Context
Dogfood dependency graph output was dominated by noisy common-symbol edges.
Evidence
deps reported very large edges through symbols such as Regex.
SymbolExtractor.cs -> BoundedRegex.cs and LanguageReferenceExtractionSupport.cs -> ReferenceExtractor.PatternTypeReferences.cs dominated the top output.
- The graph already appears to know about dependency noise symbols, but JSON/edgelist output still surfaced them prominently.
Additional dogfood evidence
hotspots --group-by file --exclude-tests surfaced suspicious common-symbol hotspots such as CommandExitCodes.cs with about 2820 refs across only 2 symbols and LspServer.GetInt32 with about 1405 refs.
- Model records such as
SymbolRecord and FileRecord, tiny helpers such as WorkerOutputBuffer.AppendLine, and generic command/error constants can dominate rankings despite not being the most useful architectural review targets.
- This appears related to common-name/reference fan-out noise, not only deps graph output.
Risk
Noisy edges and hotspots hide real architectural dependencies and make graph/hotspot output less useful for audits.
Suggested direction
Apply common-symbol/noise filtering consistently across graph and hotspot output modes, or expose an explicit noise profile flag. Add tests that keep tiny constants, DTO records, and common helper names from dominating top-level architectural rankings.
Context
Dogfood dependency graph output was dominated by noisy common-symbol edges.
Evidence
depsreported very large edges through symbols such asRegex.SymbolExtractor.cs -> BoundedRegex.csandLanguageReferenceExtractionSupport.cs -> ReferenceExtractor.PatternTypeReferences.csdominated the top output.Additional dogfood evidence
hotspots --group-by file --exclude-testssurfaced suspicious common-symbol hotspots such asCommandExitCodes.cswith about 2820 refs across only 2 symbols andLspServer.GetInt32with about 1405 refs.SymbolRecordandFileRecord, tiny helpers such asWorkerOutputBuffer.AppendLine, and generic command/error constants can dominate rankings despite not being the most useful architectural review targets.Risk
Noisy edges and hotspots hide real architectural dependencies and make graph/hotspot output less useful for audits.
Suggested direction
Apply common-symbol/noise filtering consistently across graph and hotspot output modes, or expose an explicit noise profile flag. Add tests that keep tiny constants, DTO records, and common helper names from dominating top-level architectural rankings.