Synthetic support shadow validation

Run the C7 engineering fixture with one command from the repository root:

CARGO_INCREMENTAL=0 cargo test -p duduclaw-gateway c7_synthetic_shadow_harness --lib -- --nocapture

The default test-thread stack is sufficient (verified 2026-09: the harness completes in ~90s with no RUST_MIN_STACK set, which is also how CI runs duduclaw-gateway’s test suite). If a future change to this fixture ever grows its stack usage past the default, RUST_MIN_STACK=33554432 raises the worker thread’s stack size — optional, not required today.

The test prints one C7_SYNTHETIC_SHADOW_SUMMARY= JSON line. It contains only case counts, review decisions, and failure codes; it does not include ticket IDs, source text, or tenant data. A passing test demonstrates that the current backlog and ticket-SLA shadow engines can process a deterministic 21-day sequence with forecasts committed before each day’s outcome source exists. It then checks a missing score, a reviewed correction of the final day’s aggregate and ticket outcome, and source removal. Missing, stale, and revoked evidence must suppress complete-window review; saved historical screens retain their original meaning until a linked source is revoked.

Historical dates and ingestion times are staged only in the #[cfg(test)] fixture. Production policy, forecast, and score methods still use the wall clock and reject late commitments. This fixture is synthetic engineering evidence. It does not measure skill on real support data, calibrate uncertainty, authorize model promotion, or authorize staffing action. The Decision Lab’s synthetic engineering validation remains the dashboard view for a current local fixture; this test’s output is deliberately separate from operational evidence.