Double-loop learning is what happens when a person or organisation does more than repair an error: it examines and, when warranted, changes the assumptions or governing rules that made the error likely. If a school repeatedly misses an assessment deadline, sending extra reminders is a useful single-loop correction. Asking whether the deadline, workload design or definition of success is itself sensible is a different question. Neither response is automatically superior in every situation.
The distinction comes from Chris Argyris and Donald Schön’s work on organisational learning. A single loop modifies actions to meet an accepted objective; a double loop tests whether the objective, rule or underlying theory of action should change. Real double-loop learning requires observable evidence that a governing assumption was challenged and decisions changed, not simply a meeting in which people used the word reflection. This article shows how to recognise each loop, gather evidence without blaming people, run small reversible tests and judge what really improved.
The original 1977 Argyris article on double-loop learning, his 1991 analysis of defensive learning and his 2002 research on case-based double-loop teaching anchor the explanation. All student, team, timetable and business numbers below are invented teaching cases with explicit denominators, not reports about eduKate or any real organisation.
Contents: choose a learning route
Start with the two questions, the original theory, espoused and enacted rules, governing variables, the value of single-loop learning and causal mapping.
For applied diagnosis, visit the 90-assignment example, misleading indicators, defensive routines, intent and consequence, private inferences, decisive meetings and safe evidence.
For practical action, use bounded pilots, causal limits, rate accounting, decision logs, decision rights, personal learning, group learning and knowledge transfer.
Finish with routine traps, classroom choices, common breakdowns, ethical boundaries, the transfer exercise and answer, action cycle, FAQs, research sources and verified reader pathways.
One error, two fundamentally different questions
Start with a visible mismatch between the intended result and what actually happened. Perhaps students finish practice questions but cannot explain them independently. A single-loop response asks which immediate actions should change so the current measure improves: add an explanation sheet, extend practice time or offer feedback. These can be effective and should not be belittled simply because the loop is single.
The second question asks whether the programme is pursuing the right measure at all. If its governing assumption equates fast completion with understanding, then extra drilling may strengthen a misleading indicator. Double-loop learning challenges that assumption and considers measuring independent reasoning, delayed recall or novel-problem transfer. It may then revise what lessons optimise, not merely how quickly students complete the same worksheets.
The two loops are distinguished by the level of correction, not by how dramatic the change looks. Installing expensive software may leave the objective and governing rule untouched, making it single-loop change. Quietly changing a criterion from worksheet speed to reliable explanation may alter the governing theory, even without buying a new tool. Size of budget and organisational status are poor classifiers.
A practical diagnosis records the original intended outcome, the current action, the rule that selects the action and the evidence behind the rule. The intervention is double-loop only when the rule or governing value is actually examined and, if evidence warrants, revised. Mere dissatisfaction, brainstorming or criticism of a manager is not enough.
Why Argyris called it learning rather than just change
Chris Argyris examined how organisations detect and correct mismatches between intentions and outcomes. In his 1976 article on single- and double-loop models of decision making and the September 1977 Harvard Business Review piece, the distinction concerns corrections made inside existing governing values versus corrections that put those values into question. His framework also asks why intelligent, well-qualified people can defend arrangements that repeatedly disappoint them.
An organisation may hold an elaborate discussion about a failure yet learn little if its conclusions cannot change decisions. Someone must translate observations into revised action strategies, rules or evidence requirements. The changed strategy must then be tried and its consequences examined. Otherwise the organisation has a new vocabulary for the old routine rather than demonstrable organisational learning.
The approach matters because organisations can be skilled at improving the performance of a failing system. A sales group might learn to process objections faster while never asking whether the product fits customers’ needs. A school might reduce the number of late submissions by simplifying assignments in ways that diminish the learning objective. Fast local corrections can coexist with an untested global assumption.
The original work is a conceptual and research tradition, not a guarantee that any workshop labelled double-loop will produce measurable gains. Later research has noted that the idea is popular but hard to implement consistently. A responsible reader should treat the framework as a disciplined method of questioning and testing organisational reasoning, with evidence quality as a separate obligation.
Espoused theories and theories actually used
People often state ideals that differ from the rules revealed by their decisions. A department may say it rewards accurate reporting but repeatedly praises teams that meet a deadline even when they conceal uncertainty. The publicly stated rule is its espoused theory. The priorities actually governing decisions constitute a theory-in-use, inferred cautiously from repeated behaviour rather than a single embarrassing incident.
These two levels can diverge without anyone deliberately lying. Incentives, social expectations, time pressure and vague instructions may cause behaviour to follow a rule people would reject if it were stated plainly. A leader may endorse independent reasoning but interrupt every meeting to provide the final answer. Observing the pattern creates an opening for an honest question about what behaviour the environment reinforces.
Do not try to infer someone’s private motives from one error. Compare repeated decisions, recorded criteria and what happens when someone raises contrary evidence. Ask people which constraints they believe they face. The purpose is to make the actual governing logic discussable, not to declare that a colleague’s intentions are false because an observer dislikes the outcome.
Suppose a team states that all safety issues may be escalated immediately, but its case logs show repeated pressure to close reports before review. The double-loop question concerns incentives, authority and accepted evidence for closure. It is not enough to send the same escalation reminder again if the unstated rule penalising escalation remains in force.
The governing variables behind an action
A governing variable is a value, constraint or objective that a decision process tries to preserve. Examples include never missing a shipment date, protecting a person’s dignity, maximising test scores or controlling budgets. Some governing variables are explicit policies; others emerge from repeated decisions. They can be legitimate, harmful, conflicting or simply outdated as circumstances change.
The distinction is clearer when actions are written as conditional rules. If weekly complaints exceed a threshold, add customer-service shifts. That rule can be executed perfectly while leaving unexamined the assumption that complaint handling, rather than product defects, is the right bottleneck. A double-loop inquiry asks what the threshold represents, why it was chosen and whether the desired outcome should be product reliability instead.
Changing a governing variable is not equivalent to abolishing standards. A safety-critical requirement to verify identity before releasing records may be ethically and legally non-negotiable. In that setting, a team can question workflow and assumptions while keeping the core protection intact. A method that encourages questioning does not override professional, legal or safeguarding duties.
To make governing variables visible, write four items in one decision record: the observed outcome, the action selected, the rule connecting them and the value the rule protects. If colleagues disagree about the last item, that disagreement is useful information. It means the organisation may be enforcing an unstated priority that needs clarification before a correction is chosen.
Single-loop learning is not a mistake
Single-loop corrections are essential when the objective is valid and the cause is understood. If a working temperature monitor has a loose cable, fixing the cable is better than spending a month reconsidering the meaning of temperature. If students understand a concept but misread one instruction, clearer wording can be sufficient. The aim is fit between intervention and diagnosis.
An organisation that treats every issue as philosophical may become slow and evasive. It can hold repeated workshops about mission and values while failing to restore a broken process. Double-loop learning adds a level of reasoning when the governing assumptions may be wrong; it does not eliminate the need for routine feedback, competent maintenance and timely help.
Use a proportional decision rule: correct an obvious local fault, document it and watch for recurrence or patterns across contexts. Escalate to assumption review when corrections fail, side effects grow, the metric conflicts with the stated purpose or knowledgeable people consistently see contrary evidence being dismissed.
The two loops can operate together. A hospital-like fictional workflow can fix a clearly broken reminder today and separately examine whether an unsafe workload standard creates persistent missed checks. Immediate protective action should not wait for a strategic review. The deeper review should not be abandoned merely because a temporary fix improved the next day’s count.
Make a causal map before changing rules
To avoid vague brainstorming, separate outcomes from actions, rule-selection criteria, incentives and contextual constraints. For a missed deadline, outcomes might include actual completion times and quality; actions might include reminder messages; rules might define fixed cutoff times; constraints might include access to equipment or a conflicting timetable.
Write possible causal paths as hypotheses, not accusations. For example, an early cutoff may increase missed work among learners who can access a device only after a particular hour. A reminder may help some of those learners, but it will not remove the access constraint. The causal map should include alternative explanations, such as task difficulty, illness or an inaccurate record of when work was available.
Distinguish what was observed from what someone inferred. We observed 18 late submissions in a set of 90 eligible assignments. We infer that the cutoff is the primary cause only if additional evidence supports that link. Without checking the timing of access and the nature of the tasks, changing the deadline is a plausible experiment rather than a demonstrated cure.
Good learning begins when the group can say which observation would change its mind. If evidence that devices were accessible all afternoon would weaken the access hypothesis, specify how to check it. If the hypothesis cannot be revised under any possible finding, it is functioning as a protected belief rather than a testable explanation.
Numerical worked case: late work and its denominator
Consider an invented tutorial programme tracking ninety eligible assignments during a comparable teaching interval. Eighteen are submitted after the agreed cutoff. Its late-submission rate is 18/90=0.20, or 20 percent. This percentage is not a judgement about any particular learner: it is a count of events divided by a clearly defined population of assignment opportunities.
The team first introduces reminders while leaving the cutoff unchanged. In a second invented interval, twelve of ninety assignments are late, giving 12/90≈13.33 percent. The rate is 6.67 percentage points lower than the first interval. Relative to the original 20 percent rate, that is about a one-third reduction, because (20−13.33)/20≈0.3335. Relative improvement and percentage-point change are different quantities.
Now the team tests a changed governing rule: submission windows reflect documented device access and still preserve equivalent feedback time. In a third invented interval, six of ninety are late, or 6/90≈6.67 percent. The sequential arithmetic is clear, but it does not demonstrate that the rule change alone caused the improvement. The students, tasks, teacher effort and other circumstances might also have changed.
The double-loop component is not the lowest percentage. It is the explicit reconsideration of what the deadline is for, which constraints learners face and how success should be defined. A fair evaluation would also inspect learning quality, feedback timeliness and whether different learner groups benefited. If the new window merely made work so easy that everyone finished quickly, the late-rate improvement would not prove educational value.
Record the counts and definitions before drawing a lesson. Were there really ninety eligible assignment opportunities in each interval? Were extensions included or excluded by a consistent rule? Were incomplete submissions counted as late or missing? These denominator choices can change a headline without changing any behaviour, so they belong in the decision record.
When optimising one indicator damages the purpose
A metric is useful only within the purpose it represents. A support desk may reduce average call duration by hurrying difficult callers off the line. A class may increase the number of completed exercises by assigning simpler work. These moves can improve the measured number while worsening the user outcome that justified the work in the first place.
Before choosing a new correction, identify at least one balancing measure. If the objective is learning, pair completion with independent transfer or delayed recall. If the objective is customer service, pair response time with correct resolution and repeat contacts. If the objective is safe delivery, pair throughput with safety checks and error rates.
A better indicator is not always one scalar number. A small scorecard can preserve different aspects of a decision, such as timeliness, accuracy, equity and cost. The weights among them are themselves a governing choice and may need revision when the mission changes. A seemingly objective composite score can conceal value judgements behind arithmetic.
Ask whose perspective is missing from the indicator. Students, front-line staff and service recipients may notice harms that the organisation’s dashboard never records. Include observations from those affected while still checking their generalisability. Double-loop learning can challenge the metric definition without treating every anecdote as conclusive evidence.
Why a sensible organisation can defend a failing rule
Argyris examined defensive routines: patterns of reasoning and communication that protect people from embarrassment or threat but obstruct public testing of assumptions. A team can sincerely want better outcomes while discouraging precisely the questions that would reveal why those outcomes are poor. This is not a claim that every disagreement is psychological defensiveness.
Consider a manager who says every assumption may be challenged, then consistently responds to criticism by asking who made the mistake rather than what the process rewarded. Team members may learn to present problems as isolated human errors. Their reports become polished but less informative, and the organisation’s apparent agreement increases while its evidence quality declines.
Defensive routines may be interpersonal, but they can also be structured into systems. A performance bonus based only on on-time closure encourages closing difficult cases prematurely. A public ranking of teams may encourage selective reporting. Correcting these incentives may do more than lecturing staff about openness.
An effective response does not accuse a person of being defensive as a substitute for analysis. It creates permission and procedural protection to examine a claim, tests the evidence for and against it, and makes decision criteria transparent. Leaders can model uncertainty by showing which observations would cause them to revise their own position.
Separate someone’s intention from the impact of a routine
One person may intend to help students become independent while a routine trains them to wait for the teacher’s answer. Another may intend to protect quality while a review system discourages early error reporting. Double-loop inquiry does not require assuming malicious motives; it asks whether the way work is organised consistently produces consequences different from the intended ones.
Use concrete examples in conversation. Instead of asserting that a colleague does not trust staff, identify the recurring occasions when instructions were revised after staff had already decided. Ask whether the rule requiring all decisions to return to one approver is necessary, and what risk it is supposed to control. These are observable subjects for discussion.
People’s explanations are data too. A person may reveal a constraint observers did not see, such as a legal sign-off obligation or a previous fraud incident. The inquiry becomes more accurate when the proposed governing rule can incorporate that context rather than treating resistance as proof that the person has something to hide.
An outcome can also be mixed. A central approval rule may reduce unauthorised decisions and simultaneously slow harmless routine work. The double-loop task is to choose the risk boundary deliberately, perhaps preserving approval for sensitive actions while delegating routine ones with clear audit trails. The design is a decision, not a psychological label.
Use private inferences as questions, not as evidence
Argyris’s action-science methods include exercises that surface what people think during a conversation but do not say aloud. A practical adaptation is to keep two columns during a rehearsal: observable words and actions on one side, and the inferences or concerns they prompted on the other. The second column is not a transcript of external facts.
Suppose a supervisor says a project is on track. A team member silently thinks the supervisor is hiding a delay. The spoken statement is observable; concealment is a hypothesis. A respectful test would ask which milestones have been completed, how on-track is defined, and what independent evidence supports that status.
This separation helps prevent a common self-confirming cycle. Someone infers that questions are unwelcome, asks no questions, sees no open disagreement and concludes that the team suppresses debate. But the lack of discussion may have several explanations. Running a small, well-framed evidence request is more informative than treating the first private inference as a proven group norm.
Privacy and dignity matter. The aim is not to force employees to disclose every private thought or use a reflection worksheet to interrogate them. Instead, people can voluntarily identify assumptions relevant to the shared decision, consider alternative interpretations and ask testable questions. Personal therapy and managerial surveillance are outside this learning framework.
A meeting that actually changes the reasoning process
Start a review by stating the question the group is authorised to decide. Present the observed discrepancy, the current objective and the rule that governs the team’s response. Let participants distinguish verified observations from conjecture. Give them access to the underlying counts and definitions before arguing about causes. This sequence reduces the temptation to compete with confident stories.
Invite alternative explanations with equal standards of evidence. One person may think students need better reminders, another may think assignment deadlines ignore device access, and a third may think the tasks are too large. Each hypothesis should imply a distinctive observation or intervention. The meeting is productive when people can explain how one proposed test would discriminate between explanations.
State the decision point. The group might keep the original cutoff, run a temporary flexible-window trial or redesign the feedback schedule. For each option, write the value protected, foreseeable risks, who has authority to approve and what result would prompt reversal. A discussion without a decision, experiment or justified deferral is not yet demonstrated double-loop learning.
End with an exact record: what was challenged, what evidence was used, which governing assumption remained or changed, who owns the next observation and when it will be reviewed. Avoid a vague commitment to communicate better. Such a statement is hard to test and lets the next team mistake enthusiasm for evidence.
Make contrary evidence safe enough to surface
Double-loop learning depends on information that may threaten reputations, budgets or established practice. If a team member risks punishment whenever an assumption is questioned, the information needed for revision may remain invisible. Procedural safeguards are therefore part of the mechanism, not a pleasant optional addition.
Ask what happens after someone raises an inconvenient fact. Is it recorded? Does a decision maker respond to the substance? Can the person appeal a mistaken interpretation? Are staff able to report safety issues through protected channels? A policy saying speak up has limited force when those practical protections do not exist.
Feedback procedures must also guard against unfounded accusations. Challenging a rule is different from declaring colleagues negligent without evidence. A fair design lets a person explain context and keeps the distinction between observed actions and inferred motives. Independent reviewers may be appropriate when senior people control the process under question.
The changed governing rule might be that reporting an unresolved issue counts as responsible work rather than as failed performance. Test that change through actual case handling, not only attitude surveys. If reported issues increase after a safe channel opens, the rise may reflect improved detection rather than worsening behaviour. Interpretation requires context.
When to pilot a changed rule rather than announce a revolution
A double-loop hypothesis may be right in one context and risky elsewhere. Before changing an entire institution, define a bounded pilot that is legal, ethical, reversible and measurable. Identify which rule will change, what will remain constant, how affected people will be informed and what outcomes might indicate harm.
In a fictional homework programme, the team might pilot flexible submission windows in two comparable class groups, while preserving the same learning objectives and marking rubric. It should not deny students entitled accommodations or randomise essential safeguarding protections. A small trial is useful only when the chosen design does not expose people to an unacceptable risk.
To compare conditions, record both opportunity counts and learner characteristics. If one group has thirty assignments across ten students and another has forty across twelve students, late counts alone cannot be compared. A rate per eligible assignment may answer one question, but independence and repeated measures also matter. Multiple assignments from one learner are correlated observations, not thirty independent pupils.
Predeclare a review date and the conditions for stopping the pilot. A changed rule that unexpectedly delays feedback or increases teacher workload beyond an agreed limit may need modification. Reversing a pilot after clear contrary evidence can be a success in organisational learning; it shows the new assumption was tested rather than protected.
Why before-and-after improvement rarely proves the cause
A sequence of outcomes can be persuasive while leaving causal uncertainty. Perhaps a team adds reminders, changes the deadline and receives easier assignments in successive weeks. Observing better completion at the end does not identify which change made the difference. Timing is an observation, not a mechanism.
When feasible, use a comparison group or a staggered rollout with reasonable safeguards. If a simultaneous comparison is impossible, at least record alternative explanations, stable definitions and trend data from more than one time point. A tiny pilot may be appropriate for learning how a new rule operates but not for making a confident claim about an average causal effect.
Consider an invented example: a pilot group completes 24 of 30 tasks on time and a comparison group completes 21 of 30. Their observed proportions are 80 and 70 percent, a ten-percentage-point difference. The sample is small and may have different learners, task difficulty or teaching conditions. The calculation is accurate; the causal claim that the policy raised completion by ten percentage points is not established by the count alone.
Record uncertainty honestly. In this exercise, the unit is a task opportunity rather than an independently sampled person, so a simple binomial confidence interval treating all tasks as independent may also be misleading. If many tasks belong to a few learners, consider clustering or report descriptive rates with an explicit limitation. Statistical sophistication cannot rescue a weak assignment procedure.
Track corrected errors and newly visible problems
A successful organisational intervention may initially increase the number of reported problems. If workers learn that near misses are welcomed as evidence, previously hidden issues may enter the system. A headline count of reported incidents may rise even as the underlying process improves. Distinguish incidence, detection and reporting behaviour.
Use paired measures. In an illustrative support service, 60 complaints arise from 300 completed orders in a month, a rate of 20 percent. In another period, 45 complaints arise from 300 orders, a rate of 15 percent. The change is five percentage points, or a 25 percent reduction relative to the initial complaint rate. Do not claim customers were 25 percent happier; that outcome was not measured.
Now suppose orders increase to 500 while complaints reach 50. The raw count rose from 45 to 50, but the complaint rate becomes 50/500=10 percent. This could indicate improvement or changed reporting, order composition or problem severity. The denominator must be visible, and users should still inspect repeat complaints and unresolved harm rather than relying solely on one rate.
For double-loop learning, also track whether challenged assumptions actually changed decisions. A useful record shows the prior rule, evidence reviewed, revised rule if any, affected cases and an independent check after a suitable interval. An improved rate without any change in governing logic might be valuable single-loop progress; call it what it is.
Write a decision log that distinguishes the loops
A good decision log begins with a specific mismatch and records the evidence source, denominator and affected group. It then identifies the action used under the current rule and why that rule was originally adopted. Historical rationale can reveal legitimate constraints that current participants might overlook. Do not erase that context simply because the present result is disappointing.
The next field is the governing assumption under review. For instance, every learner can submit work before 17:00, or all quality problems originate in insufficient worker effort. Record support and counter-evidence separately. If the premise is rejected, the log should name the replacement rule and who is authorised to change it.
Then record a small test, the expected effect, the balancing indicators and conditions under which the organisation would change course. If there is no testable difference between the old and new rules, the supposed revision may be merely cosmetic. A new slogan that selects the same actions under the same evidence is not evidence of a new theory-in-use.
Finally, preserve follow-up decisions and independent verification. If a new manager asks months later why a rule changed, the record should answer without relying on one person’s memory. It should also show whether the change helped, harmed or remained inconclusive. In that sense, double-loop learning benefits from sound knowledge transfer but is distinct from simply archiving documents.
Who can change the rule, and who bears the risk?
An individual can notice a flawed objective yet lack authority to revise it. An organisation may therefore need an escalation route that reaches the level where governing rules are set. If staff are held accountable for an outcome but cannot influence the constraints producing it, telling them to take ownership is not a substantive solution.
Map the decision rights before a pilot. Who can change assessment timing? Who may change a customer complaint definition? Which safeguarding or privacy rules are fixed by law or professional obligation? If the right owner is absent from the discussion, the group may create recommendations that cannot be enacted or make unauthorized changes.
Power also affects evidence. Front-line observations may be discounted because they come from people with low status, even when those people see the recurring failures. A fair review gives claims weight according to verification and relevance, not the seniority of the speaker. It also protects dissent from retaliation and allows senior leaders to explain genuine external constraints.
When a core protection cannot change, double-loop thinking can still question surrounding assumptions. A school should not relax its duty to protect students in order to improve an administrative metric. It can instead ask whether staffing, tool design and communication routines make the protective duty easier to perform. A revised strategy is legitimate only within the constraints that remain binding.
Double-loop reasoning in personal study and decisions
The framework works at individual scale, although an organisation has additional complications involving shared rules and authority. A learner who fails to remember terminology might increase flashcard repetitions, a single-loop correction. A deeper inquiry asks whether memorising isolated definitions is the right objective when the task actually requires using words accurately in a novel explanation.
A person can write their own theory-in-use as a conditional statement: I begin studying when I feel ready, or I treat correct recognition as proof I can produce an answer. That makes the rule discussable. The learner can then test a replacement, such as short scheduled retrieval without the cue present, rather than simply applying more of the same ineffective study routine.
Suppose an invented learner recognises 16 of 20 terms on a multiple-choice quiz but can independently define only 8 of the same 20. Recognition accuracy is 80 percent, independent production is 40 percent; neither fraction by itself demonstrates why the difference occurred. A change in learning strategy should aim at the task that matters, with a fresh delayed test rather than repeatedly using the same twenty items.
Do not pathologise a student because their first strategy fails. Learners have different constraints, and a weak result may reflect poor explanations, task design or inadequate time as well as an unsuitable personal rule. Double-loop inquiry should make available alternatives clearer, not turn every struggling learner into the sole cause of a system problem.
When a group has learned, not just its members
One individual may revise their belief while the organisation continues to reward the old routine. Organisational learning becomes more plausible when a rule is changed, understood, enacted across the relevant roles and maintained after a particular advocate leaves. Shared procedures, decision rights and evidence access help a revised theory-in-use survive staff turnover.
A team meeting that produces agreement but no changed practice is an incomplete outcome. Check later decisions: when a similar case arose, did people apply the new rule? Could they explain why? Were exceptions recorded rather than hidden? Did the process allow another round of questioning if the new rule produced unforeseen effects?
Collective intelligence requires room for disagreement, but not permanent indecision. Different people hold partial information. The group can combine it by making assumptions explicit, asking which observations support them and choosing tests whose outcomes matter for the decision. Repeated slogans about teamwork are not a substitute for observable learning.
Beware an opposite failure: a team may overinterpret one dramatic event and rewrite every policy. Shared learning needs evidence standards and proportionality. Sometimes the strongest conclusion is that the original objective was sound but one local procedure failed, and that no change to governing values is justified. The ability to retain a rule after testing it also demonstrates disciplined thinking.
Learning a new rule is not the same as remembering the old one
Organisations can lose knowledge when experienced people leave. The verified Institutional Forgetting guide owns the reader job of preserving know-how and verifying handover. Double-loop learning asks a different question: whether the knowledge or policy being transmitted should continue to govern decisions.
A perfect archive can preserve a mistaken procedure. Conversely, a brilliant redesign can fail if the next team cannot find the reasons for the change. Useful practice links the two: retain original evidence, the assumption that was questioned, the replacement rule and the later test result. New staff should learn why a rule exists rather than only memorise its current wording.
Suppose a support worker inherits a rule requiring three approvals for every refund. The archive explains which fraud cases originally motivated it. A later review finds that a subset of low-value, verified cases could safely use a simpler process while sensitive cases retain strict controls. The learning lies in revising the risk-based decision boundary; the record enables continuity.
This distinction prevents a common publishing collision: a guide on knowledge handovers is not the same as a guide on how an organisation challenges the governing reasons behind a routine. Linking the two helps readers move from preserving information to deciding what information should still control work.
Do not confuse routine improvement with escape from a competency trap
A competency trap occurs when an organisation becomes particularly efficient at a routine and continues to exploit it even when a better alternative might exist. The related Competency Traps specialist guide addresses exploration and the reinforcing feedback behind established routines.
Double-loop learning is a complementary diagnostic discipline. It asks which objective, metric or assumption makes the efficient routine appear unquestionably valuable. A team may keep optimising average call duration because that number is easy to measure, even when customer resolution has become the more important outcome. Revising the metric and decision criteria can create room for better alternatives.
Neither framework implies that organisations should always abandon familiar procedures. Some established routines are reliable because they embody genuine learning and control essential risks. Escaping a competency trap requires evidence that the current routine’s apparent advantage comes from the evaluation system or accumulated practice rather than intrinsic suitability for the present job.
Put the distinction into a pilot: compare the current routine with an alternative on the real customer outcome and a safety or cost constraint. Log which assumption was under test. If the new alternative wins only by receiving more staff, clearer data or easier cases, the result does not establish that the governing routine alone was flawed.
A classroom diagnostic that can reveal a governing assumption
An educator might observe that learners can follow worked examples but struggle when a familiar concept appears in a new setting. A simple correction gives more guided practice. A double-loop inquiry asks whether the programme’s prevailing definition of understanding has been reduced to recognising a rehearsed form.
One diagnostic is to contrast near-transfer questions with a genuinely changed-context question. For example, after teaching proportions using recipes, ask learners to reason about scale on a map with different units and an unknown quantity in a different position. A learner who can imitate the recipe steps but not identify the underlying ratio may need conceptual reconstruction, not just more repetition.
The teacher can revise the governing criterion for mastery: students must explain which quantities correspond, justify the operations and solve one independent transfer item after a delay. This is not proof that every learner should receive the same assessment format. Language access, disability accommodations and curricular requirements still need attention.
Evaluate with comparable tasks and honest denominators. Record how many learners meet each stated criterion, not merely the average number of steps completed. If a new assessment is much harder, scores might initially fall even while the diagnostic becomes more informative. Reconsidering the measurement can be an improvement without an immediate rise in the original headline number.
Seven ways a double-loop exercise can fail
First, the group may change only its vocabulary: the same approval rule receives a new name. Second, participants may debate personal intentions rather than test actions and evidence. Third, a dominant speaker may suppress alternative explanations. Fourth, the proposed test may lack a clear outcome measure. These failures make the exercise appear reflective without changing how work actually happens.
Fifth, a team may choose a trial whose results cannot be interpreted because tasks, participants and denominators change unpredictably. Sixth, a beneficial local correction may be delayed while leaders wait for a grand strategy. Seventh, a new rule may be announced but not granted authority, resources or monitoring. Each failure calls for a different repair, not another generic request for openness.
Fix terminology-only change by writing down exactly which decisions would differ under the new rule. Fix evidence-free debate by separating observations from inferences. Fix hierarchy effects by allowing alternative hypotheses to be proposed safely and checked against common criteria. Fix weak evaluation by specifying the denominator, timing and balancing outcomes before the trial.
Fix scope and implementation by naming a legitimate owner, a bounded pilot and a follow-up point. Preserve emergency protections throughout the review. The most useful repair question is specific: what would we see the team do differently when the next similar case appears?
Ethical limits of questioning everything
An instruction to challenge assumptions is not permission to disregard rights, privacy, examination integrity, safeguarding or professional standards. Some values are explicit boundaries rather than variables to optimise away. A programme may question how to fulfil those duties efficiently, but it must not risk people’s welfare to demonstrate managerial cleverness.
Double-loop facilitation can be misused to pressure staff into admitting fault or to rationalise a decision already made. That would reverse the spirit of evidence-based inquiry. Participation should have clear purpose, confidentiality limits and accessible routes for contesting interpretations. Where a serious allegation is present, proper independent procedures matter more than a workshop format.
Organisational power can make nominally open questions coercive. If the people defining success also judge the employment of those supplying inconvenient evidence, workers may withhold information. Decision makers should consider separate channels and protections, not simply demand courage from those with the least authority.
Respect uncertainty and disagreement. A strongly held principle can survive a challenge when evidence and obligations support it. An unpopular practice can be necessary under constraints outsiders have not seen. Rigour means neither assuming every tradition is right nor declaring every existing rule irrational.
Independent transfer: a different workplace and an explained answer
Try a new situation. A fictional service company completes 240 repair visits per month and records 36 repeat visits because a first repair did not resolve the issue. Its headline repeat-visit rate is 36/240=15 percent. The manager adds reminders to workers and the next similar month records 30 repeats out of 240 visits, or 12.5 percent. Is that automatically double-loop learning?
No. The evidence shows a local improvement in the stated metric, with a difference of 2.5 percentage points, but reminders leave the governing objective and explanation intact. The rate alone cannot establish whether the reminders caused the change, either. A single-loop correction may be useful and successful without being double-loop.
Next, workers report that a target to minimise minutes on site rewards temporary fixes. The organisation tests a revised rule prioritising verified first-time resolution, with time and cost tracked as balancing constraints. That rule challenges a governing assumption: shortest visit equals best service. The act of testing and potentially revising the decision criterion is the double-loop component, not the number of meetings.
Suppose a pilot later records 18 repeat visits among 240 comparable visit opportunities, or 7.5 percent. This is 7.5 percentage points below the original 15 percent. It is a descriptive result, not proof of causal effect without examining case mix, measurement, time and staffing. Also check customer satisfaction, safety and cost: lowering repeat visits by refusing difficult customers would fail the real objective.
Transfer the idea again to a classroom. If a learner repeatedly guesses in mathematics, a single-loop fix may add more examples. A double-loop inquiry asks why the learner believes recognising a pattern is sufficient proof of understanding and tests an independent explanation requirement. The job is to make the rule guiding action explicit and testable in a new context.
A usable action cycle for students, teams and leaders
Begin by writing one observable mismatch in a sentence with the unit and denominator. Avoid beginning with a judgement about someone’s intelligence or commitment. Add the intended purpose and who benefits if it is achieved. If participants cannot agree on the purpose, that difference itself may be a governing question.
Record the current correction routine and what it assumes. Ask which observations might reveal the assumption is false or incomplete. Make a brief list of competing explanations, including one that would vindicate the current rule. This safeguards against using the framework as a pretext for a preferred reorganisation.
Choose the smallest ethical and authorised test that could change the decision. Keep the current safety boundaries, write outcome and balancing measures, specify the interval and who reviews the result. In education, preserve access arrangements and genuine learning requirements. In business, preserve contractual and privacy duties.
At review, separate observed counts from their interpretations, account for changing contexts and record what did or did not move. If evidence remains weak, continue cautious measurement rather than promoting a confident story. If the governing assumption is convincingly wrong, update the decision rule and make sure affected people can actually enact it.
Finally, test transfer: does the new rule help in a different class, team or period without importing new harms? Save the reasoning alongside the procedure so that successors can question it again when conditions change. A rule that can be revised by evidence is more valuable than one whose survival depends on the prestige of its author.
Frequently asked questions
Does double-loop learning mean changing every rule?
No. It means deliberately testing governing assumptions when evidence warrants. Sometimes a correct rule survives scrutiny and a local process repair is sufficient. Frequent arbitrary rule changes can damage reliability. Evidence, authority and safety constraints govern whether a deeper change should occur.
How do I distinguish a double-loop change from a big single-loop change?
Ask whether the underlying criterion or governing value changed. Buying new software or restructuring teams can be large but single-loop if both pursue the same unexamined objective. Quietly redefining mastery from fast completion to independent explanation can challenge a governing assumption.
Is a retrospective meeting automatically double-loop learning?
No. A meeting can identify errors and propose fixes while leaving governing rules untouched. Look for an explicit challenged assumption, evidence bearing on it, an authorised decision or pilot and later verification that the altered rule changed relevant actions.
What is the theory-in-use?
It is the governing pattern that best explains observed decisions, contrasted with the ideals people publicly endorse. It should be inferred cautiously from several decisions and relevant constraints, not used as a label for an individual’s hidden motives after one incident.
Why do intelligent teams resist feedback?
Argyris described defensive routines that protect people or groups from embarrassment while obstructing testing of their assumptions. Power, incentive structures and real organisational risks can also limit openness. Do not treat every objection as defensiveness; test claims and create fair reporting procedures.
How should we measure a pilot?
Define the relevant unit, denominator, comparison and balancing outcomes before the trial. Distinguish counts from rates and percentage points from relative changes. Consider confounding, repeated observations from the same people and unequal opportunities. Report uncertainty instead of attributing all change to the new rule.
Can a learner use this without an organisation?
Yes. A learner can ask whether their study strategy is serving the correct goal. Repeating flashcards is a single-loop adjustment; challenging an assumption that recognition equals mastery and testing independent transfer questions is a possible double-loop inquiry.
What if an assumption is legally required?
Do not violate legal, safeguarding, assessment or professional duties. Identify which constraints are binding, and examine surrounding workflow and resource assumptions within those boundaries. Seek authorised expert guidance when interpretation of an obligation itself is unclear.
How is this different from competency traps or institutional forgetting?
Competency traps concern reinforcing efficiency in a routine; institutional forgetting concerns the loss of knowledge or verified handovers. Double-loop learning owns the specific question of challenging and revising the governing rules behind actions. The topics can interact without being interchangeable.
Original research, intellectual provenance and limitations
The original framework is presented by Chris Argyris in Double Loop Learning in Organizations, Harvard Business Review, September 1977. His earlier original research on single- and double-loop decision models appeared in Administrative Science Quarterly 21(3), 1976, DOI 10.2307/2391848. The examples in this article are new invented exercises, not reproduced corporate anecdotes from those papers.
For evidence about how capable professionals can defend existing beliefs, consult Argyris’s Teaching Smart People How to Learn, Harvard Business Review, May–June 1991. His later Double-Loop Learning, Teaching, and Research, Academy of Management Learning & Education, 2002, discusses case-based practice and research rather than promising guaranteed organisational results.
For an original critical discussion of practical obstacles, see Myeong-gu Seo’s Overcoming Emotional Barriers, Political Obstacles, and Control Imperatives in the Action-Science Approach, Academy of Management Learning & Education, 2003. A later 2023 systematic review discusses difficulties defining, measuring and implementing double-loop learning; it is included transparently as synthesis rather than misrepresented as an original Argyris experiment.
These references establish a conceptual learning framework and debated implementation methods, not a validated universal numerical effect size. No invented classroom, customer-service or repair-case percentage should be cited as an empirical finding. Dated policy or safeguarding requirements must be checked against their relevant current authorities rather than inferred from organisational-learning theory.
Verified related eduKate specialist reader routes
Continue through the verified How Intelligence Works hub for wider coverage of individual and collective reasoning. Use Institutional Forgetting when your problem is preserving know-how across handovers, not challenging the governing purpose behind that know-how.
For the separate question of why a group becomes very effective at a routine even when exploration might help, visit Competency Traps — Efficient Routines, Exploration and Organisational Learning. This double-loop article focuses on making the underlying rule discussable and revisable with evidence.
A useful final transfer assignment is to describe an actual decision without identifying anyone who should not be named. Write its observed outcome, governing assumption, a possible local repair, a possible assumption-changing correction, and a fair test with denominators and constraints. Then ask a peer to explain what evidence would overturn the preferred explanation.