10. Sequencing — Choosing the Pattern When Nobody Tells You Which Week It Is
Take the headers off the cards. That is the whole change this week, and it is enough to drop your performance by half.
For nine weeks you have practised inside a frame that told you the answer before you began. The page said meta-model challenge, and then gave you eleven sentences to challenge; your only question was how fast and how cleanly. The page said reframe in sixty seconds, and the only uncertainty was which of six frames. Every rep you have banked so far has been banked with a label attached to it, and the label was doing work you did not notice it doing. It was doing retrieval. You were never selecting a pattern. You were executing a named one.
Live conversation never names it. Someone says something, and in the four hundred milliseconds before your reply is due, you have to answer a question the workbook has never asked you: which of the nine things I know how to do is the thing this moment wants. That question has its own latency, its own error rate, and — this is the part that catches people who have drilled well — its own failure modes, which are not the failure modes of any individual pattern. You can have a two-hundred-millisecond challenge and a four-second selection, and the four-second selection is what the room experiences. Speed in the wrong pattern is not speed. It is confident wrongness delivered promptly.
So say it plainly: pattern selection is a separate capability, it is not a by-product of pattern execution, and no amount of additional single-pattern drilling will produce it. You could run another thousand meta-model reps and your selection accuracy would not move, because selection is not a harder version of execution — it is a different task with a different input. Execution takes a labelled stimulus and produces language. Selection takes an unlabelled stimulus and produces a label. Drilling the first has never taught anyone the second, and if you have felt competent in practice and clumsy in life, this gap is where the discrepancy lives. It is not nerve. It is a skill you have simply never trained.
The ladder, and the order it actually runs
There are four questions, and they run in this order every time:
Is there an outcome? Is there rapport? Is the state usable? Is the language challengeable?
The order is not arbitrary and it is not preference. Each question determines whether the next one means anything.
Start with outcome, because without a stated destination you have no criterion for whether any intervention helped. If the person cannot say what they want — not what they want to stop, not what they want less of, but what they want — then every subsequent move you make is unscoreable. You can reframe brilliantly and have no idea whether the reframe served them, because served them toward what has no answer. A practitioner working without an outcome is not steering; they are producing effects and hoping the client likes them. This is the most common reason a session feels good and changes nothing.
Rapport second, because an intervention is not a piece of language, it is a piece of language from someone. The same challenge — "everyone?" — is either an invitation or an attack depending entirely on the state of the relationship carrying it, and you learned in Week Seven that the relationship is not a fixed background condition but something that breaks and repairs several times inside twenty minutes. Checking rapport is checking whether the channel is currently able to carry load.
State third, because a nervous system past a certain threshold of arousal cannot do the cognitive work the middle patterns require. Ask someone in a heavily loaded state to interrogate their own outcome and they will produce compliance — words that satisfy your question and commit nothing. Reframes offered into a loaded state are received as dismissal, every time, no matter how good the reframe is.
Language fourth and last, because the meta-model is the finest-grain tool you own and it is the one everyone reaches for first. It works on surface structure, and surface structure is the least load-bearing thing in the room. When you lead with it you are sanding a joint that has not been glued.
Now the part that reorganises the ladder, and it is the reason people who know the four questions still waste sessions: you read the ladder top-down, and you repair it bottom-up. Reading down tells you which rungs are broken. But you cannot build an outcome across broken rapport, and you cannot restore rapport with someone whose state has gone past usable. So having read down and found, say, no outcome and broken rapport, you do not start at outcome because outcome came first in the reading. You start at the deepest break. Read for diagnosis, act at the foundation. Practitioners who run the ladder as a to-do list in reading order spend twenty minutes constructing an outcome with someone who stopped being with them in minute four, and they come out of it with a beautifully specified goal that belongs to nobody.
The ladder has an edge past which it does harm, and you should know where it is before you rely on it. Run as a live checklist — four questions consciously asked, in sequence, while the person is talking — it consumes the exact attention that Week Two spent four weeks buying you. You will be checking rung three while they say the sentence that mattered. The ladder is a structure for review and for the first twenty seconds of a conversation; past that it must be running under the floor or not at all. If you can feel yourself running it, you are not listening.
Why this week will feel like going backwards
Skill retrieval is bound to the conditions of its encoding. This is not a metaphor. Godden and Baddeley put divers through a word list learned either on land or underwater in 1975, and tested recall in both environments; what was learned underwater came back underwater. Recall degraded when the retrieval context did not match the learning context, for material that had nothing to do with water. Context binds itself to memory whether or not the context is relevant.
The label is a context. A pattern drilled exclusively under its own heading gets encoded with that heading as part of the retrieval cue, and it stays reliably available exactly as long as the heading is present. This is why your Week Four performance was strong and your first live meta-model challenge, weeks later, arrived three seconds late or not at all. Nothing was wrong with the pattern. The cue that reliably fired it was a word printed at the top of a page, and life does not print it.
The fix is interleaving, and the fix is unpleasant. Shea and Morgan showed the shape of it in 1979: learners practising movement sequences in blocked order — all of one, then all of the next — outperformed randomly-ordered learners during practice, and underperformed them on retention and transfer. Practising in a jumble makes you worse now and better later. Kornell and Bjork found the same inversion in 2008 with people learning to recognise painters' styles: interleaved exposure produced better ability to identify new paintings than blocked exposure did, and most participants reported afterwards that blocking had helped them more. They were wrong about their own learning in the direction of comfort. The fluency of blocked practice feels like competence and is mostly the absence of a retrieval demand.
Be honest about the limits of this. The contextual interference effect is robust in laboratory motor tasks and less settled for complex real-world skills, where several studies find it weaker, and where there is a live argument that novices below a threshold of basic proficiency are simply overloaded by randomisation rather than helped by it. That argument is the reason this week is Week Ten and not Week Three. You have the individual patterns to a workable floor. Now they get scrambled.
Expect day three to be the worst day. Expect to want, badly, to go back to a page with headings on it. The wanting is the mechanism working.
Twenty minutes with the labels off
What follows is one session, condensed to its decision points. Details are changed; the sequence of choices is not. The client — call him Tomas — is a deputy head at a secondary school, forty-one, three months into a department that lost its head of faculty in the spring. My annotation follows each turn: what I read, what I chose, and what the ladder said afterwards.
T: It's just been chaos since Ines went. Everyone thinks I'm difficult.
Fork one. I challenged the universal quantifier — "everyone?" — at roughly the eight-second mark. It is the best-drilled pattern I own, the sentence is textbook, and my mouth got there before my ladder did. He said, "Well. Not everyone. Obviously." Flat. Then he explained, for four minutes, in detail, why he was not in fact difficult. Wrong fork, and instructive: the challenge was technically correct and structurally premature. Rung one was empty — he had not said what he wanted — and rung two was untested, we were eight seconds into a first session. A meta-model challenge across untested rapport with no outcome present does not open the generalisation. It recruits the person to defend it. I bought four minutes of defence with one clean sentence.
T: …so it's not that. It's that nobody's actually running the department and it's landed on me.
Me: Landed on you.
T: Yeah. Landed. Like — I didn't apply for this.
Backing all the way down to verbatim return. Rung two work, not rung four. Giving a word back exactly as spoken, with no challenge attached, is the cheapest rapport instrument in the kit and the one people abandon first because it feels like doing nothing. His "yeah — landed" came back warmer than anything in the previous four minutes.
T: I want it to stop feeling like this.
Me: What would you want instead?
T: I want the department to be functional.
Rung one, opened. And then immediately: a nominalisation with a deadline attached, exactly as Week Nine warned. "Functional" is not a destination, it is a mood in a noun costume. Correct next move: interrogate it. I asked what he would see on a Tuesday morning in a functional department. He said, "I'd know what Priya was doing without having to ask her."
That sentence is worth the whole first ten minutes. It is sensory, it is specific, it names a person, and it is testable by Wednesday.
T: But she won't tell me. She just — she's decided I'm Ines's replacement and she's not having it.
Fork two. I offered a reframe: that her resistance might be loyalty to Ines rather than a verdict on him. He said, "That's one way to look at it." Politely. Wrong fork. That flat five-word receipt is the sound of a reframe declined, and Week Eight taught you to hear it. The error was rung three: I had not checked state. His shoulders had come up two sentences earlier, his breathing had shortened, and I was offering an act of cognitive flexibility to a nervous system that had just narrowed. Reframes require slack. He had none. What I should have run — what I ran two minutes later, having wasted the interval — was a state move: slowing my own speech below his, dropping the pitch at the end of sentences instead of lifting it, and asking a question with no work in it. His shoulders came down over about forty seconds. Then the reframe would have landed. I never re-offered it, which was a second, smaller error.
T: I keep going in there ready for a fight. I can feel it in the corridor before I've even opened the door.
Fork three, and the one I am least comfortable writing down. I set an anchor. The material was genuinely anchorable — a repeatable location, a reliable state, a clear before-and-after boundary at the door. It was also, by any honest reading of the ladder, not what the moment wanted. He had just handed me a self-report about a corridor, which is a description of an outcome not yet specified: he had told me what he did not want to feel and had not been asked what he wanted to be able to do when he opened the door. Rung one was open but shallow. I went to rung three because anchoring is my favourite pattern. It is kinaesthetic, it produces a visible effect inside four minutes, and running it makes me feel like a practitioner. The anchor held, incidentally. He used it. That is what makes this the dangerous kind of error — it worked, and working is not the same as being correct, and a session full of effective wrong choices leaves a person feeling helped and pointed nowhere.
The recovery drill, which matters more than the selection drill
You will select wrong. Seventy percent is the target and thirty percent is therefore the plan. What separates a practitioner from someone with a good hit rate is not the hit rate; it is what the miss costs.
A pattern abandoned badly is experienced by the other person in one of two ways: as a topic change they did not authorise, or as a test they failed. Both are rapport events, and both cost more than the wrong pattern would have. The dropped-pattern feeling has a specific signature — you start something, sense it going wrong, and go quiet mid-structure. From inside, that is judgement. From outside, it is withdrawal.
Four moves, in this order, and the whole thing takes under six seconds.
Own it out loud, and locate the fault in yourself. "I've asked that badly — let me come at it from somewhere else." Not "let's move on," which is a decision you made about them. The attribution is the entire mechanism: a practitioner who visibly owns a misstep is a practitioner it is safe to be wrong in front of, and you are about to ask them to be wrong in front of you for an hour.
Finish the sentence you started. Do not leave a hanging structure. An incomplete question sits in the room and the other person keeps trying to answer it. Close it, even badly.
Hand the material back verbatim. Return the last thing they said in their own words and restart from there. This does two things at once: it proves you were listening through the fumble, and it gives the new pattern a stimulus that is demonstrably theirs, not one you constructed while recovering.
Do not explain the pattern. Never say "I was going to run a reframe there but the state wasn't right." You have just made the session about your technique, and a client who learns to watch for your moves stops producing usable material. The recovery is invisible or it is not a recovery.
Drill this deliberately. Twenty times, in low-stakes conversation, deliberately start a pattern you know is wrong and recover from it. That is the rep. You are not practising being right; you are practising the four seconds after being wrong, which is the only part of the miss you control.
Your turn: forty unmarked stimuli
Build the deck before you build the discipline. Forty index cards, or forty lines in a file. Each carries one utterance and nothing else — no heading, no rung, no difficulty marker. Draw them from your Week Two transcripts, which is why you kept them. Mix in deliberate near-misses: sentences that look like meta-model material but are actually rapport events, sentences that look like outcome work but are state problems wearing goal language.
Shuffle. Randomise difficulty deliberately — do not sort easy-to-hard, because sorting reintroduces the structure you are trying to strip.
For each card, in order, and with a clock: read it once. Within three seconds, write down two things — the rung it fails on, and the pattern you would run. Then, and only then, write the first line you would actually say. Three columns on the page: rung, pattern, opening line. Leave the fourth column blank; you will fill it during scoring.
You are scored on column two, not column three. This is deliberate and it will feel wrong, because column three is where your craft is visible. Execution quality is not being measured this week. A perfect challenge on a card that wanted a state move scores zero. A clumsy, hesitant, badly-phrased state move on that same card scores full marks. You are training the retrieval, not the delivery; you have nine weeks of delivery banked and one week of retrieval.
Scoring needs an honest key, and an honest key is not a single right answer. Most live utterances admit two defensible patterns and a few admit three. So write the key before you run the deck — for each card, list every pattern you would accept and the rung each addresses — and mark yourself correct if your choice is on that list and your named rung matches the rung you assigned it in the key. Both halves. Choosing the right pattern for the wrong reason is a coincidence, not a skill, and it will not repeat.
The fourth column is for misses. On every card you got wrong, write one sentence: the recovery line you would have used. Every miss carries a recovery or the miss is not finished.
The success condition for this week is seventy percent selection accuracy across forty unmarked stimuli, and a written clean recovery on every single wrong selection. Not ninety. Live selection is genuinely underdetermined — the material often supports more than one good road, and a scale that demands ninety is a scale you will meet by writing a permissive key. If you find yourself above eighty-five, suspect your deck before you celebrate. Either the cards are too clean or the key has grown generous. Rebuild the deck from harder transcript.
The failure mode, and what your log will actually show you
Run the week. Then, on day three, before you have finished, stop and count.
Go through your log and tally the distribution: how many times each of the nine patterns was your column-two answer. You need thirty interventions minimum for the count to mean anything. If any single pattern accounts for more than about a third of your selections, you have a hammer, and everything downstream of it is compromised.
Here is what the tally is really for, and it is not the hammer. Look at the pattern sitting at the top of your distribution and ask a question that has nothing to do with technique: do you enjoy running it?
You will find that you do. That is the finding. The pattern over-represented in your log is not the pattern the situations called for — it is the pattern you like being inside. Anchoring, for me, because it is physical and it works fast and running it makes me feel like a practitioner rather than a person having a conversation. For you it will be something else; reframing, usually, because reframing is the one that feels clever, or the meta-model, because it is the one that feels rigorous. The specific pattern does not matter. What matters is that preference has been arriving dressed as diagnosis, and you have been experiencing it, from the inside, as fluency. It is fast. It is smooth. It fires without deliberation. Every subjective marker you have for competence lights up — and those markers are measuring how often you have done the thing, which is exactly what preference produces.
This is the most durable incompetence in the field, and it survives decades of practice specifically because practice strengthens it. The pattern you over-select gets more reps than any other, which makes it faster, which makes it more available, which makes it get selected more. Skill and preference feed each other, and the feedback loop is invisible from inside because at no point does anything feel like a mistake. You are not choosing badly. You are not choosing at all — you are retrieving your most-practised move and experiencing the retrieval as a judgement. The client is helped, often. The session goes fine, mostly. And a decade goes by in which you have run one pattern at everything and called it a practice.
The tally catches it in three days because it does not ask you how you decided. It asks what you did, and counts.
The counterpart finding is quieter and worth as much: look at the bottom of the distribution. Two patterns will be at or near zero. You will have an explanation ready — that the material did not call for them, that they suit a different kind of client. Do not accept it. Your deck was randomised. The material called for them at roughly the rate it called for anything, and you did not see the call, because you cannot select a pattern you are not carrying. Absence in the log is not evidence about the world. It is evidence about you.
So this is the week, whole. Build the forty cards from your own transcripts and write the key before you draw a single one. Run the deck cold, three seconds a card, rung and pattern first and language second, and let the language be bad — it is not what is being measured. Score selection only, mark a card correct only when the pattern and the rung agree with the key, and write a recovery line into the fourth column for every miss, because thirty percent of forty is twelve recoveries and twelve recoveries is a drill in itself. Then, on day three and again at the end, count the distribution and look for your own hand in it.
And when you find the pattern you have been over-running — you will — do not resolve to use it less. Resolutions do not change latency. Take the two patterns sitting at the bottom of your log, build a new deck of twenty cards that can only be answered by those two, and drill them blocked for three days until they are as fast in your mouth as your favourite. Then put them back in the shuffle and run the forty again. The point is not to stop reaching for what you love. It is to make the reach for everything else cost the same, so that when the choice comes it is a choice.
Practice — Chapter 10
The label was doing more work than you knew. For nine weeks the material arrived pre-sorted — this is a Widen drill, this one is Repair — and what you practised was execution. Execution is the easy half. In a live conversation nobody hands you the header, the material arrives mid-sentence, and the first thing you have to produce is not a move but a reading.
So the rep changes shape this week. You are given unmarked material. You diagnose, you run, and when the diagnosis is wrong — it will be — you recover without making the recovery the new subject of the conversation.
The nine, with their tells. You know these. What you may not have held in one place is the trigger for each, which is the only part that matters when the label is gone.
- Reflect — return the sentence, slightly compressed. Tell: the speaker has not yet heard themselves say it.
- Separate — split what happened from what it meant. Tell: an interpretation is being carried as though it were an observation.
- Widen — move from the incident to the pattern. Tell: it is the third time, and they are still describing it as the first.
- Narrow — move from the abstraction to the next physical action. Tell: agreement with no verb in it.
- Steelman — state the absent or opposing position better than its holder did. Tell: someone is winning an argument against a cartoon.
- Exception — ask about the time it went differently. Tell: totalising language. Always. Never. Everyone.
- Stake — name what this person stands to lose. Tell: heat out of proportion to the object. A twenty-minute argument about a font.
- Hold — say nothing. Tell: they are one sentence away from the real sentence.
- Repair — name your own miss and re-enter. Tell: the room just went flat, or tight, and you are the reason.
The worked example
Here is the material. Read it before you read anything below it.
Theo: I'm going to stop putting things in the Thursday review. Honestly. I put the migration plan in three weeks ago, nobody read it, then Priya raised the same thing in her own doc and suddenly it's a priority. It's fine. I just — there's no point. Everyone in that meeting is performing for Marcus anyway.
Step one: inventory the tells, before choosing anything.
Do not go looking for the right pattern. Go looking for tells, and let the count tell you how contested the reading is. In four sentences:
- "three weeks ago" and "the same thing" — a comparison across instances. Possible Widen.
- "Everyone in that meeting is performing" — totalising. Possible Exception.
- "performing for Marcus anyway" — an interpretation of motive, delivered as fact. Possible Separate.
- "I'm going to stop putting things in the Thursday review" — a decision with no date and no first action. Possible Narrow.
- "It's fine. I just —" — the sentence breaks. He didn't finish it. Possible Hold.
- The heat is high for a document-routing complaint. Possible Stake.
Six candidates in four sentences. That is normal. Unmarked material is always over-determined; the fantasy that live conversation offers one obvious pattern is exactly what the labels in Chapters 1 through 9 were quietly teaching you.
Step two: stop trying to be right.
Here is the move that carries the whole chapter. You cannot reliably pick the correct pattern from one paragraph, and the effort to do so is the main thing slowing practitioners down at this stage. So change what you are optimising.
Rank the candidates not by how likely they are to be right but by how cheap they are to be wrong.
A wrong Reflect costs you four seconds and produces more material. A wrong Hold costs you a beat of silence, and silence is rarely a debt. A wrong Widen tells a person who is describing a specific injury that you have already filed it as a category, and it takes real work to come back from that. A wrong Stake — naming what you think someone is afraid of losing — can be read as an accusation, and if you have misread it, you have now put a false motive in the room under your name.
The patterns are not equally reversible. Order them that way:
Cheap to be wrong: Reflect, Hold, Exception.
Moderate: Separate, Narrow, Steelman.
Expensive: Widen, Stake.
Then run the cheapest pattern that is plausibly right, and read the response as your second diagnostic. You are not diagnosing. You are instrumenting — spending a low-cost move to buy information you could not have got by thinking harder in your own head.
Step three: run it.
The cheapest plausible candidate here is Hold. His sentence broke at "I just —". Something was behind it.
You: (nothing. Four seconds.)
Theo: I just don't want to be the person who has to be loud to be heard.
That single line is worth more than any inference you would have made. Note what it did: it eliminated Widen (his complaint is not really about a repeated process failure) and it eliminated Exception (the totalising sentence about Marcus was scenery, not the load). It promoted Stake from a guess to something close to a statement — he told you what he stands to lose, which is a self-image, not a document.
Step four: get it wrong.
Suppose you had not held. Suppose the "three weeks ago… the same thing" structure pulled you, as it pulls most people, into the expensive pattern:
You: This isn't the first time, though, is it? There's a pattern here with how work gets attributed on this team.
Watch what that does. It is true, quite possibly, and it is still wrong, because he came to you with an injury and you handed him a taxonomy. The likely reply is a flat one:
Theo: Yeah. Anyway. It's not a big deal.
That flatness is the tell for Repair. Not "he disagreed" — disagreement is fine, disagreement is engagement. Flatness. The temperature dropped and the material stopped arriving.
Step five: repair without making the repair the subject.
The failure mode of Repair is a practitioner who gets so interested in their own miss that the other person ends up managing their discomfort. Keep it to one sentence, and let the second sentence be the re-entry.
You: I jumped to the general thing and you were telling me about a specific one. Say the Priya part again.
One clause of acknowledgement, no apology tour, and the door reopened at the exact point you closed it. Then you are back at step three, now with better information and one candidate eliminated.
The whole sequence, visible: Hold → (information arrives) → Stake → Narrow. Or, on the wrong branch: Widen → flatness → Repair → Hold → Stake. The second is longer by two moves and remains a good conversation. That is the actual claim of this chapter: at this level of skill, choosing wrong is not expensive. Choosing slowly is, and choosing irreversibly is.
Now the edge, in the same breath.
Cheap-first is a heuristic, not a law, and it inverts in a specific condition: when the window is closing faster than the information arrives. Someone whose finger is over send. Someone who has already drafted the resignation. There, a Reflect is not cheap — it costs you the window, and the window was the whole asset.
United Airlines Flight 173 went into a Portland suburb on 28 December 1978 with the crew absorbed in a landing-gear indicator problem and the fuel running out. The flight engineer's warnings grew more pointed as the minutes went, but they stayed in the register of hint and observation. They never became a demand with a number and a deadline in it — never became Narrow. The cheap, deniable, socially safe pattern was run repeatedly into a situation that required the expensive one on the first attempt. The accident is one of the cases that pushed the industry toward what became crew resource management, and the training that followed is, read one way, an institutional answer to exactly this chapter's question: how do you know which pattern this week calls for, and what makes you reach for the wrong one anyway.
The answer that case gives is uncomfortable. You will reach for the cheap pattern not because you diagnosed cheap, but because cheap is safe for you. Hold that suspicion. When the cost of delay is high and rising, invert the order deliberately and run the expensive pattern first, with the reversibility spent on purpose rather than lost by default.
There is a second edge, closer to home. Run cheap-first for long enough with no escalation and you become a very skilled mirror with nothing behind it. People can feel that. They will describe you as a good listener and stop bringing you anything that matters. The cheap patterns are for acquiring the read. If you never spend the read, you were not being careful. You were hiding.
The exercises
1. Name the tell. Below are three unmarked lines. For each, write the pattern you would run and — this is the graded part — the exact words in the line that triggered it.
(a) "It went fine. I mean, he said it was fine."
(b) "Every single time we ship something, marketing finds out from Slack."
(c) "Yeah, no, do whatever you think. I'm good with it."
Check: Your answer is good if the trigger you quote is fewer than six words and you could defend it to someone who chose differently. If your justification requires a paragraph of context you invented, you have inferred rather than read. For (c), if you named a pattern without noticing that "do whatever you think" is agreement with no verb in it, go back to Chapter 4 for ten minutes.
2. Two candidates, one choice. Take the Theo transcript. Assume you did not hold, and assume the sentence had ended cleanly at "It's fine." — no break, no "I just." Now Reflect and Separate are both live. Write the actual sentence you would say for each, then choose one and justify the choice in two lines using reversibility, not likelihood.
Check: A good answer has two speakable sentences, not two descriptions of sentences. Read them aloud; if either sounds like a technique, rewrite it. The justification is good if it names a specific cost of being wrong — what the other person would have to do to recover from your move — rather than asserting that one is "more appropriate."
3. The trap. Unmarked:
Dana: She never once asked me what I thought. Not once in eighteen months. And now there's a "development conversation" on my calendar for Tuesday and I'm supposed to what, be grateful?
The totalising language points loudly at Exception. Say why running it here would land badly, and name what you would run instead.
Check: You have it if you can articulate that "never once… not once in eighteen months" is not a distortion to be corrected but a claim about being unseen — and that asking for a counterexample makes you the second person in the story to dispute her account of her own experience. Bonus rigour: notice you have a hard date. Tuesday. Something in your sequence eventually has to become Narrow, and you should be able to say roughly where.
4. Recovery drill. Here is a practitioner running the wrong pattern:
Sam: The whole launch was a mess and I own most of it.
You: When has a launch gone the way you wanted? What was different that time?
Sam: …I don't know. Ages ago, I guess. Does it matter?
Write the repair sentence and the re-entry sentence. Two sentences total. No third.
Check: The repair sentence is good if it names what you did, not how you feel about it — "I went looking for a bright spot and you were telling me something hard" passes; "sorry, that was a bad question, I do that" does not, because it recruits Sam into reassuring you. The re-entry is good if it hands the floor back at the precise phrase you interrupted: own most of it. If your two sentences are longer than his original line, cut.
5. Live, this week. Run one full sequence in a real conversation and log it afterward in this form: material → tells I saw → what I ran → what came back → what I ran next. One page. Do it within an hour of the conversation, before the log becomes a story about how well it went.
Check: A good log contains at least one tell you noticed and did not act on, and says why. A log in which every observation led to a move is not a log of a conversation; it is a log of you performing a method at someone. Second test: if what came back is recorded as a paraphrase in your voice rather than roughly their words, you are already editing, and the log has lost most of its diagnostic value.
6. The long one. Keep a sequencing log for two weeks — every conversation where you ran anything, three lines each. At the end of week one, count. You are looking for two numbers, and neither is a score.
The first is your default: the pattern you run most, probably by a wide margin. The second is your blind: the pattern you have not run once. Almost everyone has one. It is not a gap in your skill — you practised all nine and you can execute all nine. It is a gap in your reaching, which is a different thing and made of something other than competence.
Then, for week two, take the default away. Do not run it. When the situation calls for it and your hands go there anyway, you must run your second-best candidate instead and log what happened.
Check: You will not finish this today, and you should not try to. The week-one count is the easy half and it will be roughly what you expect. The finding worth having comes in week two, and it is usually this: your default was not being selected on the merits. It was being selected because it is the pattern that keeps you comfortable — the one that never leaves you exposed, or never lets the conversation go somewhere you cannot steer. When you can write one honest sentence about what your default protects you from, this chapter is finished. Not before.