wize-research · diff

git:20260613.a8c0f41 to git:20261019.207e46c

24 added, 90 removed. Audit A to A.

---

- description: "1-analysis: Research"

+ description: "Use quando precisar pesquisar algo mas não souber qual tipo de pesquisa se aplica."

globs:

alwaysApply: false

---

# Research

- # Research

- **Goal.** Surface evidence that hardens the brief and trigger map. Frame the questions Pepper actually needs answered, do focused passes (market / competitors / analytics / interviews), and write the synthesis in a way Maria Hill can quote in the PRD.

- Pepper drives. Peggy edits prose. Output lands in `.wize/planning/research.md` with raw materials in `.wize/knowledge/research/`.

+ # Research (Dispatcher)

- ## Inputs

+ **Goal.** Classify the user's research question and delegate to the correct variant. This skill does NOT execute research directly — it routes to one of the specialized research skills.

- - Open questions in `.wize/planning/brief.md`.

- - Hypotheses in `.wize/planning/ux/trigger-map.md`.

- - External access (browser, MCPs) and/or attached materials (decks, transcripts).

+ ## Classification guide

- ## Outputs

+ Analyze the user's question for these signals. Score each category by keyword hits; delegate to the highest-scoring category.

- - `.wize/planning/research.md` — short synthesis (≤ 2 pages) referencing sources.

- - `.wize/knowledge/research/{slug}.md` — one file per major source (raw notes, links, attached PDFs).

+ |---|---|---|---|

+ | **Market** | concorrência, concorrentes, mercado, competidor, pricing, preço, posicionamento, cliente, segmento, TAM, SAM, market share, G2, Capterra | competitor, market, pricing, positioning, customer, segment | `wize-market-research` |

+ | **Domain** | domínio, indústria, setor, regulatório, regulação, compliance, legislação, lei, normativo, PESTLE, tendências do setor | domain, industry, sector, regulatory, compliance, legislation | `wize-domain-research` |

+ | **Technical** | técnico, tecnologia, stack, biblioteca, framework, arquitetura, ferramenta, linguagem, banco de dados, API, protocolo, desempenho, escalabilidade | technical, technology, stack, library, framework, architecture, tool, language, database, API, protocol, performance, scalability | `wize-technical-research` |

## Steps

- ### 1. Frame the questions

- Convert open questions + hypotheses into a numbered list. Each question is:

- - **Specific** — "What's the median onboarding time for the top 5 competitors?" not "How is onboarding usually done?".

- - **Answerable in a finite pass** — if it takes weeks, decompose.

- - **Tagged with category** — `market / competitive / analytics / interview / desk-research / regulatory`.

- ### 2. Choose the framework that fits each question

- | Category | Recommended frame | Tools |

- |---|---|---|

- | Market sizing | TAM/SAM/SOM, top-down + bottom-up reconciled | Statista, public filings, gov registries |

- | Competitive | Porter's 5 + feature matrix + pricing scan | competitor sites, G2/Capterra, Pricing/Plans pages |

- | User research | Jobs-to-be-Done (Christensen / Klement) | 5–8 interviews (semi-structured, 30min) |

- | Behavioral | 5W2H + analytics funnel | product analytics (Amplitude/PostHog), session replays |

- | Compliance/legal | Reg list + clause map | gov sites, legal counsel review |

- | Macro context | PESTLE (political/economic/social/technological/legal/environmental) | desk research |

- ### 3. Run focused passes

- One question, one pass. Don't bundle. For each:

- - Define what "enough evidence" looks like before starting.

- - Cite or it didn't happen. Links + access date.

- - When numbers conflict across sources, write down both and flag.

- ### 4. Synthesize

- In `research.md`, for each original question, write:

- ```markdown

- ### Q1. {{the question}}

- **Finding:** one sentence answer.

- **Confidence:** high | medium | low

- **Why we believe it:** 2–3 lines + citations [1][2][3].

- **Implication for the brief/PRD:** what changes (or doesn't) because of this.

- ```

- Don't include raw quotes here — link to `.wize/knowledge/research/{slug}.md`.

- ### 5. Flag the gaps

- End the synthesis with a "Still open" section. List questions we couldn't answer + the cheapest next step (one interview / one analytics query / one expert call). Route back to Wizer with the gap.

- ## Output template (synthesis)

- ```markdown

- ---

- status: ready-for-prd

- owner: Pepper Potts

- created: YYYY-MM-DD

- ---

+ ### 1. Read the question

- # Research — {{project_name}}

+ Extract the user's research topic or question. If the user provided a long message, identify the core research intent.

- ## Q1. {{question}}

- **Finding:** …

- **Confidence:** medium

- **Why:** … [1]

- **Implication:** Brief constraint #3 holds; PRD scope can drop feature X.

+ ### 2. Classify

- ## Q2. …

+ Match keywords against the classification guide above. Score each category by keyword hits. Pick the highest-scoring category.

- ## Still open

- - Q5: pricing elasticity in segment S — needs 1 hour with sales lead.

+ ### 3. Delegate or ask

- ## Sources

- [1] {{title}} — {{url}} (accessed YYYY-MM-DD)

- [2] interview-acme-cto-2026-05-30 — see knowledge/research/acme-cto.md

- ```

+ - **Clear match (one category dominates):** Delegate to the corresponding variant. Load that skill and hand off the user's question intact — do not rephrase or summarize.

+ - **Ambiguous (multiple categories tie or no clear signal):** Ask one clarifying question:

- ## When to use which framework (heuristic)

+ > Isso é pesquisa de mercado, domínio ou técnica?

- - **JTBD** when you're about to write personas; JTBD is sharper than archetypes.

- - **Porter** when picking a market to enter or a defensible moat.

- - **PESTLE** for projects with regulatory or geopolitical exposure.

- - **5 Whys + analytics funnel** when a metric is bad and the cause isn't obvious.

+ Wait for the user's answer, then delegate to the chosen variant.

- ## Anti-patterns Pepper rejects

+ ## Anti-patterns

- - Synthesis without citations.

- - "We surveyed users" with no n, no method, no script.

- - Single-source claims with high confidence ("everyone is using X" because of one blog post).

- - Using brand-name competitor research as user research — they're different.

- - Continuing to research when the decision is already made. Stop and ship.

+ - Executing research directly instead of delegating.

+ - Asking more than one clarifying question.

+ - Guessing when the signal is ambiguous — ask, don't assume.

## Hand-off

- > Research is in `.wize/planning/research.md`. Q3 still open — sales lead has a slot tomorrow. Hill can start the PRD on what's confirmed.

+ > Pesquisa classificada como {{category}}. Delegando para {{variant}}.