Originally written in French. Translated by AI — the meaning has been preserved, not the prose.
Angle
A product decision register draws its power from three refusals, and each one guards against a distinct way in which corporate documentation ordinarily fails. It refuses anything that says what to build, because a rule brought down to feature level stops being derivable for the cases nobody had anticipated. It refuses anything that prevents no future rediscussion, because a register is consulted from memory and past fifteen or so entries nobody consults it at all. It refuses to be corrected, because a decision is a choice-and-context pair, and rewriting deletes the context while believing it is only replacing the choice. What the company gets in exchange for these three refusals is not saved time but a single answer where four functions used to apply four: support, partnerships, product and sales stop settling each according to its own rationale. And the price of the setup is read by subtraction — the explanations given fifty times over, the arbitrations reopened for every large prospect, the dependence on the memory of whoever was in the room.
Synthesis
Taken separately, these ideas look like writing advice. Assembled, they describe a setup whose every useful property is a restriction.
The starting point is a diagnosis, not a method. A product company knows what it did and no longer always knows why: the question of a dispute between customer and supplier on a marketplace comes back at the roadmap committee, with sales, in support, in the specs. That recurrence is not a run of badly handled cases, it is the sign of a rule missing at the level where it would apply. What the absence costs is not first meeting time: it is that each function settles according to its own logic, and that four incompatible decisions apply simultaneously to dispute policy, support messaging and product priorities.
Then comes what makes the remedy work, and it consists of three limits. Altitude first: "in this type of situation, here is the rule" carries because the sentence commits to no feature — a rule that says what to build is a disguised specification, and a less precise one than a real specification. Then scarcity, which is not a discipline but the condition of use: fifteen entries fit in the head of a Product Manager and a sales director, fifty a quarter fit nowhere. Immutability last: the rule on supported languages, replaced two years later because translation costs collapsed, is not corrected — the new decision is written alongside it, and the old one goes on explaining why the refusal was reasonable at the time.
A fourth piece keeps these limits from turning into rigidity. Writing down what would have to happen for the decision to change closes the subject today without closing it for good, and moves the reopening from a balance of power to a fact to be established: an Italian prospect who insists does not open the door, a demonstrated change in translation costs does.
There remains the question of where it is filed, which decides whether any of the rest serves. A cross-cutting rule written inside a Jira card is accurate and unfindable for the people it concerns; it inherits the ticket's audience and lifespan, which is to say nothing at all. Nothing signals that disappearance, since the decision has not disappeared.
What the whole brings into view, and what none of these ideas carries alone: an organization's memory is not built by writing more. It is built by a setup whose every rule is an exclusion, and which is judged on the discussions it makes disappear — a balance sheet structurally unfavourable to read, since the cost of writing is immediate and visible while the rediscussions avoided leave no trace.
Tensions / contradictions
The first tension bears on the reach of the remedy. A written register produces a single answer only if the teams know it and apply it; nothing in the setup guarantees that, and an ignored rule leaves the diverging readings intact while adding a document that claims the opposite.
Second tension, between two ideas assembled here. One holds that an old artifact describes a vanished context and misleads whoever picks it up; the other holds that an old decision preserved with its context stays useful years later. The distinction maintained here is that the first carries an intention and the second a dated rationale — but the boundary is thinner than it looks as soon as a decision contains conditions for reassessment, which are an intention.
Third point, left open: the scarcity threshold is set by experience, not by measurement. No observation says at how many entries a register stops being consulted, and the number put forward varies with the size of the organization without our knowing by what law.
Questions
- What is it, in an organization, that makes a decision register actually get consulted at the moment the question arises, rather than dug up afterwards to justify what has already been answered?
- Which structuring decisions are already written into a team's tickets and specifications, and by what means can they be spotted there?
- Whose job is it to write a cross-cutting rule when it concerns support, sales and product in equal measure, and none of the three has a mandate over the other two?