Marketing Engine · plain-English guide
How to read Researcher
Researcher grounded evidence pipeline. This guide explains what it produces, what each part means, and how it stays honest — so you can act on the output with confidence.
Takes
Topic + angle + research rules
Gives
Dossier + evidence table
The data it uses: The topic, angle, and your research rules (including required proof types).
Watch how it works
How to read the output
Dossier
The organized evidence — what is actually known about the topic, sourced.
Evidence table
Each claim with its source, confidence, and recency. This is what the writer is allowed to use.
Proof mix / gap report
Whether the required proof types (named examples, expert quotes, statistics) were found — and honestly, what is missing.
The terms it uses
- Proof mix
- A required count of named companies, expert quotes, and stats — the proof that earns trust and gets cited.
- Confidence
- How strong a piece of evidence is: primary source and recent beats a secondhand mention.
- Frontier
- Forward-looking best practice or prediction — allowed only when sourced, never invented.
How it stays honest
Every agent separates what it measured from your data from what it inferred. You always know which is which.
Measured
- ✓Sources found + their content
- ✓Proof counts + recency (computed)
Inferred
- ~Synthesis
- ~Which proof matters
Questions people ask
Does it ever invent a source or quote?+
No. Anti-hallucination is the hard rule — every verifiable claim traces to a real source or it is left out, and gaps are reported honestly.
Now read the real thing.
Run Researcher on your own data and compare the output to this guide. If it matches, you can trust it.