Deployment gates, state coverage, and key-person risk.
A single-dimension deep review of how change reaches production safely — staged gates with named exit evidence, tests that exercise accumulated state, and whether the architecture and runbook survive the loss of any one person.
Best fit
Teams whose deploy process lags research, whose tests pass while incidents recur, or whose stack only one engineer can operate.
Evidence standard
Methodology grounded in a working, benchmarked market-making system (prototype-proven). The review is read-only: no production access, no capital moved. Findings fold into the parent Trading Systems Diagnostic; no live-production track-record or P&L assurance is claimed. The state-space and key-person findings also draw on delivered engagements (reported outcomes, not independently verified).
What this review covers
- 01Staged promotion gates with named exit artifacts and multi-role sign-off.
- 02Canary verification and rollback as a rehearsed pipeline stage.
- 03State-persistent test coverage — behaviour under accumulated state, not only fresh state.
- 04Continuous invariant checks in production.
- 05Architecture that matches runtime, with a runbook and a second operator.