graphify

Purpose

graphify turns a folder of raw research into a structured, evidence-backed graph the rest of the Discovery skills can interrogate. The graph is a map of what the corpus actually says — it does not extend, summarize, or infer beyond what the sources support.

It does NOT generate insights. It does not "complete" the picture. It does not silently fill in plausible-sounding entities. Where the corpus is silent, the graph is silent (or surfaces the silence as an explicit Assumption node). Where the corpus is thin, the graph says so prominently — the thin state is the primary state, not a problem to paper over.

Procedure

1. List every file in sources/. Skip sources/README.md — it is the workspace boilerplate that mogkit init ships, not a real source. Also skip any file without a readable text extension (.md, .markdown, .txt). Note any skipped non-trivial files in meta.notes.

2. For each file, read it end to end. Parse the YAML frontmatter for type (e.g. interview, ticket, prd, memo, research, transcript, note, other). If a file has no type, treat it as untagged and note it in meta.notes.

3. Compute meta:

   • generated: current ISO-8601 timestamp.
   • sourceCount: number of files contributing to the graph.
   • sourceTypes: unique types observed.
   • health: derived per the rule below.
   • notes: any skipped files, ambiguities, parse warnings.

   Health rule:

   • thin if sourceCount < 8 OR uniqueTypes < 2 (where untagged counts as zero distinct types).
   • rich if sourceCount >= 20 AND uniqueTypes >= 3.
   • developing otherwise.

4. Branch on health before extraction:

   • If health is thin: extract entities normally but state prominently in graph.md that the corpus is thin and the downstream skills will gate confidence accordingly. Do not invent additional entities to pad the graph. A small honest graph is the correct output.
   • If health is developing or rich: proceed to full extraction.