Review Mentorship Pairing Workflow

Pair junior reviewers with seniors on real reviews, calibrate with shared rubrics, and graduate reviewers by demonstrated finding quality.

beginner~20 mincode-reviewmentorshipdeveloper-experience

Objective

Grow the reviewer bench through deliberate pairing on real reviews, so review capacity scales with the team instead of concentrating.

Environment

Senior reviewers with mentoring capacity
New or junior engineers to onboard into review
A tracking place for review findings and calibration

Steps

1. Pair on real reviews, not exercises

Assign each junior as second reviewer alongside a senior on live PRs, with the senior's review as the calibration reference.

Expected state: Every junior accumulates real-review experience with reference outcomes.

2. Compare findings before the senior submits

Have the junior review first and privately record findings, then compare against the senior's submitted review, discussing every divergence.

Expected state: Divergences become teaching moments with concrete cases.

3. Calibrate against a shared rubric

Maintain a rubric of finding quality: severity, specificity, and constructiveness, scoring both reviewers' findings to align standards.

Expected state: Rubric scores converge between junior and senior over paired reviews.

4. Rotate pairs and domains

Rotate juniors across seniors and code areas, preventing single-style cloning and broadening domain coverage.

Expected state: Each junior has paired with multiple reviewers across at least two domains.

5. Graduate by evidence

Promote a junior to first reviewer when their private findings match or exceed the senior reference across a series, not by tenure.

Expected state: Graduation decisions cite the finding-quality record.

Expected outcome

A growing bench of calibrated reviewers who find real issues at senior-recognized quality, with review capacity scaling as the team grows.

Verification

Graduated reviewers's solo reviews show sustained finding quality; the paired-review divergence rate declines over each onboarding.

Reviewers are grown on real diffs, in pairs

Review skill is pattern recognition over failure modes, and it grows on real code with a reference to diverge from. Pairing juniors as second reviewers, comparing private findings against the senior’s submitted review, turns every PR into a calibration session.

pairing session (PR #5188, junior B + senior A):
  B reviews privately first (20 min):
    [1] blocking: retry loop has no cap
    [2] suggestion: rename `d` -> `delayMs`
    [3] (missed) -- division by zero when
        `interval == 0`
  A submits:
    [1] same as B[1]
    [2] A: "blocking: interval=0 crashes -- guard"
  divergence discussion:
    B[3] missed -> pattern taught: "always test
    the zero/empty boundary of every number"

rubric (finding quality 1-5): severity accuracy,
  specificity, constructiveness
  B series: 2.9 -> 3.6 -> 4.1 -> 4.2  graduate

Divergence is the curriculum

The missed finding is the lesson: not an abstract principle but the specific boundary check the junior walked past, attached to real code they will pass again. Divergence discussion is where senior intuition gets articulated into transferable patterns.

Graduation by record, not calendar

Tenure-based reviewer promotion produces entitled approvals; evidence-based promotion produces trusted ones. The series of private-versus-reference comparisons is a record that makes the promotion defensible to both the graduate and the team relying on them.