Understanding AI Spend + Proof
What the report tracks, how token usage and proof receipts are connected, and what to act on when the numbers are off.
What this report shows
The AI Spend + Proof report combines two signals that are usually tracked separately: how many tokens your team used, and how many of those runs produced verified proof receipts. The combination matters because spend without proof is harder to defend than spend that is traceable to verified output.
Token spend by member
Each bar shows one team member's token usage for the selected period. Hover on a bar to see the breakdown by route weight: how much of their spend was low-weight (efficient), medium-weight, and high-weight (complex or manual override).
A member with high spend but a high proportion of low-weight tasks is running efficiently at scale. A member with moderate spend but a high proportion of high-weight tasks may be working on complex work that requires review of the routing decisions.
Route distribution
This section shows how the team's tasks were distributed across route weights. The expected shape depends on your codebase, but most teams see the majority of tasks in low and medium weight, with high-weight tasks representing the most complex or novel work.
A shift toward high-weight across the team can indicate task descriptions that are not giving Avorelo enough signal to route efficiently, or a period of genuinely complex work.
Route weight is set automatically based on task scope, intent signals, and prior session receipts. Override rates above 15% may indicate routing configuration worth reviewing.
Proof receipts
The receipts table shows recent runs with their verification status. Verified receipts exited cleanly with tests passing and scope confirmed. Needs review exits indicate something was not confirmed before the run closed.
Unverified exits are not failures. They are signals. A run that exits without a clean receipt is telling you that the output was not fully confirmed before the session closed. That may be intentional (exploratory work, draft state) or it may indicate a gap in the validation step.
What to act on
- More than 2-3 unverified exits in a period: check whether the validation step is being skipped
- High-weight route proportion above expected: review task descriptions for routing signal quality
- One member's spend significantly higher than others with similar task counts: investigate if scope creep is happening at the task level
Open AI Spend + Proof report
View the full team report for the current period