Extraction: rationale node labels are a raw 80-character slice of the docstring, producing truncated mid-word titles and filenames ending in `..md`

[irvinolave-alt]

Found on graphifyy 0.9.26 (uv tool install, Python 3.11, Linux), running graphify <dir> --backend claude-cli. Unlike #2204 and #2205 this one is not export-specific — it is in the extractor.

Summary

_extract_python_rationale / _extract_js_rationale build the node label by slicing the raw docstring to 80 characters before normalizing whitespace. Three consequences:

1. the slice lands mid-word (... solo dev, ... GET articles/consulta). Si);
2. the docstring's newline + indentation survives as a run of literal spaces inside the label, because the slice happens before the \n → replacement collapses anything;
3. a docstring ending in . at the slice boundary yields an Obsidian filename ending in ..md.

These labels are not export-only — they are the label in graph.json, so they also show up in GRAPH_REPORT.md, in graphify query output, and in explain.

Where

graphify/extract.py:1075 (Python) and graphify/extract.py:1212 (JS/TS) — identical line:

label = text[:80].replace("\r\n", " ").replace("\r", " ").replace("\n", " ").strip()

The slice is applied to text, before the replacements. Reversing the order would fix (2) but not (1) or (3).

Reproduction

def list_products(...):
    """Lista/busca productos del catálogo de The Best (GET /articles/consulta).
       Si no se pasa consulta, devuelve el catálogo completo.
    """

produces, in graph.json:

{"label": "Listabusca productos del catálogo de The Best (GET articlesconsulta).      Si",
 "file_type": "rationale", "source_file": "mcp-thebest/server.py",
 "source_location": "L187", "_origin": "ast"}

and, in the Obsidian vault, the file:

Listabusca productos del catálogo de The Best (GET articlesconsulta).      Si.md

Twelve of my fourteen rationale nodes were affected. A second real example, where the truncation cuts mid-sentence and the trailing period makes a double-dot filename:

000005 - TIENDA PIURA' - '000005'..md

Expected

A rationale label that reads as a title: whitespace collapsed, cut on a word boundary, no trailing punctuation artifact.

Suggested fix

Normalize first, then slice on a word boundary:

flat = " ".join(text.split())                 # collapses \r\n, \n and indentation runs
label = textwrap.shorten(flat, width=80, placeholder="…")

textwrap.shorten never cuts mid-word and gives a visible truncation marker, so explain/query output stops implying the docstring ended there.

Bigger suggestion, separable

Even fully normalized, a docstring makes a poor node identity. Every rationale node has exactly one rationale_for edge to the thing it explains, so it could be labelled after its target — rationale: list_products(), disambiguated by source_location when one symbol carries several. That is a nicer graph (rationale nodes sort next to their subject, and the label is stable across docstring edits, so the node ID stops churning), and it makes the Obsidian vault navigable. I renamed mine that way by hand and it was a large readability win. Happy to open this as a separate enhancement issue if you'd rather keep the bug narrow.