Expectation Ledger v0.1.0-alpha.1

Expectation Ledger v0.1.0-alpha.1

The first public alpha of Expectation Ledger introduces an in-process expectation
layer for AI agent control loops: predict before acting, compare the observed result,
and open a circuit breaker when a high-stakes contradiction occurs.

This release is intended for developers experimenting with agent reliability and
control-loop safety. APIs may change before a stable release.

Included

• Core types: Prediction, PredictionOutcome, PolicyRule, BreakerState
• Severity routing: result × stakes → severity → policy action
• Circuit breaker: open on high-severity contradiction, close on verified match
• Pluggable prediction sources (rule-based default shipped)
• Hermes adapter: context block injection + response frame overlay
• OpenClaw adapter: pre-model context packet assembly
• Tests for severity routing, breaker open/close/block lifecycle, classification, stability

Install

git clone https://github.com/tier4research/expectation-ledger.git
cd expectation-ledger
python -m pip install .
ledger test

Compatibility

• Python 3.10 or newer
• No required runtime dependencies
• No network service or hosted control plane

Not included yet

• Frequency-based prediction sources
• kNN retrieval for similar-prediction lookup
• Breaker persistence backends (currently in-memory; serialize yourself)
• Full integration example with a live Hermes engineering loop

Verification

• 19 automated tests pass.
• The installed ledger test smoke check passes.
• Source distribution and universal Python wheel build successfully.
• The release tree contains no credentials, private runtime state, or generated logs.

See the README for usage, comparisons, and framework integration.