Your Notes Never Tell You When They Stopped Being True

The answer you found was true in April

Somebody who has been on the account for six weeks joins a customer call and repeats what the notes say, that the customer is blocked on single sign on and will not expand until it ships. The customer says they worked around that in May and have been waiting on export permissions ever since. The call recovers. Nobody was careless, and the note was a fair account of a real conversation.

It simply stopped describing the situation at some point after it was written, and nothing about it changed when that happened. No flag went up, no strikethrough, no move to an archive. It sat in the same place with the same date at the top, and it came back in the search results looking exactly as current as it did the week it was made.

Every company has a drawer full of these. They are not gaps, and they are not wrong in the way anyone means when they call a record wrong. They are correct reports about a world that has moved on, and there is no mark on them that says so.

A note is written about a moment and read as a fact about the company

Almost everything a team writes down is phrased in the present tense, because the person writing is describing what is in front of them. The customer wants monthly billing. Pricing is waiting on the legal review. Priya owns the migration. All accurate as typed, and all of it a report from one particular afternoon.

Read eight months later, those same sentences no longer read as reports. They read as facts about how the company works, which is a much stronger claim than anyone was making at the time. The date at the top helps less than it looks like it should, because a date tells you when somebody wrote a sentence and says nothing about how long the sentence was good for. Those are two different pieces of information and only one of them ever gets recorded.

That is the whole mechanism. A record carries its moment of creation and never carries its expected shelf life, so the reader has to supply the second one from context they usually do not have.

Reasons keep. States do not. One sentence usually holds both.

Take a line that could appear in any operator's notes. We are staying on annual billing because their finance team cannot process monthly invoices from a vendor under a certain size.

Two claims, one sentence, wildly different lifespans. The reason is a constraint inside somebody else's finance function. It might hold for years, it explains a whole family of decisions, and knowing it saves the next person from re-deriving it from scratch. The state, that we are currently on annual billing, can flip next week with one approval from either side.

Notes do not separate them. The two arrive fused, get filed as a single fact, and then age at the speed of the faster half. So one of two things happens. Either the sentence gets binned as out of date and the durable reasoning goes out with it, which is how a team ends up relearning the same constraint every eighteen months. Or the expired half gets carried forward on the credibility of the half that is still true, which is worse, because the part that no longer holds is now travelling with a reason attached and sounds better researched than whatever anybody says to contradict it.

If you only change one habit after reading this, separate the two when you write. State what is currently the case, then state why, and let them be two sentences that can die on different days.

The correction almost always exists somewhere in the company

This is rarely a hole in what the organisation knows. Somewhere there is a newer record, every bit as accurate as the old one: a line in a different set of meeting notes, a message in a thread that four people were on, a sentence somebody said on a call last month and everybody present absorbed without comment.

The company holds both facts. What it does not hold is any connection between them. Records are written forward. A new one does not reach back and annotate the thing it supersedes, because the person writing it is not thinking about an eight month old note and in most cases has no idea the note exists. Nobody is being lazy. There is no moment in normal work at which anyone is handed the job of retiring an old record, and no notification that would tell them there is one to retire.

So the two sit in the same searchable pile, each faithful to its own moment, with nothing at all to indicate which one is the survivor. Whichever surfaces first becomes what the company believes.

This is not the same problem as two records disagreeing

Worth separating carefully, because the fixes are different. One familiar failure is a single decision that spawns a chain of copies, notes to a ticket to a line in an update to a row in a plan to a slide, with each copy written for a new reader and drifting a little further from what was said. There the records are all describing the same moment and some of them are less faithful than others. There is a right answer and the work is finding the version closest to the room.

Here every record is faithful. Neither is a bad copy of the other, and no amount of tracing lineage helps, because they are not versions of one thing. They are honest descriptions of two different moments in the correct order, and the only thing that went wrong is which one you happened to open.

It is also not the case of something that quietly stopped and left an absence behind. Nothing here is missing. It is all written down, searchable, dated and readable. The axis that failed is time.

Search ranks on wording, and the dead record is usually better written

When a subject is new to a company, somebody explains it. The background, the constraint, what was considered, what was settled. It gets a proper few paragraphs, because whoever reads it next is going to need them.

By the time the position actually changes, everyone involved already carries the context, so the change gets recorded in about nine words. Moving them to monthly, finance signed off. Perfectly clear to the people in that conversation and nearly invisible to anything that searches.

Now put those two artefacts in front of a search. The thorough document matches on more terms, reads as authoritative, has structure and headings, and sits in a place that looks like where knowledge goes. The one line that retired it is buried in a thread and matches almost nothing. They are not competing on equal terms, and the old one wins on both wording and completeness. The better a record is written, the longer it keeps being believed after it stops being true.

Text keeps the confidence it was written with

A record holds its original certainty exactly and forever. A firm statement written in March reads in November with precisely the weight it had in March, and a careful hedge stays as careful, which means the hedge cannot warn you about anything except what the writer already doubted. Nothing in the medium fades.

Numbers are the sharpest version. A figure reads as though it was measured, so it does not sound like a claim from a date, it sounds like an amount. Repeat a number out of an old note in a room where nobody knows its origin and it becomes the working number, and the next person to use it is quoting you rather than the note. Two hops later the source is unrecoverable and the number is in a deck.

This is also why the flinch never arrives. A wrong record sounds like a wrong record. An old one sounds like a competent colleague telling you something.

The people who find it are the people least able to date it

Anyone who was in the room does not search. They already know the current state and they know the old note is behind without ever opening it, so their knowledge is never applied to the record that needs it.

The people who search are the ones without the context. Someone who joined in July. Someone from another function picking up an account for a quarter. Someone covering a week of leave. A new hire reading back through the history to get oriented, which is the exact behaviour every company asks for. They receive the strongest available record and have no way to weigh it, because weighing it requires knowing what happened after it, which is the thing they came here to learn.

The result is that a record is trusted most precisely where it is least likely to be checked. That is not a training problem. Nobody can be taught to doubt a document they have no basis for doubting.

A gap produces a question. An old record produces an answer.

When you find nothing at all, you go and ask somebody. The cost is a few minutes and a small amount of friction, and the outcome is usually right, because you end up talking to a person who knows. A gap announces itself and it is recoverable.

An out of date record ends the search instead. You found something. It reads well, it has a date, it came out of your own company's notes, so you act on it. And then it travels, into a slide, into an email to a customer, into a plan for the quarter, into an answer you give a colleague who now believes it too and will repeat it without knowing where it came from.

By the time the fact gets contradicted it has been restated in four places with your name behind it, which makes the correction socially expensive as well as slow. The original note cost nothing to write and nothing to store. What it cost was one person's afternoon eight months later, plus whatever was decided in between on the strength of it.

Why a review process does not hold

The standard answer is a review cycle. Assign owners, put a last reviewed stamp at the top, do a quarterly pass. It fails for three reasons that have nothing to do with anyone's diligence.

First, a review date records that somebody looked. It does not record that anything is current. The look is shallow by construction, because the reviewer is checking a document against their own memory rather than against the world, and if they had current knowledge of every claim in it they would not have needed to read it.

Second, review load scales with the number of records. A team that captures more of its work gets more to audit, so the practice buckles at exactly the size where the problem starts to bite, and the first thing dropped is the long tail where most of the decay lives.

Third, the records that go stale worst are not the ones anybody thinks to review. Reviews happen to handbooks, policies and onboarding docs. They do not happen to the notes from a customer call in March, and the notes from the customer call in March are what the account team actually reads.

What tells a live record from a retired one

The signal that separates them sits outside the record, in what was said afterwards.

If the note you are reading is the last thing anybody in the company said about that customer's billing setup, it is probably still the state of play, however old it is. If there have been four conversations about it since and your note is the oldest of the five, it is almost certainly history. Same document, same date, same wording, and the correct response is opposite in the two cases. Everything that determines which case you are in is in the surrounding record and none of it is in the document.

That is worth noticing because it is computable rather than editorial. It does not ask anyone to maintain, revise or delete anything, and it does not need a human to have realised that the old note went out of date. It needs the record to be organised by subject as well as by document, so that finding one mention of a thing shows you where that mention sits in the run of mentions instead of handing it over as a free standing fact.

A mention that arrives with its position in a sequence gets read for what it is, which is testimony from a date. A mention that arrives alone gets read as the truth. The reader's judgement is fine in both cases. Only one of them gives them something to judge with.

Where Driffle fits

Two things have to hold, and neither of them is a discipline anyone has to sustain.

Capture cannot depend on somebody deciding to file an update, because the conversation in which a fact quietly changes is never the conversation anyone writes up. It is a passing line on a call about something else. Driffle keeps meeting notes without putting a bot in the meeting and takes context from what is on screen, so a change mentioned in passing lands in the record at the same quality as a change announced in a planning session. If the nine word update is captured at all, it can compete with the polished document that it replaced.

Then retrieval has to answer with a sequence rather than a single best match. Asking what the current state of an account's billing is should return the most recent thing anybody said about it, with the older and fuller account underneath as history rather than beside it as a rival. The old note keeps its value, since the reasoning in it may well still hold, and it stops being able to pose as the present.

Neither of those requires the team to notice anything. That matters, because noticing is the step that never happens. The only person who could flag that a record has expired is somebody who knows both the record and the newer fact, and if that person existed and was reading, the problem would have resolved itself already.

A memory of the work, not a list of who repeated an old number

Being out of date is not a mistake, and none of this is worth doing if what it produces is a way to check who quoted a stale figure in a meeting. Nobody in the story did anything wrong, including the person who repeated the note on the customer call.

The useful artefact is the sequence of what was said about a subject and when, held so that anyone entitled to see the subject can see its order. Access follows the work rather than the reader's seniority, and the record answers what the current position is and what it used to be, not who was behind.

There is a real deletion question underneath this, and it is not answered by clearing out old records. The state in an old note expires and the reason in it often does not, so a purge throws away the half a team most needs and keeps nothing it was worried about. Retiring a record and ending its ability to look current are separate things, and only the second one is what anybody actually wants.

A test you can run this week

Pick three facts your team states with confidence and repeats often. A number that appears in the deck. A claim about what a specific customer wants. A constraint everyone quietly routes around.

For each one, find the earliest place it was written down. Then find the most recent time anybody said anything at all about that subject. If the second date is months later than the first and nobody can tell you what was said, you are not holding a fact. You are holding a record, and you do not yet know whether it is current.

Then ask the one question that makes the difference visible. What would have to have happened for this to have stopped being true, and would we know if it had?