AETRE & The Governed Agent

AETRE & The Governed Agent: Unified Decision-Theoretic Operating System

"Governing Autonomous Intelligence in the Age of Abundance."
A high-performance, mathematically unified decision-theoretic operating system uniting macroeconomic proposal/portfolio triage (AETRE) and microeconomic execution governance (The Governed Agent).

Release status: experimental public alpha. The software and mathematical simulations are testable, but the demonstration records are synthetic and do not establish prospective effectiveness in a live conference, grant, or investment workflow. Use outputs as decision-support diagnostics, not as autonomous acceptance, rejection, funding, or investment decisions.

Based on the research series by Clayton Gray (2026):

1. The Innovation-Absorption Gap: How Artificial Intelligence Can Accelerate Idea Production Faster Than Complementary Institutions Adapt
   Clayton Gray (2026a) — SSRN: 7161458
2. The Admission Frontier: An Economically Regulated Decision-Theoretic Runtime and Fail-Closed Verification Gate for Autonomous Agents
   Clayton Gray (2026d) — distributed within the replication bundle The Implementation Frontier: Capital Allocation, Deliberative Stopping, and Verified Agent Gatekeeping (Zenodo: 10.5281/zenodo.22814799)

This software is archived under its own DOI, separate from the papers above: AETRE: Adaptive Epistemic Triage & Recall Engine (Zenodo: 10.5281/zenodo.22098366).

---

The Problem: The Innovation-Absorption Gap

When Artificial Intelligence makes idea and action generation cheap ($c_{\text{gen}} \to 0$), proposal and execution volume ($N$) explodes. However, downstream evaluation, laboratory validation, code review, and human gatekeeper capacity ($K$) remain strictly finite.

This creates three critical pipeline pathologies:

1. The Kingman Delay Explosion: When evaluator utilization $\rho = \lambda / \mu$ approaches saturation ($\rho > 0.85$), wait times shoot up non-linearly according to Kingman's Heavy-Traffic equation: $$E[W_q] \approx \frac{\rho}{1-\rho} \cdot \frac{c_a^2 + c_s^2}{2} \cdot \frac{1}{\mu}$$
2. The Asymmetric Payoff Trap: In heavy-tailed domains like venture capital, breakthrough discovery, and agentic code patches (Pareto index $\alpha \approx 1.25$), consensus-seeking scoring systems penalize high-variance, transformative outliers in favor of safe, incremental proposals.
3. The Finite-Capacity Recall Ceiling (Proposition 1): Without active epistemic triage, true breakthrough recall asymptotically decays towards zero as arrival rates surge: $$R_N \le \min\left(1, \frac{K_N}{H_N}\right) \to 0 \quad \text{as } N \to \infty$$

---

Two-Layer Decision-Theoretic Architecture

AETRE unifies Macroeconomic Pipeline Triage (managing institutional review bandwidth and portfolio recall) with Microeconomic Execution Governance (safeguarding agentic execution runtimes and pull-request verification).

========================================================================================
 MACRO LEVEL: Institutional Proposal & Portfolio Triage (AETRE / Gray 2026a)
========================================================================================
                     Incoming Submissions / Dealflow (N)
                                     │
                                     ▼
        ┌─────────────────────────────────────────────────────────┐
        │ 1. Bayesian Value-of-Information (VOI) Engine           │
        │    - Closed-form normal VOI & Pareto heavy-tailed VOI   │
        │    - Direct-Pass, Fast-Drop, or Deep-Review allocation  │
        └─────────────────────────────────────────────────────────┘
                                     │
                                     ▼
        ┌─────────────────────────────────────────────────────────┐
        │ 2. Kingman Heavy-Traffic Capacity Governor              │
        │    - Dynamic queue throttling when ρ → 1.0              │
        │    - Preserves reviewer quality; deters burnout         │
        └─────────────────────────────────────────────────────────┘
                  │                                     │
                  ▼                                     ▼
        [ Admitted Cohort (K) ]             [ Exploration Audit Pool ]
        Optimal Conviction Allocation       Horvitz-Thompson H_hat_D Unbiased Audit

========================================================================================
 MICRO LEVEL: Autonomous Execution & Verification Gate (Governed Agent / Gray 2026d)
========================================================================================
                     Autonomous PRs / Candidate Actions
                                     │
                                     ▼
        ┌─────────────────────────────────────────────────────────┐
        │ 3. Tripartite Review Boundary & Bellman DP (Road A)     │
        │    - Computes Expected Value of Verification (V*)       │
        │    - Classifies AUTO_EXECUTE, REQUIRE_REVIEW, or REJECT │
        │    - Fail-Closed: Rejects on malformed input/divergence │
        └─────────────────────────────────────────────────────────┘
                                     │
                                     ▼
        ┌─────────────────────────────────────────────────────────┐
        │ 4. Heterogeneous cμ-Rule Knapsack Controller            │
        │    - Value density sorting: ρ_i = Δu_i / c_i            │
        │    - Budget-constrained admission under shadow prices   │
        └─────────────────────────────────────────────────────────┘
                  │                                     │
                  ▼                                     ▼
        [ Auto-Merged / Dispatched ]        [ Escalated to Human Gatekeeper ]

---

Repository Structure

.
├── Cargo.toml                  # Workspace manifest (AGPL-3.0)
├── crates/
│   ├── aetre-core/             # Pure Rust decision engine:
│   │   ├── src/voi.rs          #   - Bayesian VOI & heavy-tailed Pareto
│   │   ├── src/governor.rs     #   - Bellman DP & Tripartite Review Boundary (Road A)
│   │   ├── src/knapsack.rs     #   - Heterogeneous cμ-rule knapsack controller
│   │   ├── src/queues.rs       #   - Kingman heavy-traffic capacity governor
│   │   ├── src/audit.rs        #   - Horvitz-Thompson exploration audit
│   │   └── src/staking.rs      #   - Super-linear anti-sybil staking
│   ├── aetre-cli/              # Native CLI for simulations, bounds & backtests
│   └── aetre-mcp/              # Model Context Protocol server (24 tools, 4 resources, 3 prompts)
├── python/
│   └── governed_agent/         # Python reference runtime for agent execution governance
├── scripts/
│   └── pre-commit-governed-gate.py  # Standalone pre-commit verification gatekeeper
├── tests/                      # Python unit & regression suite (79 tests)
├── benchmarks/                 # Verification benchmarks and institutional queue sweeps
├── examples/
│   ├── datasets/               # Held-out review and dealflow test splits
│   ├── proposals.json          # Benchmark evaluation candidates
│   └── mcp_config.json         # Claude Desktop & Cursor connection template
├── .pre-commit-hooks.yaml      # Pre-commit hook definition for git integration
├── .github/workflows/
│   ├── ci.yml                  # Rust & MCP server automated verification
│   └── governed-gate-template.yml # Reusable GitHub Actions agent PR gating workflow
├── CITATION.cff                # Dual academic citation metadata
├── Dockerfile                  # Production container definition
├── fly.toml                    # Serverless Cloud deployment config
├── DATASETS.md                 # Fixture provenance and third-party data guidance
├── LICENSE                     # GNU Affero General Public License v3.0 text
├── LICENSING.md                # AGPL licensing and publication status
└── README.md

---

Quickstart & CLI Usage

1. Run the Rust Test Suite & Verification

cargo test --workspace
cargo clippy --workspace --all-targets -- -D warnings

2. Run the Python Reference Governance Suite

python -m unittest discover -s tests

3. Run the Standalone Governed Agent Pre-Commit Gate

Fast, fail-closed verification gate for agentic code modifications:

python scripts/pre-commit-governed-gate.py --all-files

4. Run the Macro Monte Carlo & Dealflow Benchmarks

# Multi-regime academic triage simulation (500 replications)
cargo run -p aetre-cli -- benchmark --replications 500

# Venture Capital Pareto dealflow benchmark (10,000 deals, α = 1.25)
cargo run -p aetre-cli -- vc-benchmark --deals 10000 --budget 100 --alpha 1.25

# Theoretical Proposition 1 recall ceiling bound
cargo run -p aetre-cli -- bound --arrivals 5000 --capacity 200 --high-rate 0.067 --csv

5. Simulate Institutional Heterogeneous Agent Queues

Simulates density-greedy $c\mu$-rule knapsack vs. FIFO across varying institutional review capacities:

python evaluate_institutional_queues.py

---

Model Context Protocol (MCP) Integration

AETRE provides a high-performance native JSON-RPC 2.0 Model Context Protocol (MCP) server implementing 24 Tools, 4 Resources, and 3 Pre-Configured Prompts for Claude Desktop, Cursor, and autonomous agent sidecars.

Configuration (Claude Desktop / Cursor)

Install the binary (cargo install aetre-mcp), then add to your claude_desktop_config.json. The server speaks stdio by default:

{
  "mcpServers": {
    "aetre": {
      "command": "aetre-mcp"
    }
  }
}

Local HTTP / Container Mode

cargo run -p aetre-mcp -- --serve --headless

HTTP mode binds to 127.0.0.1:8080 by default and does not enable cross-origin browser access. For container deployment, set AETRE_BIND_ADDRESS=0.0.0.0 and set a strong AETRE_HTTP_SERVER_TOKEN. Non-loopback startup fails closed when that token is absent. POST clients must send it in the X-AETRE-Server-Token header. Also place the service behind a TLS reverse proxy. The bundled Dockerfile supplies the bind address and runs as a non-root user.

Comprehensive 24-Tool Catalog

Macroeconomic Pipeline & Portfolio Tools (AETRE)

1. aetre_system_catalog: Diagnostic catalog of registered tools, resources, and algorithms.
2. aetre_triage_proposal: Full three-stream triage routing (FAST_DROP, VOI_QUEUE, AUTO_PASS).
3. aetre_calculate_voi: Closed-form Gaussian Value-of-Information ($V^*$) calculation.
4. aetre_check_governor: Kingman heavy-traffic queue delay ($E[W_q]$) forecasting and capacity throttling.
5. aetre_exploration_audit: Horvitz-Thompson unbiased discovery rate estimator ($\hat{H}_D$) for rejected pools.
6. aetre_evaluate_staking: Anti-sybil quadratic staking schedule for incoming proposals.
7. aetre_proposition_1_bound: Asymptotic recall bound ($R_N \le \min(1, K_N / H_N)$) verification.
8. aetre_correlated_posterior_update: Multi-agent reviewer consensus correlation debiasing.
9. aetre_heavy_tailed_voi: Pareto power-law ($\alpha \approx 1.25$) expected value of information for extreme outcomes.
10. aetre_quadratic_staking: Continuous super-linear deposit curves to eliminate volume spam.
11. aetre_heterogeneous_queues: Heterogeneous task duration and multi-server queue delay modeling.
12. aetre_author_preflight_benchmark: Pre-flight variance and review risk diagnostic for manuscript drafts.
13. aetre_simulate_benchmark: End-to-end multi-policy institutional pipeline Monte Carlo simulation.
14. aetre_batch_triage: Bulk dataset triage for high-volume portfolio operations.
15. aetre_recall_scaling_curve: Empirical recall scaling curves across arrival volumes.
16. aetre_congestion_matching: Bipartite reviewer-candidate matching under capacity constraints.
17. aetre_sequential_stopping_rule: Wald sequential likelihood ratio multi-round review termination.
18. aetre_platt_calibrate: Empirical score recalibration via Platt sigmoid transformations.
19. aetre_empirical_bootstrap: Non-parametric bootstrap confidence intervals for triage policies.
20. aetre_frontier_sweep: Multi-dimensional ROC and cost-utility frontier sweep.

Microeconomic Agent Runtime Governance Tools (Governed Agent)

21. governed_bellman_triage: Evaluates agent action triage via Bellman dynamic programming and computes Expected Value of Verification ($V^*$).
22. governed_review_boundary: Computes the tripartite decision boundary ($\Delta u = u_{\text{auto}} - u_{\text{review}}$) classifying candidates into AUTO_EXECUTE, REQUIRE_REVIEW, or REJECT.
23. governed_knapsack_admit: Density-greedy $c\mu$-rule knapsack controller packing candidates by value density ($\rho_i = \Delta u_i / c_i$) under budget $K$.
24. governed_gate_pr: Road A fail-closed pull request gatekeeper evaluating verification, costs, blast radius, and test regressions.

---

Autonomous Agent Pre-Commit & CI Integration

The Governed Agent gatekeeper can be integrated into any autonomous coding agent pipeline (Claude Code, Cursor, GitHub Actions, pre-commit):

Pre-Commit Integration (.pre-commit-config.yaml)

repos:
  - repo: local
    hooks:
      - id: governed-gate
        name: Governed Agent Verification Gate (Road A)
        entry: python scripts/pre-commit-governed-gate.py
        language: python
        types: [python]

GitHub Actions Pull Request Gate

See .github/workflows/governed-gate-template.yml for a complete workflow that fails closed on any unverified autonomous pull request.

---

Research publication status

AETRE is an independent open-source research project. No new commercial licenses, paid pilots, managed hosting, consulting, or paid support are currently offered. The project website is paused. Reports and questions can be submitted through this repository's issues; see SECURITY.md for private security reports.

The code remains available under AGPL-3.0-or-later, including for commercial use that complies with that license. This publication status does not revoke rights already granted or alter any existing written agreement.

---

Citation & Academic Reference

If you use AETRE or the Governed Agent runtime in your research or production systems, please cite:

@article{gray2026innovation,
  title={The Innovation-Absorption Gap: How Artificial Intelligence Can Accelerate Idea Production Faster Than Complementary Institutions Adapt},
  author={Gray, Clayton},
  journal={SSRN Electronic Journal},
  year={2026},
  doi={10.2139/ssrn.7161458},
  url={https://ssrn.com/abstract=7161458}
}

@software{gray2026admission,
  title={The Admission Frontier: An Economically Regulated Decision-Theoretic Runtime and Fail-Closed Verification Gate for Autonomous Agents},
  author={Gray, Clayton},
  year={2026},
  publisher={Zenodo},
  doi={10.5281/zenodo.22814799},
  url={https://doi.org/10.5281/zenodo.22814799}
}

---

License & Contact

The code is licensed under AGPL-3.0-or-later, including for commercial use subject to its terms. See LICENSING.md. Third-party components retain their own terms; the locked Rust dependency inventory and bundled notices are in THIRD_PARTY_NOTICES.md.

Bundled demonstration datasets are synthetic fixtures, not empirical validation corpora. This includes the institutional queue examples and the 15-task gatekeeper demonstration. See DATASETS.md for their scope and provenance. Evaluation fingerprints are reproducibility identifiers, not proof of external validation or a guarantee of safe execution.

• Author & Maintainer: Clayton Gray
• Project questions: GitHub issues
• Private security or privacy reports: [email protected]