v3.1.0-beta.1

What's Changed

• Chore/repo cleanup by @kpifinity in #103
• Fix/docs stale stragglers by @kpifinity in #104
• feat(evals): SKI Evals — the verdict-accuracy evaluation suite by @kpifinity in #105
• fix(evals): close the real-model output-contract gap the first eval r… by @kpifinity in #107
• chore(evals): nightly run dumps raw model output; raise token budget by @kpifinity in #108
• feat(conformance): implement the L3 ledger tamper-resistance rig (5 o… by @kpifinity in #109
• fix(evals): the real-model scores were measuring a timeout, not a model by @kpifinity in #110
• feat(conformance): implement the L3 air-gapped boot rig (6 of 6) by @kpifinity in #111
• feat(benchmarks): performance benchmark suite by @kpifinity in #112
• fix(evaluator): degrade, never crash, on schema-violating model output by @kpifinity in #113
• docs(crosswalks): EU AI Act crosswalk - article-level control mapping by @kpifinity in #114
• fix(evals): prompt v4 - stop the model re-scoping the pre-scoped snapshot by @kpifinity in #115
• fix(verifier): ground observations in the measurement by @kpifinity in #116
• feat(backends): vLLM backend - production GPU inference with guided decoding by @kpifinity in #117
• feat(deploy): Helm chart - constraints enforced at render time by @kpifinity in #118
• feat(schemas): extract ski-schemas - the wire contract defined once (RFC 0003 PR 1) by @kpifinity in #119
• feat: beta hardening - /metrics contract, predicate parity, working quickstart by @kpifinity in #120
• fix: no unverified CLEAR; assertions grounded in the cited obligation by @kpifinity in #121
• chore(release): bump to 3.1.0-beta.1 by @kpifinity in #123

Full Changelog: v3.1.0-alpha.2...v3.1.0-beta.1