Rule
Verification has a price. Measure what verified-done costs (tokens, minutes, review effort) and drive it toward the cost of unverified-done, without giving up the verdict.
Rigor is tiered by risk, because proof you can't afford is proof you'll stop running. A deterministic check is nearly free. The default judgment is one fresh reviewer from the author's own model family, and you choose a cross-model review when the risk justifies its cost. Spend the expensive tiers where mistakes are expensive, and let the cheap tiers carry the routine. Metered cycles answer the question every unattended run must face: was the outcome worth the tokens?
Ceremony has a price too. Plans, audits, verdicts, and dashboards earn no capability credit on their own; the change they gate is what counts. Never buy green by weakening acceptance, loosening a test, or widening a tolerance. An honest null, a blocked run, or an incomplete result stays visible in the numbers.
Put It to Work
- Track the cost of each verification tier alongside the outcomes it caught.
- Match rigor to risk: deterministic checks for routine work, a fresh same-family reviewer by default, cross-model review for irreversible or high-blast-radius changes.
- Before creating a process artifact, name who consumes it and what decision it gates. If nobody does, skip it.
- Tie metrics to goal movement: proof quality, recurrence, lead time. Agent activity alone is not a goal.
- Reject metrics that reward token burn, session count, code volume, or artifact count by themselves.
- Report nulls and failures with the same weight as wins.
Failure Signal
- Every change gets the most expensive review, so verification quietly gets skipped under deadline.
- The dashboard improves while the product does not.
- Plans and audits pile up while the code under them stays the same.
- A test got loosened until it passed.
- Nobody can say what a verified change costs versus an unverified one.
- A metric cannot tell you whether to continue, stop, or roll back.
Done Looks Like
Each verified change carries a known cost, that cost trends toward the unverified baseline, and every measured number, nulls included, helps decide whether the loop is making delivery better.
In AgentOps 3.8: Skill Eval measures whether a skill helps a named task or needs revision or removal, because passing structural conformance cannot show benefit.