Skip to content
Archgate Studies

Sentry PR Friction and ADR Standardization

Sentry PR Friction and ADR Standardization

Section titled “Sentry PR Friction and ADR Standardization”

Abstract

Section titled “Abstract”

This study evaluates where review back-and-forth is concentrated in getsentry/sentry and identifies which Architecture Decision Records (ADRs) plus enforceable rules could reduce repeated discussion cycles.

Scope

Section titled “Scope”
  • Window: last 90 days
  • Sample: 500 merged PRs
  • Metrics: review events, merge latency, changed file count, churn
  • Deep dive: highest-friction PRs by review-event volume and merge latency

Baseline Findings

Section titled “Baseline Findings”
  • Median time-to-merge: 4.12h
  • P75/P90 time-to-merge: 21.84h / 69.54h
  • Median review events per PR: 2 (mean 3.19)
  • Formal CHANGES_REQUESTED: 1%
  • Small PR median TTM: 2.07h
  • Large PR median TTM: 21.44h

Coverage caveat

Section titled “Coverage caveat”

The baseline above is merge-biased. It captures friction in merged PRs but misses:

  • high-discussion PRs that were closed without merge
  • open PRs with substantial discussion that later went stale

Those two cohorts often contain failed-consensus signals and unresolved decision ambiguity.

Missing-Signal Extension: Abandoned and Stale PRs

Section titled “Missing-Signal Extension: Abandoned and Stale PRs”

To capture those signals, we added:

  • closed-unmerged PRs in the same 90-day window
  • open PRs stale for 14+ days and 30+ days

Closed-unmerged traction

Section titled “Closed-unmerged traction”
  • Queried: 500 closed-unmerged PRs (GitHub API list cap)
  • Median discussion volume: 2
  • High-discussion abandoned share (>=10 discussion items): 2.4%

Examples:

  • #109150: 45 discussion items, closed unmerged
  • #108406: 30 discussion items, closed unmerged
  • #109526: 24 discussion items, closed unmerged
  • #111112: 22 discussion items, closed unmerged

Open stale traction

Section titled “Open stale traction”
  • Open PRs sampled: 286
  • Stale 14+ days: 36
  • Stale 30+ days: 9

Examples:

  • #108533: 42 discussion items, stale 20 days
  • #109781: 6 discussion items, stale 19 days

Interpretation update

Section titled “Interpretation update”

Measuring merged PRs alone can understate governance friction. Abandoned high-traction PRs and stale discussed PRs should be tracked as first-class outcomes in future runs.

High-Friction Examples

Section titled “High-Friction Examples”

Recurring Discussion Themes

Section titled “Recurring Discussion Themes”
  • API contract and compatibility semantics
  • Nullability/type safety and error paths
  • UX flow behavior invariants
  • Testing evidence expectations
  • Permission and reliability guardrails

Proposed ADR Pack

Section titled “Proposed ADR Pack”
  1. PR Slice Boundaries and Risk Budget
  2. API Contract Evolution Protocol
  3. Test Evidence Matrix by Change Type
  4. UI/Flow Behavioral Invariants
  5. Permission and Reliability Guardrails

Reproducibility

Section titled “Reproducibility”