DEPENDENCY-CHAIN MAP Stage 1: Code review — capacity: reviewers self-report as not bottlenecked; average wait 4 hours Stage 2: CI test suite — capacity: 12 minutes fixed runtime, automated, no indication of queuing Stage 3: Manual QA sign-off — capacity: turnaround has grown from same-day to 2-day over the past month Stage 4: Deploy — capacity: 5 minutes, no indication of queuing Dependency: strictly sequential, each stage must fully complete before the next starts (as stated). BINDING CONSTRAINT: Stage 3 — Manual QA sign-off CONFIDENCE LABEL: VERIFIED. The evidence directly shows queuing at this specific stage: turnaround time has doubled-plus (same-day to 2-day) over a defined period, coinciding with a stated 30% PR volume increase, and the input explicitly states PRs are "visibly piling up" before QA. This is direct backlog evidence at a named stage, not an inference from general slowness. By contrast, code review's 4-hour wait is NOT flagged as the constraint despite sounding slow — the reviewers themselves report no backlog pressure, and no queuing evidence is given for that stage, only a wait-time figure that may simply reflect how the process normally runs. EXPLOIT / SUBORDINATE / ELEVATE PLAN EXPLOIT (get more from QA sign-off as it currently exists, no added headcount): identify whether all PRs currently require the same depth of manual QA sign-off, or whether some categories (e.g., documentation-only changes, previously-reviewed patterns) could use a lighter-weight check within the existing QA process — this increases effective throughput of the same QA capacity without adding people. SUBORDINATE (what everything else should defer to QA's pace): code review should NOT be further sped up right now — a 4-hour review wait feeds directly into an already-backlogged QA stage, so shrinking the 4-hour wait to 1 hour would only make PRs arrive at QA faster without QA's capacity increasing, worsening the visible pile-up rather than helping it. This is a concrete case of a stage that looks improvable but should NOT be optimized while it isn't the constraint. Similarly, deploy automation improvements (already fast at 5 minutes) are not worth engineering time right now. ELEVATE (only after exploit and subordinate, if still insufficient): consider adding QA sign-off capacity (additional reviewer, or splitting sign-off duties) only if the exploit step (differentiated QA depth by change type) doesn't close the gap. This is the last resort here, not the first move, precisely because it costs headcount or budget that the other two steps might make unnecessary. REPEAT NOTE: once QA sign-off capacity matches current PR volume, watch for the constraint to shift — likely back to code review, given its 4-hour wait is currently tolerated only because it isn't yet the binding constraint. Re-run this analysis once QA's queue is resolved rather than assuming the pipeline is now fully unconstrained. ```