Developing judgment in others

The support lead comes to you with a recommendation. "All three of my calls said the same thing. Enterprise leads lose decisions after recurring meetings. Three customers, about 20% of revenue. We should build summaries this cycle."

You can see the problem in four seconds. Three requests arrived through one account manager inside one month. That is one channel, not a sample. Nobody has tested whether the loss of decisions is real, or how often it happens, or whether summaries would fix it. Meeting type is not even captured in the event data.

So you say: "Three requests through one account manager is not prevalence. Go and test whether the underlying problem is real before we spend a cycle on it."

That is correct advice. It is also the more expensive of the two ways to fail.

The support lead now does exactly what you said. They come back with a better version of the same argument, and they wait for you to tell them whether it passes. Next month they bring the next thing, unexamined, because checking it is your job now. You have made yourself the quality function for reasoning you did not do, and every improvement in their output arrives through you. That is dependence, and it scales to exactly one person: you.

The other failure is quieter and more common. You are busy, the revenue number is large, the support lead is credible, and you say: "That's a strong signal — let's look at scoping it." Now the reasoning has gone unexamined and it is wearing your authority. When it turns out that three requests were three requests, nobody will be able to say which part of the argument was load-bearing, because nobody inspected it. That is abdication.

Both failures feel good at the time. Giving the answer feels like mentorship. Waving it through feels like trust. Neither one leaves the other person more able to catch the same gap alone next month.

Three paths lead out of the same moment. Two of them are the trap:

flowchart TD
  W["Someone brings reasoning<br/>with a weak joint"] --> G["Give the answer"]
  W --> A["Approve it"]
  W --> E["Expose the reasoning:<br/>a question they must answer,<br/>the call stays with them"]
  G --> GD["Dependence<br/>every improvement arrives<br/>through you, and scales to one"]
  A --> AD["Abdication<br/>unexamined reasoning now<br/>wearing your authority"]
  E --> ED["The next case arrives<br/>with the gap already found"]

Notice what the two trap paths have in common. In both, the weak joint is handled by you. Only the third path leaves it handled by the person who will meet it again next month.

Ask yourself: when someone on my team brings me a recommendation, do they arrive with the weak joint already named — or do they arrive expecting me to find it? Eight weeks of the second pattern is a diagnosis of your coaching, not of their ability.

Three moves are available when someone brings you reasoning that is not good enough yet.

Move	What you do	What the person learns	When it is the right move
Give the answer	Supply the conclusion and the reason	Your conclusion — not how to reach one	Decision is due today, or the consequence is one-way, or they cannot get the information in time
Approve	Accept without inspecting	That weak reasoning passes here	Only when the stated evidence standard is genuinely met
Expose the reasoning	Ask questions that make the gap visible to its author, then leave the decision with them	How to find the gap themselves	Default, when the class will recur

The third move is the one this lesson is about, and it is harder than it looks, because most attempts at it are the first move in disguise.

The counterfeit: the answer in question form

"Do you really think three requests through one account manager is a pattern?"

That is not a question. It is a verdict with a question mark attached. The person hears the verdict, agrees, and learns nothing except that you had already decided. Rhetorical questions are the most common counterfeit of coaching, because they let a leader feel patient while still supplying the answer.

The repair is not softer wording. It is a question whose answer the other person holds and you do not:

The difference is not politeness. The weak version transfers a verdict. The strong version transfers the method for producing one.

Three counterfeits recur, and they fail in different directions:

Counterfeit	What it sounds like	What the person learns
The verdict in question form	"Do you really think three requests is a pattern?"	That you had already decided, and the conversation is a formality
The unanswerable question	"What is the true prevalence of this problem?"	That this is a refusal wearing curiosity; it produces paralysis, not work
The interrogation	"What is the sample? The window? The channel? The population?"	To comply with a checklist. You get a better artifact and no better judgment

What makes a question calibrated

A calibrated question meets three conditions at once:

1. They can answer it from what they already hold, or can get within a day or two.
2. They have not looked at it yet. If they have, you are quizzing.
3. They cannot answer it without touching the weak joint. The gap becomes visible to them in the act of answering.

A working bank of questions

Use the exposure that is weakest to pick the question. These are wordings you can say as they are written.

When the evidence boundary is weak

• "Who did we not hear from, and how would we know if they disagreed?"
• "Walk me through how each of these three requests reached us. What was the path?"
• "If a fourth customer had said the opposite this month, where would that have shown up?"

When the mechanism is weak

• "Say we build it and it works perfectly. What is that person doing differently on a Tuesday?"
• "What has to be true about their meetings for this to help — and which of those things have we actually checked?"
• "If the real cause were something else entirely, what would it be?"

When alternatives are missing

• "What are we not doing if we do this?"
• "Give me the strongest version of the case for leaving the roadmap unchanged. I want it argued, not listed."

When uncertainty is unstated

• "How confident are you, as a number?"
• "What would move that number to ninety?"
• "What do you expect to see in six weeks if you are wrong?"

When ownership or reversal is missing

• "What would bring you back to reverse this?"
• "Who carries it if this is wrong, and have they heard the case?"

When the person is defensive, or you have no read yet

• "What is the part of this you are least sure about?"

That last one is the highest-yield question in the set. It invites the person to volunteer the weak joint rather than defend the strong one, and it works even when you have misjudged which exposure is missing.

Silence is part of the instrument

After you ask, stop talking. The useful thinking happens in the pause, and the pause will feel longer to you than to them. Leaders who cannot sit through it fill it with a hint, and the hint is the answer.

Here is the whole move at the length it actually runs — two questions, one silence, and a decision that never changes hands:

sequenceDiagram
  participant S as Support lead
  participant Y as You
  S->>Y: Three customers, about 20% of revenue.<br/>We should build summaries this cycle.
  Y->>S: Walk me through how each request reached us.
  Note over Y,S: silence — the pause is<br/>part of the instrument
  S->>Y: All three came through one account manager.
  Y->>S: What is the part of this you are least sure about?
  S->>Y: Names the untested claim, and keeps the call

Read the last two messages together. You did not supply the weak joint; they did. That is the only outcome that survives your absence next month.

Reviewing without taking the decision back

Three rules make review safe to receive.

1. Review the reasoning, not the conclusion. Say which exposure is missing. Do not say which answer you would pick. The moment you name your preferred answer, the rest of the conversation is about pleasing you.
2. Label your own view when you give it. If they ask what you think — and they should be able to — answer, and mark it: "This is my read, not an instruction. The call is still yours." Withholding your view entirely is not coaching either. It is a puzzle.
3. If you take the decision back, say so out loud. "I am taking this one. Here is why, and here is the question I would have wanted you to answer." A takeback that is named teaches something. A takeback disguised as a suggestion teaches that ownership here is decorative.

Exposing your own reasoning

Judgment transfers by watching reasoning happen, not by hearing conclusions announced. Most leaders show their team only finished positions, which are the least instructive part of their thinking.

The practical version is small: narrate one live decision a week, out loud, before you have resolved it. Include the option you are tempted by and why you distrust the temptation. Include a past call you got wrong, with what you knew at the time — not the moral, the evidence. PJ-05 gave you the discipline of judging a decision from what was knowable then; this is where you use it in front of other people.

The test

The measure of coaching is not whether the decision improved. It is whether the next decision arrives with the gap already found.

Over eight weeks, watch for one thing: does this person bring you recommendations with the weakest exposure already named and addressed? If yes, judgment is transferring. If they bring you polished arguments and wait, you are still the quality function.

Boundary

Coaching is the wrong instrument in three situations, and using it anyway is its own failure.

When time has run out. If the decision is due today and they cannot close the gap by then, ask the question, give the answer, and label both: "I am deciding this one. The question I would have asked you is X — hold onto it." Coaching someone into a deadline they will miss is not development.

When the gap is information you hold and they do not. Questions asked across an information gap are not calibrated; they are a guessing game with a power difference. Give the information, then ask.

When the problem is will, not capability. Coaching develops judgment. It does nothing about a person who is not doing the work, and running a fourth coaching conversation with someone who has not acted on the first three is a way of avoiding a harder conversation. That conversation is LD-06.

One more honest limit: this is slower per decision. A coached decision costs more than a given one. The return comes only if the decision class recurs. If this exact question will never come back, giving the answer is the correct use of everyone's time, and dressing it up as development wastes it.

Read the Noted case if you have not already.

Take Signal 2: three enterprise customers asked for automated meeting summaries, and the person bringing it is the support lead — who, the case tells you, recorded several of those customer conversations and is closest to what people actually said before it became a summary.

Their recommendation to you, in their words: "All three of my calls said the same thing. They lose decisions after recurring meetings. Three customers, about 20% of revenue. We should build summaries this cycle."

What the case gives you and they may not have weighed: all three requests came through the same account manager within one month; the underlying claim has never been tested; meeting type is not captured in the product's event data. What the case also gives you: this person holds something you do not — the raw conversations, before summarization.

Your task. Build the coaching move, not the answer.

1. Name the single weakest exposure in their reasoning. Use one of the six — choice, evidence, uncertainty, alternatives, owner, revision trigger. Not "the whole thing is thin."
2. Write three questions, verbatim, in the order you would ask them. Each must be answerable by this person from what they hold or can get in two days. At least one must use the fact that they, not you, have the raw calls.
3. For each question, write what a good answer sounds like, what a defensive answer sounds like, and what you do next in each case. "Ask a follow-up" is not a next move; write the follow-up.
4. Write down the question you were tempted to ask that is actually your verdict in question form. Keep it in the artifact and mark it rejected. Naming your own counterfeit is the point of this step.
5. Decide the authority level you are operating at, using the LD-02 table, and say whether the call stays with the support lead after the questions. Then check: does anything in your plan quietly move it to you?
6. Write the exact sentence you would say if you do take the decision back. Plain words. If you cannot write it without softening it, you are planning a disguised takeback.
7. Write your own exposure move: one live decision of yours you will narrate to this person this week, including the part you are unsure about and one past call you got wrong with what you knew at the time.
8. Commit to the dependence test. Write what you expect to see in eight weeks that would tell you judgment transferred, and what you would see if it did not.

Expect to be pushed on: whether your three questions are genuinely open or whether at least one is a verdict wearing a question mark; whether any question is unanswerable given that meeting type is not captured, which would make it paralysis rather than coaching; and whether step 5 survives inspection — most first drafts return the decision to the leader through a phrase like "then bring it back to me and we will decide together."

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

• One weak exposure named from the six, not "the whole thing is thin".
• Three verbatim questions, each answerable by the support lead from what they hold or can get in two days, and at least one built on the raw calls they have and you do not.
• For each question, what a good answer sounds like, what a defensive answer sounds like, and the actual next sentence you would say in each case.
• The rejected counterfeit present and labelled as yours.
• An authority level from the LD-02 table, the call left where you said it stays, a plain takeback sentence, and a dependence test with a date on it.
• The most common weak move is a verdict with a question mark — "do you really think three requests is a pattern?" It is weak because your conclusion is already in the sentence, so they learn what you decided and not how to notice it next month.

Produce artifacts/LD-03-judgment-coaching-plan.md using the template below.

Write it for yourself, to be used in a specific conversation this week. It is not a development framework and it is not for a performance file. If it contains no verbatim question wording, it will not survive contact with a real conversation, because the moment of pressure is exactly the moment you will revert to giving the answer.

This attaches to the decision-rights map from LD-02. The evidence standard you wrote there is what these questions are asked against — it is what makes a question a request for reasoning rather than an expression of your taste.

# Judgment coaching plan · <person or role> · <decision>

- **Date:** <YYYY-MM-DD>
- **The decision they hold:** <stated as a choice, with its alternatives>
- **Their current position:** <in their words, not improved>
- **Authority level (LD-02):** <1-5> — <does the call stay with them after this
  conversation? yes / no>
- **Will this decision class recur?** <yes / no — if no, coaching may be the
  wrong use of the time>

## The weak joint

- **Weakest exposure:** <choice | evidence | uncertainty | alternatives | owner
  | revision trigger>
- **Why it is weak, in one sentence:** <specific, not "it is thin">
- **What they hold that I do not:** <the information advantage they have — the
  questions should use it>

## Questions, in order

| # | Verbatim question | Answerable from what they hold? | Which exposure it touches |
|---|---|---|---|
| 1 | "<exact words>" | <yes / by when> | <exposure> |
| 2 | "<exact words>" | <yes / by when> | <exposure> |
| 3 | "<exact words>" | <yes / by when> | <exposure> |

## What each answer means

| # | A good answer sounds like | A defensive answer sounds like | My next move in each case |
|---|---|---|---|
| 1 | <…> | <…> | <the follow-up, written out — not "probe further"> |
| 2 | <…> | <…> | <…> |
| 3 | <…> | <…> | <…> |

## Rejected question

<The question you were tempted to ask that is your verdict with a question mark
attached. Write it here and leave it marked rejected. Naming your own
counterfeit is the point.>

- **Tempting question:** "<exact words>" — **rejected because** <…>

## If the reasoning still does not meet the standard

- **Does the call stay with them?** <yes / no>
- **If no, the exact sentence I will say:** "<plain words naming the takeback and
  the reason. If you cannot write it without softening it, you are planning a
  disguised takeback.>"
- **What I will not do:** <the supportive-sounding move that would quietly
  relocate the decision — "let's decide it together", "send it over and I'll
  finish the analysis", "run the final version past me">

## My own exposure move

- **Live decision I will narrate this week:** <a decision of mine, before it is
  resolved>
- **The part I am unsure about:** <stated as I will say it>
- **A past call I got wrong:** <what I knew at the time, not the moral>

## Dependence test

- **Review date:** <YYYY-MM-DD, roughly eight weeks out>
- **Judgment transferred if I see:** <the person arrives with the weakest
  exposure in their own case already named>
- **Judgment did not transfer if I see:** <polished arguments brought to me for a
  verdict; or fewer escalations with no sign the reasoning changed>

## Gaps

<Everything marked `<unknown>` above, listed with who could close it.>