The Starko AI ledger: runs, outcomes and cost

Every Starko AI run is recorded with what triggered it, what came of it and what it cost. Two pages show this: Settings → Fix agent for outcomes and Settings → Usage for cost and tokens.

Fix agent

The page opens with the headline numbers for the period you pick (7, 30 or 90 days): merged pull requests, merge rate, time to merge, billed cost per merged pull request, runs, median run duration and automatic revisions.

fix agent ledger

The Outcomes bar splits runs into merged, pull request open, closed unmerged, reverted, no fix, needs info, manual, superseded and running. Below it, Starko lists the most common reasons a run produced nothing mergeable, in the agent's own words, so you can fix the cause: a missing repository, an ambiguous ticket, a test command that does not run.

The table lists each run with its ticket, project, trigger (button, mention, autonomy policy, crew, continuation, review revise), task kind and size, outcome, risk, CI result, tokens and billed amount. Tabs switch to pull request reviews, projects and evaluation cases.

Usage

The Usage page shows cost and tokens across the workspace for the last 24 hours, 7, 30 or 90 days, as a daily chart and a breakdown by project, by day or by trigger.

settings usage

Cost is the list price of the metered tokens. The plan's included quota and discounts are not applied here; the Billing page shows what you actually pay.

Tokens shows billable and metered tokens.

Limits shows how far each project is from its monthly automated-run budget.

Reading the numbers

A low merge rate with many "needs info" outcomes means tickets lack steps to reproduce. Tighten the ticket template.

Many closed unmerged pull requests usually point to missing project instructions or a test command that does not verify the right thing.

Time to merge falls when automated follow-ups are on and reviewers leave actionable comments.

Budgets

Each project has an autonomy policy and a monthly cap on automated runs; manual runs never count against it. Each plan includes a monthly agent run budget, and a single ticket can consume at most three runs before a person has to step in. See Plans, credits and usage.

Did this answer your question?