Decision infrastructure for agent systems and operators.
Policy evaluation · recoverable execution · evidence-bound workflows
I turn operational problems into inspectable systems. Each project states its authority boundary, tests its failure paths, and separates what the evidence proves from what it does not.
Featured project · Agent infrastructure · Decision methods · Experiments
A runnable Node.js reference path across policy evaluation, recourse-gated execution, and dispute evidence replay.
Follow a synthetic refund through the policy decision and execution outcome, then inspect the optional dispute-evidence simulation. The components use pinned versions and state their limits in the repository.
| Project | What it provides |
|---|---|
| Constitutional Agent Testbench | Deterministic Python policy evaluation with precedence tracing for structured agent responses. |
| Consequence Rail | JavaScript execution rail that reserves a remedy, checks the postcondition, and emits signed settlement receipts. |
| MandateBound | TypeScript evidence-readiness simulation and deterministic dispute replay for UCP/AP2 agentic commerce. |
| TraceCanary | Offline Python CLI and desktop GUI for detecting privacy regressions in OTLP GenAI trace exports. |
| Agent Team | Reusable Python skill for bounded specialist workflows with independent audit and schema-backed evaluation. |
| Project | What it provides |
|---|---|
| Crypto Research Desk | Research-only workflow with four specialist lanes, independent risk review, and a fixed four-horizon forecast contract. |
| Corridor Lab | Desktop GUI and CLI for comparing fictional cross-border payment routes across cost, speed, liquidity, and failure assumptions. |
| Unconventional Moves | Reusable decision skill for practical, non-obvious approaches with reversible 48-hour tests. |
| The Smallest Agreement | Local browser workshop for finding the lowest-cost structured agreement meeting an approval threshold and optional protected constraints. |
| Project | What it provides |
|---|---|
| Common Cart | Local-first pooled-purchase simulator that matches aggregate buyer constraints to synthetic offers without exposing buyer records. |
- Define the user, decision, and authority boundary.
- Build the smallest inspectable system that can refuse invalid inputs.
- Test the normal path, failure path, and recovery path.
- Publish only claims that the artifacts can reproduce.
I direct problem selection, product direction, requirements, business judgment, evaluation, rights review, and final acceptance for every project published here.