synthesis-map

Purpose

After a round of discovery interviews, the temptation is to leap straight to solutions: "we should build X, Y, Z." That move skips the synthesis step — clustering what users were trying to do and what was getting in their way — and the resulting backlog ends up being a list of solutions to vaguely-articulated problems. The strongest product organizations synthesize opportunities first, then map solutions under each.

This skill produces the opportunity layer. It clusters the interviews into the underlying jobs and pains that, if addressed, would move the named outcome — each cluster traceable to specific quotes in specific interviews. The PM then decides which opportunities to pursue and what solutions to consider under each.

It does NOT produce solutions, features, recommendations, prioritization, or any version of "and here's what to build." That step happens in a separate working session, with the PM and the team.

Procedure

1. Read graph/graph.json for context. If it does not exist, treat the batch as standalone and note that the synthesis cannot draw on prior graph structure.
2. Read meta.health (if present). Cold-start branch: on a thin corpus, even a fresh batch may not be enough to produce a stable opportunity map. State this and produce a sparse map honestly — do not generate opportunities a single interview hints at as if they were established.
3. Identify the outcome the synthesis is mapping toward. Sources of this:
   • The Outcome nodes in the current graph (e.g. trial-to-paid conversion, week-1 activation).
   • The strategic memo or PRD intent, if one is in sources/.
   • If unclear, ask the PM to name the outcome before proceeding. An opportunity map without a target outcome is unmoored.
4. Read every file in the fresh batch (typically the most recently added interviews; the PM may list them explicitly). Extract candidate opportunities — underlying jobs, needs, or pains that, if addressed, would plausibly move the outcome.
5. Cluster opportunities. Rules:
   • Cluster only when ≥2 sources independently describe the same underlying job or pain. A single-source candidate stays as a Single-source opportunity so its evidentiary thinness is visible — do not promote it into a cluster of one.
   • Cluster by the user's job, not by the surface mechanic. "Importing comments fully", "preserving deal history", "keeping audit trail" might cluster as "preserve historical context when switching tools."
   • Distinguish opportunities from solutions. "Decide admin questions before inviting the team" is an opportunity. "Add a four-question setup wizard" is a solution to that opportunity. This skill only emits the former.
6. For each cluster, attach provenance: 2–4 verbatim quotes drawn from the interviews, with file paths. The provenance must include at least two distinct source files.
7. For each cluster, note its plausible link to the outcome — one short sentence on the causal story connecting this opportunity to the named outcome. If the link is not plausible, the cluster is noise; cut it.
8. Append a Single-source opportunities section listing standalone candidates with their one source. These are not clusters; they are leads for the next round of interviews.
9. Emit the output contract.