skills/engagement-setup/references/discovery-methods.md

# Discovery methods

## Work backward from a decision

For each material question, record its decision implication, available evidence, gap, method, and intended output. Choose a proportionate sample and scope. Interviews can explain mechanisms and perspectives; transactional data can measure patterns; observation can reveal workarounds; surveys need a defined population and response-coverage check.

Request data with the exact fields, grain, period, units, definitions, format, and purpose needed. Identify the likely owner and due date as proposals if unconfirmed. Start with existing extracts and documents before asking the client to build new reports. Record missing and unusable data without assuming bad intent.

## Interview guide

Open with purpose, time, and the agreed treatment of attribution or confidentiality. Do not promise confidentiality beyond what the engagement permits. Ask about outcomes, obstacles, decisions, prior attempts, what should be preserved, and examples that test the current hypothesis. Follow useful threads rather than reading every question.

For process work, ask for a recent case from start to finish, exceptions, queues, handoffs, errors, and what the systems record. For leaders, ask about priorities, constraints, trade-offs, success, and decisions they can make. Seek counterexamples and ask which evidence would change their view.

Capture quotations accurately and distinguish them from paraphrases or interpretation. Do not infer that a small set of interviews represents an entire population. Check whether multiple accounts are independent or repeat the same source.

## Synthesize

For each material theme, state the claim, supporting and contrary evidence, limitation, and implication. Separate observed facts from candidate causes and questions for the next phase. Quantify only where the data supports it; do not convert an interview frequency into a prevalence estimate without a sampling basis.

Produce a concise current-state assessment, hypotheses to test, and prioritized next analyses. Preserve what works as well as what fails. The discovery output should reduce uncertainty and guide effort, not merely document that meetings happened.