[Tristan-Stoltz-ERC]

Summary

Adds a result-ledger contract and an explicitly not-run JSON fixture for the shipyard evaluation lab.

• Defines terminal outcomes separately: feasible, infeasible_proven, budget_exceeded, timed_out, invalid_input, incomplete, invalid_candidate, evaluator_error, and not_run.
• Requires every assigned episode to remain visible, including failures and non-started runs.
• Specifies paired-comparison identity, independently recomputed metrics, uncertainty reporting, Pareto-frontier rules, proof labels, and publication/industry comparison gates.
• Includes a synthetic example marked not_run; it is not a benchmark result.

Why this matters

A frozen disturbance corpus prevents input leakage; an append-only result ledger prevents outcome leakage and survivorship bias. Together, these make it harder to manufacture apparent superiority by dropping timeouts, treating heuristic failure as infeasibility, or comparing different test episodes.

Validation / claim boundary

Documentation and fixture only. No Rust implementation or CI pass is claimed. The fixture is illustrative, and no policy was executed. This contract improves evaluation integrity but does not establish optimizer quality, model validity, or industrial uplift.

Stacked on PR #2742; related protocol PR #2735 and implementation ticket issue #2743.