# Exemplar Tools

The engineering stack that governs itself.

**[exemplar.tools](https://exemplar.tools)**

## What is this?

Exemplar is the umbrella site for a 23-tool open-source engineering stack
that forms a closed governance loop — from constraint specification through
code generation, deployment, production observation, and contract tightening.

### The Stack

| Pillar | Projects | Domain |
|--------|----------|--------|
| **Trust & Privacy** | Signet, signet-eval, Agent-Safe | [signet.tools](https://signet.tools) |
| **Intelligence** | Kindex, Apprentice, Chronicler, Transmogrifier | [kindex.tools](https://kindex.tools) |
| **Governance** | Pact, Constrain, Baton, Arbiter, Sentinel, Ledger, Cartographer, Stack Smoke, Aegis, Covenant, Scram, Witness, Vigil, Exemplar | — |
| **Audit & Quality** | Webprobe, Advocate | — |

### The Loop

```
SPECIFY -> BUILD -> DEPLOY -> OBSERVE -> TIGHTEN -> (back to SPECIFY)
```

Production incidents don't just get fixed — they generate tightened contracts
that make the bug class non-recurring.

## Development

```bash
npm install
npm run dev
```

## Deployment

Deployed to [Fly.io](https://fly.io) via Docker + nginx.

```bash
fly deploy
```

## License

MIT

---

## Extended Notes for Agents

### Project Inventory

**Trust & Privacy (signet.tools)**
- **Signet** — Personal sovereign agent stack. Cryptographic vault, ZK proofs,
  Ed25519 identity, XACML-derived policy engine. Four protocols (SCP, VAP, SNP,
  SVP). Rust, 473 tests.
- **signet-eval** — Deterministic policy enforcement for AI agent tool calls.
  Rust, 88 tests.
- **Agent-Safe** — SPL authorization language. Six SDK implementations,
  sub-microsecond evaluation.

**Intelligence (kindex.tools)**
- **Kindex** — Persistent knowledge graph for AI-assisted workflows. 37 MCP
  tools, hybrid FTS5 + graph search. Python, 930 tests.
- **Apprentice** — Adaptive model distillation. Routes between frontier API and
  local fine-tuned model, progressively shifting traffic as correlation proves
  quality. Python, 2,628 tests.
- **Chronicler** — Event collection and story assembly.
- **Transmogrifier** — Register-aware prompt translation for model- and
  task-specific output quality.

**Governance**
- **Pact** — Contract-first multi-agent software engineering framework.
  Ten-phase pipeline: Interview, Shape, Decompose, Contract, Test, Validate,
  Implement, Integrate, Polish, Diagnose. Python, 1,814 tests.
- **Constrain** — Interactive constraint elicitation. Python, 259 tests.
- **Baton** — Cloud-agnostic circuit orchestration. Python, 804 tests.
- **Arbiter** — Trust enforcement with append-only ledger.
- **Sentinel** — Production attribution via PACT keys; spawns fixer agents and
  pushes tightened contracts upstream. Python, 177 tests.
- **Ledger** — Schema registry and data obligation manager. Python, 276 tests.
- **Cartographer** — Stack adoption and compatibility scanner.
- **Stack Smoke** — Cross-stack smoke tests for suite installation and
  interoperability.
- **Aegis** — Resource-budget primitive for timeouts, fallbacks, and tagged
  blocking calls.
- **Covenant** — Contract validation runtime with per-environment policy and
  violation budgets.
- **Scram** — Emergency kill-switch service for catastrophic failure
  containment.
- **Witness** — Human-in-the-loop coordination primitive with approval,
  rationale, and audit.
- **Vigil** — Sub-threshold anomaly detection across event streams and traces.
- **Exemplar** — Governed code review service demonstrating the entire stack.
  Python, 1,950 tests.

**Audit & Quality**
- **Webprobe** — Site state-graph auditor with browser metrics, security checks,
  visual review, and interactive testing.
- **Advocate** — Six-persona adversarial review engine for code, design docs,
  and architecture decisions.

### The Closed Loop Pipeline

1. **Specify** — Constrain, Ledger, and Covenant extract project boundaries,
   schema obligations, and runtime contracts.
2. **Build** — Pact generates contracts, tests, and implementations; Stack
   Smoke verifies integration.
3. **Deploy** — Baton orchestrates deployment with self-healing topologies;
   Aegis budgets blocking calls and Scram provides last-resort rollback.
4. **Observe** — Sentinel, Arbiter, Vigil, Witness, and Kindex monitor
   production and coordinate judgment.
5. **Tighten** — Production incidents feed back to tighten contracts,
   eliminating bug classes.

Cross-cutting layers:
- **Trust Layer** — Signet provides cryptographic policy enforcement across
  every stage.
- **Intelligence Layer** — Kindex provides persistent knowledge context to
  every stage.

### Research Foundation

Built on 26 active research papers and 8 books by Jeremy McEntire, published at
[perardua.dev](https://perardua.dev). Key works:
- Organizational Physics of Multi-Agent AI (OT-4)
- Ambient Structure Discovery (OT-5)
- Emergence: Programming paradigm for constraint-shaped agency (OT-6)
- Structural Compression Theory (Monograph)

### Key Numbers

- 26 active research papers and 8 books providing the theoretical foundation
- 15,000+ automated tests across the stack
- 23 proper tools
- 6 programming languages (Python, Rust, JavaScript, HTML, TypeScript,
  multi-language SDKs)

### Capabilities Manifest

A machine-readable capability descriptor is available at
`/.well-known/capabilities.yaml` following the spec at
`https://perardua.dev/specs/capabilities`.

### Infrastructure

- Astro (static site) + Tailwind + React
- Built to `dist/`, served by nginx in a container on Fly.io
- Region: `dfw`

### Links

- Website: https://exemplar.tools
- GitHub: https://github.com/jmcentire
- Signet: https://signet.tools
- Kindex: https://kindex.tools
- Research: https://perardua.dev
- Author: Jeremy McEntire (https://perardua.dev)
- License: MIT