Most games don't die in production. They die in pre-production because nobody defined what "done" looks like. A pipeline isn't a checklist—it's a contract between disciplines that turns ambiguity into ship dates.
The Five-Stage Contract
Every studio renames these, but the dependencies don't change. Skip the gate reviews and you'll pay interest in crunch.
| Stage | Gate Criteria | Indie Timeline | AAA Timeline |
|---|---|---|---|
| Concept | One-pager + risk matrix signed off | 1-2 weeks | 3-6 months |
| Pre-production | Vertical slice playable, tech risks retired | 1-3 months | 12-24 months |
| Production | Content complete, zero P0 bugs | 3-12 months | 2-4 years |
| Polish | Cert submission build stable | 1-2 months | 6-12 months |
| Live | Day-1 patch ready, telemetry dashboards live | Ongoing | Ongoing |
Where Pipelines Fracture
Unity and Unreal handle the heavy lifting, but the seams between tools eat schedules. Three fracture points appear in every postmortem:
Second: design specs written in Confluence that engineers can't parse. Third: QA finding "design intent" bugs because acceptance criteria lived in someone's head.
The CI/CD Backbone
Modern pipelines treat builds as artifacts, not events. Every commit triggers validation. Here's a minimal GitHub Actions workflow that catches the basics before humans waste time:
Outsourcing Without Losing Control
External art and co-dev studios scale capacity, but they need tighter interfaces than internal teams. Define deliverables as acceptance tests, not asset lists.
| Deliverable | Acceptance Criteria | Review Cadence |
|---|---|---|
| Character model | Tri count ≤ 25k, 4k textures, LODs 0-2, rig passes auto-validation | Twice weekly sync + async PR reviews |
| Environment tile | Modular snap grid compliant, lightmap UVs < 0.5% overlap, streaming ready | Weekly playtest build drop |
| VFX prefab | GPU instancing compatible, < 2ms/frame at 1080p, parameterized for LOD | Per-effect sign-off |
Metrics That Prevent Surprises
Dashboards replace status meetings. Track these four and you'll see trouble two sprints out:
| Metric | Healthy Range | Action Threshold |
|---|---|---|
| Build success rate | > 95% | < 90% for 3 days |
| Mean time to merge | < 4 hours | > 8 hours |
| Open P0/P1 bugs | < 5 per dev | > 10 per dev |
| Content completion velocity | Steady ±15% | Two sprints declining |
Ship It
Your pipeline is a product. Version it. Retro it. Delete steps that don't prevent defects or accelerate decisions. Next sprint: pick one fracture point, instrument it, and cut the feedback loop in half. That's how you ship.
✦










