Summary
In supermodel audit and the underlying impact endpoint, every file's blastRadius.transitiveDependents and affectedFiles come back as 0, even when directDependents has a meaningful non-zero count. This makes blast-radius output misleading — large central files appear to have no downstream reach.
Observed on Swift but likely not language-specific — it looks like the reverse-reachability step isn't propagating past the first hop.
Environment
supermodel 0.6.10, macOS arm64- Language observed: Swift (monorepo with multiple SwiftPM targets)
- Likely affects any language where direct edges exist but transitive closure is computed
Reproduction
Run supermodel audit on a codebase with clear file-level dependency chains. The audit report's Impact Analysis table shows:
| File | Direct | Transitive | Affected Files |
| Sources/.../TemplateServiceImpl.swift | 97 | 0 | 0 |
| Sources/.../FinancialIntegrationStoreImpl... | 75 | 0 | 0 |
| Sources/.../OverviewTabView.swift | 66 | 0 | 0 |
| Tests/.../EthicalWallMutationGuardTests.swift| 66 | 0 | 0 |
| ... (every row has Transitive=0)
The raw impact JSON (from ~/.supermodel/cache/<impact-hash>.json) confirms:
{
"metadata": {
"totalFiles": 382,
"totalFunctions": 5153,
"analysisMethod": "reverse_reachability_call_graph"
},
"impacts": [
{
"target": { "file": ".../AuthPolicy.swift", "type": "file" },
"blastRadius": {
"directDependents": 0,
"transitiveDependents": 0,
"affectedFiles": 0,
"riskScore": "low"
},
"affectedFunctions": [],
"affectedFiles": []
},
...
]
}Every impacts[].blastRadius.transitiveDependents is 0 across all 382 files analyzed, and every affectedFunctions / affectedFiles list is empty.
Expected
For a target file with, say, Direct=97, transitiveDependents should include the downstream reach (callers of callers), and affectedFiles should be a non-empty list unless the file genuinely has no consumers beyond immediate imports.
Hypothesis
- The first hop of the reverse-reachability walk is producing
directDependents correctly (those numbers appear believable relative to the codebase).
- The closure step past hop 1 either:
- Terminates immediately (seed set not being expanded), or
- Produces a set that is later filtered down to empty before serialization.
- Not language-specific; would likely repro on any language where the graph has call/import edges.
Happy to share the cached impact JSON (with identifiers redacted) and the underlying graph JSON if useful.
Summary
In
supermodel auditand the underlying impact endpoint, every file'sblastRadius.transitiveDependentsandaffectedFilescome back as 0, even whendirectDependentshas a meaningful non-zero count. This makes blast-radius output misleading — large central files appear to have no downstream reach.Observed on Swift but likely not language-specific — it looks like the reverse-reachability step isn't propagating past the first hop.
Environment
supermodel 0.6.10, macOS arm64Reproduction
Run
supermodel auditon a codebase with clear file-level dependency chains. The audit report's Impact Analysis table shows:The raw impact JSON (from
~/.supermodel/cache/<impact-hash>.json) confirms:{ "metadata": { "totalFiles": 382, "totalFunctions": 5153, "analysisMethod": "reverse_reachability_call_graph" }, "impacts": [ { "target": { "file": ".../AuthPolicy.swift", "type": "file" }, "blastRadius": { "directDependents": 0, "transitiveDependents": 0, "affectedFiles": 0, "riskScore": "low" }, "affectedFunctions": [], "affectedFiles": [] }, ... ] }Every
impacts[].blastRadius.transitiveDependentsis 0 across all 382 files analyzed, and everyaffectedFunctions/affectedFileslist is empty.Expected
For a target file with, say, Direct=97,
transitiveDependentsshould include the downstream reach (callers of callers), andaffectedFilesshould be a non-empty list unless the file genuinely has no consumers beyond immediate imports.Hypothesis
directDependentscorrectly (those numbers appear believable relative to the codebase).Happy to share the cached impact JSON (with identifiers redacted) and the underlying graph JSON if useful.