What an evidence-backed market research workflow looks like
An evidence-backed workflow lets you start at a recommendation and walk back to the pages that justify it. The six steps below are the minimum that makes that possible, and they suit anyone who researches the same market more than once.
Write the goal as a question with a decision attached
“Research Meridian” has no end. “How does Meridian position onboarding for agencies, and does the claim hold up?” does. A usable goal names the decision it feeds and can finish in one of two ways: answered, or declared unanswerable with what is missing.
Record each source with its type and retrieval date
Keep the URL, the kind of source (vendor page, customer review, forum thread, analyst note) and the date you read it. The date matters more than people expect. A pricing page read in March says little about October.
Split pages into claims
A claim is one proposition that can be true or false: “Meridian advertises onboarding in a day.” Claims are easier to check, compare and reuse than paragraphs of summary.
Link each claim to its evidence
For every claim, note which sources support it, which contradict it and which only add context. Most workflows skip this step. It is where a disagreement becomes visible before it is averaged into a confident sentence.
Write insights only from stored claims
An insight is an interpretation, so it needs its inputs listed. Record how fresh the underlying evidence is, what remains unknown, and who approved it.
Attach the trail to the action
A battlecard line, a positioning option or a campaign brief should point to the insight and claims behind it. When someone challenges it in a client meeting, you open the sources instead of searching your memory.
| Step | What Fieldnote's team has |
|---|---|
| Goal | Does Meridian's “live in a day” onboarding claim hold up for agencies with 20 or more clients? |
| Sources | Meridian onboarding page (read 2026-10-02), 6 customer reviews, 1 changelog entry |
| Claims | “Meridian advertises onboarding in a day” (vendor page); “setup took three weeks” (review) |
| Evidence | The review contradicts the vendor claim; the changelog adds context about a new import tool |
| Insight | Meridian's speed claim is contested for larger agencies. Freshness: recent |
| Action | Battlecard line: “Ask how many client accounts need importing.” Linked to the insight |
Where this breaks down
- The goal stays vague, so every source seems relevant.
- Claims are written as opinions (“Meridian is slow”) instead of checkable statements.
- Evidence is linked after the conclusion is written, which turns it into decoration.
- Nobody records what was searched for and not found.
How Sigblaze follows these steps
Sigblaze stores the goal, sources, claims and evidence relationships as separate records, and insights and actions keep a link back to the claims that produced them. In the private alpha a person approves an insight before it is reused or exported.
Questions
How many sources does a claim need?
There is no fixed number. One vendor page can support a claim about what the vendor says. A claim about how customers experience the product needs sources from customers.
This piece contains no external statistics. Fieldnote and Meridian are fictional companies, and the figures in the examples are synthetic.