EPIC: Evaluation, release and daily-driver quality

Programme: #1

Own resettable fixtures, protocol/security/chaos/performance suites, outcome/anti-wrapper benchmarks, signed distribution, updates, backup/restore, diagnostics and operational readiness.

## Child issues
- [ ] #101 EVAL-01 — Resettable fixtures and protocol tests
- [ ] #102 EVAL-02 — Security, chaos and performance suites
- [ ] #103 EVAL-03 — Outcome benchmarks and anti-wrapper tests
- [ ] #104 EVAL-04 — Signed distribution, updates and operations

## Cross-cutting gates
Fixture truth is independent of model self-grading; nondeterministic tasks use repeated trials; security/recovery regressions block release; deterministic UI latency is measured separately from model latency; irreversible live actions are never replayed by eval tooling.

Start by reading child issues #101–#104, which divide the evaluation, release, and operations work into resettable fixtures, test suites, benchmarks, and distribution updates. Done requires completing those child scopes while preserving the listed cross-cutting gates for fixture independence, repeated trials, security and recovery, latency measurement, and safe handling of irreversible actions.

Written by the indexing model from the issue text.