Fan-in review [Beta]
Automatically compare multiple implementations and get a recommendation for the best one.
When you run multiple agents on the same ticket, fan-in review automatically compares their work and recommends the best implementation. Instead of reviewing every diff yourself to decide which approach wins, you get an AI-powered recommendation with detailed reasoning.
How it works
Fan-in review triggers automatically when you launch two or more implementations on a ticket:
- Each implementation runs to completion
- Each implementation is evaluated independently
- Once all evaluations finish, a separate Compare Implementations agent analyzes the results
- The best implementation is highlighted with a recommendation badge and reasoning, at the top of the ticket page
Evaluation criteria
The fan-in review agent evaluates each implementation on:
- Feature completeness — Did it fully implement the ticket requirements?
- Code quality — Test coverage, linter compliance, code style
- Correctness — Do tests pass and does the implementation run without errors?
The agent uses Superconductor's internal evaluation Codex model to analyze all evaluation data and produce a structured recommendation, including strengths of the recommended implementation and weaknesses of the others.
Credentials
Fan-in review runs using Superconductor's internal API key — you don't need to provide any additional credentials. This is similar to how Benchmark evaluations work. Even if you don't have OpenAI/Codex credentials configured for your project, fan-in review still works.
What happens when...
A new implementation is launched
If you launch additional implementations after a recommendation has been generated, the existing recommendation is marked as outdated. A new fan-in review automatically runs as part of the new launch, comparing all implementations including the new ones. Once the new review completes, it replaces the outdated recommendation.
The recommended implementation is deleted
If you delete the recommended implementation, the recommendation is removed entirely. Fan-in review does not automatically re-run in this case — launch a new implementation to trigger a fresh review.
No implementation meets the bar
If the agent determines that none of the implementations adequately fulfill the ticket requirements, it will not recommend any of them and will explain why in the reasoning.
Feedback
This feature is in beta and under active development. We'd love to hear how it's working for you — reach out to us with feedback, questions, or feature requests.