# Provider maturity

Clawmeter uses a binary provider maturity model. An integration is either
**experimental** or it is not. Experimental means the live-user confidence is
below 90%, the producer
  contract is undocumented or unstable, or known semantic defects remain.

Current classifications are explicit metadata, not an inference from whether
an audit covered a provider:

| Providers | Maturity |
|---|---|
| Claude, Codex (`openai`), Gemini | not experimental |
| Alibaba, Antigravity, Grok (`xai`), Kimi, Kimi K2, Copilot, OpenRouter, JetBrains, Synthetic, z.ai | experimental |

The experimental group reflects the current provider audit's documented
contract or semantic risks. Antigravity has been validated against one live
consumer account, but its CLI-owned credential storage and internal quota API
still need live validation on macOS and Windows. Grok is experimental because
its live endpoint has omitted the usage percentage even while the account was
exhausted. The other group retains this project's existing live-validation
evidence. Maturity describes confidence in the integration, not whether
credentials were found or whether polling is enabled. The `providers`
inventory remains the place to see setup and polling state; quota rows and the
tray intentionally do not carry maturity labels.

Removing the experimental flag requires a deliberate metadata change backed by
focused tests and current evidence. CodexBar is useful executable prior art,
not provider authority. No live entitled-account validation was available for
the current experimental classifications.