Inquiry-Change Learning Loop

Version

v5 · 2026-09-07 · History

Prime #

1383

Origin domain

Research Methods

Subdomain

action research → Research Methods

Related primes

Feedback, Iteration, Learning

Core Idea

An inquiry-change learning loop is an investigation in which finding out and changing things are the same act. A model that is currently held — part explanation, part working practice — selects a deliberate change in a live setting; the consequences of that change are recorded as evidence about how the setting works, not merely as a report on whether the change was delivered; that evidence durably revises both the model and the action the model will next select; and the revised model travels forward together with the setting the change has already altered. ^[1]

What separates the loop from investigation in general is that the inquirer sits inside the system under study and pays for knowledge in irreversible alterations to it. A bench experimenter borrows a system, perturbs it, and hands it back; a loop of this kind permanently rearranges the ward, the classroom, the assembly line, or the service it is learning about, so the next cycle asks its question of a setting the previous cycle built. Knowledge and change are not two deliverables that happen to arrive together — each is the other's instrument. ^[2]

Six commitments carry the load: a retained model that selects action; deliberate change in a real operating setting; systematic observation of the consequences; durable revision of the understanding; use of that revision to choose what to do next; and recurrence with practical and epistemic state carried across together. Remove any one and something else results — regulation, repetition, a one-off pilot, a routine audit — that can be hard to tell apart from outside. The target may also move: the problem frame is itself revisable, and a pass whose main product is a better question has not failed. ^[3]

Structural Signature

held model → deliberate change in a live setting → consequence observed as evidence → durable revision of model and next action → both states carried forward → next change

The formula is a cycle rather than a pipeline, and its distinguishing arrow is the one that leaves the setting altered. Each pass hands its successor two things at once, a changed world and a changed understanding of it, which is why the same roles can be checked in a hospital, in a compiler's tuning process, and in a fisheries quota regime without translating between them. ^[4]

Recurring features:

• A model that exists and is committed to before the change, in a form specific enough to be embarrassed by an outcome.
• Change enacted where consequences are real and borne by the setting's occupants, not in a sandbox built to be discarded afterwards.
• Observation instrumented ahead of the change, so that the record functions as evidence rather than as retrospective justification.
• Revision that outlives the cycle — written into the model, the standard work, or the tuned parameters — rather than living only in the memory of whoever ran the pass.
• Two kinds of state carried in lockstep: the altered setting and the updated understanding, neither reset between passes.
• Warrant that accrues across passes, so no single cycle has to carry the whole inferential burden alone. ^[5]
• A revisable target: the loop may narrow a fixed gap, but it may equally close a pass by redefining which gap was worth closing.

What It Is Not

The name invites a reading far broader than the thing it picks out. "Learning by doing" is not enough: a crew that gets better at a task through repetition is accumulating skill, but unless someone held a model, changed something on purpose, watched what followed, and wrote the answer back into the model, no loop of this kind ran. Trial and error is not sufficient either, because error-driven search can proceed happily without any retained explanation of why a given variant failed. ^[6]

The loop is not confined to organisations, to the social sciences, or to human-run methodologies, even though its best-known named forms are all three. Nothing in the structure requires a committee, a facilitator, or a published write-up; it requires only that a change-selecting model be revised by the consequences of its own changes and then carried on. Nor does the loop entail participation. Whether the people affected by a change help design it is a real and often decisive question, but it is a commitment layered onto the structure rather than a part of it.

The loop also promises nothing about improvement. It is a way of learning, not a guarantee of getting better; a sequence of passes can leave a setting worse off and still have been a properly run loop, provided the model absorbed what went wrong. It does not require convergence on a stable answer, does not require later cycles to be smaller than earlier ones, and does not require one continuous project — a loop can be resumed years later by different people if the model and the changed setting were both preserved well enough to restart from.

Finally, it is not a licence for undisciplined tinkering. Changing things and seeing what happens satisfies none of the observation, revision, or carry-forward commitments; it is the loop's most common counterfeit precisely because it produces the same visible activity and the same air of energetic engagement.

Broad Use

Clinical and health services. Improvement cycles on a hospital unit, treatment protocols amended as a cohort progresses, and pragmatic trials embedded in ordinary care all push a change through the setting under study, then read that setting to choose the next change. ^[7]

Software and operations. Deploy-and-observe practice — ship to real traffic, instrument the result, revise the internal account of how the system behaves, and let that revision pick the next change — is this structure with a build pipeline attached. So is capacity tuning, where each configuration change simultaneously serves live load and probes the load's response curve.

Education and instructional design. Design-based research revises a teaching design and the theory behind it in the same classroom across successive enactments, and the classroom that receives the third enactment is one the first two reshaped.

Agriculture, ecology, and resource management. Adaptive management of a watershed, a forest stand, or a fishery sets harvest or flow levels partly to meet an objective and partly to resolve which model of the system is right, then carries the resolved model into the next season's altered setting. ^[8]

Product and market discovery. Continuous discovery treats each release as an assay on demand, with the market that responds to the fourth release having been partly constituted by the first three.

Engineering commissioning and control. Bringing a plant, a network, or a controller into service involves deliberate excursions whose purpose is jointly to move the system toward its operating point and to identify the parameters that govern how it gets there.

Clarity

The confusion this prime dissolves is the assumed choice between doing something and finding something out. Practitioners inherit a division in which research happens under protected conditions and operations happen for real, so a change must be either an experiment or an implementation. That division makes two kinds of waste look inevitable: operational changes that generate no transferable understanding, and studies whose findings arrive about a setting that no longer exists. Naming the loop shows that a single change can be both, and that treating it as both is a design decision taken in advance rather than a bonus noticed afterwards. ^[9]

It also settles what counts as a result. If inquiry and change are separate activities, a change that fails is a failure and a study that finds nothing is a disappointment. Inside the loop the unit of success is the pass, not the change: a pass that enacts a change, records that the model's prediction was wrong, and revises the model has delivered precisely what it exists to deliver. The corresponding failure is quieter and easier to miss — a pass whose change worked, whose consequences were never read, and whose model therefore finishes exactly where it started.

Manages Complexity

What the loop lets you stop tracking is the complete causal picture of the setting. Because the model's job is only to select the next change and to be wrong in a way that shows, it may be crude, local, and openly partial; the setting supplies the detail the model omits, and it supplies it by responding. An analyst who would otherwise need a full specification of a ward, a market, or a plant before acting needs instead a model precise enough to make one falsifiable commitment. ^[10]

It also retires the space of untried alternatives. A branching analysis of what might have happened under every unchosen option outgrows inspection almost immediately; the loop replaces it with two carried objects — the setting as it now stands and the model as now revised — which between them compress the entire history into something you can look at. Nothing before the current pass has to be replayed, because whatever earlier passes established is either written into the model or built into the setting. The price of that compression is that the discarded branches are genuinely gone, which is why a record of what was tried and rejected must be kept deliberately if it is to be kept at all.

Abstract Reasoning

The loop licenses a specific drill, run once per pass. State the model in force before the change, in language that could turn out to be wrong. Derive what the change should produce, and name in advance the observation that would count against it. Enact the change and record what actually happened. Then partition the discrepancy across three sources: the model was wrong, the change was not executed as specified, or the setting moved for reasons of its own. Only the first licenses revision of the model; the second sends the same change back for another attempt; the third demands better instrumentation before either conclusion is drawn. Finally, derive the next change from the revised model rather than directly from the discrepancy — the step most often skipped, and the one that separates cumulative inquiry from chasing the last surprise. ^[11]

The same structure supports a classification test. Given a candidate process, ask which of the six roles is unfilled. A change with no antecedent model is intervention without inquiry. A model revised by observation with no change enacted is ordinary empirical study. Recurrence with no carried epistemic state is repetition. Naming the missing role explains a case more usefully than ruling the whole of it in or out.

Knowledge Transfer

What travels between substrates is the role structure and its failure modes, not the contents of any pass. Carry across the requirement for a committed prior model, the discipline of naming a disconfirming observation before acting, the coupling of practical and epistemic carry-forward, and the three-way partition of discrepancy; someone who has run improvement cycles on a ward holds genuinely transferable equipment for tuning a control system, and the transfer runs just as well in the other direction — though it has to be made deliberately, since a structure shared under dissimilar surfaces is rarely noticed unprompted. ^[12]

What does not travel is nearly everything that makes a pass concrete. Cycle time varies by orders of magnitude between substrates and determines what kind of learning is even possible within a pass. Tolerance for irreversibility differs sharply: a configuration change can be rolled back, a felled stand of timber and a discharged patient cannot, and a loop designed for the reversible case will be reckless in the other. Measures, effect magnitudes, and local causal texture do not port at all. Neither does the governance arrangement — who may authorise a change, who bears its consequences, and who is entitled to declare the model revised are settled by each substrate's institutions, and importing those answers along with the structure is how the pattern picks up commitments it does not actually carry.

Examples

Formal/abstract

Consider a controller regulating a plant whose gain is unknown. Its input has two jobs that pull against each other: drive the output toward the setpoint, and excite the plant enough that the input-output record identifies the unknown gain. A purely regulating input settles the plant and, in settling it, stops informing — a quiet plant reveals nothing, so the estimate stops improving at exactly the moment the controller starts performing well. A purely probing input identifies the gain quickly while holding the output far from where it should be. The standard resolution designs the input to carry a probing component sized by current uncertainty: large while the estimate is poor, shrinking as it sharpens. Each interval the controller acts, records, updates the estimate, and hands both the plant's new state and the sharpened estimate to the next interval. ^[13]

Mapped back: the model is the parameter estimate, the deliberate change is the input including its probing component, the evidence is the measured response, the durable revision is the updated estimate, and the carried state is the pair — plant state and estimate — moving forward together. The pull between regulating and probing is the formal shadow of the prime's defining feature: one action must both do the work and produce the knowledge, and optimising it for either purpose alone degrades the other.

Applied/industry

A hospital unit with a persistently high central-line infection rate starts from an explicit model: infections are driven mainly by variation in insertion technique rather than by patient case mix. That model selects the first change, a mandatory insertion checklist on one shift. Observation is instrumented before the change — line-days, infection events, and checklist completion are recorded separately, so a fall in the rate can be told apart from a fall in reported events. The rate drops, but checklist completion also drops after week three, and the case-mix record shows the fall concentrated on a shift whose admission profile changed at the same time. The model is revised: technique variation matters, sustained compliance is itself a mechanism, and case mix is not the negligible term the first model assumed. The second pass acts on what the first pass built — a unit that now has a checklist, staff who know they are being counted, and a revised model — by moving the checklist from mandate to two-person verification, and adds case-mix stratification to the measurement. ^[14]

Across four passes the unit ends with a lower rate and, separately, with a defensible account of which components carried the effect and which decayed. Neither product could have been had without the other: a study without the changes would have described a unit nobody was improving, and the changes without the measurement would have left the unit unable to say what to protect when staffing shifted. ^[9]

Mapped back: every pass fills all six roles, and the case exposes what distinguishes this prime from a trial. The population in the fourth pass is not the population that entered the first; it has been reshaped by the loop's own earlier changes, so the clean comparison a randomised design would have preserved is simply not available. What the loop offers instead is warrant assembled across passes, each cycle's carried model narrowing what the next one has to explain.

Structural Tensions

T1 — Intervening consumes the counterfactual. Every pass spends the comparison it would need in order to prove itself. Because the change lands in the only copy of the setting there is, the unchanged version stops existing at the moment of the change, and later passes run on a world the earlier ones made. The warrant available is therefore cumulative and mechanistic rather than contrastive: consistency across passes, dose-response within one, and a model that keeps predicting. Anyone who wants contrastive proof must build a comparison in deliberately, and pay for it in reach.

T2 — Loop speed against depth of learning. Short cycles multiply passes and make error cheap, but they truncate the window in which slow consequences appear; a change judged good in a two-week pass may be exactly the one that degrades over a quarter. Long cycles see the slow effects and pay by learning almost nothing per unit of calendar time, while the setting drifts far enough that attribution collapses. There is no cycle length that is right for a setting, only one right for the mechanism under study — and a single loop usually studies several mechanisms at once.

T3 — Who owns the change when the inquirer is also the actor. The role fusion that gives the loop its power also removes the person who would ordinarily check the work. An inquirer who chose the change, executed it, and now reads its evidence has every incentive to read generously and no structural counterparty who does not. Splitting the roles restores scrutiny but reopens the gap the loop exists to close, because whoever holds the model stops feeling the consequences. Real arrangements sit somewhere between, and where they sit is a design choice rather than a staffing detail.

T4 — The system adapts to being studied. A setting under observation is not the setting that was there before. Staff who know a metric is counted attend to it; a market that has absorbed three releases anticipates the fourth; a stand stressed by last season's harvest answers this season's harvest differently. The loop therefore measures a moving compound of the mechanism it wants and the setting's reaction to being measured, entangled by design rather than by sloppiness. Repeating an earlier pass to check it is not replication, because the second run enters a setting the first one taught.

T5 — Carried state compounds error as readily as insight. The carry-forward that makes warrant accumulate cannot tell a good revision from a bad one. A model wrong in the first pass selects the second pass's change, which alters the setting in ways that make the wrong model fit better, and by the fifth pass the loop is confidently exploring a region its own first mistake created. Nothing internal detects this, because every pass is locally consistent. Only a periodic check against something the loop did not itself produce breaks the compounding, and that check lies outside the loop.

T6 — Local warrant against transportable claim. What the loop knows best is the setting it remade, which is precisely the setting least like anyone else's. Four passes of tuning yield a model finely fitted to one unit's staffing, equipment, and history, and part of that fit is to conditions the loop installed itself. Publishing the result as a general finding overstates it; withholding it as merely local wastes it. An honest report separates the mechanism that might travel from the configuration that will not, and the loop's own evidence cannot make that separation.

Structural–Framed Character

Inquiry Change Learning Loop sits at the midline of the structural–framed spectrum — mixed-framed, aggregate 0.5, with every diagnostic at 0.5. The skeleton is a cycle with carried state: a retained practical and explanatory model selects a deliberate change in a real setting; the consequences are measured as evidence, not implementation outcomes; both the understanding and the next action are durably revised; and changed practical state and revised understanding move forward together instead of resetting, so warrant accumulates even as the setting evolves.

Human-practice-bound at 0.5 explains the reading best: the loop presupposes an inquirer who holds the model, chooses the change, and carries state across passes, and its named realizations — PDSA, build–measure–learn, design-based research, pragmatic trials, action research — are methodologies people run. What keeps it off the framed pole is that the required roles stop short of the participatory and social-science commitments belonging to action research.

Vocabulary travels partially at 0.5 — plan-do-study-act and build–measure–learn carry an improvement-and-design register, while model–change–evidence–revision states the same roles without it. Evaluative weight is 0.5: deliberate change, systematic observation, and durable revision read as prescriptions as much as descriptions. Institutional origin is 0.5, its homes being named methodologies rather than a formal regularity. Import-versus-recognize is 0.5: a field already running such a cycle recognizes the structure; elsewhere invoking the prime imports the methodology framing.

The grade asks that every role hold before the name is used: feedback supplies a return path without durable inquiry, iteration repeats without measuring evidence, and nothing here delivers the assignment, control, or isolation of experimental design.

Substrate Independence

Inquiry-Change Learning Loop is a highly substrate-independent prime — composite 4 / 5 on the substrate-independence scale. Its core is a cycle whose distinguishing arrow leaves the world altered: a held model selects a deliberate change in a live setting, the consequences are recorded as evidence about how that setting works, the model and the next action are durably revised, and both states travel forward into the following pass. The roles check out unchanged in plan-do-study-act cycles, build-measure-learn iteration, design-based educational research, continuous product discovery, pragmatic trials, and action research. What keeps it off the top rung is a commitment the signature cannot drop: someone must hold a revisable model and sit inside the system being altered, which confines instances to managed human practice.

• Composite substrate independence — 4 / 5
• Domain breadth — 4 / 5
• Structural abstraction — 4 / 5
• Transfer evidence — 4 / 5

Relationships to Other Abstractions

Current abstraction Inquiry-Change Learning Loop Prime

Parents (3) — more general patterns this builds on

• Inquiry-Change Learning Loop is part of Feedback Prime

  The loop contains feedback because observed consequences must return to alter the next model-guided action.

• Inquiry-Change Learning Loop is part of Iteration Prime

  The loop contains iteration because revised practical and epistemic state is carried into a subsequent pass.

• Inquiry-Change Learning Loop is part of Learning Prime

  The loop contains learning because evidence must durably update the retained model or capability that selects later action.

Children (1) — more specific cases that build on this

• Action Research Domain-specific is a decomposition of Inquiry-Change Learning Loop

  Action research is the participatory social-science form of the neutral inquiry-change learning loop, adding researcher-participant overlap, a dual deliverable, and a participation-rigor frontier.

Hierarchy paths (4) — routes to 4 parentless roots

• Inquiry-Change Learning Loop → Feedback

Show alternative paths (3)

Neighborhood in Abstraction Space

Inquiry-Change Learning Loop sits in a sparse region of abstraction space (81^st percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely rather than landing on a neighbor.

Family — Event Phases & Staged Recovery (10 primes)

Nearest neighbors

• Regime Change — 0.70
• Learning — 0.70
• Progressive Refinement from Core Model — 0.69
• Learning Curve Effects — 0.69
• Refinement — 0.69

Computed from structural-signature embeddings · 2026-09-10

Not to Be Confused With

The nearest neighbour is Feedback, which this prime contains rather than competes with. Feedback requires only that a system's outputs influence its inputs; a thermostat, a servo, and a price mechanism all qualify, and none of them holds a model, entertains an explanation, or is any wiser at the end of the day than at the start. The loop demands that the returning signal do epistemic work — that it change a retained account of how the setting behaves, and that the changed account survive to choose the following action. Cut that requirement and you have regulation: responsive, possibly excellent, and permanently ignorant. A control system that hunts around a setpoint for years exhibits feedback in full and never once conducts an inquiry.

Iteration is the second constituent, and the second thing mistaken for the whole. Iteration repeats a step with some state carried between passes, which the loop also needs; what iteration does not require is that anything be measured or that any understanding be revised. A team on its fourteenth redesign has iterated fourteen times, and if none of those passes recorded consequences against a prior expectation, the fourteenth designer knows only what the first one did plus a longer list of things that felt unsatisfactory. Iteration supplies recurrence and carried state; it is indifferent to whether the carried state includes anything learned.

Learning is the third constituent and the most easily over-credited. Learning is the durable, experience-driven update of an internal state that later behaviour depends on, and it happens constantly without deliberate change in a real setting: from observation, from instruction, from simulation, from other people's mistakes. The loop is the special case where the experience that drives the update was manufactured on purpose by the learner, in the setting the learner also has to keep running. That constraint is what generates the loop's characteristic problems, none of which trouble learning in general.

Experimental Design is the sharpest contrast and the one most often used as a stick. Experimental design earns causal identification through assignment, control, and isolation: a comparison group that did not receive the change, randomisation to break confounding, and enough insulation that the manipulated variable is the only one moving. The loop deliberately forfeits all three. Its change goes to the live setting rather than to a treatment arm; the untreated version of that setting ceases to exist; and later passes are confounded by earlier ones on purpose, since carrying the change forward is the point. What it buys with that forfeit is relevance to conditions no protected design reproduces, and warrant assembled across passes rather than delivered by one.

Intervention in the causal-inference sense is a single operation, not a cycle: fixing a variable's value while severing its normal upstream causes and preserving its downstream effects. It is the atom the loop's action step is built from, and it is silent about everything the loop adds. Intervention says nothing about whether a model preceded the act, whether anyone measured what followed, or whether the next act differs from this one. A one-shot policy change enacted on a hunch and never evaluated is a perfectly good intervention and not the beginning of a loop.

Monitoring is the mirror image of the missing-change failure. Monitoring watches a system's state continuously to detect deviation and trigger a response, and it can be highly instrumented, statistically careful, and entirely passive with respect to the model it uses. A monitored system generates exactly the observation stream a loop would want, but the deviations it detects are treated as events to be handled rather than as evidence about whether the model that defined "normal" was right. Add a deliberate model-selected change and a revision path and monitoring becomes the observation stage of a loop; without them it remains surveillance.

Finally, Intervention-Induced Model Invalidation names a hazard the loop lives inside rather than a rival structure. That prime describes what happens when a model is used to choose an intervention and the intervention alters the very data-generating process the model described, so its pre-intervention parameters no longer transport. For most modelling practice this is a failure to be avoided; for this loop it is the standing condition, since every pass invalidates part of the model that selected it. The difference is one of stance: invalidation is the diagnosis, and the loop is the discipline built to keep working while it is true.

Solution Archetypes

No catalogued solution archetypes reference this prime yet.

References

[1] Lewin, Kurt. "Action Research and Minority Problems". Journal of Social Issues 2(4), 1946. Formulates research in which planning, action, and fact-finding about the result of the action form one recurring spiral inside a live social setting. registry ↩

[2] Schon, Donald A. The Reflective Practitioner: How Professionals Think in Action. Basic Books, 1983. Characterises the practitioner's on-the-spot experiment, in which the inquirer stands inside the situation and alters it in the act of investigating it. registry ↩

[3] Argyris, Chris, and Donald A. Schon. Organizational Learning: A Theory of Action Perspective. Addison-Wesley, 1978. Distinguishes single-loop learning, which corrects error inside a fixed frame, from double-loop learning, which revises the governing variables themselves, so a pass whose product is a reframed problem counts as learning rather than failure. registry ↩

[4] Deming, W. Edwards. The New Economics for Industry, Government, Education. MIT Center for Advanced Engineering Study, 1993. Sets out the plan-do-study-act cycle in which a prediction is made, a change is run, the result is studied against the prediction, and the accumulated knowledge is carried into the next turn of the cycle. registry ↩

[5] Langley, Gerald J., Ronald D. Moen, Kevin M. Nolan, Thomas W. Nolan, Clifford L. Norman, and Lloyd P. Provost. The Improvement Guide: A Practical Approach to Enhancing Organizational Performance. 2^nd ed. Jossey-Bass, 2009. Specifies measurement planned before the change and degree of belief accumulating across a sequence of cycles rather than within any single one. registry ↩

[6] Arrow, Kenneth J. "The Economic Implications of Learning by Doing". The Review of Economic Studies 29(3), 1962, 155-173. Models learning by doing as productivity gain accruing from cumulative experience, a mechanism that carries no retained explanatory model of why any particular attempt succeeded or failed. registry ↩

[7] Taylor, Michael J., Chris McNicholas, Chris Nicolay, Ara Darzi, Derek Bell, and Julie E. Reed. "Systematic review of the application of the plan-do-study-act method to improve quality in healthcare". BMJ Quality & Safety 23(4), 2014, 290-298. Reviews the use of iterative PDSA cycles on hospital units, where a change is pushed through the setting under study and the setting is then read to choose the next change. registry ↩

[8] Walters, Carl J. Adaptive Management of Renewable Resources. Macmillan, 1986. Establishes management actions chosen partly to meet a harvest or flow objective and partly to discriminate among competing models of the resource system. registry ↩

[9] Institute of Medicine. Best Care at Lower Cost: The Path to Continuously Learning Health Care in America. National Academies Press, 2013. Argues against the separation of research from operations and for care delivery designed so that the same activity produces both the improvement and the evidence about it. registry ↩a ↩b

[10] Box, George E. P. "Science and Statistics". Journal of the American Statistical Association 71(356), 1976, 791-799. Presents the iterative model-data loop in which a model known to be wrong earns its keep by being specific enough to produce a checkable discrepancy. registry ↩

[11] Rossi, Peter H., Mark W. Lipsey, and Howard E. Freeman. Evaluation: A Systematic Approach. 7^th ed. Sage, 2004. Distinguishes theory failure from implementation failure as separate explanations of a disappointing result, each carrying a different remedy. registry ↩

[12] Gick, Mary L., and Keith J. Holyoak. "Analogical problem solving". Cognitive Psychology 12(3), 1980, 306-355. Shows that a structurally analogous source is rarely applied to a surface-dissimilar target without an explicit hint, so transfer of a shared role structure is a deliberate act rather than an automatic one. registry ↩

[13] Feldbaum, A. A. "Dual Control Theory, I-IV." Automation and Remote Control 21, 1960, 874-880; 22, 1961, 1033-1039, 1-12, 109-121. Establishes that one control input must serve both regulation and identification, and that the probing component has to be sized against current estimate uncertainty. registry ↩

[14] Pronovost, Peter, Dale Needham, Sean Berenholtz, David Sinopoli, Haitao Chu, Sara Cosgrove, et al. "An Intervention to Decrease Catheter-Related Bloodstream Infections in the ICU." New England Journal of Medicine 355(26), 2006, 2725-2732. Reports a checklist-based central-line intervention across Michigan ICUs with infection rates measured per catheter-day before and across the rollout, the instrumented change-and-read structure the worked example follows. registry ↩