1. Purpose of this note

This note situates the Agora Commonplace Protocol (ACP) within the current research frontier of:

  • agentic AI systems
  • retrieval-augmented reasoning
  • provenance and auditability
  • scientific AI workflows
  • governance and verification systems

The goal is not to claim novelty prematurely, but to determine whether ACP represents:

  1. a known architecture,
  2. a recombination of existing ideas,
  3. or a structural departure from current systems.

The literature reviewed suggests ACP is closest to a recombination of emerging patterns into a coherent constitutional system, rather than an entirely unprecedented primitive.

However, the particular combination ACP proposes appears rare.

2. The Core Problem ACP Solves

Most current AI systems treat knowledge as generated text.

Typical architecture:

query → retrieval → model → answer

The answer is ephemeral.

Even when grounded in documents, the reasoning chain is rarely stored or governed.

ACP takes a different approach.

ACP assumes that knowledge systems should operate like institutional memory systems, not conversational engines.

Its architecture is:

artifact
   ↓
claim
   ↓
scenario
   ↓
decision_event

This is a governed reasoning stack, not a generation stack.

3. Major Research Directions Adjacent to ACP

Across the reviewed literature, five distinct research frontiers appear.

Each touches part of ACP but not the whole system.

3.1 Claim-Level Reasoning and Verification

Several recent papers move toward claim-level decomposition.

Typical pipeline:

question
 → generate answer
 → decompose answer into claims
 → verify claims
 → update answer

The SimulRAG paper is a strong example.

It proposes:

  • decomposition into atomic claims
  • simulator-based claim verification
  • uncertainty-guided verification
  • selective claim updating

The key idea:

long-form answers should be verified at the claim level, not the document level.

This aligns strongly with ACP's Claim primitive.

However, these systems still produce ephemeral text outputs.

Claims are not:

  • persistent objects
  • governed entities
  • institutional knowledge nodes

ACP therefore extends this pattern by making claims first-class data structures.

3.2 Evidence-Grounded Retrieval Systems

Another large research cluster focuses on RAG architectures.

Standard RAG systems attempt to reduce hallucination by grounding answers in retrieved documents.

However, most RAG pipelines remain limited:

retrieve documents
 → concatenate context
 → generate answer

Recent work attempts to improve this through:

  • simulator retrieval
  • tool invocation
  • iterative retrieval loops
  • agentic orchestration

But these systems still treat evidence as context, not as governed artifacts.

ACP changes this assumption:

Evidence is not temporary context.

Evidence becomes immutable artifact surfaces.

3.3 Provenance and Attribution Systems

A growing research area focuses on model provenance and source attribution.

These papers attempt to answer questions such as:

  • Which data produced this model output?
  • Which sources influenced a response?
  • Can provenance be statistically proven?

Some approaches construct provenance sets that identify the most likely sources of model behavior.

This research is important but operates at a different layer.

These papers address:

model ← training data

ACP instead addresses:

claim ← artifact

This difference is crucial.

ACP governs knowledge production, not model training.

3.4 Agent Governance and Typed Workflows

Another emerging direction attempts to constrain autonomous agents.

These systems introduce:

  • typed action schemas
  • execution validation
  • workflow graphs
  • runtime guards

For example:

plan
 → validate
 → execute
 → verify

These architectures resemble governed execution frameworks.

However they still focus on action correctness, not knowledge integrity.

ACP shares the governance philosophy but applies it to reasoning artifacts rather than actions.

3.5 Multi-Agent Scientific Systems

A particularly interesting research cluster attempts to build AI scientist systems.

These often include:

  • literature retrieval
  • hypothesis generation
  • simulation testing
  • result synthesis

These systems resemble early attempts at automated research workflows.

But they still treat:

  • hypotheses,
  • claims,
  • and conclusions

as temporary outputs.

They are rarely stored as durable knowledge objects.

ACP differs by treating reasoning outputs as institutional records.

4. What the Frontier Is Missing

Across the research landscape, several capabilities appear repeatedly.

But they are rarely integrated.

Common elements:

  • retrieval grounding
  • claim verification
  • uncertainty estimation
  • multi-agent orchestration
  • provenance tracking
  • execution validation

What appears rare is a system that integrates them under one governing substrate.

ACP attempts to do exactly that.

5. ACP’s Distinct Structural Properties

ACP introduces several architectural commitments that are unusual.

5.1 Artifact-First Architecture

ACP begins with artifacts.

Artifacts are immutable surfaces representing evidence.

Every claim must trace to an artifact.

This inverts the standard AI design:

text generation → citations

ACP enforces:

artifact → claim

5.2 Immutable Evidence Layer

Artifacts are append-only.

Once ingested, they cannot change.

This allows:

  • deterministic reasoning
  • reproducible claims
  • stable provenance

This resembles version control systems more than AI pipelines.

5.3 Governed Claim Promotion

Claims move through a lifecycle:

raw
 → triage
 → stable
 → canonical_candidate
 → canonical

This introduces institutional curation, not just automated inference.

5.4 Constitutional Reasoning Primitives

ACP restricts reasoning to four primitives:

ArtifactVersion
Claim
Scenario
DecisionEvent

This creates a bounded reasoning grammar.

Many research systems allow arbitrary structures.

ACP intentionally limits the system’s conceptual vocabulary.

5.5 Advisory-Only AI Agents

ACP explicitly prevents AI agents from altering governance state.

Agents may:

  • read artifacts
  • propose claims
  • generate scenarios

Agents may not:

  • mutate canonical knowledge
  • bypass governance rules
  • modify artifacts

This design philosophy resembles constitutional governance models.

6. ACP’s Docking Harness Concept

One particularly distinctive idea is the Docking Harness (DH).

The DH connects:

GitHub
Atlas
Aalam

Where:

  • GitHub provides code state
  • Atlas performs audit
  • Aalam performs reasoning

Each system operates within strict boundaries.

The harness enforces separation between:

reasoning
verification
execution

Few research systems appear to enforce this separation explicitly.

7. Why Small Teams Can Reach These Ideas

It is reasonable to ask whether a solo developer or very small team could discover these ideas.

The literature suggests that small teams are common in architectural exploration.

Reasons:

  1. Architectural ideas require conceptual coherence more than large compute budgets.
  2. AI-assisted coding tools dramatically amplify individual productivity.
  3. Many frontier ideas emerge through design iteration rather than pure research.

There are documented examples where:

  • a single engineer
  • supported by AI coding assistants

developed complex agent workflows.

Large teams dominate model training.

But architectural design remains accessible.

8. The Likely Future Convergence

The research frontier suggests several trends will likely converge.

Claim-level reasoning

More systems will operate on atomic claims.

Evidence grounding

Models will increasingly rely on external verification surfaces.

Governance layers

Agent systems will introduce stronger constraints.

Provenance tracking

Outputs will include structured provenance graphs.

Institutional knowledge stores

Some systems will begin storing reasoning outputs as durable knowledge.

ACP appears positioned near the intersection of all five.

9. Risks for ACP

Several risks remain.

Over-structuring

Too much rigidity could slow iteration.

Governance complexity

Governance systems often grow difficult to maintain.

Developer friction

Strict primitives may frustrate developers.

Performance overhead

Verification layers can slow reasoning loops.

Phase 6 and Phase 7 are designed to address these issues incrementally.

10. Final Assessment

The literature reviewed suggests:

ACP is not identical to any known architecture.

However, it sits at the intersection of several emerging research directions.

Its distinctive feature is not any single component.

Its distinctive feature is integration.

ACP combines:

  • artifact immutability
  • claim-level reasoning
  • governance enforcement
  • provenance tracking
  • agent orchestration
  • institutional memory

into a single coherent system.

If successful, ACP would function less like an AI application and more like knowledge infrastructure.