PR Splitting Cost Simulator

Adjust parameters to see how mandatory PR splitting affects your team's review pipeline.

Split ratio 1×
1 (natural)5×
Dependency fraction 40%
0%100%
Reviewer pool 3
110
Outputs vs baseline
Baseline (split ratio = 1)
Current settings
⚠ Queue runaway: PR arrival rate exceeds reviewer capacity. The review queue will grow without bound at these settings.
Model assumptions ▶
Defaults are conservative estimates. The qualitative result — that splitting increases all six costs — holds across the full range of each assumption. Drag any slider to test your own numbers.
Glossary
Split ratio
Parameter
How many sub-PRs a natural unit of work is divided into. 1 = no splitting — the PR is sized by the work. At 3×, a single feature becomes three PRs. This is the variable under the split-PR policy's control.
Dependency fraction
Parameter
Percentage of sub-PRs that are blocked on a prior sub-PR in the same split set. If you split a feature into A → B → C, B and C are dependent. Higher values mean more of your splits create sequencing constraints.
Review speed exponent (β)
Assumption
Controls how much faster reviewers process smaller sub-PRs. β=0: no speedup — each sub-PR takes the same time regardless of size (pessimistic). β=1: perfect linear scaling — splitting in half halves review time (optimistic). Default β=0.5 reflects partial benefit: sub-PRs are faster but fixed overhead doesn't shrink proportionally.
Reviewer pool
Parameter
Number of engineers available to review PRs. Adding reviewers is the only lever that increases throughput capacity. A small pool amplifies every queue-depth effect of splitting.
PRs in review
Output — queue stock
Steady-state count of PRs waiting for or undergoing review. Splitting multiplies PR arrival rate directly. More PRs in flight = longer average wait, more context-switching for reviewers.
Open dependency chains
Output — blocked work stock
Count of PR sequences where a later PR is blocked waiting for an earlier one to merge. Each chain represents an engineer who cannot proceed until the reviewer clears the blocker. Grows linearly with split ratio × dependency fraction.
Reviewer time
Output — reviewer cost
Total reviewer capacity consumed per day (reviewer-days/day). Splitting multiplies PR count but also speeds up individual reviews (controlled by β). The net effect depends on whether the speed gain (μ × s^β) outpaces the arrival rate increase (λ × s) plus rework. At β=1, splitting is neutral on reviewer load; at β<1, it increases load.
Engineer wait
Output — engineer cost
Engineer-days lost per day to blocked dependency chains. An engineer whose PR B is waiting on PR A's merge cannot ship, cannot rebase safely, and context-switches to other work — fragmenting their own output.
Integration conflicts
Output — coordination cost
Conflicts requiring manual resolution per PR. This grows superlinearly with split ratio (exponent 1.5) — each additional sub-PR touches shared interfaces and files, multiplying coordination surface faster than the number of PRs.
Production incidents
Output — coupling risk proxy
Incidents per week caused by integration conflicts that escape review. Used as a concrete proxy for coupling failure — when reviewers assess sub-PRs in isolation, cross-cutting bugs that only appear at merge are more likely to reach production.