Evidence for #4074: retro dispatched on clean-approval dep bump with no novel findings (segment-bridge #293)

[fullsend-ai-retro]

What happened

PR #293 is a single indirect Go dependency patch bump by MintMaker bot (2 files changed: go.mod and go.sum). The review agent ran once (workflow 29134393510), approved unconditionally in ~3.5 minutes at 01:22 UTC with zero findings, and applied the ready-for-merge label. The PR was auto-merged at 07:01 UTC by red-hat-konflux[bot]. No human review was needed or performed. On merge, the retro agent was dispatched (workflow 29143798518).

This is the second consecutive day a retro was dispatched on an identical workflow pattern — PR #288 was retro-analyzed on July 10 and produced three evidence issues (#3975, #3976, #3977) all covering pre-existing upstream issues. PR #293 is even cleaner than #288 (no force-push rebase, single review run, no human intervention), leaving zero novel observations for the retro agent.

What could go better

The retro agent consumed compute and tokens to analyze a workflow that had no rework, no missed findings, no human-agent disagreement, and no novel patterns. The previous day's retro on the same workflow class (PR #288) already exhausted the observation space. Every clean bot-dep-bump PR that triggers a full retro run is pure overhead.

Confidence: High. This is the same pattern described in #4074 with a new concrete data point. The class of change (bot-authored, indirect Go dep patch bump, single review with unconditional approval, zero findings, no human comments) is highly predictable and contributes no learning signal to the retro process.

Proposed change

Implement the skip criteria proposed in #4074: when a closed/merged PR meets all of the following, skip retro dispatch:

1. Author is a known bot (red-hat-konflux[bot], dependabot[bot], renovate[bot])
2. Only go.mod/go.sum (or lockfile equivalents) changed
3. Review agent produced exactly one approval with zero findings
4. No human review comments or change requests exist
5. No code-agent activity occurred

This would have prevented the retro dispatch on PR #293 and the functionally identical retro on PR #288 the previous day.

Validation criteria

After the skip criteria are implemented:

1. Retro is NOT dispatched for the next MintMaker indirect Go dependency patch-bump PR in konflux-ci/segment-bridge (or any repo with the criteria enabled) that receives a clean single-approval review.
2. Retro IS still dispatched when a dependency PR has human comments, review agent findings, multiple review runs, or code-agent involvement.
3. Token spend attributable to retro on bot-dep-bump PRs drops to zero for the qualifying class.

---

Generated by retro agent from konflux-ci/segment-bridge#293