Volume
Count eligible pull requests and updates per month. Distinguish new reviews from reruns of the same change.
GUIDE · REVIEW ECONOMICS
Budget for the review workflow you actually run.
A price per model request does not describe the cost of reviewing a repository. Review depth, repeated pushes, retries, and the time spent checking findings all affect the result.
Start with three inputs.
Depth
Decide which changes need additional independent reviewers or a deeper model profile. Small documentation changes and permission changes need different attention.
Useful outcomes
Record completed reviews, accepted findings, and maintainer time. A cheap run that never finishes is still a cost.
Separate model spend from the total service bill.
Estimate monthly model spend as eligible pull requests × average charged review rounds per pull request × average model cost per round. If your round estimate already includes every reviewer, do not multiply by reviewer count again.
Add the subscription or management fee, compute and storage charges, and any direct provider charges. Keep onboarding and migration work separate so a first-month setup charge does not look like recurring review usage.
| Input | Example |
|---|---|
| Eligible pull requests | 80 per month |
| Charged review rounds | 1.5 per pull request on average |
| Model cost per complete round | $0.50, including its reviewers |
| Estimated model spend | 80 × 1.5 × $0.50 = $60 per month |
| Other costs | Add the actual subscription, compute, provider, and setup charges |
Use round limits as well as spending limits.
Independent reviewers provide separate perspectives on one change. Review rounds determine how often the system reviews a pull request. Raising both can multiply work quickly, especially while an author is pushing frequently.
Set an explicit round limit and decide when a maintainer may request another review. Begin with fixer commits disabled if you are evaluating findings; correction attempts have their own work and cost.
Redgold exposes monthly repository automation and per-pull-request spending limits. Recorded usage controls subsequent model calls. In-flight work and accounting delays can overshoot, and a model-spend limit does not cover unrelated charges.
Review quality and cost together.
Run the calculation with your real repository volume and written prices. Then compare the pilot with the review workflow your team already uses. A budget is a decision aid, not evidence that reviews save money.
- Keep incomplete runs and failed stages in the denominator when reporting completion rate.
- Inspect a sample of accepted and rejected findings; count a useful finding only after maintainer assessment.
- Compare cost and waiting time for similar changes before moving to a deeper profile.
- Include the time spent configuring the bot, investigating failures, and checking false positives.