Getting started
Overview
Understand Performance Lab's real-player evidence workflow, supported decisions, explicit non-passing states, and bundled Core boundary.
Performance Lab is a reproducible performance decision system for Unity. It captures real player evidence, records metric capability and environment identity, aggregates repetitions, manages reviewed baselines, evaluates budgets, analyzes BuildReports, and exports canonical CI/report evidence without requiring an account or telemetry.
Minimum supported Unity Editor: 6000.0.79f1. Earlier 6000.0.0f1 receipts are retained as historical evidence only.
Supported workflows
- Scenario and Run Center — create run profiles and execute local desktop runs from one surface.
- Immutable desktop repetitions — build once, inventory the exact player bytes, and verify those bytes before every repetition.
- External-device capture kits — build Android, iOS, or WebGL once, collect a bounded queue of portable
.kplabbundles, then import the exact expected inventory. - Metrics — CPU, GC, GPU, memory, draw calls, batches, loading transitions, custom
ProfilerMarkertimings, registered extension metric providers, and explicit capability states includingNotApplicable. - Statistics — minimum, median, p90, p95, p99, maximum, raw/included/excluded counts, repetition distributions, and explicit outlier policy.
- Baselines and decisions — reviewed immutable baseline revisions, history, rollback, environment matching, absolute/relative budgets, A/B, and contiguous A-B-B-A.
- BuildReport analysis — build duration, output size, files, warnings/errors, options, environment fingerprint, baseline comparison, budgets, and evidence export.
- Automation — the same run service is available through
KiwiPerformanceBatch.Executefor unattended CI. - Recovery — interruption preserves authenticated partial evidence; recovery retries as a new reviewed run instead of silently resuming a benchmark mid-state.
Guides
- Scenario, Run Center, and complete workflow
- Statistics, baselines, A/B, and A-B-B-A
- Android, iOS, WebGL, and portable bundles
- CI and batch execution
- Extension metric providers
- Cancellation and recovery
- Update, rollback, removal, and migration
- Samples
- Troubleshooting
- Support bundle
- Limitations and evidence interpretation
- Product credits
- Distribution terms
- Third-party notices
Evidence rule
A missing, unsupported, unavailable, permission-denied, unknown, invalid, environment-mismatched, incomplete, or checksum-invalid requirement is not a pass. A NotApplicable metric is retained as explicit sample-free evidence, is not compared or budgeted as a supported measurement, and never becomes a pass. Treat only complete supported evidence from the intended player, build type, target, and reviewed environment as release evidence.
Included Core component
Performance Lab includes the exact net.kiwistudios.releasecore 26.8.5 package as a required bundled internal component. Release Core is not a separately published product. Its shared package identity allows compatible Kiwi products to use one verified Core installation.
Support: support@kiwistudios.co
Still stuck?
Bring the exact evidence with you.
Include Performance Lab and Unity versions, the scenario and Run Profile, target and build type, environment identity, focused logs, and only the smallest replayable evidence needed to reproduce the problem.