Wedge · 1/5

Discovery

Customer research as a versioned, compounding asset.

knowledge engine: shipped

The problem

Discovery is the wedge where most PM tooling rots into a Notion graveyard of half-summarized interviews. The pattern is universal: the team runs research, the notes sit somewhere, and three months later nobody can answer 'what did we actually learn?' without re-reading thirty documents. The strong product teams treat their research corpus as a versioned asset that grows; the weak ones treat each interview as a one-off.

The level-up path

Discovery is the only wedge where mogkit ships a full knowledge engine: a stateful, multi-step system that operates on your corpus of sources and a generated knowledge graph. Everything you collect compounds.

The loop:

interview-guide → (collect sources) → graphify → assumption-audit
      ↑                                              ↓
      └──────── prd-interrogate ← discovery-query / synthesis-map

You add interviews, tickets, and notes to sources/. graphify turns the corpus into a schema-valid graph with provenance on every claim. discovery-query answers questions against it. assumption-audit ranks what you’d be betting on without evidence. prd-interrogate makes your next PRD sharper by interrogating the evidence behind it first — it never writes the PRD; you do.

Visual setup walkthrough

Install mogkit and scaffold a workspace.
```
npx mogkit init my-research
cd my-research
```
You’ll see sources/, graph/, knowledge/, a CLAUDE.md orienting every Claude Code session, and .claude/skills/ with all 13 skills installed.
Add your first source.
```
mogkit add
```
Pick a file (an interview transcript, a support ticket export, a doc you took notes in). Tag its type. It lands in sources/ with frontmatter.
Check the corpus health.
```
mogkit status
```
On your first day this will say thin. That’s the correct, useful answer — the skills will gate their confidence on it.
Open the workspace in Claude Code and run graphify. The skill reads everything in sources/, builds the graph in graph/graph.json, and writes a human-readable graph/graph.md. On a thin corpus, it tells you so loudly. That is the honest result.
Interrogate. Once you have ~8 sources across 2+ types, run discovery-query, assumption-audit, or — when you’re about to write a PRD — prd-interrogate. The output is always a reasoning scaffold, not a finished doc.

What the skills will not do

They will not write your PRD. They will not infer entities your corpus doesn’t mention. They will not fabricate a rich graph from a handful of documents. The thin-corpus state is loud and honest on purpose — it is the primary state for most workspaces, and pretending otherwise builds a PM tool that hallucinates research.

Skills to install

graphify

advanced

discovery engine discovery

Reads every file in `sources/` and produces a schema-valid `graph/graph.json` plus a human-readable `graph/graph.md`. Every node and edge carries provenance; unsourced claims become explicit `Assumption` nodes. Rates corpus `health` and says so loudly when it is thin.

discovery-query

intermediate

discovery engine discovery

Answers a discovery question by interrogating the graph — returning grounded findings with provenance, the explicit gaps in the evidence, and the discovery questions that would close them. Refuses to answer beyond what the corpus supports, and names every gap.

assumption-audit

intermediate

discovery engine discovery

Reads the graph and ranks every claim with single-source or zero-source backing by risk — surfacing what the team would be betting on without evidence, each tied to the decision it would affect. Surfaces risk; does not resolve it.

prd-interrogate

advanced

discovery engine discovery

Turns a PRD intent into an interrogation — what your evidence supports, what is assumed, what is unvalidated, and who you haven't talked to — so you write a sharper PRD yourself. Never produces the PRD.

interview-guide

intermediate

discovery engine discovery

Reads the graph (specifically its gaps) and produces a discovery interview guide targeted at the biggest current gaps — non-leading, JTBD-grounded questions, each tied to a named gap so the next conversation actually moves the graph.

synthesis-map

advanced

discovery engine discovery

Takes a fresh batch of interviews and turns them into an opportunity map — the input to an Opportunity Solution Tree. Maps opportunities (the underlying jobs and pains that would move outcomes), not solutions; keeps provenance on every opportunity.

interview-coach

beginner

standalone discovery

Coaches the interviewer based on a discovery-interview transcript — where they led the witness, where they accepted an answer without a follow-up, where they pitched instead of listened, and the highest-value follow-up questions they missed. Teaches the craft; never just summarizes the content.

All of these install automatically when you run npx mogkit init.

Material

A curated path. Not a link dump.

Continuous Discovery Habits book

by Teresa Torres

The single best book on running weekly customer discovery as a habit, not a project. The Opportunity Solution Tree comes from here.

The Mom Test book

by Rob Fitzpatrick

How to run customer interviews without getting lied to. Short, brutal, indispensable.