One Noun, One Number, One Date
A key result you cannot repeat from memory after one read is not one key result. It is two, and at the grade date that becomes an argument instead of a score.
A key result is one noun, one number, one date. If it does not fit in about twelve words, it is carrying reasoning that belongs somewhere else.
The wordiness is not a style problem. It is the visible symptom of a key result that has not been decided yet.
Why long ones are worse, mechanically
Long key results feel more rigorous. They are usually less rigorous, for a reason that has nothing to do with taste: a key result with two clauses has two truth values, so it cannot be graded.
At the grade date you are left arguing about partial credit. I found this described exactly in a retrospective from an old archive, where a company graded its own quarter at 34% and wrote down why the number was not worth much:
I had to make a lot of judgement calls because some of the key results were just not articulated in ways that are easily gradeable.
The breakdown in that retro is the teaching. The objective with a single number attached graded instantly. The one about craftsmanship needed a pile of judgement calls and scored 12.5%. That correlation is not a coincidence: a key result you can grade is a key result you can steer toward mid-quarter.
There is a second cost, and it is the one people underrate. A key result nobody can recall is a key result nobody steers by. The whole point of the form is that a person in the middle of the quarter can hold the target in their head and notice they are off it. Prose defeats that.
The rules
- Twelve words. A working ceiling, not a law. Over it, look for what to move out.
- One number. Exactly one thing moves. Two numbers is two key results.
- No connectives. An “and”, a “with”, a colon, or a “not X” contrast each mean you have two key results, or a key result plus a note.
- The text never argues. Why it matters, what it is gated on, and what it replaces all live outside the row. The row states the fact; the note carries the case.
- Baseline is a value and a date, not a sentence.
- Read-aloud test. Say it once, look away, say it again. If you cannot, it is not finished.
The shape
[verb] [one noun] [from X to Y] [by date]
Any part is droppable except the noun and the measure.
Worked examples
Good ones, all from real archives:
| Key result | Words |
|---|---|
| Place 25 graduates | 3 |
| Keep blended acquisition cost under $700 | 6 |
| 64 leads identified by 21 September | 6 |
| Every page completes under 200ms by 1 June | 8 |
And before-and-afters from my own portfolio, which is where these rules came from:
| Before | The problem | After |
|---|---|---|
| Smart import ships: a new account processes the latest episode per show, not the full backlog | 16 words. A colon, a spec, and a “not X” contrast | New-account import capped at 1 episode per show |
| Cost per new account is measured and published weekly, with a written ceiling | Three deliverables in one row: measure, publish, set a ceiling | Cost per new account published weekly plus Per-account cost ceiling adopted |
| The warm candidates get one approach, gated on the first objective being complete, with a written verdict | Carries a gate that the operating rules already state | 4 candidates approached plus Written verdict on the channel by 20 Nov |
The third is the one worth remembering. The gate was already written in the operating rules. The key result was duplicating a constraint that lived somewhere better. Wordiness is often duplication, not detail.
What to do with what falls out
Write the long version first if that is how the thinking comes out. Then run the six rules over each row and push everything that fails into one of three homes:
- A second key result, if it is a separate thing that moves
- The note under the objective, if it is reasoning, sequencing, or a contrast
- The operating rules, if it is a gate or a stop condition
Nothing gets deleted. It gets relocated. The row that survives is the one you can grade.
The failure shape that reliably scores zero
One more thing from that old retro, because it is the most common bad row and it looks respectable:
“Start collecting data on X” is the most common failure shape, and it reliably scores zero.
It has no owner-visible consequence, it can always slip a week, and at grading time it is binary and usually false. If instrumentation genuinely is the work, make the key result the decision the data enables, with a date, rather than the collection.
Get updates
Occasional notes on what's happening at Enginery. No spam, no marketing.