v0.6.3 — Current provider models

Refresh the balanced provider defaults to current models:

• OpenAI: gpt-6.1-sol (previously gpt-5.5).
• Anthropic: claude-sonnet-5-5 (previously claude-sonnet-5).
• DeepSeek: deepseek-flash, selecting V4.1 Flash, instead of the retired deepseek-v4-flash alias.

Explicit model overrides, provider selection, structured-output adapters, and deterministic effect dispatch are unchanged. Configuration examples and live DeepSeek cases now use the current IDs. Historical 0.6.2 qualification evidence is preserved.

Validation: 238 offline tests, Python 3.12–3.14 CI, strict typing/lint, documentation and isolated packaging checks, and a clean dependency audit. Required release qualification passed six live contracts: OpenAI GPT-6.1 Sol Plan and operational Q-LLM, DeepSeek Flash Plan and operational Q-LLM, and explicit Flash/Pro Plan cases. No selected provider was skipped; Anthropic was excluded because no API key was configured.

The same checksummed wheel and source distribution were promoted through TestPyPI and PyPI. The public PyPI package was then installed outside the checkout and passed the operational reference conformance profile and the six live provider checks. Attached evidence records the exact commit, artifacts, models, SDK versions and results.

This qualification uses synthetic data and in-memory effects. It does not establish a performance benchmark, Anthropic live qualification, GPT-6 Astra/Claude Opus 5.5 qualification, or production host-adapter acceptance.

Release workflow · PR #59