Skip to content
Postcept
Verified Completion Rate

Verified Completion Rate

Verified Completion Rate is the percentage of claimed AI-agent completions that are independently verified as correct against the system of record. It gives teams a reliability metric for action-taking agents that goes beyond eval scores, traces, and self-reported success.

Definition

Definition

Verified Completion Rate is the percentage of an agent’s claimed completed actions that are independently verified as correct against the system of record.

Formula

Verified Completion Rate = verified completions / claimed completions

Example dashboard

For an agent that claimed 1,000 completed refunds and cancellations, Postcept verifies each one against the system of record and surfaces the real reliability number, plus exactly where the gap came from.

Verified Completion Rate

81.2%

Claimed completions
1,000
Verified correct
943
False completion
31
Duplicate / mismatch
18
Policy failed
8

VCR over time

87.0%97.0%

Why claimed success is not enough

Agent success rates and eval scores measure what an agent or benchmark thinks happened. They are computed from the agent’s own outputs, so they inherit the agent’s blind spots: a confident tool call that silently failed still counts as a success.

Verified Completion Rate is computed from the system of record instead. It is the difference between an agent that looks reliable and one you can prove is.

How VCR improves agent operations

Because every gap is attributed, false completion, duplicate, mismatch, or policy failure, teams can see exactly which failure modes are dragging the number down and fix them. As Postcept catches recurring patterns and they are addressed, Verified Completion Rate climbs and stays measurable over time.

Each verified action also leaves a signed completion receipt, so the metric is backed by evidence, not estimation.

Measure your Verified Completion Rate.

Connect your systems and start measuring verified completion across every high-risk action your agents take.