Scope arguments are memory arguments

Nearly every dispute between two companies about what was agreed is a dispute about a record that was never written. By the time money is involved, both sides remember honestly — and differently.

On 12 March, on a call that ran nine minutes over, an Atlas producer and a Meridian product manager agreed something about the mobile app. Everyone hung up satisfied. Neither wrote it down, because it hadn’t felt like a decision — it had felt like a clarification.

On 4 June, Atlas invoiced for three additional weeks of build. On 9 June, Meridian held back $41,200 pending “clarification of the March agreement.”

Both companies now go looking for evidence. Both find some. Neither finds enough.

The fork

What happened on that call is not complicated, and it is not anybody’s fault. A decision was made verbally and then filed twice, in two different companies, by two people carrying different concerns into the room.

Atlas’s producer heard a scope change and noted it as one: client wants mobile in launch — will need to quote. Meridian’s product manager heard a clarification of something they thought was always implied, and noted it as such: confirmed mobile is in scope for launch. Both notes are accurate summaries of what that person understood. Neither is a record, because a record requires the other party to have seen it.

Four months of work then proceeds on two subtly different premises. Nothing breaks, which is the dangerous part. The divergence only becomes visible when it touches money.

How one unrecorded decision becomes an invoice disputeA single decision node in March splits into two diverging lines — one recorded by the agency as new billable scope, one recorded by the client as a clarification already in scope. The gap between them widens across April and May while work proceeds, and the two lines converge in June on a disputed invoice.one nine-minute conversation · four monthsOne decision, spoken aloudit didn't feel like a decision at the timeAtlas files it as new scopeMeridian files it as a clarificationthe gap neither side could seework proceeds on two different premisesinvoice disputed$41,200 held back12 mar13 marapr — may4 jun9 jun
The forkThe cost of the missing record isn't paid in March, when writing it down would have taken ninety seconds. It's paid in June, at the point where the two versions finally have to be the same.

Nobody in this story is lying

It’s worth being precise about the mechanism, because the usual framing — one side is chancing it — is wrong often enough to be useless.

Memory is reconstructive. You don’t retrieve a meeting, you rebuild it, and you rebuild it out of the parts that mattered to you at the time. The producer remembers the commercial implication because managing scope is her job. The product manager remembers the user outcome because shipping the right thing is his. Four months on, each recalls a coherent, sincere, materially different meeting.

Then the incentives arrive. Nobody sets out to reinterpret March in their own favour, but the reinterpretation happens anyway, because the ambiguity has to resolve somewhere and each side’s private notes point one way. Add a procurement team that wasn’t on the call, and an account manager whose bonus depends on the quarter, and the disagreement stops being about the mobile app.

Memory isn't a record. It's a reconstruction, performed by someone with an interest.

The evidence is real, and useless

Here’s what happens when the two companies go looking for the truth.

Atlas has the producer’s note, which is internal, and therefore worth roughly nothing as evidence. There’s an email from 14 March that mentions “the mobile piece” without defining it. There’s a Slack thread with the relevant sentence in it, in a channel Meridian has never had access to. There is a recording of the call, ninety-four minutes long, and nobody is going to watch it — and if someone did, they’d find the exchange was four sentences of pronouns.

Meridian has a version of the spec with tracked changes, which shows mobile in a bullet list, but the file is spec_v4_FINAL_rev2.docx and no one can establish whether Atlas ever received that revision.

Everything needed to settle this exists. None of it is a record, because a record has properties that none of these have: both parties saw it, at the time, and neither could change it afterwards.

And the argument is expensive well beyond the invoice. Two weeks of senior time on both sides. A legal review that costs more than the disputed line. A relationship that quietly moves from partner to counterparty, which nobody says out loud but everybody starts behaving as though were true. Renewal conversations get shorter after this. Not because anyone behaved badly, but because it happened once and neither side can be confident it won’t happen again.

Write it down once, where both sides can see it

The fix is unglamorous and old: decide in the open, record it jointly, and don’t let anyone rewrite history quietly. What’s been missing is somewhere neutral to put it.

A shared, append-only decisions logThree entries in a shared decisions log. Entry twelve says launch scope excludes the mobile app and is marked superseded. Entry thirteen covers data residency and is in force. Entry fourteen adds the mobile app to launch scope with three extra weeks and forty-one thousand two hundred dollars, is in force, and supersedes entry twelve. Both companies have signed every entry, and nothing has been deleted.the decisions log · append-only · signed by both companies№12Launch scope excludes the mobile app12 Mar · rationale: fixed launch date takes prioritysupersededATME№13Data residency: EU tenant only2 Apr · rationale: Meridian legal, DPA clause 7in forceATME№14Mobile app added to launch scope12 Mar, recorded same day · +3 weeks · +$41,200in forceATME№14 supersedes №12 — and №12 is still there
Supersede, don't editThe March decision isn't corrected or deleted — it's superseded by a later one, with both companies' sign-off on each. Six months on, the question "what did we agree, and when did it change?" has one answer instead of two archives.

There’s a version of the 12 March call where someone opens the decision log while everyone is still on the line, types one sentence, and both sides click sign-off before the next meeting starts. It takes ninety seconds. It is not a satisfying story, and it saves $41,200 and a working relationship.

The disputes worth having are about what to do next. Almost none of the ones companies actually have are.