/edikt:sdlc:discovery

Capture a structured discovery document for a feature you don't know enough about yet.

Use this when an idea has unknowns that matter enough to resolve before writing a PRD. It's a peer to /edikt:brainstorm, not a wrapper — discovery has its own sections (Known, Unknown, Kill Criteria, Discovery Plan) and evolves independently.

When to use which

Tool	Use when
/edikt:brainstorm	Open-ended exploration. You're not yet sure what you're solving.
/edikt:sdlc:discovery	You have a candidate feature, but the unknown-to-known ratio is high. Reduce uncertainty before scoping.
/edikt:sdlc:prd	You already know the feature. Now define what done means.

A brainstorm can graduate into a discovery. A discovery can graduate into a PRD.

Usage

/edikt:sdlc:discovery realtime collaboration for the editor
/edikt:sdlc:discovery BRAIN-007                   # lift an existing brainstorm
/edikt:sdlc:discovery                             # asks what to explore

What the command asks

Four questions, one at a time. Each is intentionally hard to skip — the answers are the document.

1. What do you know for certain? Facts, data, prior research, confirmed user statements. Be precise: "3 of 12 users said X" beats "users seem to want X."
2. What are you most uncertain about? Three to five unknowns, ordered by impact on the decision — which one, if resolved, most changes whether you build this. Not alphabetical, not easiest-first.
3. What would change your mind? Kill criteria. "If fewer than N% of users X, we stop." Concrete and measurable where possible.
4. What's the smallest experiment that reduces the biggest uncertainty? The discovery plan. Aim for the cheapest signal that moves the biggest unknown.

Output

Writes DISCOVERY-NNN-<slug>.md to docs/product/discovery/:

---
type: discovery
id: DISCOVERY-001
title: "Realtime collaboration for the editor"
status: active
graduates_to: null
---

## Context
<1-2 paragraphs — what this is about, why it matters now>

## Known
- 12 of 30 customer interviews mentioned losing edits during multi-user sessions
- Existing CRDT library benchmarked at 50ms p99 with 10 concurrent users

## Unknown
1. **Whether last-write-wins is acceptable for our 95th percentile workflow** — drives the entire engineering approach
2. **How much network jitter our enterprise customers tolerate** — sets the offline-first decision
3. **Whether undo across sessions is table-stakes** — large UX surface area

## Kill Criteria
- If fewer than 30% of pilot users edit collaboratively in week 1, we stop
- If CRDT memory cost exceeds 100KB per active doc, we pick a different approach

## Discovery Plan
| # | Experiment | Method | Success signal | Timeline |
|---|------------|--------|----------------|----------|
| 1 | Last-write-wins prototype | 2-day spike with 5 internal users | <2 reported conflicts/day | Week 1 |
| 2 | Memory benchmark | Yjs synthetic load | <100KB at 1k ops | Week 1 |

## Assumptions Register
| Assumption | Confidence | Source |
|-----------|-----------|--------|
| Users want simultaneous editing | medium | 12/30 interviews |

## Outcome
_Filled when discovery concludes._

The document has its own status (active, concluded, killed) and its own revision flow. You update Known/Unknown as findings come in — the document is alive while the question is open.

Graduating to a PRD

When uncertainty is reduced enough to commit, run:

/edikt:sdlc:prd DISCOVERY-001

This pre-populates the PRD's Problem from your Known section and the Open Questions from anything still in Unknown. The discovery's frontmatter gets graduates_to: PRD-NNN so the trail is bidirectional.

Why discovery exists separately from brainstorm

Brainstorms are open-ended. Specialist agents join as topics emerge. The artifact looks like a working memo.

Discoveries are structured uncertainty reduction. The four sections are not optional — they're what makes the document useful three months later when someone asks "why did we decide X?"

The duplication is deliberate. Each can evolve independently of the other.

What's next

• /edikt:brainstorm — open-ended exploration
• /edikt:sdlc:prd — graduate this discovery into a PRD
• PRD v2 Deep Dive — what comes after discovery concludes