Xerify 0.3.0 — Jev verification

Xerify 0.3.0 adds TypeSafe Jev alongside LLM verifiers: bounded evidence, a different invocation provider, and the same confirmed | refuted | unclear contract through CLI, library, and MCP.

• Use --to jev or pin a Jev model. Typed probabilities, confidence, model identity, and decision policy are preserved; below-threshold results abstain as unclear.
• Explicit --timeout 0 disables the lifecycle deadline while preserving cancellation and byte limits.
• Cursor verification now includes a concrete JSON output example. Four separate live integration checks passed; strict schema validation remains in place.
• New OpenAI/Anthropic/Jev benchmark documentation includes six directions and 72 selected historical measurements. These are synthetic-label comparisons, not production accuracy or calibration estimates. Cursor diagnostics remain separate.
• Documentation and release media are available in English, Turkish, German, French, Spanish, and Simplified Chinese. README and site use the new LLM/Jev animation; repository links now point to Verhex.
• Dotenv exclusions and checks cover Git, npm, source archives, site output, and container contexts. The Hono lockfile update clears the production dependency audit.

Validation: 293 automated tests, clean package installation and CLI/MCP smoke checks, generated schemas, brand asset checks, and the release/supply-chain audit. No API keys or raw provider responses are included.

npm install --global xverify-cli@0.3.0
xerify --help

Jev setup · Benchmark methodology · Türkçe