Should failed pull requests become part of the learning archive?

Rejected pull requests and their review discussions contain reasoning the final code cannot show: which abstraction was too brittle, where a security assumption failed, or why maintainers rejected a seemingly clean design. That is valuable material for newcomers. A GitHub PR can preserve line comments, review summaries, suggested edits, and request-changes decisions, so the path to the answer is often more educational than the answer itself.

But permanent searchability changes contributor behavior. Someone may be less willing to submit a rough experiment if an awkward first draft is attached to their name forever. I’d favor an educational archive separate from official project history, with author-controlled redaction, optional expiration, and administrator archiving for sensitive cases. Curated lessons could live in Discussions or documentation instead of every failed PR being treated as canonical.

Does preserving failure improve technical learning, or create an unfair reputational burden? Disagree, and share examples from real development communities.

I’m sympathetic, but I’d keep the raw record by default. The messy thread often contains the security constraint or operational edge case that gets lost in a polished write-up. The better safeguard is a short maintainer summary marking what remains valid, not pretending the failed design never existed.

My concern is security-sensitive material. Redaction after the fact is useful, but it may leave people guessing what was removed, and the original comment can still have an edit trail. I’d separate public teaching examples from private review records, with a clear path for contributors to request removal.

The evidence supports educational value, not the stronger claim that permanence improves learning. Studies of review suggestions and Apache pull requests indicate that discussions can preserve rationale and induce refactoring, while suggestions can also increase resolution time. That argues for measuring whether archives help newcomers before making them universal.

Expiration sounds humane, but it can erase exactly the historical context maintainers need when the same proposal returns two years later. I’d make expiration opt-in for ordinary experiments and mandatory for personal data or sensitive security details. Searchability should be earned by usefulness, not assumed from the platform’s defaults.