Token benchmark

Generated by benchmarks/RoslynLens.TokenBenchmark. For every single-type source file, it compares the full file (what an agent pays to read when it opens the file) against the JSON returned by the real get_public_api tool. Both sides are counted with the GPT-4o (o200k_base) tokenizer whose vocabulary ships embedded in the package, so the run is offline and reproducible.

  • Solution: RoslynLens.slnx
  • Types compared: 94 (skipped 74 unresolved/empty)
  • Median reduction: 73.1 %
  • Pooled reduction (size-weighted): 77.4 %
  • Total tokens: 68,795 source -> 15,565 lens

Largest 20 types by source size

Type Source tokens Lens tokens Reduction
ValidateGranitConventionsTool 2,690 76 97.2 %
ComplexityAnalyzerTests 2,640 1,208 54.2 %
SymbolResolver 2,461 596 75.8 %
FindDeadCodeTool 2,129 80 96.2 %
FindSurprisingDependenciesTool 2,036 70 96.6 %
WorkspaceManager 1,905 510 73.2 %
GetCommunitiesTool 1,653 70 95.8 %
GetModuleDependsOnTool 1,572 74 95.3 %
SolutionDiscoveryTests 1,529 696 54.5 %
TypeStructureHelperTests 1,510 454 69.9 %
SymbolResolverTests 1,441 493 65.8 %
DetectCircularDependenciesTool 1,421 70 95.1 %
GetTypeOverviewTool 1,325 66 95.0 %
FindIsolatedSymbolsTool 1,297 72 94.4 %
GetComplexityMetricsTool 1,291 76 94.1 %
HardcodedSecretDetectorTests 1,240 525 57.7 %
GetTestCoverageMapTool 1,163 70 94.0 %
FindGodNodesTool 1,124 74 93.4 %
GetDiagnosticsTool 1,119 68 93.9 %
HardcodedSecretDetector 1,052 120 88.6 %