How you know it
The fact column is not homogeneous. Two sentences can both be facts and be worth completely different amounts, and the difference is not visible in the sentence. It is in how the fact reached you.
Tag every entry in your fact column with one of four sources.
Observed. You saw it, were present, or read the primary document yourself. The date on the invoice, because you opened the invoice.
Reported. A named person told you, on a date. This is not weaker than observation in any general sense, and in most professional work it is the bulk of what you have. It is different, because it carries a second person's split inside it, and their inferences arrive in your fact column wearing your fact column's clothes.
Documented. It is written down in a system of record. Worth distinguishing from reported, because documents persist and can be re-read, and worth distinguishing from observed, because a document records what someone entered, not what happened.
Assumed. You have been treating it as true and cannot say how you came to. This category is the one that matters. Almost every situation contains at least one, it is almost never the one you expect, and it is usually load-bearing.
Take each line of your fact column and ask: if someone asked me to prove this in ten minutes, what would I produce? Where the honest answer is "I would explain why it must be true", you have found an assumption. Move it.
The reason to tag rather than rank is that the tag tells you what to do when the fact turns out to be wrong. A wrong observation means you misread something. A wrong report means a person's account did not match, which is information about the situation and sometimes about the person. A wrong document means the system of record is unreliable, which is a much larger problem than the one you were working on. A wrong assumption means nobody ever checked, and the useful question is why the process allowed it.
Reported facts deserve a second field: what the person was in a position to know. Someone reporting what they did is in a strong position. Someone reporting what a third party decided is passing on a report, and it has now been through two people's inference. In a chain of three, the original event is usually recoverable and the reason for it is usually not.
This is not a counsel of suspicion. It is the difference between "the client said they are unhappy with the timeline" and "our account manager said the client seems unhappy with the timeline", which are different facts with different next actions, and which get written down identically by almost everybody.
A recurring failure in organisations that keep good records: the record is treated as the event. A ticket marked resolved at 14:05 establishes that someone clicked resolve at 14:05. Whether the customer's problem ended is a separate fact that the system may never have held. When a document and a report conflict, the useful move is not to pick the more official one. It is to ask what each was created for, because a field that nobody's work depends on is filled in accordingly.
There is a specific professional habit worth building here, and it costs almost nothing. When you write anything down for later use, put the source in the same line. Not in a separate provenance column that you will not maintain. In the sentence: partner confirmed by phone, 4 July, spoke to the finance officer. The cost is four seconds at the moment of writing. The cost of reconstructing it in November is an hour, and the reconstruction will be wrong in the specific way that suits whatever you now believe.
Your notes say the regulator has confirmed the extension. Applying the source tags, what should you check first?
Which of these belongs in the "assumed" category?
Notes are kept with your account, alongside your progress and your gate claims. The lesson itself is readable without one.
