Calibrated product judgment

A PM finishes an opportunity case. The room asks the only question that matters: so what do you think we should do?

The PM says the evidence is not conclusive yet, and proposes more discovery. This sounds careful. Everyone nods. Six weeks later the account manager escalates, the founder makes the call in a hallway, and the roadmap changes without anyone writing down why.

The PM did not avoid a decision. The PM transferred it to whoever was willing to speak with confidence. Refusing to take a position does not preserve rigor; it hands the position to the least calibrated person in the room.

The opposite failure is just as common and easier to admire. Another PM says "enterprise clearly needs this" and gets funded. The claim was built from three requests through one account manager, plus a memory of a previous company where something like it worked. Neither input was wrong. Both were presented as stronger than they were, and nobody could tell which part of the position would break first.

Both PMs failed in the same way. A position was stated without the inputs it rests on, and a confidence was implied that nobody could check. The first implied confidence near zero, which was false. The second implied confidence near certainty, which was also false.

Both failures run through the same gap, and both end in the same place — a roadmap that moved without a recorded reason:

flowchart LR
  C["Opportunity case<br/>evidence is thin"] --> P{"Position stated<br/>with its inputs?"}
  P -->|"no — 'we need<br/>more discovery'"| V["Position vacuum"]
  V --> R["Roadmap changes,<br/>reasoning unrecorded"]
  P -->|"no — 'enterprise<br/>clearly needs this'"| F["False certainty<br/>nobody can check it"]
  F --> R
  P -->|"yes — choice, inputs,<br/>confidence, update condition"| K["A position someone<br/>else can act on"]

The middle node is the one to watch. A vacuum does not stay empty; it is filled by whoever speaks next with confidence, and that person is usually not the one holding the evidence.

Ask a PM who has just given a recommendation: what is your confidence, and what specific observation would move it? If the answer is "we would need more data," they have not taken a position. If the answer is "nothing would change my mind," they are not reasoning.

Judgment is not intuition, and it is not the average of the evidence. It is a stated position with visible inputs and a checkable confidence.

Four inputs feed a product position. Each one is legitimate. Each one lies when it is used alone.

Input	What it contributes	How it misleads when used alone
Evidence	What was observed, inside a stated population, window, source, and limit	Says nothing outside its boundary, but reads as if it does
Accumulated exposure	Fast hypotheses from products and users you have seen before	Transfers a pattern from a context that no longer holds
Strategic context	Which outcome the company is currently trying to buy	Quietly overrides evidence and becomes a wish
Accountability	Who carries the consequence, and therefore who must choose	Authority gets mistaken for correctness

PF-03 gave you the discipline for the first row: a claim carries product weight only when its population, window, source, and limit are visible. PJ-01 asks you to do the same thing for the other three. Where did the pattern come from? What is the strategy actually paying for? Who eats the cost if this is wrong?

The four inputs combine into one position. What keeps that position from hardening into false certainty is that the confidence is stated separately from the choice, and carries the observation that would move it:

flowchart TD
  E["Evidence<br/>population · window · source · limit"] --> POS["Position<br/>one live alternative"]
  X["Accumulated exposure<br/>patterns from products before"] --> POS
  S["Strategic context<br/>what the company is buying"] --> POS
  A["Accountability<br/>who carries the cost"] --> POS
  POS -->|"stated separately,<br/>never merged"| CF["Confidence<br/>one word you can defend"]
  CF -->|"names what moves it"| U["Update condition<br/>one observation, one direction"]
  POS -.->|"inputs hidden"| FC["False certainty"]

The dashed edge is the failure. It is not a different diagram — it is the same position with the four source nodes deleted, which is what "enterprise clearly needs this" is.

What a calibrated position contains

A position is calibrated when a reader can check it without asking you questions:

1. The choice, stated as one of the live alternatives from your decision statement — not a topic and not a direction.
2. The inputs, separated by type, so a reader can see which rows are doing the work. A position resting mostly on accumulated exposure is not disqualified; it is disclosed.
3. A confidence, stated in words that carry meaning. "Moderate" means you would be unsurprised to be wrong. "High" means you would owe someone an explanation.
4. The strongest counter-case, written well enough that the person who believes it would recognize their own argument.
5. The update condition — the specific observation that would move your confidence, in which direction, and by roughly how much.

Here is the same position on Noted's enterprise summaries signal, written both ways. Both take a side. Only one of them can be checked:

The difference is not caution. Both are positions. The second one names which input is load-bearing, so a reader can attack the right part — and it names the observation that would move it, so the team knows where to look.

Confidence and certainty are different things

Preserving uncertainty does not mean lowering your position. It means keeping the position and the confidence separate.

Statement	Position	Confidence	Usable?
"We need more data before deciding."	none	none	No — the decision still gets made, elsewhere
"Enterprise clearly needs this."	build	implied certainty	No — no reader can check it
"Build it. Moderate confidence. Three requests through one channel is not prevalence, so I would drop to low if a second channel showed nothing."	build	moderate, with a named test	Yes

The third statement can be wrong. That is what makes it useful. It tells the team where to look to find out.

Boundary

Calibration has two real limits.

Low confidence does not license inaction. Some decisions expire. When an option set is shrinking, "wait for more evidence" is itself a position with a cost, and it must be stated and priced like any other. A PM who calls waiting "not deciding" is hiding a choice inside a process.

You cannot calibrate against a scale you never check. Stated confidence drifts unless it is compared to what actually happened, over a run of decisions. A single decision cannot tell you whether your "high confidence" means anything. Until you have that record, treat your own confidence labels as claims you owe evidence for, not as measurements. PJ-05 starts building that record, and EV-07 uses it to update belief.

Calibration also does not substitute for expertise. Stating moderate confidence on a question that belongs to your engineering lead does not make your answer more usable than theirs. It makes it better labelled.

Read the Noted case if you have not already.

Take Signal 2: three enterprise customers asked for automated meeting summaries.

Bring your PF-06 opportunity case. Whatever it concluded, assume the head of product now wants your position, not your case: one sentence they can act on, and a confidence they can check.

Your task. Take a position on the enterprise summaries opportunity, and make it checkable.

1. State the choice you are taking, as one of the live alternatives. "Build the summaries capability this quarter," "make no roadmap change yet," and "run a bounded test of the underlying claim first" are three different positions with three different costs.
2. Separate your inputs into the four rows. Write what each one actually contributes here, and mark which row your position mostly rests on.
3. For the evidence row, apply PF-03: state the population, window, source, and limit of every claim you use. The 20% revenue figure is a contract value, not a churn risk. Say which one you are treating it as, and why.
4. For the audience, apply PF-05: enterprise team leads are a behavior-and- context boundary, not a contract size. Say which behavior you mean.
5. State your confidence in one word, and one line saying why that word and not the one above or below it.
6. Write the strongest counter-case. Write it so that the founder, who made the earlier sequencing decision, would recognize it as fair.
7. Name the update condition: one observation, its direction, and what it would do to your confidence.

Expect to be pushed on: whether "run a bounded test first" is a real position or a way to avoid taking one; whether you are treating three requests through one account manager as three independent data points; and whether the 20% revenue figure entered your reasoning as evidence of value or as pressure.

Committed to a position? Check it against what a strong answer holds

• The choice is one of the live alternatives, stated so that it could turn out wrong. "Run a bounded test first" counts only if the answer says what the test would settle, by when, and what the roadmap does while it runs. Without those three, it is a deferral wearing a position's clothes.
• The four inputs are separated and one is marked as load-bearing. If that row is accumulated exposure, the answer says which product or pattern the exposure came from, so a reader can judge whether the context still holds.
• The evidence row carries PF-03's fields: three accounts, one account manager, one month, an untested underlying claim, meeting type not captured. The 20% is treated as contract value unless the answer can say what makes it exposed.
• The confidence is one word, with a line on why not the word above or below. "High" on this evidence has to survive the observation that one channel is not three independent points.
• The counter-case contains the strongest fact for the other side — written so the founder, who made the earlier sequencing call, would recognise it as fair — and the update condition is one observation with a direction and a rough size of the move it would cause.
• The most common weak move is "we need more discovery" with no position attached. It is weak because the decision still gets made, in a hallway, by whoever speaks next with confidence — and that person is rarely the one holding the evidence.

Produce artifacts/PJ-01-calibrated-product-view.md using the template below.

Write it for the person who has to act on your position without being able to ask you a follow-up question. That is the real test of calibration: the document carries the confidence, not your tone of voice.

This extends your Product Decision Case. PF-01 gave you the decision and its alternatives. This artifact commits to one of them, in public, with the uncertainty attached rather than removed.

# Calibrated product view · <short name of the position>

- **Date:** <YYYY-MM-DD>
- **Product:** <product>
- **Decision this position answers:** <the decision statement it takes a side of>
- **Owner of the consequence:** <person or role>
- **Specialist authority:** <who holds expertise this position depends on>

## Position

<One sentence. A choice from the live alternatives, not a topic and not a
direction. "Build X this quarter" is a position. "Explore X" is not.>

**Alternatives displaced by this position:**

1. <option not being taken>
2. <option not being taken — including "no change yet" where that is real>

## Inputs

Mark the row this position mostly rests on.

| Input | What it contributes here | Load-bearing? |
|---|---|---|
| **Evidence** | <what was observed> | <yes / partly / no> |
| **Accumulated exposure** | <the pattern, and where you saw it before> | <yes / partly / no> |
| **Strategic context** | <what the company is currently trying to buy> | <yes / partly / no> |
| **Accountability** | <who carries the consequence, and what that obliges> | <yes / partly / no> |

## Evidence boundaries

| Claim | Source | Population | Window | What it cannot establish |
|---|---|---|---|---|
| <claim> | <source> | <who, as behaviour and context> | <when> | <limit> |

<If a figure entered your reasoning as pressure rather than as evidence, say so
here. That is a disclosure, not a confession.>

## Confidence

- **Confidence:** <low | moderate | high>
- **Why this word and not the one above it:** <one line>
- **Why this word and not the one below it:** <one line>

## Strongest counter-case

<Write the best argument against your own position, in the words of the person
who holds it. If they would not recognise it as their argument, you have written
a strawman and the position is not yet tested.>

**Why you still hold your position:** <one or two sentences>

## Update condition

| Observation | Direction | Effect on confidence |
|---|---|---|
| <a specific, observable thing> | <if it happens / if it does not> | <moves to low / moderate / high> |

<"More data" is not an observation. Name what would be seen, where, and by when.>

## If this position is wrong

- **Cost:** <what it costs, to whom>
- **Reversibility:** <reversible | costly to reverse | one-way>
- **Earliest point the error becomes visible:** <date or event>

## Gaps

<Everything marked `<unknown>` above, listed as open questions with who could
close them and roughly what it would cost to close each one.>