deterministic-verification

Open-source AI verification infrastructure for deterministic verification of LLM outputs, tool calls, code, schemas, and agent state before production execution.

Every tool you've tried edits the document. That's why nothing holds — the document was never the source. Meaning is. Branch it, simulate a change before it exists, bind an approval to the exact state a human read, and let a merge preserve a conflict instead of inventing a number. Verified offline by a stranger. No model, no account. MIT.

GeoTask — 面向 AI Agent 的时空任务上下文引擎 | Spatiotemporal Task Context Engine for AI agents.

Turn "AI writes code" into a ticket pipeline with guardrails: product design → ticket design → agent-driven implementation with deterministic verification.

🏛️ Deterministic rejection layer for computational legal claims. Verifies dates, amounts, and structured constraints; blocks unproven legal outputs.

Reliability Harness v2 for OpenCode CLI — task contracts, least-privilege sub-agents, clean-checkout verification, and a deterministic done-gate. Orthogonal agents (scout/spec-critic/test-designer/lead/verifier/adversary) instead of 3 identical thinking clones. Compatible with Oh My OpenAgent. Not a model swap or 1:1 emulation.

Deterministic MCP Security Architecture. FrozenNamespace as Root of Trust for Model Context Protocol tool verification

Proof records linking detection, validation, and evidence-backed claims.

Deterministic validation firewall that verifies AI-generated proposals against ground-truth state using immutable rules. Zero dependencies.

Autonomous Multi-Agent Software Factory governed by Finite State Machines (FSM), deterministic verification gates, execution budgets, and Git worktree isolation.

Reliability Harness v2 for ZCode (GLM-5.2 / ZAI) — task contracts, least-privilege sub-agents, clean-checkout verification, and a deterministic done-gate. Orthogonal agents (scout/spec-critic/test-designer/lead/verifier/adversary) instead of 3 identical thinking clones. Mythos-inspired, not a model swap or 1:1 emulation.

Validation harnesses, test cases, and reports for detection quality gates.

The DS for Art Basel 2018, Basel University Diploma certifications, to process corporate registration on TV in 1h37 in Switzerland's Canton of Zug and more. As Proxeus switched to a NPO, open source all codebase and the new maintainers rewrote an inferior DS to take credit. My superior and proven track record DS is abandonware and I take ownership.

Verification for Universal Commerce Protocol (UCP) transactions — Deterministic verification layer for UCP checkouts: catches math, state, and schema errors before payment.

Platform architecture and operational components supporting the security stack.

Deterministic local proof harness for verifying AI agent recommendation claims against source-of-truth fixtures, with append-only outcome logs.

Reproducible demo connecting repository knowledge, agent guidance, and deterministic checks for more reliable coding-agent changes.

Evidence-backed FDA recall matching for hospital inventory — combining Amazon Bedrock for ambiguity resolution with deterministic verification for explainable recall decisions.

Reliability Harness v2 for Grok Build CLI — task contracts, least-privilege sub-agents, clean-checkout verification, and a deterministic done-gate. Orthogonal agents (scout/spec-critic/test-designer/lead/verifier/adversary) instead of 3 identical thinking clones. Mythos-inspired, not a model swap or 1:1 emulation.

ProofRail: an inspectable Pokee Isaac agent run with deterministic evidence checks, negative controls, and provenance receipts.