Reducing CI Build Triage Time by 80%

A team running a 60-module Unreal project was averaging 45 minutes per CI build failure triage. The pipeline produced build logs between 8,000 and 20,000 lines. Engineers would claim the failure in Slack, open the log, and start scrolling. Most of the time was spent finding the relevant error in the noise — the actual fix was usually straightforward once the root cause was clear.

1. CI fails, Slack notification fires
2. Engineer claims the failure, opens the full build log
3. 20-30 minutes scrolling and searching for the actual error
4. 5-10 minutes cross-referencing with recent merges
5. 5-10 minutes verifying the root cause vs symptoms
6. Fix, push, wait for CI to confirm

Steps 2 through 5 were the bottleneck. The fix was rarely the hard part. Understanding the failure was.

The team added a single step to their pipeline: run Build Doctor on failure. The step captures the build log, runs diagnosis, and posts the structured findings as a comment on the failing commit.

- name: Diagnose failure
  if: failure()
  run: |
    gamibase doctor --log $BUILD_LOG --format json > diagnosis.json
    # Post findings to PR comment via existing reporting tool

1. CI fails, Slack notification fires with diagnosis summary
2. Engineer reads 3-5 ranked findings with file, line, and module context
3. Verifies the top finding against the source — usually under 2 minutes
4. Fix, push, CI confirms

The 20-30 minute log scrolling step was eliminated entirely. The cross-referencing step was reduced to confirming what Build Doctor already surfaced. Average triage time dropped from 45 minutes to under 10.

The structured diagnosis data started creating a record of the project's failure patterns. After a month, the team could see which modules failed most often, which error families were most common, and which areas of the codebase were most fragile. This was never the goal — it was a side effect of having structured data instead of raw logs.