Performance & Scalability
The Locus 1.0 release candidate was measured in an Unreal Engine 5.6 Win64 Development Editor audit using isolated synthetic fixtures. These measurements describe that test configuration; they are not a contractual capacity limit or a promise that every project and machine will produce the same values. Hardware, storage, project scale, open editor state, and source-control provider can all affect results.
Measured validation scale
Section titled “Measured validation scale”| Area | Test configuration | Measured result |
|---|---|---|
| Notes | 10,000 synthetic, test-owned Notes | Initial discovery: ~2,769 ms; search p95: ~20.557 ms; watched update maximum: ~127.043 ms; reconciliation p95: ~1.140 ms; working-set delta: ~13.21 MiB |
| Documents | 10,000 indexed Documents | Search: ~1.71 ms |
| Document editing and recovery | 512 KiB Document | Edit/recovery update: ~12.04 ms |
The 10,000-Note and 10,000-Document datasets are validation scales, not published support limits.
Reduced recurring workspace work
Section titled “Reduced recurring workspace work”The audit also measured unchanged-state work before and after targeted improvements:
| Area | Test configuration | Before | After |
|---|---|---|---|
| Documents workspace | 1,000 Documents, 5 Hz polling | ~45.637 ms/poll (~228.186 ms/sec) | ~1.343 ms/poll (~6.715 ms/sec), an approximately 97% reduction in this measured recurring workload |
| Content Browser Document projection | 10,000 Documents, unchanged index | ~4.627 ms/frame | Effectively negligible unchanged-state comparison in the audit |
| Viewport Pin query preparation | 1,000 Pins, unchanged query state | ~0.955 ms/frame | Effectively negligible query preparation in the audit |
Viewport marker projection and hit testing remain real-time work when the Editor needs them. For open Document tabs, the 512 KiB external-file fingerprint check measured ~0.360 ms per poll at two polls per second per open tab.
Idle and background behavior
Section titled “Idle and background behavior”Locus favors event-driven or revision-aware work when state is unchanged:
- Directory changes drive discovery updates instead of continuous full rescans.
- Document projection refresh is gated by relevant index changes.
- Pin query results are reused until relevant world, scope, or index state changes.
- MCP has no active server runtime while disabled. When enabled but idle, it performs lightweight session maintenance only.
These choices reduce recurring Editor work without changing the authoritative storage or source-control safety rules. For day-to-day workflow details, see Documents, Pins, and MCP Integration.