[Tristan-Stoltz-ERC]

Summary

Adds the next concrete artifact for the matched-budget shipyard recovery evaluation lab:

• tools/shipyard-evaluation-lab/DISTURBANCE-TAPE-CONTRACT.md specifies immutable event tapes, separate effective/observed/delivery times, policy-visible versus evaluator-only information, deterministic ordering, episode manifests, replay procedure, receipt fields, and adversarial negative controls.
• tools/shipyard-evaluation-lab/fixtures/synthetic-disturbance-tape-v0.json provides a small synthetic example with a delayed vehicle outage, hidden clearance fact, and delayed yard-closure observation.

Why this matters

The benchmark must compare policies on the same disturbances without allowing future truth to leak into decisions. This contract makes that requirement testable before adding more optimizers.

Validation and claim boundary

• This PR is documentation plus a synthetic JSON fixture; it does not implement a parser or executable replay engine.
• The fixture has not been validated by the repository's Rust test suite in this PR. No CI result is claimed.
• The fixture is synthetic and not calibrated to a real shipyard.
• Conformance to this contract would establish replay semantics only, not physical validity, SOTA, or industrial performance.

Related: evaluation-lab contract PR #2735, evaluation-lab issue #2734, and non-anticipative policy replay PR #2730.