Changelog

Added typed recipes for classification, extraction, search, verification, matching, reranking, routing, and screening

Added typed in-memory decision evaluation and bounded, ordered async batch execution

Added structured –pluck selection for CLI results without bypassing confidence gates

Added named decision contracts, offline compilation, async support, safe authentication checks, and agent guidance

Added an async semantic linter example that batches fourteen rules per function and reports structured findings

Added documented decision patterns and a support-triage example for routing, scoring, confidence gates, and fan-out

Added an interactive emoji analysis example using one typed bundle request per message

Added local result pipelines for selecting bundle answers and applying confidence or probability gates

Added the core Python and CLI interfaces for Noul, Choice, and Score decisions

Added reusable TOML decision declarations, local validation, and explicit confidence-gated CLI automation

Added credential management and runnable Python and shell examples

Added a playable Connect Four example with structured candidate moves

Added native async support, offline decision compilation, mixed-question bundles, and comprehensive API documentation

Improved Connect Four with deterministic tactical filtering so Jev asks only among admissible moves

Fixed bundle output handling and strengthened typed request-preview and example paths

Documented clearer application-policy guidance in the examples

Documented the Sphinx build and expanded the tested Connect Four walkthrough