Benchmarks
It maps how a codebase fits together, not just what is in it
Run on 21 real open-source projects plus Salesforce orgs, fully offline and deterministic. The figures below come straight from a reproducible pass; the structural, area, and coverage numbers are identical on every run.
Reads Python, JavaScript, TypeScript, Rust, Go, C, C++, C#, Dart and Flutter, and Salesforce Apex with SOQL.
Every repository in the run
| Repository | Language | Files | Areas | Connected |
|---|---|---|---|---|
| linux | C | 63,865 | 200 | n/a |
| node | JS/TS | 35,955 | 126 | n/a |
| vscode | TypeScript | 11,225 | 94 | 78% |
| flutter | Dart/Flutter | 10,512 | 35 | n/a |
| pytorch | Python/C++ | 9,213 | 150 | 8% |
| kubernetes | Go | 6,528 | 33 | n/a |
| AppFlowy | Rust/Dart | 2,813 | 4 | n/a |
| codex | Rust | 2,813 | 147 | 35% |
| postgres | C | 2,574 | 29 | n/a |
| zulip | Python | 2,497 | 44 | n/a |
| lwc-recipes | Salesforce | 547 | 87 | 18% |
| localsend | Dart/Flutter | 459 | 72 | 19% |
| apex-recipes | Salesforce | 275 | 43 | 29% |
| source-deploy-retrieve | TypeScript | 263 | 44 | 84% |
| coral-cloud | Salesforce | 244 | 22 | 10% |
| easy-spaces-lwc | Salesforce | 153 | 26 | 4% |
| CleanArchitecture | C# | 152 | 38 | 51% |
| ebikes-lwc | Salesforce | 147 | 26 | 13% |
| dreamhouse-lwc | Salesforce | 139 | 26 | 13% |
| automation-components | Salesforce | 84 | 11 | 15% |
| code-analyzer | TypeScript | 84 | 20 | 77% |
Coverage is 100% on every repository (every scanned source file lands in an area) with zero duplicate area names. The connected column is the share of links carrying an import or reference signal; it is shown only where the dependency graph runs, since very large repositories use a faster directory pass.
Salesforce
Feature areas, not folder dumps
On Salesforce it groups LWC, Apex, and objects into the features they form, following @salesforce/apex calls, Apex references, and SOQL into your data model. Config metadata stays in clean, named buckets. Deterministic, 100% coverage.
Methodology
In-scope files are what remains after standard excludes: .git, build output, vendored dependencies, and gitignored paths. One sequential pass on a laptop, no network and no model. The same discovery engine runs in the desktop app, the CLI, and the MCP server.