Skip to main content
Once a day, IronBee runs an LLM analysis for each project. It looks at the recent sessions your AI coding agents recorded with the IronBee CLI and reports how well they verify their changes: first-pass success, where the agent gets stuck, how effective its fixes are, and which files cause the most trouble. Verification runs that IronBee executes on its own infrastructure aren’t part of the analysis. Each analysis also produces the project’s findings and recommendations.

​
Open an analysis

Open Analyses in the sidebar and pick a project. The page shows the project’s latest quality analysis and when it was generated.
  • If the project has no analysis yet, the page says No analysis has been generated for this project yet.
  • An analysis built from few sessions carries a Limited data label.
  • Cards appear only when they have data. Hover the info icon on a card to see what it measures.
You can also reach an analysis from a recommendation’s Source Analysis link, or with Go to Analyses in the ⌘K palette. The Analyses page for a project, with the project picker, the time the analysis was generated and the Quality overview card The Analyses page for a project, with the project picker, the time the analysis was generated and the Quality overview card

​
Quality overview

The headline states how many sessions were analyzed and how many verify and fix cycles the worst session ran. Below it are six metrics: First-pass success and Verifications / session also show a weekly sparkline and the change from the previous week. Issues caught on fail shows the weekly trend of checks per verdict instead.

​
Session outcomes

  • Session outcomes: how sessions ended.
    • Pass: the last verdict passed.
    • Fail: the last verdict failed.
    • Abandoned: the last verdict failed and a new cycle started after it.
    • Completed: the session ended without a verdict.
  • Client distribution: which AI client ran each session: Claude Code, Cursor or Codex.
  • First-pass share: the share of time not spent on fixes, against rework.
The Session outcomes, Client distribution and First-pass share cards of a project analysis The Session outcomes, Client distribution and First-pass share cards of a project analysis

​
Where time goes

How session time splits across coding, verification and fixes, with the average duration of a session, an activity, a verification and a fix.

​
Issues before vs after fix

The average issue count before a fix attempt and after it. It shows whether fixes resolve the problems the agent finds.

​
Verification depth

Verification depth: on pass vs on fail compares how thoroughly the agent checked before it passed and before it failed:
  • Checks / verdict
  • Screenshots / verif.
  • Aria snap. / verif.
  • Issues / verdict

​
Fix outcomes

Fix cycles grouped by what they did:
  • Helpful: the fix reduced or resolved the issues.
  • No-effect: the issues didn’t change.
  • Backfiring: the fix introduced new issues.

​
Top blockers

The most frequent reasons the verification gate blocked the agent from finishing, such as a failed verdict, a missing verdict, or a verification that didn’t use the required tools.

​
Hot files

The files the agent edited most, with how many sessions touched them, how many fix cycles they needed, and their fail rate. The Hot files card, listing each file with its edits, sessions, fix cycles and fail rate The Hot files card, listing each file with its edits, sessions, fix cycles and fail rate

​
Weekly trend

A chosen metric by week: pass rate, efficiency, cycles or checks.

​
Outcomes by session size

A heatmap of session outcomes by duration and by number of activities. It shows whether longer or busier sessions fail more often.

​
Recent corpus

Three lists from recent verdicts:
  • Recent issues: what the agent flagged in failed verifications.
  • Recent fixes: what the agent said it did to resolve problems.
  • Recent verification checks: what the agent verified.
Use them to spot recurring patterns in failures.

Findings

Observations the analysis surfaces.

Recommendations

Directives generated from findings.