Summary
GRAPH_REPORT.md's Knowledge Gaps section always offers the same two possible explanations for an isolated node, regardless of whether the graph could ever actually support one of them:
These have ≤1 connection - possible missing edges or undocumented components.
"Undocumented components" only means anything when a semantic layer exists — document/paper/image nodes an LLM extracted meaning from, where an isolated node genuinely could be "mentioned but never explained further." A code-only graph (--code-only, or any run where semantic extraction never happened) has no such layer at all: there is nothing to be "undocumented" about, ever, on that graph. Offering it as a live possibility on every report regardless is misleading — a reader with a pure-code graph has no way to know that half the explanation offered to them can never be true.
Verified across 12 real code-only projects in this run: 0/12 ever had a document/paper/image node, and 100% of their isolated-node messages still offered "undocumented components" as if it might apply.
Root cause
report.py, the Knowledge Gaps section:
lines.append(
" These have ≤1 connection - possible missing edges or undocumented components. "
...
)This string is unconditional — it never checks whether the graph has anything that could BE undocumented.
Suggested fix
The information needed is already in the graph, no new build-time flag required — VALID_FILE_TYPES (validate.py) already distinguishes document/paper/image from code/rationale/concept:
has_semantic_layer = any(
G.nodes[n].get("file_type") in ("document", "paper", "image")
for n in G.nodes()
)
reason = (
"possible missing edges or undocumented components"
if has_semantic_layer else "possible missing edges"
)
lines.append(f" These have ≤1 connection - {reason}. ...")Patched and verified locally: regenerated all 12 real code-only projects' GRAPH_REPORT.md — every one now reads "possible missing edges" only, with no wording claiming a possibility the graph structurally cannot have.
Related
Environment
graphifyy==0.9.65 (PyPI)- Windows 11
- AST-only extraction (
--code-only --no-label), no LLM
Summary
GRAPH_REPORT.md's Knowledge Gaps section always offers the same two possible explanations for an isolated node, regardless of whether the graph could ever actually support one of them:"Undocumented components" only means anything when a semantic layer exists — document/paper/image nodes an LLM extracted meaning from, where an isolated node genuinely could be "mentioned but never explained further." A code-only graph (
--code-only, or any run where semantic extraction never happened) has no such layer at all: there is nothing to be "undocumented" about, ever, on that graph. Offering it as a live possibility on every report regardless is misleading — a reader with a pure-code graph has no way to know that half the explanation offered to them can never be true.Verified across 12 real code-only projects in this run: 0/12 ever had a document/paper/image node, and 100% of their isolated-node messages still offered "undocumented components" as if it might apply.
Root cause
report.py, the Knowledge Gaps section:This string is unconditional — it never checks whether the graph has anything that could BE undocumented.
Suggested fix
The information needed is already in the graph, no new build-time flag required —
VALID_FILE_TYPES(validate.py) already distinguishesdocument/paper/imagefromcode/rationale/concept:Patched and verified locally: regenerated all 12 real code-only projects'
GRAPH_REPORT.md— every one now reads "possible missing edges" only, with no wording claiming a possibility the graph structurally cannot have.Related
Environment
graphifyy==0.9.65(PyPI)--code-only --no-label), no LLM