How long it takes a company to write down what it stands behind

That second part is what matters most. The risky facts were the ones nobody was worried about.

Why we ran this test

Reynforce asks a company to keep the facts it stands behind in one list. We call that list a register. Each fact has an owner, which is the person who answers for it, and the date it was last checked. When an AI agent speaks for the company, it reads from this list, so it does not have to make things up. All of that only works if a company can write the list in an afternoon. If it took months, nobody would do it. So we wanted to measure it before asking anyone to buy it.

How it was done

• What was asked for — “Thirty true statements about a real product, with an owner and a date against each. One afternoon.”
• When it was run — “Run: September 2026.”
• The rule about looking things up — “Don't look anything up until you've tried.”
• The rule about owners — “Name a real owner or leave it blank.”
• The rule about the clock — “Time yourself. Note when you start and when you stop.”

Every figure on this page is copied word for word from the written plan for the test or from the notes made during it. Our build makes the page from those two documents. If a number here does not appear in one of them, our own build fails.

We have not published the list the company wrote. It describes a real company's product and prices. Publishing those to make a point about risk would be the very mistake this page is about.

What came back

• What came back — “26 rows, 60–90 minutes, an owner on every row, no dates guessed.”
• Who owns the rows — “Seven named roles own the 26 rows, one of them seven rows alone, all on 90-day intervals.”
• Whether the intervals were thought about — “The question of how fast each line actually goes out of date wasn't really asked.”
• Where the dates came from — “Rows sourced from the same document share one date”
• What the evidence pointed at — “Five rows cited the GTM messaging as their evidence”
• What a second reader found — “A second reading found four in ten minutes”

The dates came from documents, and nobody checked each fact. Facts taken from the same contract or sales pack all got that document's date. That date says when the document was last changed. It does not say when anyone last checked the fact was still true. Using the document date makes a list quick to draft. It also leaves the exact gap a register is meant to close.

Some rows held two or more facts at once. One row had four conditions in it. Another held two numbers that go out of date at different speeds. An agent can be half right about a row like that. So when Reynforce suggests a fact for the list, it has to split a row like that into separate facts first.

The four risky facts nobody flagged

We asked the person who wrote the list whether any row would worry them if a buyer pushed on it. They said none would. A second reader, ten minutes later, found four:

• A certificate that belonged to a supplier. It was written as if it covered the company itself. The company had no certificates of its own on the list, and that is the first thing a buyer's purchasing team asks for.
• Two promises with no room for exceptions. One said "all" standards were met. The other said data was "never" lost. A promise like that is broken the first time an exception happens, and one always does in the end.
• A plan for a future feature with no year on it. It was confirmed in the last month of the quarter it named. If the feature slipped, the fact would be wrong within days.

Finding these needed no special tool. It needed someone other than the writer to read the list. A register with an owner and a date on every fact gives a company that second reading every day, where this test gave it once.

What it does not show

• The company — “The product was a tech-enabled services company (outsourced SDR)”
• The person filling it in — “The person filling it in knew every answer.”
• The measurement column — “how_hard was left blank and reported afterwards.”
• What nobody was asked — “None of them was asked.”

There is a second question nobody has answered yet. Writing the list took an afternoon. Whether the owners keep checking their facts on the schedule they chose is a separate question. That is what decides whether the list stays true over time. None of the owners has been asked. We say so here so you don't have to ask.

What we changed because of it

Helping a company write good facts mattered more than making the list easier to fill in. Reynforce now does four things when it suggests a fact. It splits a row that holds more than one fact. It flags a promise with no room for exceptions. It flags a plan for the future with no date. It tells a document's date apart from the date someone checked the fact. All four came from this test.

What the next test will do

We will run it with a software company. The first test was with a services company. The person will score how hard each row was while they fill it in. Last time they had to remember it afterwards. And we will ask one owner directly whether they would check their facts again on the schedule they chose.