A research report earns its value when a reader can ask “What supports this?” and trace the answer to original material. Interviews, surveys, web pages and industry summaries do not produce insight simply by sharing a folder. This guide uses a fictional community makerspace reviewing tool reservations to cover source synthesis, desk research, report delivery and Skill selection. It contains no real participant quotations, market estimates or completed research findings.
Define the decision the report supports
The question is whether to trial reminders and a waitlist next quarter, not to prove that users need a new feature. Specify participants, current workflow, time range, constraints, audience and non-goals. Opening a new location or selling user information is outside scope. A new idea in an interview does not automatically expand the project. Start with a research brief, evidence standard and missing-input list before recommendations.
Break the decision into observable questions: where people abandon reservations, whether reminders address forgetting, whether a waitlist could reduce unused slots and what staff work it introduces. Decide whether the gap is research design, additional external evidence, source cleaning or drafting. A writing tool cannot resolve an undefined sample by producing fluent conclusions.
Assign evidence identifiers and metadata: type, origin, collection or publication date, applicable period, permitted use, completeness and verification state. Keep interview T-01, survey S-01, web source W-01 and operational record B-01 distinct. Preserve raw copies and version transcripts, coding and drafts. Sanitize internal material and inspect external models, synchronization, logs and sharing destinations separately.
Stage one: turn interviews and feedback into traceable evidence
Check completeness and review transcription of dates, negation and specialist terms. Separate direct quotes, researcher observations, participant explanations and analyst inferences. A cleaned paraphrase should not acquire quotation marks as if verbatim. Preserve source identifiers and locations for representative quotes. The report need not publish full transcripts, but important themes must remain traceable to context.
Create a coding guide from the research questions, defining inclusion, exclusion, boundary examples and counterexamples. A passage may have several themes; one person repeating a point is not three users. State whether counts refer to people, events or fragments. Silence does not establish absence of a need. Have analysts review an overlapping pilot batch, record disagreements and adjudication, then expand coding.
Treat interviews, support tickets, surveys and observations as different evidence types. Waiting might mean page response, a physical queue or uncertainty about rules, not necessarily performance. A support agent's paraphrase and direct observation have different roles. Record that distinction and limitations on each finding card. Quantitative and qualitative sources can complement one another, but a small interview set does not estimate population incidence.
A finding card contains the observed pattern, supporting records, counterexamples, possible explanations, relevant context and follow-up questions. “Improve user experience” is not a finding, and an unexplained priority rank is not a method. A minority view may expose an important boundary rather than noise to remove. Review fidelity first and actions second.
Worked example: reservation abandonment has competing explanations
T-03 describes missing a slot, T-06 describes uncertainty about cancellation and B-02 shows unused equipment periods. These are illustrative identifiers, not actual quotes. Do not combine them into “everyone abandons because reminders are missing.” Build separate forgetting, unclear cancellation and changing-plans hypotheses, then identify evidence needed for each.
If another interview says reminders are excessive, keep it as counterevidence. Compare optional reminders, clearer cancellation instructions and a bounded waitlist trial rather than searching only for favorable quotations. If operational records lack cancellation reasons, report that attribution gap. Do not infer users' psychology from unused slots. Follow-up questions or a controlled trial may be next steps, not automatic messages to all users.
A bounded request is: “Use only approved-notes and codebook-v1. Separate quotes, observations and inferences. Give each theme evidence identifiers, counterexamples and unanswered questions. Do not combine sources into person counts or fill unknowns. Do not invent quotes, contact participants, edit reservations or upload unauthorized records. Return ten coding suggestions and sample finding cards before the rest.”
Stage two: trace desk research to original sources
Use an issue tree to find public rules, comparable organizations' workflows, product documentation and applicable statistics. Record queries, URLs, publication and access dates, geography, population and definitions. Five articles repeating one announcement are not independent corroboration. Source authority depends on the claim: an organization's page establishes how it describes a feature, not necessarily that the feature works.
Keep an atomic evidence ledger with one checkable claim per row, a direct source, location, interpretation, limits and verification action. Preserve units, denominators, periods and currencies. Separate plans, forecasts and historical observations. If the original report is unavailable, mark secondary evidence rather than claiming a full-text check. Use paid or restricted material only through existing authorization, not bypasses.
For conflicts compare definition, date, scope, sample and calculation. Registered members differ from monthly active users; reservations differ from actual collections. Preserve both claims and the difference. Document a reconciliation when justified, otherwise retain the conflict and name the next evidence task. Newer numbers or a larger publisher are not automatically correct for the question.
Industry work can examine value chains, business models, competition, drivers and risks, but must define its market boundary. If using TAM, SAM or SOM, state eligible customers, constraints, price basis and assumptions. A broad sector total is not achievable revenue for this project. Policy or contract meanings need appropriate professional verification; fluent reporting does not make an output legal or investment advice. Keep unavailable fields unresolved rather than manufacturing a complete market table.
Stage three: make the move from findings to recommendations inspectable
Organize the report around question and method, sources and sample, findings, explanations and alternatives, option comparison, recommendations, limitations and validation. Link every conclusion to finding and evidence identifiers. Give recommendations their own value, cost, dependencies, adverse possibilities and test. Label facts, interpretations, forecasts and decisions separately so perceived usefulness cannot become a guaranteed utilization gain.
Compare feasible reminder and waitlist options, including retaining the current workflow or gathering more evidence first. Define conditions for revisiting a choice, such as extra complaints, staff burden beyond an agreed limit or insufficient trial uptake. Explain confidence through coverage, source consistency and method limits instead of inventing a precise probability. Qualitative findings can motivate hypotheses without producing generalizable causal estimates.
MADR is an open-source software-architecture decision-record format available under MIT or CC0. Its approach to context, alternatives, outcome and consequences can inform a separate decision appendix. It is not an interview-analysis method or certification. Reuse the simple versioned record approach without installing a service, while organizing the actual research report according to its own question.
When reviewed material is already in an authorized Notion workspace, Notion Research & Documentation fits the drafting checkpoint for synthesizing read pages with facts, caveats, dates and source links. It needs working tools, readable sources and a writable destination, followed by inspection of the resulting page. Without tools, the deliverable is a draft. It does not conduct new interviews, independently verify external evidence or establish sample representativeness, and is not an automatic industry-research engine.
Freeze stages and accept the final artifact
Version the brief, manifest, coding ledger, findings and report separately, recording reasons and impacts of changes. When a source updates, review affected claims instead of regenerating the entire report and losing references. Ask a colleague to choose important conclusions, follow identifiers to originals and explain the limits. A broken chain returns to verification; more footnotes do not fix it.
Check decisive numbers against sources and definitions, and quotes against their records. Separate not found from not applicable. Inspect title, tables, links and appendices in the delivered file, ensuring exports retain counterevidence and notes and omit sensitive fields. Saving, public sharing and notifying others are separate actions. Producing a readable draft does not mean anyone received it.
Discover and refresh matching Skills
Search SkillPI by subject, material type, method, network need and output, distinguishing synthesis, desk research, evidence organization and writing. Check fixed source, license, dependencies, credentials and data flow before comparing candidates on one small fixture. Evaluate theme fidelity, conflict retention and source locations rather than download counts or blanket security-review claims.
For discovery assistance, verify the official SkillPI Skill Finder source and host fit, confirm installation and discovery, then search for the target capability. Discovery is not authorization to install every result or proof of execution quality. No matching verified source records for the nine source-article names were established here, so installation links are not guessed. The linked verified note-synthesis capability appears only at its appropriate step. As the catalog grows, rerun the fixture and record versions and reasons for added or replaced associations.
FAQ and final checklist
Can interviews alone support a report? They can support qualitative findings with sample context and limitations, not population claims. Which conflicting source should win? Align definition, date and provenance first; preserve unresolved conflicts and follow-up work. Can missing figures be filled from general knowledge? Not silently: a separately authorized estimate needs its method and assumptions, and unknown is an acceptable value.
Must every quote be public? No: retain controlled identifiers and necessary representative excerpts, checking permissions and identifiability before publication. When is desk research needed? When external support is missing, gather evidence before polishing prose. How should efficiency be measured? Compare rework, source-tracing errors and total delivery time rather than report length. Deliver the brief, source ledger, coding, findings, options, report, decision appendix and pending list so the decision has both support and boundaries.