Your modeled scenario
What this could mean
Platforms:
Operating modes:
Platform and operating-mode selections are descriptive. They do not alter the illustrative rates in 0.20.0.
The takeaway
Where it goes
Modeled monthly spend allocation
Monetary finding assumptions
Signal-associated spend for review
Remaining useful-work scenario
Total annualized spend
Assumptions (adjustable)
The 26% starting mix is illustrative: 7.5% across monetary finding kinds and 18.5% across neutral signals. These rates are scenarios, not findings, benchmarks, maximums, or savings promises. Categories are modeled as separate slices; a real audit must reconcile overlap. The 50% combined ceiling is only an interface bound for this modeler.
Monetary finding assumptions
Xerg can report these three as monetary findings when the required runtime evidence is present.
Potential waste and optimization signal assumptions
These rates model spend that a review might determine is recoverable. In an Xerg audit they remain neutral signals, not measured waste.
Monetary finding assumptions:
Signal-associated spend for review:
Conditional combined scenario: If review confirms signal-associated spend is recoverable.
View all modeled values
| Allocation | Model group | Share of current spend | Monthly amount | Annual amount |
|---|
What you won’t see in the bill
The model call is the visible line item. Cost per completed task is shaped by everything around it.
Xerg treats retry waste, exact no-progress tool loops, and net-negative cache lifecycles as monetary findings only when the required evidence is present. Seven other patterns remain neutral signals in an audit. This modeler lets you explore how much signal-associated spend might prove recoverable, but it does not classify that spend as measured waste.
- Plan
- Tool
- Finish
- Plan
- Tool
- Retry
- Repeated tool
- Growing context
- Finish / fail
Every extra model step is still metered, whether or not it advances the outcome.
A 2026 study found identical-task coding-agent runs varied by up to 30× in total tokens, while “higher token usage does not translate into higher accuracy.” Read the study →
Xerg replaces these assumptions with three evidence-strict monetary findings, seven separate neutral signals, detector coverage, and prioritized fixes from your runtime evidence. Signal spend remains unclassified until intent, outcomes, and counterfactual evidence support a conclusion.
Replace the scenario with evidence
Find the recoverable spend.
Xerg audits the runtime data you already have, attributes the costly patterns, and shows what to fix first.