Marketing & GrowthFree
Category Narrative Brief With Sourced Claims
Every claim in the narrative split out, graded by who said it and whether they gain from it being true, then sourced or cut.
Halewell came out of a two-day offsite with nine pages in the five-act shape every narrative template produces. Split into claims that can each be true or false on their own, that is 34 claims. Eleven carry a citation, which is what made the draft feel researched. Grade the eleven by who wrote the thing being cited and three survive: six point at vendor-commissioned surveys, two point at Halewell's own earlier blog posts. So 3 of 34 claims, 8.8%, rest on a source with nothing to gain.
The statistic the shift rests on appears six times under four different-looking sources. All four trace back to one commissioned survey of 214 people, 131 of whom, 61.2%, were the sponsoring vendor's own customers, and the sponsor is the competitor Halewell exists to displace. The shift itself had no series behind it at all. Four candidates were tested and one cleared: two readings, 31 months apart, same definition, moving 18.6 points against a 2.0 point margin. Act two got rewritten.
Then the interview corpus. The category term appears in 19 of 22 transcripts, 86.4%, which the team read as validation. The interviewer said it first in 17 of those 19, so two buyers raised it unprompted, 9.1%. A phrase the draft never contained was volunteered by 14. Five claims cut, four narrowed, three labelled as Halewell's own bet, leaving 29 of which 26 carry a dated source. Feeds the page copy and the positioning canvas.
Why a narrative assembled internally cannot be checked
Agreement is not the same evidence as a volunteered answer, and the size of the gap is measured. Pew Research asked the same question two ways after the 2008 election. Offered the economy as one of five options, 58% picked it. Asked openly, only 35% volunteered it. One question produced a number 23 points higher than the other about the same electorate. A discussion guide that names your category in question two is the closed version of the question.
Four citations can be one study. The people whose job is synthesising evidence treat that as a known hazard, because duplicate publication can introduce substantial biases if studies are inadvertently included more than once. A single study is often reported across several articles and abstracts. Nobody made the survey's sponsor announce itself either. An equity research report has to disclose whether the analyst received any compensation from the subject company in the previous year. A vendor's market survey carries no equivalent rule.
A shift is a claim about direction, so it needs two readings of one series taken the same way both times. The Census Bureau is blunt about this for its own data: only compare 1-year estimates with 1-year estimates, and because variables and geographies change over time, comparisons should be made with caution or not made at all. That single rule killed three of Halewell's four candidate series, two on changed wording or a missing second reading, one on a 0.4 point move inside a 1.6 point margin.
How it works
Split the draft
Nine pages become one row per claim, each phrased so it can be true or false alone.
Grade the sources
Every citation followed back to its origin, then marked disinterested, interested, or cited to yourself.
Test the shift
Candidate series screened for two comparable readings and a movement bigger than the published margin.
Check the words
Transcripts sorted by who used each term first, so agreement and volunteering separate.
What you get
- The draft split into atomic claims, one row each, tagged to the act it sits in.
- Every citation graded on whether whoever published it gains from the claim being true.
- Each repeated statistic followed back until the chain ends, with the sponsor named.
- The shift tested against candidate series for two readings, one definition, real movement.
- Interview mentions split by who said the term first, so prompted agreement is separated out.
- A brief and a deck where each factual claim footnotes its source and that source's date.
Common questions
Is this not just adding footnotes to the narrative we have?
The footnotes change the narrative, which is the point. Halewell's shift had no series behind it, so act two got rewritten around the one claim that could be measured. Its stakes act got rewritten in a phrase the draft never contained. Five claims went in the bin. A sourcing pass that leaves the story where it was has not been run properly.
Our draft already cites eleven sources. Why is that not enough?
Because six of the eleven were surveys paid for by vendors, four of those by a direct competitor, and two more pointed at Halewell's own blog posts, which is a claim cited to an earlier version of itself. Grading by who published the thing rather than counting citations takes eleven down to three, which is 8.8% of the draft.
What counts as a disinterested source?
Someone whose position does not improve if the claim is true, and who put a date on it. A statistical agency, a standards body, a regulator's own text, a trade body publishing against its members' interest, a buyer's own words in a transcript. Not a competitor's commissioned survey, not your own earlier post, not a trade article quoting either of those.
We have 22 interviews saying the category resonates. Does that not settle it?
It settles that buyers will agree when asked. The term appears in 19 transcripts, but the interviewer said it first in 17 of them, and 15 of those 17 then agreed. Two buyers brought it up themselves. Meanwhile 14 volunteered a phrase nobody on your side had used, which is the stronger signal, and it feeds the customer language pack.
What happens to claims we believe but cannot source?
They get cut or labelled, and which one depends on the act. Halewell's five cuts all fell in acts two, three and four, which are claims about the world and need external evidence. The three that survived unsourced sit in act five, on one slide headed as the bet Halewell is making, so a buyer sees where evidence stops and argument starts.
Can a buyer actually open the sources?
That gets checked, because a source a rep cannot hand over is not much use on a call. Of Halewell's 26, 21 are freely readable and five sit behind subscriptions. A free substitute existed for three, leaving two flagged in the brief as needing the analyst licence. So 24 of 26 can go straight to a sceptical buyer.
What do we end up with?
A Sheet with one row per claim carrying act, source, publisher, date, interest grade and verdict. A Doc with the five acts rewritten around what survived. A deck where each factual slide footnotes source and date. Competitor claims go through the teardown and the axes through the positioning map.
Category Narrative Brief With Sourced Claims
Fill in the form and your workspace opens with the work already underway.