⚙ Pipeline Playground
Build it. Break it. Fix it. Ship it.
You have inherited a broken pipeline.
PR #101 | Lint (30s) | Unit (2m) | API (3m) | E2E (12m) - Selenium, sequential | Merge
Developers are waiting ~28 minutes for feedback on every PR.
Your job: make it faster without making production less safe.
Mission 01 — The Monolith Bottleneck
| Company | FinBank |
| Team | 5 developers |
| Application | Banking API |
| Release frequency | 10 PRs/day |
Mission objectives
| PR feedback | < 6 min |
| Estimated Runner Cost | ≤ $400 |
| Critical escapes | = 0 |
Available tools
Move stage · Parallelize · Change runner tier
PR
MERGE
RELEASE
Runner Tier
Cost model (MVP: Per-Run). Estimated Runner Cost = Σ(stage duration × runner rate) + tier overhead, for one full run of 20 PRs. Moving or parallelizing a stage changes wall-clock time, not total compute - each stage still executes exactly once per PR - so cost here only moves via Runner Tier. Modeling cost drops from triggering expensive stages less often (e.g. moving them past every PR push) needs a trigger-frequency layer this MVP doesn't build yet.
Terminal
Live Activity
⚠ Flaky result detected on for
Pipeline Debrief
Before / After
Why?