The dataset is currently empty. Upload or create new data files. Then, you will be able to explore them in the Dataset Viewer.

WAIL v1.3 — Execution Control & Governance Evidence

Official Verification & Evidence Repository

WAIL is an AI, Agent & MCP Execution Control & Governance Layer designed to observe, control, govern, and verify AI execution across models, agents, and MCP tools.

Control what you can’t control. Prove what actually happened.

This repository is the evidence publication destination for WAIL v1.3, including execution traces, governance enforcement evidence, cryptographic verification records, and reproducible test reports.

Explore WAIL:

Execution Control & Governance

WAIL operates around existing AI applications without replacing their orchestration frameworks or requiring a gateway.

Its execution lifecycle follows:

OBSERVE → DETECT → DECIDE → ACT → VERIFY → PROVE

Key capabilities include:

  • Runtime observability, anomaly detection, and behavioral baselines.
  • Controlled observe, retry, and reroute actions.
  • Agent, model, and MCP execution tracking.
  • Continuous Governance Enforcement across active executions.
  • Cryptographically signed evidence and chronological governance records.
  • On-premises and private-cloud deployment architecture.

Verification Domains

V-01 — Mid-Execution Governance Enforcement

Verification of authorization and policy changes during active execution, including REVOKE handling, intervention in affected operations, and prevention of unauthorized subsequent dispatch.

Evidence must distinguish local enforcement from independently confirmed remote termination.

V-02 — Agent, MCP & Model Execution

Verification of connected execution relationships across Agent Runs, MCP tool operations, and model requests, including chronological evidence and signed audit records.

V-03 — Controlled Runtime Recovery

Verification of controlled retry and reroute behavior, provider transitions, execution authority, and observable recovery outcomes.

V-04 — Signed Evidence & Ledger Integrity

Verification of cryptographic signatures, evidence chronology, ledger linkage, checkpoint integrity, and trusted-head validation within the documented trust model.

Evidence Publication

Verification packages are organized by domain and may include:

  • Test reports and execution summaries.
  • Sanitized execution traces and signed records.
  • Cryptographic manifests and SHA-256 checksums.
  • Verification scripts and reproduction instructions.
  • Explicit verification scope, limitations, and results.

A verification claim is supported only by the artifacts and checks documented in its corresponding evidence package.

A recorded decision is not proof of enforcement.

Verification Principles

WAIL distinguishes between:

  1. A control decision.
  2. An attempted action.
  3. An observed local enforcement outcome.
  4. An independently verified remote outcome.

These states must not be represented as equivalent without supporting evidence.

Hugging Face Verification Architecture

WAIL maintains two complementary resources on Hugging Face.

Verification Lab — Technical Presentation

WAIL Verification Lab

Provides an accessible overview of WAIL's execution control architecture, governance enforcement, verification domains, and evidence methodology.

Evidence Repository — Technical Artifacts

WAIL v1.3 Evidence Dataset

Serves as the repository for version-specific verification artifacts, integrity manifests, test reports, and reproduction instructions.

The Verification Lab presents the technical context; the Evidence Repository provides the underlying artifacts for independent inspection.

Resources


WAIL — AI, Agent & MCP Execution Control & Governance Layer

This repository is dedicated to verifiable technical evidence, not marketing-only performance claims.

Downloads last month
48

Space using wailinfra/wail-v1.3-evidence 1