Your Team Has Four Records of the Same Decision

Two people, two screens, two different answers

You are on a call with a customer who is unhappy. They believe the migration support was included in what they signed. Your implementation lead believes it was scoped separately and quoted at the time. Both of them are reading off a screen while they say it. Neither of them is guessing, and neither of them is wrong about what their screen says.

The account executive is looking at the email thread from March. The implementation lead is looking at the statement of work. Somewhere there is also a line in the CRM, a slide from the renewal review, and the notes from the call where this was actually agreed, which nobody has opened in six months and which would probably settle it.

The failure people expect from a work record is a gap. You go looking for something and it is not there. That one is at least honest about itself. The failure that costs more is the opposite shape. There are several records. They all exist, they are all findable, they all sound authoritative, and they do not say the same thing.

A decision does not produce a record. It produces a chain of them.

Something gets decided in a forty minute conversation, usually in the last four minutes of it, often after the substantive argument is over and everyone is half packing up. That is the moment. Everything after it is a copy.

Somebody writes it in their notes that afternoon. A day later it becomes a ticket, because work needs somewhere to live. On Friday it appears in the weekly update as one line, phrased for people who were not in the call. The following week it becomes a row in the quarterly plan, because planning pulls from the update rather than from the call. A month later it is a bullet on a slide in a review deck. And at some point, because a customer asked, a version of it goes out in an email.

That is six artefacts from one conversation. Only the first one was made by a person who was in the room. Every other one was made from a previous artefact by somebody solving a different problem, and each of them is a perfectly reasonable thing to have produced. Nobody set out to create six competing accounts of a single decision. Each step was one person doing the obvious next thing.

There is a related problem that gets more attention, which is the labour of running that relay, the human cost of carrying the same fact from the call to the ticket to the update because no tool does it for you. This is about what the relay leaves behind. Not the work of making the copies, but the copies themselves, sitting in six places with nothing on any of them to say which one came first.

Each hop is a rewrite with a new reader in mind

If the copies were mechanical duplicates none of this would matter. They are not. Every step in that chain is a person compressing the previous version for an audience that is different from the last one, and compression is a set of choices about what to keep.

Written for a founder or an exec, a decision gets shorter and more certain, because a summary with four qualifications in it reads as indecision and gets sent back. Written for a customer, a conditional turns into a commitment, because the whole point of the email is reassurance and a hedge does the opposite. Written for the team that was in the call, the reasoning drops out entirely, because everyone already has it and repeating it would be strange.

So the drift has a direction. Each version loses the material that is hardest to carry and keeps the material that travels well, and the same categories fall off at every hop. Numbers get rounded. Names of people get replaced by the name of a team. Dates get softened into months. And the condition attached to the decision, the if that was said out loud in the room, is the single most fragile part of the whole thing.

The condition is the first thing to fall off

Frederic Bartlett ran the experiment that documents this, and he ran it almost a century ago. In the serial reproduction studies in Remembering, published in 1932, one person read a passage and wrote out their version from memory. A second person read that version and wrote theirs. A third read the second, and so on down a chain. Nothing was spoken. Every link was written, which makes it a closer analogue to a company than it might sound.

What came out the other end was heavily abbreviated. Unfamiliar detail vanished. Bartlett recorded that names and numbers rarely survived more than a few reproductions, and that the material got smoothed into a shape an ordinary reader would accept without asking questions. The subjects were generally satisfied with their work, and believed they had passed on everything important while dropping only inessentials.

One of his observations lands on this problem exactly. Describing what happened to qualified statements, he wrote that when a generality is expressed with saving clauses, "the saving clauses tend to disappear even if the generality is retained". A qualified claim goes through a chain of careful people and comes out the far end unqualified, with nobody along the way having decided to remove the qualification.

Run that against a real decision. What was said in the room was that the trial gets extended for this account if legal clears the amended terms, and only through the end of the quarter. Four hops later, the version a customer is holding says the trial was extended. Both sentences are about the same event. One of them is a commitment and the other is a commitment with a door in it, and the door is the part that did not travel.

Bartlett also noted that nobody looking at a single reproduction could predict the cumulative effect of losing small details one version at a time. That is the part worth carrying over. Compare any two adjacent versions of one of your team's decisions and the difference will look like nothing. Compare the first and the last and you have two different agreements.

The experiment is not a perfect analogue and it is worth saying so. A folk story is not a statement of work, and his chains ran through strangers rather than colleagues with shared context. A company has correction mechanisms a laboratory chain does not. It also has a force the experiment deliberately removed, which is an audience with expectations, and that force pushes in one direction rather than scattering.

Every version reads as complete

This is the part that keeps it invisible for years. A record does not tell you it is a copy. The slide does not carry a note saying it is the fifth restatement of something said in a call in March. The ticket does not open with a caveat about being a summary of a summary. Each artefact presents itself as a whole, self-contained account, because that is what an artefact does.

So you open one, it answers your question, and you close it. You do not go looking for a second version, because nothing suggested there was one and you already have your answer. Everyone in the company does this, all day, with high confidence, and the confidence is warranted from inside any single document.

Noticing the conflict requires reading two versions next to each other. Almost nothing in an ordinary week produces that act. Nobody's job is comparing the deck against the ticket. There is no view that shows you the same decision rendered six ways. The disagreement was never hidden. Nobody has ever assembled it in one place.

This is worth separating from a problem it resembles, where two people leave the same meeting with the same words in their heads and two different pictures of what those words mean. There the language matches and the understanding diverges. Here the language itself diverges, on paper, in writing, where anyone could see it, and no reader ever holds both pages at once.

The conflict surfaces at the worst possible moment

The one thing that reliably puts two versions side by side is a consequence. A customer quotes the email back at you. Finance quotes the plan. Engineering quotes the ticket. Somebody in the room says that is not what we agreed, and means it sincerely, and is holding a document that supports them.

So a team discovers that its records disagree during a billing dispute, an escalation, a missed launch, or a renewal that has suddenly gone sideways. The discovery arrives attached to a problem, which sets the frame for the conversation that follows. The question people reach for is who got this wrong.

That framing is efficient at producing an answer and useless at producing a fix. Once the topic is fault, everyone's energy goes into defending the version they wrote, which is easy to do because every version is defensible on its own terms. The question that would actually help, which of these is closest to what was said in the room, needs a piece of information nobody has, so it gets skipped and the meeting settles on a compromise nobody can check.

Ask which one is authoritative and you will get an answer that is not true

Every team has a stated rule about this. The tracker is the source of truth. The plan is what we go by. If it is not in the CRM it did not happen. Ask anyone and you will get the rule without hesitation, and the rule is real in the sense that somebody decided it and wrote it down.

Watch what happens when a question comes up mid-call and you see the operating rule instead. In practice a team resolves a conflict with whichever of these is nearest to hand. The most recently updated artefact, on the reasoning that it reflects the latest thinking. The version written by the most senior person in the chain. The account given by whoever is in the room and sounds most sure of themselves. Or the version the other party is holding, which ends the argument by making it expensive to continue.

None of those is a bad instinct. Each of them is a sensible heuristic for a person who needs an answer in the next eight seconds and has no way to tell the copies apart.

Recency and accuracy point in opposite directions

Here is the uncomfortable part. Each of those four tie-breaks selects against the thing you want.

The most recently updated artefact is the one furthest down the chain. It was assembled from other artefacts, most likely by somebody who was not in the original conversation, because the people who were in it moved on to the next thing months ago. Recency in a record tells you when somebody last touched a copy. It says nothing about proximity to the source.

The most senior author is usually the person least likely to have been in the working session where the detail got settled. Seniority in the chain correlates with distance from the room, which is exactly backwards from what the heuristic assumes.

Certainty is worse still, because certainty grows with every hop. Hedges are the first casualty of summarisation, as Bartlett's chains showed and as anyone who has written an exec update knows from the inside. The most confident version available is reliably the most processed one. Confidence in a work record is a measure of how many times it has been rewritten, and people read it as a measure of how well it is known.

And the version the other party is holding is the one that left the building, which means it was written to persuade somebody. That is the version with the fewest qualifications in it by design.

Put those together and the default tie-break is close to a procedure for selecting the least reliable copy in the set. Not occasionally. As a tendency, with a mechanism behind it. This is a different failure from a single claim losing its hedge and hardening into a fact through repetition. Here there are four complete, well-formed, mutually exclusive accounts, and the rule the team uses to choose between them prefers the one that has travelled furthest from the moment.

Many of these are a proposal and a decision, filed identically

A large share of record conflicts are not two accounts of the same moment at all. They are accurate records of two different moments in the same conversation.

At minute twelve someone proposes doing it in two phases and the room seems to like it. At minute thirty eight, after an objection nobody anticipated, the decision lands somewhere else. One person wrote down the first thing, because it was the clearest statement in the call and it sounded settled. Another wrote down the second. Both notes are faithful. Neither note says which minute it came from, or whether the thing it describes was a suggestion on the table or a conclusion the room had reached.

The reverse happens too and is harder to catch. A decision gets genuinely reversed two days later, in a corridor or a two message exchange, and the reversal produces no artefact because reversals rarely do. The stale record is then the only written account in existence, and on paper it wins every argument it is brought into.

Telling these apart needs two fields that almost no record carries: when within a conversation something was said, and what standing it had at that point. A note with a date on it and nothing else cannot distinguish a proposal from a conclusion, and most notes have a date on them and nothing else.

Why designating a source of truth does not hold

The standard answer is to pick a winner in advance. Declare the tracker authoritative, tell everyone, and treat everything else as commentary. Teams do this sincerely and it fails for reasons worth being precise about, because the failure is not laziness.

The first reason is that the designated surface is itself downstream. A ticket is a restatement. Making it authoritative promotes a summary over the conversation it summarises, which means the policy is instructing people to trust the copy. That works fine for anything the tracker was built to hold, which is the state of work. It works badly for the reasoning and the conditions, which were never said in a ticket and arrived there secondhand if they arrived at all.

The second reason is that a rule requiring a detour loses to a surface that is already open. Somebody on a call needs the answer in seconds, and the CRM tab is right there while the tracker is two clicks and a login away. They answer from what is in front of them, every time, and no amount of policy changes the physics of that. A rule only holds where following it is the path of least effort.

The third reason gets overlooked. Declaring one version authoritative does not delete the others. They keep circulating regardless, especially the ones that went outside the company, and an external party has never read your internal policy about which of your systems is canonical.

The useful property is distance from the room

Authority is a policy, and policies decay the moment they become inconvenient. Distance is a fact about a record, and facts hold up under pressure in a way that designations do not.

What settles a conflict between two documents is which of them sits closer to the moment the thing was actually said, rather than which system it happens to live in. A record written by somebody in the conversation while it was happening sits at distance zero. A record assembled a week later from that record sits one step out. A slide made from the plan that was made from the update sits three steps out, and its author may never have spoken to anyone who was there.

If you know the distance, the conflict resolves itself in about four seconds and without anyone having to be wrong. The nearer record describes what was said. The further one describes what somebody understood, later, for a purpose, and that is often useful information in its own right about how the decision was being communicated.

Almost nothing a team uses carries this. Tools record who typed and when they typed, which is a fact about the copy rather than about the decision. A document last edited yesterday by a director tells you nothing about whether that director was in the call.

What a record has to carry so the copies stop competing

Five things, and the fifth is the one that changes the shape of the problem.

Which conversation it came from, and roughly when within that conversation, so a proposal can be told apart from a conclusion reached twenty minutes later. Who said it, which is frequently not the person who wrote it down and is the detail most reliably lost at the first hop. What standing it had at that point, meaning whether the room had settled it or was still turning it over. What it was conditional on, kept as the if-clause somebody actually spoke rather than as a cleaned up caveat, because a cleaned up caveat is already one rewrite into disappearing.

And for every later restatement, which record it came from. That last one sounds like bookkeeping and is the whole thing. A restatement that carries a line back to its origin is no longer a competing account. It is a view of one. The slide stops being a fifth opinion about what was decided and becomes a pointer at the conversation, and the question of which document wins stops being a question, because there is only ever one document and a set of things that reference it.

This is also the difference between a record that survives being rewritten and one that does not. Copies are going to get made. People will keep writing summaries for their own audiences and they should, because an exec update genuinely does need to be shorter than the call. Summaries existing has never been the problem. The trouble is that a summary and its source become indistinguishable once they are both sitting in the same search result.

This is a capture problem before it is a discipline problem

None of the five can be added later. By the time the deck exists, the moment it descends from is either recorded somewhere or it is gone, and if it is gone then no amount of care from the person making the deck can put a line back to it. The origin has to be captured at the point the origin happens, which is the one moment nobody knows the decision will matter.

That rules out the obvious remedy, which is a rule about how people write things down. Every such rule asks a person to do extra work at the moment they are least inclined to, and to make a judgement they are not equipped to make, which is whether this particular Tuesday conversation is the one that will be argued about in November. Most of them are not. A handful will be, and there is no way to know which in advance.

The only version of this that holds is one where the first record exists without anybody having decided to make it. That means capture that runs across the work as it happens, the calls and the screens and the threads, without a bot sitting in the meeting announcing itself and without somebody having to nominate the conversation as important. Then the summaries people write afterwards have something to point at, and the chain has a first link that is not a copy of anything.

That is the work Driffle is built to do. The value sits in the later versions being able to resolve back to a moment, so that six months from now, when a customer and a colleague are reading two different sentences off two different screens, there is a third thing to check that neither of them wrote.

Reconciling records, not keeping a file on people

There is an obvious way for this to go wrong, and it is worth naming before someone builds it. A system that stores every version of every decision and can show you which ones diverged is, read from a certain angle, a record of who misreported what. Run it that way for a month and you will have a tidy history of every time a person's summary drifted from the source.

That reading destroys the thing it is measuring. People who believe the comparison is about them will stop putting restatements in writing. They will move to saying it on a call, or sending it in a message they can delete, or phrasing it so vaguely that no comparison is possible. The copies disappear, and because the copies were often the only written trace, the record goes with them.

So the output has to be about the record. Which version is nearest the source, what the nearest version says, and what changed on the way out. Not who wrote the version that was furthest. A team also needs to be relaxed about the fact that drift is normal and mostly harmless, since the aim is having a way to check when it matters rather than eliminating variation between documents, which is neither possible nor desirable.

The same restraint applies to scope. A record of what was said in a working conversation is a different object from a record of everything a person did that day, and the second one is not needed to solve any of this.

A test you can run this week

Pick a decision your team made in the last month that at least three people have written down somewhere. A pricing exception, a scope cut, a hiring call, a date that got committed to a customer. Something with consequences attached, not a preference about tooling.

Now open every version you can find, at the same time, on one screen. This is the unusual part. You have almost certainly never done it for anything, and the act of assembling them takes longer than you expect, which is itself part of the answer.

Then check three things. Do they say the same thing, allowing for length. Does any of them carry the condition that was attached in the room, and can you remember whether there was one. And can you tell, from the record alone, which conversation each version descends from and who in that conversation said it.

One more, and it is the useful one. Notice which version you instinctively trusted as you read. Then work out why you trusted it, and whether the reason had anything at all to do with how close that document was to the moment the decision was made.