← All notes
Project Scoping7 min

A preference is not a constraint

Scoping notes fill up with lines that sound binding and are not. "We'd prefer the field app to work offline." "It would be nice if editors could rearrange the homepage without a deploy." "Ideally the OCR queue and the field queue share one inbox." Those are preferences — valuable signal about direction — until someone treats them as non-negotiable and the estimate absorbs work the room never agreed to protect. Constraints are different: they bound what can ship. Mixing the two columns is how a soft want becomes an invisible hard requirement, or a real limit gets bargained away like a nice-to-have.

Two columns, one failure mode

A constraint is a boundary the build has to respect or the outcome fails the reason the project exists. App Store review timing before a demo. A print vendor that only accepts a specific PDF. A medical workflow that cannot invent data when OCR confidence is low. A Sanity schema that must keep empty fields from blanking a page. Those are not taste. They are conditions.

A preference is a direction the stakeholders would like if the calendar, the seam, and the operating conditions allow it. Dark-mode polish. A second campaign landing. Shared inboxes across queues that behave differently under load. Useful to hear early. Dangerous when the same bullet list makes preference and constraint look identical.

We have watched both directions fail. A preference gets promoted into the acceptance criteria without an owner or a cut rule, and the last two weeks drop the seam spike to keep the soft want. Or a real constraint gets softened in a workshop because it sounded expensive, and production invents the limit the hard way — usually on a device, a printer, or a webhook the demo never touched.

Write the label while the sentence is fresh

In the scoping notes we mark each material line as constraint, preference, or assumption. Constraint means: if this moves, the ship date or the minimum coherent product has to move with it, in writing. Preference means: ranked against the cut list; it can leave without breaking the reason for the date. Assumption means: not yet verified — with an owner and a trigger when it breaks, same discipline as elsewhere.

The label has to land the same day the sentence appears. Once a preference sits unlabeled next to three constraints for a week, the room starts treating the whole list as equally hard. Slack one-liners accelerate that blur: "we need offline" lands with the same force as "the booth date is fixed," and nobody asks which kind of need it was.

For BuilderHelp-style platforms, "field and dashboard share one session" might be a constraint if the operating model forbids a second account. "One shared notification inbox for every surface" is often a preference wearing constraint clothing — the surfaces have different failure modes, and forcing one queue invents scope the estimate did not price.

Constraints get owners; preferences get rank

A constraint without an owner is how boundaries become folklore. Someone has to say what happens if the constraint bites — who cuts scope, who moves the date, who signs the trade. Write that next to the constraint, not in a later steering call when the calendar is already tight.

A preference without a rank is how the cut list stays empty. Preferences belong on an ordered list against the minimum coherent product: what can leave, in what order, without breaking the ship reason. That is the same cut-list discipline we use for fixed dates — preferences are usually the first candidates to leave, not the last.

Do not "protect" a preference by calling it a constraint to win an argument. That move hides the trade. If the preference truly became a constraint, say what changed: a buyer commitment, a compliance rule, a demo that cannot proceed without it. Then update the cut list and the estimate together.

Mid-build asks inherit the same labels

New asks arrive as clarifications, trades, or changes. They also arrive as preferences smuggled in as musts. Before the build absorbs them, we ask: is this a new constraint, a ranked preference, or still an assumption about someone else's system?

If it is a preference, it enters only through the swap rule — something of similar size leaves, same day, written down. If it is a constraint that was missing from scoping, that is a scope correction, not a squeeze. Treating a newly discovered constraint as a four-hour favor is how hour thresholds fail; the seam moved even if the ticket looked small.

Preferences that keep getting re-litigated every standup are a signal the rank was never agreed. Stop debating the sentence. Reopen the cut list, put the preference on it, and decide whether it outranks something already in the build. Argument without rank is how soft wants burn calendar.

What we check before the estimate hardens

Open the scoping notes and split every hard-sounding line into constraint, preference, or assumption. For each constraint: owner, what fails if it moves, and whether it belongs in date acceptance. For each preference: rank against the minimum coherent product and a clear leave condition. For each assumption: owner and falsify trigger.

If the constraints column is empty and the preferences column is long, the estimate is pricing taste, not boundaries — shrink ambition or find the real limits. If everything is marked constraint, the room is using the word to avoid ranking; force a cut list until at least three outcomes can leave without killing the ship reason.

The milestone is a short constraints list the build can be held to, and a ranked preferences list the cut list can actually use. Until those labels exist, "must" and "would like" are the same ink — and the calendar will invent the difference later.

Questions

How do we tell a preference from a constraint in the room?
Ask what breaks if it does not ship with the first slice. If the ship reason still holds and only ambition shrinks, it is a preference. If the date, compliance, or operating model fails, it is a constraint — and it needs an owner.
Can a preference become a constraint later?
Yes, when something external changes: a buyer commitment, a review gate, a compliance rule. Write the promotion the same day, update the cut list and acceptance, and treat it as a scope correction — not a silent squeeze.
How is this different from an assumption?
An assumption is an unverified belief about the world ("API access lands in week one"). A preference is a ranked want. A constraint is a binding boundary. All three need labels; only constraints bind the ship without further proof or ranking.

Sources

  1. How to Scope a Software Project — exclusions and constraints — Scope includes explicit exclusions and real-world constraints; preferences without rank become disputed changes during build.
  2. Project constraints — boundaries vs stakeholder wants — Constraints bound planning; stakeholder preferences influence scope but are not the same as non-negotiable limits.
  3. Related: A fixed date is not a fixed scope
  4. Related: An assumption is not a signed decision
  5. Related: A non-goal is not a phase-two ticket

Have something to build?

Tell us what you're working on and we'll tell you honestly whether we're the right fit.

Work with us