feedback-to-findings

Synthesize supplied reviews, tickets, surveys, or interview notes into traceable themes and evidence-backed findings. Use for analysis of existing feedback, not recruiting participants or generating f

SKILL.md

GitHub

Feedback to findings

Explain what the available feedback supports, who it describes, and what remains uncertain. Keep source observations distinguishable from interpretations and proposed responses.

Establish coverage

Identify the question, source set, dates, sampling method if known, and unit of analysis: messages, incidents, respondents, or accounts. Use existing context and ask only for gaps that change the interpretation. If no feedback is available, ask for it rather than inventing themes.

Preserve source identifiers and locations while avoiding unnecessary personal details in the report. Keep original material intact. Note partial exports, missing segments, selection effects, and sources that cover different populations or periods.

Extract and group

Extract concrete observations and requests with enough context to retain their meaning. Use short faithful excerpts or clearly labeled paraphrases. A requested feature is evidence of a request; its underlying need is an interpretation until supported by context.

Group observations by shared problem or situation, allowing a record to support multiple themes. Preserve minority views, positive experiences, exceptions, and counterexamples. Do not merge conflicting experiences into a false consensus. Separate a person's experience from their speculation about everyone else.

Before reporting frequency, distinguish repeated messages from independent people and repeated reports of the same incident. Deduplicate only with a defensible rule; similar wording alone does not establish the same person. If identity is unknown, count records and disclose that unique respondents cannot be established.

Use explicit denominators for rates. Multi-label theme percentages may overlap and need not sum to 100%. Do not infer population prevalence from a convenience sample or treat silence as satisfaction. Keep severity, frequency, and confidence separate: a rare consequential issue can matter more than a common inconvenience.

Develop findings

For important themes, explain the pattern, affected context, supporting sources, contradictory evidence, and practical implication. Keep confidence tied to evidence quality and coverage instead of arbitrary numerical scores.

Separate observed behavior, reported experience, inferred need, and suggested action. Rank findings by the requested decision and supported impact, not solely mention counts. Avoid claiming that a proposed solution has been validated by feedback about a problem.

Deliver

Provide a concise synthesis with traceable examples and material coverage limits. Use a compact evidence table when volume warrants one. Identify specific follow-up questions or checks where they could change a decision. If evidence is thin, report a provisional pattern or no supported pattern rather than manufacturing findings. Contacting participants, creating tickets, or editing source systems requires authorization beyond synthesis.