What Ripcord caught — and what it would have caught

Every consequential agent action this period, scored by expected loss. Shadow-mode entries were observed without gating: the "would have caught" section is what a live gateway stops on day one.

Your agents, ranked

By expected loss averted: the near-misses each agent generated that Ripcord and a human caught in time.

Agent conduct

Doing the job is not the same as doing what you asked. Every off-mandate move, every re-attempt after a rejection, every runaway loop gets classified and counted here. Benchmarks tell you what a model can do; this ledger tells you what it did.

Risk posture by department

Corridor grades reflect the intervention-worthy exposure this period.

What we'd change

Derived from your own ledger. Accept these, and next month's report gets quieter.

Would have caught

Shadow-mode observations whose risk tier would have held or blocked the action.

Caught

Actions a human rejected, stopped mid-flight, took over, or rolled back — each one a loss that didn't happen.