Two posts from the same founder, eleven days apart. The first said Amazon suppresses listings that run more than seven images. The second said his return rate on a hero SKU went from 11.4% to 7.2% in eight weeks after a scale frame moved to slot two.
The second claim came out of a report he pulled, with a window and a date. The first came from a panel he sat in at a conference in March. Both were written in the same voice, at the same level of certainty, with the same sentence rhythm. A reader had no way to tell which one he knew and which one he had heard.
That is the failure this system exists to fix. Not inaccuracy. Flatness. An archive where every claim sits at the same altitude of confidence, so the reader either discounts all of it or trusts all of it, and both readings are wrong.
Why founders hedge wrong in both directions
There are two ways to get this wrong, and we see them in roughly equal numbers.
The over-hedger qualifies everything. "In my experience," "I think," "probably," "it seems like" on every sentence, including the ones where they pulled the number themselves. The real findings get buried under the same caveats as the guesses. The archive reads as someone who isn't sure of anything, which is not true, and which a buyer notices.
The under-hedger qualifies nothing. Every claim lands as fact. It works for months, right up until an operator who actually knows the mechanic reads the post about the seven-image penalty, checks, and files the founder as someone who repeats things. That reader never comments. They just stop reading, and if they were a prospect, they stop being one.
Both failures have the same root cause. The hedge gets decided at the moment of writing, by mood. Tired and cautious that day, everything gets softened. Confident after a good week, everything gets asserted. The status of the claim itself, what it is actually based on, never enters the decision.
The four tiers
We grade every load-bearing claim into one of four tiers when it enters a bank, before a draft exists. Each tier has a standard hedge and, more importantly, a required attachment. The attachment is what makes the tier honest instead of decorative.
Measured. You pulled the number from a report you can name, over a window you can state, on a date you wrote down. "Return rate 11.4% to 7.2%, FBA returns report, 60 days either side of the change, pulled July 3." No hedge language needed. The attachment is the window and the date, and the date goes in the post.
Observed. A pattern you have seen across a number of accounts, listings, or months, without a controlled comparison. "Across the last dozen brands we onboarded, the ones with a real scale reference in slot two had lower size-related return comments." The standard hedge is one clause: "not a controlled study." The attachment is the count. How many accounts, how many listings, over what period. Readers forgive an observation. They do not forgive an observation dressed as a measurement.
Heard. Something you believe because a credible source said it, and you have not verified it yourself. "Amazon's own help page says the trailing 48 hours of ad data is partial." The attachment is the source, by name, with a date. If you cannot name the source, the claim is not Heard. It is the next tier down, and most founders would rather not admit that.
Believed. An opinion, a mechanism you think is true, or a forecast. "I think the branded-search defense conversation is going to get expensive in Q4." The standard hedge is "I think" once, not on every sentence. The attachment is the condition that would change your mind. A Believed claim without that condition is a mood. With it, it is a position a reader can hold you to, which is exactly what makes it credible.
The tier is assigned at capture, not in the draft
This is the whole mechanism, so it deserves its own section.
If you grade claims while drafting, you will grade them by how you feel about the post. Every founder does. The claim that makes the post work gets rounded up. The claim that complicates it gets softened so the argument holds. Neither move is dishonest, and both are invisible to the person making them.
So the tier lives in the bank entry. The proof bank holds Measured claims by construction, because a proof bank entry without a window and a date is not a proof bank entry. The source ledger holds Heard claims, since that is what an external number is until you re-check it. The story bank is mostly Observed. Believed claims live in the explanation bank and the disagreement file.
When the running order pulls an entry into a post, the tier comes with it. The writer does not decide how confident to sound. They look it up.
The drift this prevents
The most common corruption in a founder archive is not a wrong claim. It is a claim that was Heard in January and reads as Measured by June.
It happens through retelling. The first time, "someone at Prosper said that Rufus reads A+ copy before it reads bullets." The second time, "the data suggests Rufus prioritises A+." The third time, "Rufus reads your A+ first, so put the argument there." Nobody lied. The claim was upgraded one notch per telling, by a writer who had stopped remembering where it came from, and the archive now contains a platform mechanic asserted as fact that nobody outside Amazon has ever measured.
Ecommerce founders are more exposed to this than most, because platform mechanics are almost always Heard. Nobody has measured the algorithm. The most repeated wrong claims in this industry, the AI-content penalty, the seven-image ceiling, the specific percentage that first-degree connections get on LinkedIn, are all Heard claims that got upgraded through repetition until they sounded like data. A founder with a tier column does not do this, because the entry says Heard, and Heard has a source attached, and the source does not say what the third retelling said.
What the reader gets
A diligence reader, the acquirer's analyst, the retail buyer, the senior operator choosing between two offers, is not assessing whether you are right. They are assessing whether you know the difference between what you know and what you believe. That is the actual competence being evaluated on a Sunday-night read-through, and a flat archive cannot demonstrate it no matter how accurate it is.
An archive with tiers demonstrates it in every post. The Measured claims carry dates. The Observed claims carry counts. The Heard claims name their sources. The Believed claims say what would change the writer's mind. A reader who has seen five of these posts has learned that the hedges mean something, which means the assertions mean something too. "I don't have a controlled study and I'd be inventing it" stops reading as weakness and starts reading as a person who would tell you the truth when it was inconvenient.
There is a cheaper benefit as well. Corrections get priced in advance. A Believed claim that turns out wrong costs nothing, because it was published as a position. A Measured claim that turns out wrong costs a correction post. The founder knows, before publishing, which posts carry that risk, which is a much better place to find out than the comment section.
The upgrade path is the content
The point of the tiers is not permanent hedging. It is the opposite. A Believed claim is a hypothesis, and the tier tells you which ones to go measure.
The strongest content pattern available to a founder runs straight through the tiers. Publish the Believed version in March: "I think the app-versus-browser split is hiding a creative problem on most catalogues." Go look. Publish the Observed version in May: "we pulled it on nine accounts, seven had a gap over two points." Pull the report on your own catalogue. Publish the Measured version in August, with the date.
The archive now shows a person who had an idea, tested it, and reported the result either way. Nobody can restate that sequence, because the sequence is the access. And the reader who arrives at the August post and reads backward finds three dated entries instead of one confident assertion, which is the difference between a record and a repetition.
The 20-minute audit
Take the last twenty posts. Mark every load-bearing claim M, O, H, or B by what it was actually based on, not by how it reads. Then mark how it reads.
The founder we described at the top ran this on 22 claims. Seventeen were written at Measured. Six actually were. Nine were Heard, written as fact, and four of those he could no longer source. That last number is the one that changes behaviour, because a claim you cannot source is a claim you cannot defend in a DM, and a founder who has been asked to defend one in a DM does not need the audit explained twice.
FAQ
Doesn't hedging cost reach? Hedging by mood does. A post where every sentence is softened performs badly because it says nothing. Hedging by tier does the opposite: the Measured claims land harder because they are not surrounded by the same caveats as the guesses. Precision in both directions reads as confidence.
How is this different from the source ledger or the disclosure ladder? Three different columns. The source ledger records where an external number came from and when to re-check it. The disclosure ladder decides how much of a number you reveal. The confidence tier records how sure you are and why. A claim can be fully sourced, fully disclosed, and still Believed.
Is opinion allowed? Believed is a legitimate tier and frequently produces the best content on the profile. The failure is not having opinions. It is dressing them as measurements because measurements sound more authoritative.
Do I run this on the archive? Once, as the audit above, to find the Heard claims that have drifted upward. After that, only on new bank entries. The decay pass finds what is no longer true. This finds what was never known, and those are different problems.
If your archive reads at one flat level of certainty and you would like it to read like a person who keeps records, that is the kind of system we build before we write a word.