The Six
Exercise 1 — Read the gap off the transcript
You just read a worked example with twelve timings in it. Without scrolling back, write down: the mean gap on deletion cards, the conversational budget, and the ratio between them.
Then write, in one sentence, what the 900-millisecond card demonstrates.
How you know it's any good: 2.9 s, 400 ms, roughly 7×. If you had to scroll back for the ratio, that is itself a datum — the numbers went in as text rather than as a fact you now own, and you have just watched the difference between reading and encoding in real time. On the last sentence: an answer about speed is incomplete. The card was fast because the response was never stored as knowledge at all. If your sentence doesn't distinguish where it was stored from how well it was known, write it again.
Exercise 2 — Take your own baseline
Run the protocol above exactly as written. Twelve cards, six with simple deletions, six clean, shuffled face-down, spoken responses, recorded, waveform-measured. Criterion set in writing before the first card turns.
Do not substitute a "close enough" version. Do not time yourself with a stopwatch you start and stop by hand — your reaction time on the stopwatch is the same 300 ms you are trying to measure, and it will land in your favour every time.
Write down: mean gap on target cards, accuracy against your written criterion, and today's date.
How you know it's any good: You have a number that is worse than you expected. That is not a joke and it is not consolation — it is the actual pass condition. A baseline that flatters you was contaminated, most likely by anticipation, a hand-run timer, or a criterion you softened after the fact. If your mean came in under 1.2 s on your first-ever pass, go back and check whether you wrote the cards yourself and remembered them. If your accuracy came in above 90%, go back and check whether your criterion was written down before or generated after. Redo it. The second baseline is the real one.
Exercise 3 — Separate the two clocks
Take three of your target sentences. For each, do two timed runs, recorded.
Run A: respond aloud as fast as a clean response will come — the baseline condition.
Run B: say aloud, as fast as you can, only the name of the pattern — "deletion" — and stop. No question, no recovery, nothing else.
Measure both. Subtract.
How you know it's any good: You should have two distinct numbers, and Run B should be substantially faster — commonly under a second once you're only naming. The difference between them is not overhead you can dismiss; it is the cost of getting from recognition to production, and it is where most of your gap lives. If A and B come out within 200 ms of each other, one of two things is true: either you are genuinely fluent on this pattern already, in which case your baseline should have been near-conversational and it wasn't — or you are not really producing in Run A, you're reciting a template you'd prepared. Listen back to the audio and check whether Run A sounds like speech or like recall. You will hear it.
Exercise 4 — Find the corridor
From your baseline, pull the three slowest target cards. For each, write out every pattern you could legitimately have run on that sentence — not the one you chose, all of them.
Now count. Then rank your twelve cards by number of legitimate options and, separately, by your measured gap.
How you know it's any good: The two rankings should correlate, loosely but visibly. Your slowest cards should be your most ambiguous ones. If they are — you have replicated the corridor effect from the worked example on your own data, using your own nervous system, and you now believe it in a way that reading it could not have produced.
If the rankings don't correlate, do not force them. You have found something. The likeliest honest explanation is that your slow cards are slow for a different reason — unfamiliar vocabulary, emotional charge in the content, a sentence whose meaning you had to work out before you could work on it. Name which. A finding that contradicts the chapter is worth more to you than one that confirms it, and you should be slightly suspicious of how neatly the worked example came out.
Exercise 5 — Break the pattern you are most proud of
Choose the Meta Model pattern you consider your strongest. The one you'd demonstrate if someone asked what NLP was.
Build twenty cards for it — twelve targets, eight clean. Run the protocol. But add one constraint: while you respond, hold a four-digit number in mind, given to you by someone else at the start of the run, and recite it aloud at the end.
Measure your gap. Compare to your unloaded baseline on the same pattern.
How you know it's any good: Your gap will lengthen. The question is by how much, and that number is your real answer to "do I own this pattern." A response running from procedural memory is largely indifferent to a concurrent load on working memory — that's most of what makes it worth having. A response assembled by deliberate search competes directly with the digits for the same narrow resource, and both degrade. So: a gap that grows by 10–15% is a pattern approaching automaticity. A gap that doubles, or a run where you lose the digits, means the pattern you are proudest of is still stored in the wrong system — and you have been mistaking fondness for fluency. Also check the digits themselves. Losing them while keeping your speed is the same finding wearing a different hat.
This exercise is designed to take something from you. If it didn't, check that the four digits were genuinely someone else's and genuinely random.
Exercise 6 — The one you will not finish today
Pick a single pattern. One. The narrowest useful one you have.
Take it to 400 milliseconds in live conversation.
Not in drill. Not with cards, not alone in a room, not against sentences you constructed. In actual conversation with an actual person who does not know they are your practice, where the stimulus arrives when it arrives and the pattern either fires inside the turn or does not fire at all. Record what you're permitted to record, or log it immediately afterward from memory — noting that memory of your own latency is unreliable in exactly the direction you'd expect, which is itself worth knowing.
Track two things weekly: your drilled gap on cards, and your hit rate in live conversation — how often the pattern fired in the turn where it belonged, versus how often you thought of it walking to your car.
Expect the second number to lag the first badly. Expect it to sit near zero for weeks after your card gap looks respectable. That lag is not failure; it is the actual curve of the thing you are learning, and the reason this exercise cannot be finished today is that procedural consolidation runs on sleep and repetition across days, not on effort within one.
How you know it's any good: Three markers, in order of appearance.
First, you notice the pattern after the conversation. That is retrieval working, too late — and it is genuine progress, not a failure, so log it as a hit-adjacent rather than a miss. Most people quit during this phase because it feels like nothing.
Second, you notice it in the conversation and choose not to run it. The pattern is now available inside the turn; you are simply exercising judgment, which is the correct next problem and one you couldn't have had before.
Third, it fires and you find out you ran it only when you hear yourself finishing the question. That is the terminal marker. There is no fourth.
Here is the part to sit with. When you arrive at the third marker, you will lose the ability to feel skilled while doing it. The pattern will have stopped being something you do and become part of how you hear — and the satisfaction of executing well, which is most of what has been carrying you through the drills, quietly stops being available. You will have to find a new reason to keep going, because the old one was made of effort and the effort is gone.
That is not a cost of the method. It is what competence is, in every discipline where the response has to beat the clock, and it is why so few people who know a great deal about this subject can do very much with it. They kept the part that felt like mastery and shed the part that was.
Log your baseline before the next chapter. Chapter 2 assumes a number on paper with a date beside it, and it will not be worth much without one.