
* continuous dynamics
* temporal decay
* memory as more than static retrieval

### Difference

Based on the PRD-108 disclosure and a quick external review, the clearest distinctions are:

1. **Mission coordination vs memory enhancement**
   * Mitra focuses on agent memory behavior
   * PRD-108 focuses on multi-agent coordination inside a mission
2. **High-dimensional shared field used directly for orchestration**
   * not just memory dynamics in abstraction
3. **Tool-facing operational interface**
   * agents call query/inject/stability tools during execution
4. **Coordinator integration**
   * field lifecycle is managed by mission execution
5. **Direct baseline comparability**
   * Redis message passing and vector field sit behind the same port

### Investor-safe takeaway

If challenged on originality, this is the safest answer:

> The closest field-based memory work is conceptually adjacent, but PRD-108 is differentiated as a mission-scoped, multi-agent orchestration system with lifecycle control, agent-callable interfaces, and direct A/B comparability against message passing.

## Comparator 5: Generic RAG / Vector Database Patterns

### Overlap

* vector store
* embeddings
* similarity search

### Difference

PRD-108 is not just "RAG for agents." It adds:

* mission scoping
* temporal decay
* reinforcement on access
* co-access strengthening
* archival thresholding
* field stability measurement
* lifecycle ownership by the orchestrator

### Investor-safe takeaway

The right comparison is not "we use vectors." The right comparison is "we turned vector retrieval into a shared coordination field with dynamics and operational lifecycle."

## What To Say

Use:

> PRD-108 combines ideas from shared workspaces, semantic retrieval, and field-like memory dynamics into a mission-scoped coordination layer for multi-agent LLM systems.

Avoid:

> No one has ever built shared semantic memory for agents.

## What Strengthens This Position Further

The fastest credibility upgrades would be:

1. a short external prior-art memo from counsel or advisor
2. a reproducible A/B appendix
3. one end-to-end mission run showing the field used in real orchestration

## Bottom Line

PRD-108 is best defended as a **novel combination in a specific applied setting**:

* shared semantic field
* orchestration-aware lifecycle
* reinforcement dynamics
* mission integration
* direct baseline comparison

That claim is credible, narrow, and materially useful in investor conversations.