"We'll go with the November launch, since legal signed off." There is a plan in that sentence and a premise, and a reader skimming it registers only the plan. The whole launch rests on that sign-off, and it arrived as a subordinate clause — which is where premises hide.
Three weeks later somebody discovers legal never signed off. Nobody lied. The premise was simply never written in a form anyone could have disagreed with.
Assumptions hide in the grammar
The load-bearing premise almost never gets its own sentence. English offers plenty of ways to move a claim past a reader without asserting it:
- Subordinate clauses. "Since the contract renews in March…" asserts the plan and presupposes the renewal.
- Definite articles. "The migration window" presupposes there is one, and that you agree which.
- Nominalisation. "After the handoff" turns an event that might not happen into a calendar fixture.
A presupposition survives a skim. Readers argue with what you assert and absorb what you presuppose, because there is nowhere to put the objection. Stating an assumption means undoing that on purpose — pulling the premise out of the clause it hides in and re-stating it as something a person can answer.
Give each premise its own line
The test is blunt. Could a reader reply "no" to this line alone, without quoting anything around it? If not, it is not stated yet.
That usually means a list, and a list works for reasons other than formatting taste. It gives each premise an address: a reader can write "3 is wrong" and be finished. Nobody can point at a subordinate clause — they must quote it, isolate the part they mean, then explain why. A paragraph of effort for a one-word correction.
One premise per line. Compound assumptions — "they're under-resourced and don't want the project" — can only be answered "half right," which both parties round up to agreement.
Say where it came from
Then attribute it. Readers weigh three origins differently.
- Told. Someone said this. Name them and when. "Priya said Tuesday the budget line closes at quarter end."
- Inferred. Nobody said it; you concluded it. Say from what. "Nothing has touched the staging repo since June, so I'm treating it as parked."
- Guessing. You have no basis and are proceeding anyway. Say so.
Most writing flattens all three into one level declarative, and the cost is asymmetric. An unmarked guess that holds costs nothing. One that fails sends the reader back to re-audit the last five things you told them, because they now know your prose does not distinguish.
Three origins beat the two-way fact-or-guess split in how to ask for a decision, because inference is where the errors live. Facts are usually right and guesses get flagged. Inferences feel like facts to whoever made them.
Make it falsifiable at a glance
A premise a reader cannot check will not get checked. "Budget is tight" is a mood; there is nothing in it to disagree with. "There is no new headcount before the next fiscal year" is a proposition — someone either knows it is wrong or does not.
The difference is a name, a date, or a quantity, ideally one your reader could verify without leaving their chair. Keep it to a line. A premise that needs a paragraph is two premises and an argument.
Put the load-bearing one first
Order assumptions by what collapses if they are wrong, not by chronology and not by how sure you feel. For each, ask what still stands if it turns out false. The premise whose failure invalidates the rest goes at the top.
This runs against instinct. People lead with what they are surest of and bury the shaky one near the end, where it feels less exposed. That is backwards — the shaky one is the only premise you genuinely need read.
A buried assumption doesn't get corrected — it gets the message rejected
The return on all of this is not tidiness.
When a premise sits on its own line, a reader who disagrees has a cheap move available. Correct line two. Everything else survives, and the correction costs neither of you any standing.
When the premise is buried in a declarative, that move does not exist. A reader who senses something is off has to find the sentence carrying it, unpick it, and explain the problem — expensive to write, and it reads as an attack on the whole message rather than a note on one line. Most will not bother. You get a non-reply, a vague "let's discuss," or a meeting where the real objection surfaces forty minutes in.
The choice is not between a clean message and a cluttered one. It is between a line edit and a rejection.
What a format can carry, and what it can't
Briefings sent through RelayLink give assumptions their own field rather than leaving them loose in the prose, and each item is tagged by origin — stated by sender or inferred by sender's AI. The reader's assistant reads the tag instead of inferring authority from how confident the sentence sounds: the general provenance argument, applied where it earns most. The list caps at ten, which forces the prioritising above.
One honest limit. That tag is self-reported by the drafting assistant. The sender's note gets its label server-side, by comparing what the human typed against the draft offered to them — but nothing checks whether your assistant was right to mark a premise as stated by you. So the assumptions list is the part of a draft to read hardest before approving, because a fluent draft sounds no different when it is guessing.
None of this needs software. Pull the premise out of the clause, give it a line, say where it came from, put the load-bearing one first. Academic collaboration requests are the clearest case where naming assumptions early pays off, since the defaults differ by field and nobody states theirs unprompted. To have your assistant build that list from a session you have already worked through, connect it.