v1.0.0

What's Changed

• Refine release workflow to match only version tags by @kelkalot in #7
• Release 1.0.0: stability, doctor, robustness by @kelkalot in #8

Full Changelog: v0...v1.0.0

v0.4.0

What's Changed

• Add trace pipeline validation and recorder by @kelkalot in #4
• Add trend command for N-way lockfile analysis by @kelkalot in #5
• Add json_mode, OTEL/Anthropic ingest, embeddings by @kelkalot in #6

Full Changelog: v0...v0.4.0

v0.3.1

What's Changed

• Make --tools optional when using traces/mined by @kelkalot in #3

Full Changelog: v0...v0.3.1

v0.3.0

What's Changed

• Add recording proxy and multi-log ingest by @kelkalot in #2

Full Changelog: v0...v0.3.0

v0.2.0

What's Changed

• Add ingest pipeline and trace-mined probes by @kelkalot in #1

New Contributors

• @kelkalot made their first contribution in #1

Full Changelog: v0.1.0...v0.2.0

v0.1.0

A capability lockfile for local models: derive deterministic probes from your tool
schemas, score them with no LLM judge, and gate CI on within-model regression.

Capabilities (11)

tool_selection, tool_discrimination, needle_in_tools, arity_robustness,
arg_validity, required_args, structured_output, format_adherence, and three
negative/safety probes — tool_restraint, tool_permission, no_hallucinated_tool.

Backends

• Any OpenAI-compatible endpoint via --endpoint (Ollama, vLLM, llama.cpp,
  LM Studio, HF TGI, OpenAI, OpenRouter, …)
• Anthropic / Gemini / Mistral / etc. via --via anyllm or --via litellm
  (optional extras: probelock[anyllm], probelock[litellm])

Also in this release

• --samples N with a statistically-aware gate (--confidence, one-sided
  two-proportion test; sub-significant drops are flagged "noisy", not failed)
• diff --format table | markdown | json | html
• probelock init scaffolding and a composite GitHub Action (action.yml)

Install

Requires Python 3.10+ (the anyllm extra needs 3.11+).

uvx probelock --help
pip install probelock