10. Strategies: The Order the Operations Have to Come In

Two people sit down to spell rhythm. Both have the same representational vocabulary — both can form a picture, both can hear a sound, both can register a feeling in the chest or the gut. Both have read the word many times. One writes it correctly without hesitation and reports having done nothing at all. The other sounds it out, produces rythem, looks at it, feels a vague dissatisfaction that does not resolve into a correction, and leaves it.

The difference between them is not vocabulary and it is not intelligence. It is order. The good speller runs a sequence that begins with a remembered image and tests it against a felt sense of rightness; the poor speller runs a sequence that begins with a sound and has no image to test against, so the feeling of wrongness has nothing to correct. Same alphabet, different grammar. That is the claim this chapter makes and then constrains: that a considerable amount of what we call ability is sequence, that sequence is in principle elicitable and notatable, and that the notation is worth having because it tells you exactly which step to install rather than telling you to try harder.

Everything in the chapter depends on three things you already have. Chapter two supplied the alphabet — the sensory channels a representation can occupy. Chapter three supplied the calibration without which you cannot tell that a step has occurred. Chapter seven supplied the operators inside a step, the parameters that make one image bright and close and another dim and far. Take away any one and the enterprise collapses: without the alphabet you cannot say what the step is made of, without calibration you cannot detect the step, without submodalities you cannot say why one person's image works and another's does not.

The unit: what a TOTE actually is

The unit of behaviour that strategy work borrows was not invented by NLP. It comes from George Miller, Eugene Galanter and Karl Pribram's Plans and the Structure of Behavior (1960), a book written to replace the reflex arc with something that could account for goal-directed action. Their proposal was the TOTE: Test — Operate — Test — Exit.

The classic illustration is a nail. You look at the nail; it protrudes. That is a test, and it fails — the nail's state does not match the standard "flush." You hammer. That is the operation. You look again. That is the second test. If the nail still protrudes, you hammer again; the loop runs until the test passes, and then you exit and go find another nail. The behaviour terminates not because the hammering ran out but because a comparison came out equal.

Two things in that structure are routinely collapsed, and collapsing them is the most common error in strategy work, so hold them apart deliberately.

The entry test is the condition that starts the sequence. It asks: is this a situation of the kind that calls for this procedure? A nail that is already flush never gets hammered, because the entry test never fires. Entry tests are how a person knows which of their many strategies to run, and they are frequently the part of a sequence that is broken. A person who cannot motivate themselves to write is often not missing a writing strategy at all; they have a perfectly good one that never gets triggered, because the entry condition is attached to a stimulus that no longer occurs.

The comparison test is what runs inside the loop and decides whether to iterate or exit. It asks: does the current state match the standard? This test requires a standard — a representation of the satisfactory outcome held somewhere and available for comparison. A loop with no standard does not terminate; it runs until interrupted by fatigue, deadline or shame. That is the formal description of a great deal of ordinary suffering. The perfectionist does not have too high a standard so much as a comparison test whose standard is not a fixed representation at all but a receding one, recomputed upward each time the operation improves the work.

Miller, Galanter and Pribram's real contribution was hierarchy: TOTEs nest. "Hammer the nail" is an operation inside "build the frame," which is an operation inside "put up the shed." The operate step of any TOTE is itself a TOTE at the level below. This matters more than it looks, because it tells you that when a strategy fails you should ask at which level it failed, and the answer is often one level down from where the complaint is stated. A person who says they cannot finish projects may finish every sub-task competently and lack only the top-level comparison test that would tell them a project was done.

The notation, in full

Strategy notation exists to make a sequence writable, and therefore comparable, teachable and checkable. It is small. Learn it once and it does not change for the rest of the book.

Each step is written as a capital letter naming the sensory channel, with superscripts naming the origin and the sub-type of the representation.

The channels:

• V — visual
• A — auditory
• K — kinaesthetic
• O — olfactory
• G — gustatory

The origin superscripts:

• i — internal. The representation is generated inside: a mental image, an inner sound, a felt sense.
• e — external. The representation comes from the sense organs: something actually seen, actually heard, actually touched.
• r — remembered. An internal representation retrieved from experience, as it occurred.
• c — constructed. An internal representation assembled from parts, not previously experienced in that form.
• d — digital. Reserved almost entirely for A^d, internal dialogue — language as language, words heard or spoken internally rather than tonal sound. It is separated out because it behaves differently from tonal auditory: it carries propositions, and propositions can be argued with.

The kinaesthetic sub-types, because K is doing three unrelated jobs and blurring them wrecks a notation:

• Kⁱ — internal, visceral: emotion, gut feeling, the felt sense of rightness or wrongness.
• K^e — external, tactile: actual sensation from the body's surface or from movement.
• K^m — meta-kinaesthetic: a feeling about a representation rather than a feeling that is part of the content. The unease about a spelling is K^m; the unease in a memory of falling is Kⁱ. This distinction does real work later.

The operators:

• → — "is followed by." A^e → V^r reads: hearing something externally is followed by a remembered image.
• / — simultaneous. V^r/Kⁱ reads: the remembered image and the internal feeling occur together, not in sequence.
• ( ) — a test, with the compared terms inside. ( V^r = V^e ) reads: the remembered image is compared with the seen image for match.
• ↺ — loop back. Written after a failed test with the step it returns to, thus: ↺ to V^c.
• ⊣ — exit. The sequence terminates and control passes up a level.
• ⌐ — polarity marker, written before a test that is passed by mismatch rather than match. Some people's comparison tests are satisfied by difference: they check whether anything has gone wrong, not whether things are right. This is not a defect but it changes what will convince them.

One convention that saves confusion: subscripts number repeated instances of the same element, so V^r₁ and V^r₂ are two different remembered images inside one sequence, not the same one twice. Without it, longer strategies become unreadable at exactly the point where reading them matters.

Elicitation without contamination

Here is where the craft actually lives, and where most modelling attempts die.

The first rule is the one every practitioner nods at and then violates within two questions: demand a specific instance. Not "how do you decide?" but "think of a particular time you decided — a specific one, with a date on it, one where you can tell me what room you were in." The general account is a theory the person holds about themselves. The specific instance is a retrieval, and retrieval reinstates something of the original sequence, which is what you are trying to observe. When someone answers a general question you are eliciting their self-concept. When they answer from a re-lived instance, the strategy runs again in front of you, partially, and it leaks.

That leakage is your primary data, and it is behavioural rather than verbal. As the person re-enters the instance you are watching for what chapter three trained you to see: the eye movements, the breathing shift, the change in vocal tempo, the small postural settle, the hand that traces something in the air to the left. You are not reading meaning off these — chapter three forbade that. You are reading boundaries. A change marks a step transition. You then ask what happened there, and now the question is specific enough to be answerable: "just then, right before you said 'and I knew' — something changed. What were you aware of?"

The second rule follows from the first. The verbal report tells you the content; the behaviour tells you the sequence. Take content from the mouth and order from the body, and when they disagree, believe the body and go looking for why.

The third technique is the one that surfaces steps a performer does not know they take, and it is the highest-leverage move in the chapter. Steps become invisible by being fast, by being reliable, and above all by being unfailing — a person cannot report a step that has never in their memory produced a noticeable event. So you break it. Not the person: the conditions. You introduce a controlled disruption and watch what they reach for.

Ask a good speller to spell a long word while you slowly move your finger across their upper-left visual field, and if the strategy is image-based their fluency will degrade and they will tell you, unprompted and now with certainty, that they "couldn't see it." Ask a fast mental calculator to do a problem while repeating "the-the-the" aloud, and if there is an auditory rehearsal step you will find it, because articulatory suppression takes it away. Hand a skilled negotiator a case with one crucial fact removed and watch what they ask for first; the first question reveals the entry test. Give an expert a deliberately wrong version of the task and watch the moment they flinch — the flinch is the comparison test firing, and its timing tells you where in the sequence the standard sits.

This is elicitation by interference, and it works precisely because it does not depend on the person's access to their own process. It depends only on their process being disruptable. A note on its limits: interference is intrusive, it degrades performance by design, and it should be run with the performer's knowledge and consent. You are not tricking anyone. You are saying, "I want to find the step you can't see, and the way to find it is to take something away and watch you miss it" — which competent people generally find interesting rather than threatening.

The spelling strategy, end to end

Spelling is the standard demonstration because it is well-bounded, verifiable against an external criterion, and shows the diagnostic value of the notation immediately.

Elicit it from a natural speller. Give them a word they know but do not write often. Watch. What you typically see: the eyes go up and to their left or hold defocused for a beat; there is a pause; then they write or spell aloud, sometimes in chunks rather than letter by letter, and sometimes they stop after writing and look at what they wrote with a small settling of the shoulders before saying "yes."

Notated:

A^e → V^r → ( V^r / K^m ) → A^d / K^e → ( V^e = V^r ) → ⊣

Read it out: they hear the word (A^e); they retrieve a stored image of it (V^r); while holding the image they get a feeling about the image, not about the word's meaning — a sense of whether that image is the right one and whether it is complete (V^r/K^m, a test); they then read the image off, letter by letter or in chunks, in inner speech and hand movement (A^d/K^e); they look at what they have written and compare it back to the image ((V^e = V^r)); match, exit.

Now the diagnostic. Elicit from someone who spells badly and you very often get:

A^e → A^d (phonetic) → K^e (write) → K^m → ↺ to A^d → ⊣ on fatigue

They hear the word and go straight to sound, spell it as it sounds, write it, and then get a feeling that something is off. That K^m is present and accurate — poor spellers are frequently quite good at knowing a word is wrong. But it has nothing to point at. The feeling is compared against no stored image, so the loop returns to the only step available, which is the phonetic one, which produces the same output again. The sequence exits on exhaustion rather than on a passed test.

The missing element is not effort and not the feeling. It is a V^r for the test to compare against, plus the ordering that puts the feeling after the image rather than after the writing. The instructional consequence is specific and testable: teach the storage of word images — look at the word, look away, rebuild it, check against the page — and put the felt check on the image rather than on the written attempt.

This is also the place to say plainly what the evidence supports. Visual-orthographic memory is a real and well-studied contributor to spelling ability; that skilled spellers rely on stored word-specific representations rather than pure phoneme-to-grapheme conversion is mainstream reading science, not an NLP claim. What is not independently established at that standard is the more specific NLP package — that the eye-accessing direction reliably indexes the retrieval, that installing the sequence in the notated form is superior to other orthographic training, or that the K^m check is what the good speller is actually doing rather than what they report doing. Grade it accordingly: the mechanism is well supported, the elicitation procedure is plausible and locally verifiable, the specific installation protocol is untested at scale.

Decision, motivation, and the convincer

Three strategy families recur often enough to be worth naming.

A decision strategy is the sequence by which a person moves from options to a commitment. It has an entry test (what counts as a decision-requiring situation), a generation step (how alternatives arise — recalled, constructed, offered by others), an evaluation loop, and an exit criterion. The clinically useful failure is a decision strategy with a generation step but no exit criterion: options keep being constructed and the loop never terminates. That person is not indecisive by temperament. They are running a TOTE with no standard.

A motivation strategy is the sequence between a representation of a task and the initiation of movement toward it. The most common structural split is direction of the driving representation. Some people run: image of the completed task, constructed, associated, with a positive K — the pull. Others run: image of the consequence of not doing it, constructed, associated, with a negative K that resolves on action — the push. Neither is superior in the abstract; each has a distinct failure. The pull strategy fails when the outcome image is not vivid enough to generate K, which is why it collapses on abstract or long-horizon tasks. The push strategy fails by requiring the aversive state to be re-manufactured every cycle, which is expensive and, over years, corrosive. Knowing which one a person runs tells you what will and will not help them, and it explains why the advice that transformed one colleague does nothing for the next.

A convincer strategy is how a person comes to believe something is so. It has two independent components, and separating them is the practical payoff.

The first is the modality — which channel the evidence has to arrive in. Some people need to see it demonstrated, some need to hear it argued, some need to handle it, some need to have done it themselves. Present evidence in the wrong channel and it does not register as evidence; it registers as talk.

The second is the demonstration mode — how much, and in what pattern, before the belief forms. There are four, and they are exhaustive enough to be worth memorising:

Number of times. The person needs a specific count of instances. Three demonstrations, or five, or two. Below the number, no conviction; at the number, conviction, and the count is often remarkably stable across domains for the same person. What you must show them: repetition. One overwhelming case will not do it.

Period of time. The person needs evidence sustained across a duration — a month of reliability, a quarter of results, a year of behaviour. Instances do not accumulate into conviction; only elapsed time does. What you must show them: consistency across a span, which means you cannot compress the process no matter how strong the case is. Trying to close them faster reads to them as pressure, and pressure resets the clock.

Automatic. The person grants provisional belief immediately and revises only on disconfirmation. What you must show them: very little. But note the corollary — the automatic convincer is paired with an equally fast un-convincer, and a single failure can undo everything. With these people your risk is not the sale; it is the first disappointment.

Consistent. The person is never finally convinced. Each occasion is evaluated afresh; yesterday's evidence does not carry. What you must show them: proof every time, indefinitely. This is exhausting to be around and can look like distrust, but it is structural rather than personal, and treating it as an insult misreads it entirely.

Notice how much practical consequence rides on a distinction so small it is usually invisible. Two people both say "I need to be convinced." One needs three demonstrations and the third one closes it. The other needs six weeks and the third demonstration in week one moves them not at all. Give the second person the first person's evidence pattern and you will read them as resistant, and you will be wrong.

Installation, and its honest range

Once a sequence is elicited and notated, installation means getting it to run — in the same person under new conditions, or in a different person entirely. Three operations do the work.

Rehearsal. The sequence is run deliberately, slowly, step by step, with each representation explicitly formed before the next. It is slower than the natural version and feels artificial, and that is correct; automaticity is what you are building toward, not starting from. Rehearsal fails when steps are skipped as obvious. If the notation says V^r/K^m, the learner must actually generate the image and actually notice the feeling about it, not simply agree that they would.

Anchoring the sequence. Chapter six's operation, applied to order rather than to state. Each step is anchored as it is generated cleanly, and the anchors are fired in sequence so that the completion of one step becomes the trigger for the next — chaining the entry condition of step n+1 to the exit of step n. What makes this more than decoration is that a well-chained sequence stops requiring conscious sequencing, which is the whole point: a strategy you have to remember to run is not yet installed.

Future pacing the entry condition. This is the step most often dropped and most responsible for installations that work in the room and evaporate outside it. A sequence with no entry test is inert. The learner must rehearse the sequence triggered by the actual cue in the actual context — not "I will spell better" but "when I am writing and I hit a word I am unsure of, that uncertainty is what starts this." Attach the strategy to the stimulus that will really be present, in the sensory form it will really arrive in.

Now the verdict, stated without the promotional gloss the field has usually given it. Strategy transfer works well on well-bounded perceptual and procedural tasks with fast, objective feedback — spelling, certain kinds of memorisation, discrimination tasks, some motor and inspection skills. It degrades as the domain widens, and it degrades sharply. The reasons are structural rather than mysterious: as the domain widens, the number of relevant entry conditions multiplies, the comparison standard becomes harder to represent, and more of the performance rests on stored domain knowledge that no sequence can substitute for. You can transfer the order of operations. You cannot transfer the twenty years of pattern-stocked memory the operations run over. A chess master's strategy notated perfectly and installed in a novice yields a novice who runs a master's sequence over a novice's board vocabulary, which is worth something and is not mastery.

What the expert-performance literature says, and why it hurts

The nearest rigorous cousin to modelling is the expert-performance tradition — the research programme most associated with K. Anders Ericsson, which studies exceptional performers by capturing performance under controlled conditions rather than by asking people how they do it. Two of its findings bear directly on this chapter, and they cut in opposite directions.

The first is friendly. Expertise, studied closely, does turn out to be largely acquired structure rather than general capacity, and much of it is procedural — sequences of perception, evaluation and action that can be described and, with the right practice conditions, trained. This is the same bet strategy work makes.

The second is not friendly at all, and it is the reason this chapter cannot end on the notation. Experts' verbal accounts of their own processes are systematically incomplete. Not occasionally wrong — systematically incomplete, in a direction. What becomes automatic becomes unavailable to introspection; what is fast is not narrated; and when a person is asked to explain what they did, they do not read out a log, because there is no log. They construct a plausible account after the fact, from their theory of the task, their memory of similar occasions, and what seems reasonable to say to the person asking. Ericsson and Simon's methodological work on protocol analysis exists precisely because of this: concurrent think-aloud during the task is treated as data, and retrospective explanation is treated with suspicion, because the two are not the same kind of report.

This strikes at elicitation directly and it should be felt as damage. Almost everything in the NLP modelling tradition is retrospective explanation, gathered by asking. That is the weakest available instrument pointed at the exact place the whole enterprise depends on.

And here is where the argument turns, because the damage is not fatal — it relocates the craft. The field has always treated installation as the hard part and elicitation as the preliminary, the interview you do before the real work. That is backwards. Installation is comparatively mechanical: rehearse, chain, future pace, test. Elicitation is where modelling actually fails, and it fails for a reason no amount of rapport or patience can fix, because a competent person's account of their competence is a reconstruction rather than a recording. The expert is not withholding the sequence and is not lying about it. They do not have it. It was never stored in a form that language can retrieve.

So the modeller's real skill is not interviewing. It is experimental design — engineering situations in which the behaviour reveals a sequence the performer is constitutionally unable to narrate. That is what the interference techniques are for, and it is why they are not a trick in the toolkit but the centre of the method. You suppress a channel and watch performance fall. You corrupt the input and watch which error they catch first. You give an incomplete case and watch which gap they close first. You put a novice's work in front of them and watch the timing of the flinch. You run the task under time pressure until the reportable overlay strips away and only the sequence is left. Each of these produces evidence that does not pass through the performer's self-model, which is exactly the property you need. Ask them afterward, by all means — but ask them to explain the behaviour you just observed, not to supply it.

The failure mode

The failure this chapter must name is the one that follows directly from the turn: installing a reported strategy that is not the expert's actual strategy.

The mechanism is unglamorous and reliable. You interview a competent person. They give you a fluent, coherent, entirely sincere account. You notate it. It is a good notation — internally consistent, well-formed, satisfying. You install it in a learner, who works hard and gets nowhere, or gets somewhere worse: the reported strategy is often a deliberate strategy the expert used years ago and has since replaced with something faster, so installing it can regress a learner into a slower method while telling them it is how the best do it. The learner then concludes they lack the talent, which is the cruellest possible reading of the situation and the most common.

There is a second-order harm worth stating. Modelling is often sold as democratising: excellence is structure, structure can be copied, therefore anyone can. When the copied structure is a fiction, that promise inverts and becomes a precise mechanism for teaching people that they, personally, are the reason it did not work. A framework that explains failure by reference to the learner's insufficiency, when the actual fault is in the model's provenance, has crossed the edge past which it does harm. That is this chapter's edge, and it is not far away.

The check that catches the substitution is behavioural, and it must be run before you teach the sequence to anyone.

Take the notated strategy back to the expert and make three predictions from it. Predict what will happen to their performance if a specific step is blocked — if the strategy is V^r-dependent, visual interference should degrade it and articulatory suppression should not. Predict where in the sequence errors will appear if the input is corrupted at a particular point. Predict the timing — where the pauses fall, which parts are fast, where the hesitation lives. Then test all three against actual performance.

A correct notation makes non-obvious predictions that come true. A reconstructed one predicts what everybody already expects and fails on the specifics: you block the visual channel of a supposedly visual strategy and performance does not move, which tells you the visual step is in the account and not in the performance. When predictions fail, do not patch the notation. Go back to interference and find what is really running, because a patched fiction is still a fiction and it is now harder to see through.

One further check costs nothing and catches a different error: run the strategy yourself before teaching it to anyone else. You have privileged access to your own failure, and where your run of the sequence stalls is diagnostic in a way no observation of another person can be.

Which is the practice.

Find someone who does well a thing you do badly — a genuinely small thing, bounded and checkable. How they remember names. How they read a room before speaking. How they decide a piece of work is finished. How they hold their temper in a meeting where you lose yours. Sit with them and refuse, from the first sentence to the last, to accept a general account. Every time they begin "well, what I usually do," bring them back: a specific time, with a date, a room, a person in it. Watch them re-enter it and watch what changes — the eyes, the breath, the tempo, the hand — and ask what happened at each change rather than asking what they do. When you reach the place where they say "I just do it" or "it's automatic," you have arrived at the interesting part, not the end; ask their permission and take something away — block the channel you suspect, corrupt the input, hurry them — and watch what falls over. Notate what you get, in the symbols, on paper. The discipline of the symbols is what stops you writing down an impression.

Then run it yourself, this week, in the real situation, and pay close attention to exactly where it breaks. It will break. Something will be missing, and the missing thing will be a step so small and so automatic for them that it never occurred to either of you to name it — a check they run without noticing, a moment they wait that you do not, an image they form before speaking that you have been speaking without. Write down the break with the same precision you gave the elicitation. Go back and hunt for what belongs there. The break is not the failure of the method; the break is the method's most useful output, because it is the only part of the sequence whose absence you can feel from the inside.

Worked Examples — Chapter 10

The strategy is not the content of thought; it is the topology of information flow across time. Two agents possessing identical neuro-semantics but different strategic architectures will generate distinct reality. The derivation below establishes how to elicit, notate, and verify that architecture using the TOTE unit, the alphabet of submodalities, and the operators defined in earlier chapters. We proceed by decomposing observable behavior into its minimal functional sequence, proving the sequence against exit conditions, and mapping the failure modes where the sequence inverts into pathology.

Worked Example 1: Derivation of the Apollo 13 CO2 Scrubber Contingency Strategy

Case: NASA Mission Control, April 1970. The spacecraft's lithium hydroxide canisters were insufficient for the combined crew of Apollo 13 and the LM Aquarius. The strategy to mitigate CO2 buildup was not a document but a dynamic sequence elicited from the flight controller team and simulated on the ground.

Objective: Derive the strategy $\sigma_{fix}$ such that the system state $x(t)$ satisfies the safety predicate $P_{safe} \equiv [\text{CO}_2 < 1.5 \text{ mmHg}]$.

Definitions:

Let $T$ denote a Test unit that evaluates a predicate against current state.

Let $O$ denote an Operator that maps state $x \to x'$.

Let $E$ denote an Exit predicate that halts the loop.

A TOTE unit is defined as $\mathcal{T} = \langle T, O, T, E \rangle$.

The sequence of TOTE units is $\sigma = (\mathcal{T}_1, \mathcal{T}_2, \dots, \mathcal{T}_n)$.

Composition is denoted by $\circ$.

Derivation:

We observe the physical constraint: The LM canisters fit the CM module, but the CM canisters do not fit the LM module without an adapter. The strategy must construct this adapter using only materials available on board (plastic bags, tape, cardboard, hose).

Step 1: Decompose the problem into discrepancy reduction.

Discrepancy $D = P_{safe} \land \neg \text{Fit}$.

The sequence must reduce $D$ to $0$.

Step 2: Elicit the sequence from the simulation.

Ground controllers observed that random assembly failed. The successful strategy exhibited a rigid temporal order. We notation this order as $\sigma_{fix}$.

$$ \sigma_{fix} = \mathcal{T}_{gather} \circ \mathcal{T}_{construct} \circ \mathcal{T}_{integrate} \circ \mathcal{T}_{test} $$

We expand each TOTE unit.

$\mathcal{T}_{gather}$:

$T_1$: Predicate $P_{parts} \equiv \{ \text{bags available}, \text{tape available}, \text{cardboard available}, \text{hose available} \}$.

If $P_{parts}$ is false, the sequence cannot proceed; this is a hard constraint.

$O_1$: Operator $\text{Locate}_{parts}$: Search cabin inventory for items satisfying $P_{parts}$.

$T_2$: Predicate $P_{gathered} \equiv \{ \text{bags}, \text{tape}, \text{cardboard}, \text{hose} \} \subseteq \text{Inventory}_{current}$.

$E_1$: Exit if $P_{gathered}$ is true. Else, loop $O_1$.

$\mathcal{T}_{construct}$:

$T_3$: Predicate $P_{shape} \equiv \text{Cylinder}( \text{cardboard} ) \land \text{Seal}( \text{cardboard}, \text{tape} )$.

$O_3$: Operator $\text{Form}_{adapter}$: Roll cardboard, insert bag, tape seams.

$T_4$: Predicate $P_{form} \equiv \text{Length} \approx 12 \text{ inches} \land \text{Diameter} \approx 3 \text{ inches}$.

$E_2$: Exit if $P_{form}$ satisfied within tolerance $\delta$.

$\mathcal{T}_{integrate}$:

$T_5$: Predicate $P_{connect} \equiv \text{Adapter} \text{ links } \text{LM}_{out} \text{ to } \text{CM}_{in}$.

$O_5$: Operator $\text{Attach}_{adapter}$: Insert constructed adapter between modules.

$T_6$: Predicate $P_{flow} \equiv \text{Airflow}( \text{LM} \to \text{CM} ) > 0$.

$E_3$: Exit if $P_{flow}$ detected.

$\mathcal{T}_{test}$:

$T_7$: Predicate $P_{safe} \equiv \text{CO}_2(t) < 1.5 \text{ mmHg}$.

$O_7$: Operator $\text{Wait}_{decay}$: Monitor decay rate.

$T_8$: Predicate $P_{sustained} \equiv \forall t \in [t_0, t_0 + 4h], \text{CO}_2(t) < 1.5$.

$E_4$: Exit if $P_{sustained}$. Else, return to $\mathcal{T}_{test}$ with reduced airflow assumption.

Mechanism:

The strategy works because each $O_i$ reduces the discrepancy defined by the preceding $T_i$ without introducing a new discrepancy that the subsequent $T_{i+1}$ cannot detect. The mechanism requires that the operator space $\mathcal{O}$ contains elements capable of satisfying the predicates $\mathcal{P}$. In this case, the space of available materials was sufficient to instantiate $O_3$ and $O_5$.

Failure Mode:

The edge of this framework is material insufficiency. If $P_{parts}$ is false, the strategy collapses. There is no operator $O$ that can create matter from vacuum. The failure mode is not a sequencing error but a domain constraint violation. A practitioner attempting to transfer this strategy to a context lacking the necessary parts will fail regardless of sequence fidelity. The insight is that strategy derivation must include a feasibility audit of the operator space before notating the sequence.

Worked Example 2: Transfer of Strategic Isomorphism from Agile Software to Public Policy

Case: The United Kingdom's Behavioural Insights Team (The "Nudge Unit"), established 2010. The team sought to apply the strategy of "Iterative User Testing" from software development to the design of government services.

Objective: Derive the transfer function $\tau$ that maps strategy $\sigma_{source}$ to strategy $\sigma_{target}$ such that $\sigma_{target}$ preserves the functional topology of $\sigma_{source}$ under the constraints of the policy domain.

Source Strategy: Agile Development.

Let $\sigma_{agile}$ be the sequence used by high-performing software teams (e.g., GitHub's CI/CD pipeline evolution circa 2015).

$$ \sigma_{agile} = \mathcal{T}_{spec} \circ \mathcal{T}_{build} \circ \mathcal{T}_{test} \circ \mathcal{T}_{deploy} \circ \mathcal{T}_{feedback} $$

Expanded:

$\mathcal{T}_{spec}: T \equiv \text{User stories defined}, O \equiv \text{Prioritize backlog}, T \equiv \text{Backlog validated}, E \equiv \text{Ready for dev}$.

$\mathcal{T}_{build}: T \equiv \text{Code compiles}, O \equiv \text{Compile/Unit Test}, T \equiv \text{Pass rate} > 95\%, E \equiv \text{Build artifact}$.

$\mathcal{T}_{test}: T \equiv \text{Integration pass}, O \equiv \text{Run integration suite}, T \equiv \text{Regression} = 0, E \equiv \text{Stable env}$.

$\mathcal{T}_{deploy}: T \equiv \text{Canary deployment}, O \equiv \text{Rollout to 5\%}, T \equiv \text{Error rate} < \epsilon, E \equiv \text{Full rollout}$.

$\mathcal{T}_{feedback}: T \equiv \text{Metrics collected}, O \equiv \text{Update backlog}, T \equiv \text{Feedback loop closed}, E \equiv \text{Next sprint}$.

Target Strategy: Policy Intervention.

We seek $\sigma_{policy}$ such that the structure is isomorphic to $\sigma_{agile}$ but the operators are instantiated in the policy domain. The UK BIT strategy (documented in their 2011 "Insights for Policymakers" framework) follows this structure.

$$ \sigma_{policy} = \tau(\sigma_{agile}) $$

We define $\tau$ as a mapping over TOTE units. $\tau(\mathcal{T}) = \langle \tau(T), \tau(O), \tau(T'), \tau(E) \rangle$.

Constraints on $\tau$:

1. $\tau(O)$ must be an operator available in the policy domain (e.g., legislation, administrative order, communication campaign).
2. $\tau(T)$ must be a predicate measurable in the policy domain (e.g., compliance rate, uptake).
3. The exit conditions must respect ethical boundaries.

Application:

$\tau(\mathcal{T}_{spec}) \to \mathcal{T}_{policy\_design}$:

$T \equiv \text{Citizen journey mapped via ethnography}$.

$O \equiv \text{Draft intervention prototype}$.

$T \equiv \text{Prototype reviewed by stakeholder panel}$.

$E \equiv \text{Intervention feasible}$.

$\tau(\mathcal{T}_{build}) \to \mathcal{T}_{policy\_pilot}$:

$T \equiv \text{Pilot design specified}$.

$O \equiv \text{Implement pilot in limited jurisdiction}$.

$T \equiv \text{Process compliance verified}$.

$E \equiv \text{Pilot data collected}$.

$\tau(\mathcal{T}_{test}) \to \mathcal{T}_{policy\_evaluation}$:

$T \equiv \text{Impact evaluated}$.

$O \equiv \text{Run RCT or quasi-experimental analysis}$.

$T \equiv \text{Effect size } \Delta > 0 \land \text{Cost-effectiveness } > \text{threshold}$.

$E \equiv \text{Decision to scale or abandon}$.

$\tau(\mathcal{T}_{deploy}) \to \mathcal{T}_{policy_rollout}$:

$T \equiv \text{Scale plan approved}$.

$O \equiv \text{National implementation}$.

$T \equiv \text{System stability maintained}$.

$E \equiv \text{Service live}$.

$\tau(\mathcal{T}_{feedback}) \to \mathcal{T}_{policy_adapt}$:

$T \equiv \text{Monitor feedback}$.

$O \equiv \text{Adjust policy parameters}$.

$T \equiv \text{Feedback loop closed}$.

$E \equiv \text{Iteration complete}$.

Mechanism:

Transfer succeeds because the topology of feedback and error correction is preserved. The exit test of the "deploy" phase in software (error rate) maps to the exit test of "rollout" in policy (system stability). The mechanism requires that the policy domain allows for small-scale experimentation. If the domain forbids pilots (e.g., immediate national mandate without testing), the isomorphism breaks, and the strategy cannot be transferred.

Failure Mode:

The edge is the "Scale Trap." If the transfer function $\tau$ ignores the non-linear dynamics of policy scale-up, the strategy inverts. A policy that works at 5% scale (canary) may fail at 100% scale due to network effects or behavioral adaptation. The failure mode occurs when $\tau$ assumes linear scaling of operator efficacy. The strongest objection is that policy contexts lack the controlled isolation of software environments; noise is part of the signal. The correction is to embed a noise-filtering operator within $\mathcal{T}_{test}$, ensuring that the exit test accounts for variance, not just mean effect.

Worked Example 3: Correction of the Theranos "Mini-Lab" Strategy Failure

Case: Theranos, circa 2011-2016. The company claimed a strategy to perform multi-analyte blood testing from a few drops of finger-prick blood, displacing venous draws and large-volume testing.

Objective: Analyze the strategy $\sigma_{Theranos}$, identify the sequence error, and derive the corrected strategy $\sigma_{correct}$.

Observed Strategy $\sigma_{Theranos}$:

Public claims and internal documents suggest the following sequence:

$$ \sigma_{Theranos} = \mathcal{T}_{claim} \circ \mathcal{T}_{build} \circ \mathcal{T}_{test} \circ \mathcal{T}_{regulate} $$

Expanded:

$\mathcal{T}_{claim}$: $T \equiv \text{Investor/Customer interest high}, O \equiv \text{Promote capability}, T \equiv \text{Funding/Adoption}, E \equiv \text{Scale}$.

$\mathcal{T}_{build}$: $T \equiv \text{Miniaturized device conceptualized}, O \equiv \text{Construct prototype}, T \equiv \text{Prototype functions}, E \equiv \text{Device ready}$.

$\mathcal{T}_{test}$: $T \equiv \text{Results generated}, O \equiv \text{Output results}, T \equiv \text{Results match venous gold standard}, E \equiv \text{Accuracy verified}$.

$\mathcal{T}_{regulate}$: $T \equiv \text{FDA approval}, O \equiv \text{Submit evidence}, T \equiv \text{Approval granted}, E \equiv \text{Commercial launch}$.

Analysis:

We derive the dependency constraints. A strategy is valid only if every Exit condition $E_i$ is a necessary precondition for the Test $T_{i+1}$ of the next unit, or if the system state satisfies the global exit predicate $E_{global}$.

In $\sigma_{Theranos}$, examine the link between $\mathcal{T}_{build}$ and $\mathcal{T}_{test}$.

$E_2$ of $\mathcal{T}_{build}$ is "Device ready."

$T_5$ of $\mathcal{T}_{test}$ is "Results match venous gold standard."

The error lies in the content of $E_2$. "Device ready" was defined operationally as "Device produces output." It was not defined as "Device produces accurate output." The sequence allowed the system to exit $\mathcal{T}_{build}$ based on a proxy exit condition (mechanical function) rather than the functional exit condition (analytical accuracy).

Furthermore, $\mathcal{T}_{test}$ was bypassed or corrupted. Evidence suggests the sequence often jumped from $\mathcal{T}_{claim}$ directly to $\mathcal{T}_{regulate}$, using results from unvalidated instances. This is a sequence violation: $\mathcal{T}_{test}$ was either omitted or its exit condition was falsified.

Correction:

We must insert a validation sub-strategy and enforce the exit test dependency.

$$ \sigma_{correct} = \mathcal{T}_{claim} \circ \sigma_{validation} \circ \mathcal{T}_{test} \circ \mathcal{T}_{regulate} $$

Where $\sigma_{validation}$ is a mandatory TOTE chain:

$$ \sigma_{validation} = \mathcal{T}_{cal\_gold} \circ \mathcal{T}_{range\_verify} \circ \mathcal{T}_{error\_budget} $$

$\mathcal{T}_{cal\_gold}$: $T \equiv \text{Correlation with gold standard } r > 0.95$ across full clinical range.

$\mathcal{T}_{range\_verify}$: $T \equiv \text{Error within CLIA limits for all analytes}$.

$\mathcal{T}_{error\_budget}$: $T \equiv \text{False positive/negative rate } < \text{threshold}$.

The exit condition $E_{validation}$ is $P_{accurate}$.

Only if $E_{validation}$ is true can $\mathcal{T}_{test}$ proceed to generate results for clinical use. The sequence is now load-bearing: the output of validation is a prerequisite for the trust in the test output.

Mechanism:

The correction works by enforcing the "Ground Truth" exit test. The mechanism requires that the validation operators have access to a gold standard reference, which may require high-volume venous samples. This creates a constraint: the strategy cannot eliminate venous draws entirely until the validation is complete. The insight is that strategy often demands the very thing it seeks to replace in order to prove it can replace it. The sequence must contain a "bridge" phase that uses the old method to validate the new, before the old method is discarded.

Failure Mode:

The failure mode of $\sigma_{correct}$ is time-to-market. The validation sequence extends the timeline and may require resources the organization lacks. The edge is "Insufficient Validation Resources." If the organization cannot access gold standard data at scale, $\sigma_{correct}$ stalls. The choice is then between launching unvalidated (fraud) or delaying indefinitely (bankruptcy). The framework names this trade-off explicitly; it does not hide it.

Problems

Drill Level

1. Notate the strategy for "Changing a Tire" as performed by a AAA-certified technician, using TOTE units. Define all predicates and operators.
2. Given $\sigma_A = \langle T_1, O_1, T_2, E_1 \rangle$ and $\sigma_B = \langle T_3, O_2, T_4, E_2 \rangle$, derive the condition under which $\sigma_A \circ \sigma_B$ is valid. Express this condition in terms of $E_1$ and $T_3$.
3. Identify the TOTE units in the strategy of "Making a Cup of Coffee" using a standard drip machine. Distinguish between the strategy of the machine and the strategy of the user.

Intermediate Level

1. Derive the strategy for "Negotiating a Hostage Release" based on the 2011 Norwegian Utøya response protocol. Identify the critical sequence constraint regarding "Building Rapport" versus "Bargaining Terms."
2. Consider the strategy of "Vaping" as marketed by early e-cigarette firms (2015-2018). Identify the failure mode where the exit test is misaligned with health outcomes. Show how this inverts the strategy into public harm.
3. Transfer the strategy of "A/B Testing" from digital marketing to "Public Policy Design" (referencing the UK BIT 2010 framework). Define the mapping function $\tau$ and the constraints on $\tau$.
4. Analyze the strategy of "The 2008 Bank Bailout" (TARP). Derive the sequence and identify the exit test. What is the failure mode if the exit test is "Systemic Collapse" rather than "Return to Solvency"?

Extension Level

1. Derive the strategy for "Deep Work" as practiced by authors of the Pomodoro technique (circa 1980s). Notate the sequence and identify the TOTE units related to "Distraction Management."
2. Consider the strategy of "Surgical Procedure: Appendectomy" using laparoscopic techniques. Derive the hierarchical strategy tree. Identify the sub-strategy for "Error Recovery" and its position in the sequence.
3. Translate the strategy of "Iterative Design" from software to "Urban Planning" (referencing the "Tactical Urbanism" movement, e.g., Park(ing) Day, 2005). Define the isomorphism and the constraints.
4. Analyze the strategy of "Climate Mitigation" as proposed in the Paris Agreement (2015). Derive the sequence of NDCs (Nationally Determined Contributions). Identify the failure mode related to "Temporal Discounting" in the exit tests.
5. Derive the strategy for "Personal Health: Weight Management" based on the "Habit Loop" model (Cue-Routine-Reward). Identify the failure mode where the reward subverts the exit test of long-term health. Show how to restructure the sequence.

Solutions

Solution 1

The strategy for changing a tire, as standardized by AAA, decomposes into three TOTE units:

$$ \sigma_{tire} = \mathcal{T}_{prep} \circ \mathcal{T}_{lift\_loosen} \circ \mathcal{T}_{change} $$

$\mathcal{T}_{prep}$: $T \equiv \text{Vehicle safe, spare available, tools present}$. $O \equiv \text{Engage parking brake, place wheel chocks, locate kit}$. $T \equiv \text{Safety check passed}$. $E \equiv \text{Ready}$.

$\mathcal{T}_{lift\_loosen}$: $T \equiv \text{Lug nuts accessible}$. $O \equiv \text{Break torque}$, $O \equiv \text{Jack vehicle}$, $T \equiv \text{Vehicle lifted, nuts loosened}$. $E \equiv \text{Ready}$.

$\mathcal{T}_{change}$: $T \equiv \text{Spare mounted}$. $O \equiv \text{Mount spare, tighten nuts, lower vehicle}$. $T \equiv \text{Vehicle on ground, nuts torqued}$. $E \equiv \text{Done}$.

The mechanism requires that $\mathcal{T}_{lift\_loosen}$ exits only when nuts are loosened while weight is on the wheel, a condition necessary to prevent rotation. The failure mode is skipping the loosening step until lifted, which makes torque application impossible.

Solution 2

For $\sigma_A \circ \sigma_B$ to be valid, the exit condition of $\sigma_A$ must imply the test condition of $\sigma_B$.

Condition: $E_1 \implies T_3$.

If $E_1$ is true, then