Hypothesis, cost, mechanism, measurement
Four fields on one page, in that order — and the mechanism field is the one that does the work, because a hypothesis with no stated causal path is a wish with a number attached.
You now have the structure and, from the metric-set module, the vocabulary. What you do not have is the artifact. Everything so far has been preparation for one page of writing, and the page is where the preparation either survives or falls apart, because a page can be read by someone who does not trust you and does not have to be polite about it.
This lesson gives you the four fields, the order they go in, and the reason the third one carries the weight. By the end you will have three cases written down rather than three cases you could describe if asked.
The four fields
One page. Four fields, in this order. Two of them are already sitting in your case file from earlier lessons.
| Field | The one question it answers |
|---|---|
| Hypothesis | What do you believe will change, and in which direction? Not how much. |
| Cost | What does this consume, including after it ships and nobody is looking? |
| Mechanism | By what physical path does the change reach the number? Who or what does something different? |
| Measurement | What will you look at afterwards, and can you actually obtain it? |
The order is not arbitrary and it is not the order you will want to write in. Most people start at measurement, because measurement is where the impressive-sounding numbers live. Starting there means picking a number first and reverse-engineering a story that reaches it, which is the exact motion this course is written to interrupt.
Hypothesis: a direction, not a magnitude
A hypothesis is one sentence with a subject, a verb, and a direction. The temptation is to put a magnitude in it, and the magnitude is what kills it, because a magnitude you have not measured is the one part of the sentence that can be shown to be false for free.
Compare two versions of the same belief about the review gate:
The review gate will reduce net revenue churn by two points.
The review gate will reduce the number of extraction errors that reach a customer, and therefore reduce the support contacts caused by them.
The first is shorter, sounds more senior, and is worse in every way that matters. It names a metric the metric-set module taught you has at least two live definitions, attaches a number nobody computed, and skips entirely the part where an error becomes a support contact becomes a renewal decision. The second is checkable at every clause. If someone wants to argue with it, they have to argue about errors and support contacts, which is a conversation you can win.
Cost: the field people quietly shrink
Cost is the only field you control completely, and it is the one most often understated, because understating it feels like making the case stronger. It does the opposite. A cost figure that is obviously low is the first thing an engineering lead attacks, and once they have won that argument they have a reason to distrust the rest of the page.
Three components, all of which belong on the line:
- Build. Design and engineering time to ship it once. State it in engineer-weeks and label it an estimate.
- Displacement. What comes off the plan to make room. The lesson on presenting to a room that will ask what you would cut is entirely about this line, and arriving without it is the most common way a good case gets deferred.
- Carry. What the thing costs every quarter after it exists — support, maintenance, the review it now requires, whatever it makes harder to change later. Software you built stays on the books.
Mechanism: the field that does the work
You wrote your mechanisms in the P&L lesson: what a person or a system does differently now, in physical terms. That was not a warm-up. Mechanism is the load-bearing field, and here is the specific reason.
Hypothesis, cost and measurement can all be waved through. A reader can accept your direction, accept your estimate, and agree to look at your metric later, and nothing about that exchange requires them to think. Mechanism is the only field that can be attacked immediately, by someone with no data, using only what they know about how the business works. That makes it the field that earns you the room’s attention if it holds, and the field that ends the meeting if it does not.
A hypothesis with no stated causal path is a wish with a number attached — and the room has heard a lot of those. When somebody says “this will improve retention” and cannot say what happens between the pixel and the renewal, the listener does not conclude that the claim is false. They conclude that the claimant has not checked, which is worse, because it applies to everything else on the page.
Measurement: the field with a trapdoor
Measurement gets a lesson of its own, immediately after this one, because the obvious approach fails in a specific and expensive way: you choose the measurement that would settle the question, discover four weeks later that nobody instruments it, and end the quarter with a shipped change and no evidence at all. For now, write the measurement you want, and mark it with a question — can we actually get this? The next lesson answers it.
The regulator’s version of the same instinct
It is worth knowing that this discipline is not a design-team invention. In February 2020 the SEC issued interpretive guidance on the performance metrics companies disclose outside their financial statements. It starts by conceding the problem the whole metric-set module was about: such metrics “can vary significantly from company to company and industry to industry, depending on various facts and circumstances.”
Having conceded it, the Commission does not respond by standardising the metrics. It responds by saying what has to travel alongside one. It would generally expect to see “a clear definition of the metric and how it is calculated,” “a statement indicating the reasons why the metric provides useful information to investors,” and “a statement indicating how management uses the metric in managing or monitoring the performance of the business.”
Then the sentence that is worth pinning above your desk. A company, the guidance says, “should also consider whether there are estimates or assumptions underlying the metric or its calculation, and whether disclosure of such items is necessary for the metric not to be materially misleading.” That is the whole ethic of this course in one clause, and it is not a suggestion about tone. The assumptions underneath a number are part of the number.
Be precise about what this does and does not support. The guidance governs public disclosure of a metric a company has already chosen to report. It says nothing about internal proposals for new work, and it does not require anybody to write a hypothesis or a mechanism. The four-field page is this course’s own construction. What the SEC supplies is independent evidence that the instinct behind it — a number is not a claim until its definition and its assumptions travel with it — is held by a party with no stake in whether design is valuable.
Where people get burned
Do not walk into a room and tell people the SEC requires this structure. It does not, and the one person present who has filed a 10-K will know it does not. The honest sentence is narrower and lands harder: “the regulator’s standard for disclosing an operating metric is a definition, a calculation, a reason and the assumptions underneath it — I have held my own case to roughly that.”
What one filled page looks like
Here is the review gate, written to the structure. Every figure in it is an estimate, marked as one, and none of it is a measured result:
Project Human review gate in the document-extraction product
— per-field confidence, source crop, approve-or-correct
before anything downstream fires.
Hypothesis Fewer extraction errors reach downstream systems, so
fewer of them surface later as customer-reported
problems. Direction only. No magnitude claimed.
Cost Build: ~6 engineer-weeks, estimate.
Displaces: the bulk-export work queued behind it.
Carry: the confidence thresholds need review whenever
the extraction model changes.
Mechanism An operator sees the low-confidence fields flagged and
corrects them in-flow. The corrected record is what
moves downstream. Errors that previously reached a
customer's finance team are caught by the operator who
is already in the document.
Measurement ??? — support contacts tagged to extraction accuracy,
if that tag exists. Open question until the next lesson.
Not claiming Net revenue retention. The chain from a caught error to
a renewal decision has at least three links this case
does not evidence.Read what the last field does for the first four. The Not claiming line is the reason a reader believes the Hypothesis line, because it demonstrates that you drew a boundary somewhere rather than claiming everything within reach. It is free to write and it is the most valuable sentence on the page.
Check your recall
Answer from memory — no scrolling back.
Retrieval check
Your case has a solid hypothesis, a defensible cost, and a mechanism that reads “the improved experience increases user trust in the product.” What is wrong with it, and what would you have to find out to fix it?
Check your answer
Nothing in that sentence happens. There is no actor, no action, and no moment at which the world is different, so there is nothing for the reader to check and nothing for you to measure later. It is a restatement of the hypothesis in warmer words.
Fixing it is not a writing exercise, which is the uncomfortable part. You have to go and find out what people actually do — which step they abandon, which ticket they file, which call they make to their account manager, which internal workaround exists because the product does not do the thing. Then the mechanism sentence writes itself, and it will be more specific and less impressive than the one you started with. That trade is the whole job.
If you cannot find out, that is a finding too. Write “mechanism unknown; here is what I would need to observe” and let the case stand on that. It is a far better position than a mechanism you made up, because it survives the follow-up question.
Hands on
Write all three pages
Done when: All three cases in VALUE-CASES.md have a Hypothesis with a direction and no magnitude, and a Cost with build, displacement and carry named separately — and at least one of the three says “no financial claim yet” in the Hypothesis field if that is the honest answer.
- Open
VALUE-CASES.md. The Route, Mechanism and Not claiming lines are already filled from the P&L lesson. Read them again before writing anything else — the hypothesis has to be the same belief the mechanism describes, and they drift apart surprisingly easily when written weeks apart. - Write each Hypothesis as one sentence: subject, verb, direction. If a number appears in it, delete the number. If deleting the number leaves nothing, the sentence was a magnitude with no belief behind it and needs rewriting from the mechanism.
- Write each Cost as three separate lines: build, displacement, carry. Guess where you have to and write “estimate” next to every guess. Leave displacement blank for now if you do not know what it displaces; that gap is the subject of a later lesson in this module.
- Do the advisory recommendation surface last and expect it to fight you. It recommends but does not execute, which makes almost every commercial claim about it weak. If the honest hypothesis is “no financial claim yet, and here is what I would have to observe to build one,” write exactly that. It is a pass.
- Bring all three into the chat. I will read the hypothesis and the mechanism as a pair and push on any case where the mechanism describes one thing and the hypothesis claims a different one — that mismatch is the most common defect in a first draft and the hardest to see in your own writing.
What this does not cover
The measurement field is still a question mark on all three pages, and that is deliberate: choosing a measurement is a different skill from writing a case, and it is governed by what your company can currently observe rather than by what would settle the argument. The next lesson is about picking a measurement you can actually get, and about the falsifier that turns a measurement from a promise into a commitment.
After that comes the displacement line you may have left blank, and finally the attribution trap — the lesson this whole course was built to reach.
Read this next — primary source
Commission Guidance on Management’s Discussion and Analysis of Financial Condition and Results of Operations (Release Nos. 33-10751; 34-88094)U.S. Securities and Exchange Commission, effective 25 February 2020 — free PDF, fetched 5 September 2026
Read the four pages on key performance indicators and metrics — they are short, and they are the closest thing in print to a specification for an honest metric claim, written by a body that can sue you for getting it wrong. The Commission does not tell companies which metrics to report. It tells them what has to travel alongside one: a definition, a calculation, a reason it is useful, a statement of how management actually uses it, and disclosure of any assumption underneath it. That list is not the case structure this lesson teaches, and the guidance is about disclosing a metric you already report rather than proposing new work. But the instinct is identical, and it is worth seeing the discipline stated by someone with no stake in design at all.
Stuck, curious, or think this lesson is wrong? Ask your teaching agent. The lessons are the scaffold; the conversation is where the learning gets unstuck.