Tier: Evidence-Backed Bug Triage

Problem: Open-source maintainers often have to prioritize GitHub issues using issue titles, descriptions, comments, and intuition before they have enough time to investigate the actual codebase. However, an issue description alone does not reveal the real technical impact of a bug, its blast radius, affected execution paths, test coverage, regression history, or whether the reported behavior can be reproduced. This creates a costly manual investigation process and can result in issues being prioritized based on how serious they sound rather than what the repository evidence actually shows. Solution: Tier is an evidence-backed bug triage system that uses IBM Bob to investigate GitHub issues directly against the repository before proposing a priority. For each issue, Bob investigates the relevant source implementation, execution paths, tests, documentation, focused Git history, and, when useful, performs a bounded reproduction. The investigation produces a structured report that Tier validates before displaying it in the priority queue. Tier separates evidence into four categories: Tested : Bob executed a command or reproduction and observed the result. Source-supported : the finding is supported by source code, tests, documentation, or local Git history. Issue-reported : the claim comes from the original GitHub issue but has not been independently verified. Unknown : evidence is incomplete, unavailable, or conflicting. Tier also keeps confidence separate from impact. A highly reproducible issue is not automatically the most important issue, allowing maintainers to consider technical impact, reach, reproducibility, regression evidence, and remaining unknowns together. Target Users: The primary users are open-source maintainers and development teams responsible for reviewing and prioritizing incoming bug reports. A maintainer reviews the proposed priority queue, opens individual issues to inspect the underlying evidence, reorders issues when necessary, and can override Bob's proposed priority based on project-specific context. The final triage result can also be exported as a Markdown brief. What Makes Tier Unique: Tier does not simply classify or summarize GitHub issues, and it does not attempt to automatically fix bugs. Its key idea is to connect issue reports with evidence from the actual repository. Our working prototype demonstrates this workflow using five real Axios issues and a pinned Axios repository snapshot. IBM Bob genuinely performs the investigations, with prompts, transcripts, Bob outputs, and run information retained as provenance. The core philosophy is: Bob investigates. The maintainer decides. Tier turns bug triage from “Which issue sounds most important?” into “What does the repository evidence actually show?”