Immunity to Change: Designing Safe Behavioral Tests

Safe behavioral tests in Immunity to Change are small, chosen actions designed to produce information about an assumption. They are not courage challenges, secret assessments of colleagues or demonstrations that the coach was right. A useful test can reveal a real constraint, mixed evidence or a reason to stop.

Robert Kegan and Lisa Lahey’s approach examines how a wanted improvement can conflict with a protective commitment. A person might want colleagues to own routine work while also trying never to appear insufficiently careful. The assumption beneath that protection may be worth examining, but the coach should not presume it is false.

Coachful is our #1 recommendation for organizing this learning workflow: an original test plan, accepted actions, actual observations and a review that preserves uncertainty. The Coachful features support the record. They do not authorize workplace risks, validate an assumption or guarantee lasting belief change.

Know what is being tested

The Harvard Graduate School of Education overview attributes Immunity to Change to Kegan and Lahey and describes attention to competing commitments and assumptions. This guide focuses specifically on designing and running a responsible test, rather than presenting the full method or reproducing its proprietary forms.

The basic logic connects a desired improvement, behavior that works against it, a protective commitment and an assumption that makes that protection seem necessary. These are different things. “I rewrite everything” describes behavior; “I must never appear careless” describes protection; “If I do not rewrite the style, leadership will remove responsibility from me” is a prediction to examine.

The client should recognize the account. A coach cannot diagnose a hidden commitment from a missed task, and not every obstacle is internal. Missing skills, unsafe conditions, unclear authority, discrimination or an impossible workload may require a different response. Kegan and Lahey’s HGSE interview distinguishes technical and adaptive elements; it does not make ordinary skill development or organizational change unnecessary.

Their 2009 book, in excerpts selected for HarvardX teaching, distinguishes gathering information about an assumption from immediately trying to improve performance. That distinction matters: a recommendation can succeed while the actual test fails to occur, and an uncomfortable response can still provide relevant information. For the interpretation stage in greater detail, see examining big assumptions.

Develop the five design qualities

Kegan and Lahey describe tests as safe and modest, actionable, taken with a research stance and directed at the big assumption. This is not the familiar goal-setting use of SMART. The practical explanations and examples below are original applications, not a reproduced guide sheet or source case.

Safe: examine consequences for everyone affected

Identify concrete duties and potential harms before considering the client’s wish to learn. An unfamiliar conversation can feel uncomfortable without being an inappropriate risk. Conversely, a confident client can propose an action that exposes other people to an unacceptable consequence.

Retain required safety, security, financial, quality and approval processes. Do not withhold a warning, remove necessary supervision, miss an obligation or misrepresent information to create a test. “Small” does not excuse a breach of responsibility. If a feared consequence would be unacceptable, look for a different source of information or do not run that action. No plan can guarantee zero risk; the task is responsible judgment within the actual context.

Modest: change enough to learn, without changing everything

A bounded action makes the result easier to interpret. Changing the task difficulty, time available, audience and support at once leaves many explanations for what happens. Reducing the scope can protect people and improve the relevance of the observation.

Modest does not mean trivial. Sending an ordinary clarification may be meaningful to a client who usually hides uncertainty. A rehearsal with the coach can prepare that action, but it cannot by itself test a prediction about a manager’s response. Choose scope according to the question, rather than requiring every test to last five minutes or fit a fixed number of slides.

Actionable: use an actual opportunity and adequate support

The client needs an action they can perform, with necessary permissions, people and resources available. An elegant hypothetical plan is not a completed test. A near-term ordinary opportunity may fit; an arbitrary 48-hour deadline is not a universal requirement.

Check dependencies before promising an action. A colleague may decline the proposed task or offer a narrower role. The client must not remove help to make their experiment more dramatic. Agree the work transparently enough for the colleague to do it properly, and let normal delivery priorities take precedence when conditions change.

Research stance: allow evidence that complicates the story

Here, research means a learning orientation, not a controlled scientific study or permission to experiment on unsuspecting people. State the prediction and decide what observation could support, challenge or qualify it before the action.

Do not build a rule under which every response confirms the assumption. Silence could reflect competing work; a request for one correction could concern the document rather than the client’s overall competence. The client’s feelings are relevant to their experience but do not provide direct access to someone else’s judgment. Keep words, actions, interpretations and unknowns distinguishable.

Test the assumption: make the action relevant

The changed behavior must bear on the actual prediction. If the client fears leaving style to a colleague, retaining necessary factual review may be appropriate. If the prediction instead concerns factual oversight, the same action would test a different question.

Check that the feared situation could actually arise in this context and that more than one result is possible. A person who rewrites the whole document before sending it has not tested leaving the wording alone. A friendly coach’s praise does not test what a particular leadership group will do. Design for useful information, without deliberately making the task fail.

Write an original test brief before acting

A concise brief can state the client’s learning question, the behavior that changes, retained responsibilities, affected people’s actual agreement, observations, stop conditions and review point. Use the client’s language and explain the record’s audience. This is an original coaching aid, not a standardized ITC map or assessment.

Consider three kinds of evidence. Direct statements can reveal an expressed expectation. Observable decisions can show whether a role, invitation or task changed. The client can also record discomfort, relief or an urge to intervene. Each answers a different question. An unchanged invitation provides some information about inclusion, not proof of everyone’s private confidence.

Agree when the proposed action must be redesigned: an unavailable colleague, changed risk, missing permission or a task outside competence can all matter. Canceling in response to new information is not automatically a protective evasion. Equally, a quiet return to the old behavior should be recognized before interpreting the result.

Follow a complete fictional test from rejected options to review

Morgan, a CTO, wants team leads to own the presentation of routine recommendations. Morgan routinely rewrites their wording after reviewing the required facts. This fictional case concerns a low-value internal comparison, not production software, security controls or real product advice.

State the protective logic and observable prediction

Morgan recognizes a commitment to avoiding any appearance of inadequate oversight. The assumption is: “If I leave a routine recommendation’s style to the lead after the required factual review, leadership will say my oversight is inadequate and remove me from chairing the next review.” The client has had uncomfortable feedback before, so the concern is not dismissed as irrational.

They distinguish two predicted events: an explicit criticism of oversight and removal from the next chairing role. Neither can be inferred solely from a short expression, a long silence or a request to edit a sentence. Morgan also expects a wording edit, but recognizes that editing alone would not establish criticism of oversight. The learning question is limited to this leadership group and this type of recommendation.

Morgan first proposes skipping all review on a major technical release. They reject that option because it removes duties, increases risk and changes far more than style. A second proposal, showing an untouched draft only to the coach, is also rejected: it might prepare the client but would not test the leadership response. Neither rejected option is carried out.

Complete the bounded arrangement

The actual opportunity is a one-page comparison for six internal training headsets at the scheduled 14 September 2034 leadership review. The budget limit is £180 including delivery. The product names and prices below are entirely fictional arithmetic for this case, not market recommendations.

Dee, a team lead, agrees to prepare the page and retain authorship of its wording. Morgan will perform the ordinary factual and budget review, ask Dee to correct any material errors and leave stylistic preferences alone. Purchasing still requires the normal compatibility check and authorized order; the coaching test does not permit buying anything.

Dee has an existing 60-minute allocation for the page, including corrections. Morgan retains a normal 15-minute factual-review slot. Esther, who chairs the 14 September meeting, confirms that Morgan remains accountable for those checks and that team leads may own the wording. She does not promise a favorable response. The ordinary meeting, not a manufactured surprise, supplies the context. The existing rota lists Morgan to chair the next review on 21 September, subject to ordinary scheduling changes.

Morgan’s completed test brief reads:

For Dee’s routine internal comparison, I will complete the factual review and leave the remaining wording unchanged. I expect leadership to criticize my oversight and remove me from the next chairing role. I will record their actual comments and the next review’s chairing decision, separately from my discomfort. No factual error, purchase, compatibility requirement or approval will be bypassed. If the page cannot meet the normal requirements, we will correct it or postpone its consideration. We will review the actual action and evidence after the 21 September meeting, rather than declare permanent change from one event.

Perform the required check without taking over the style

Dee’s draft estimates Model A at six units of £20 plus £12 delivery: £132. Model B costs six units of £28 plus £15 delivery: £183. The first draft incorrectly calls both options within budget. During the 15-minute factual review, Morgan identifies that B exceeds the £180 cap by £3 and asks Dee to correct the comparison.

Dee corrects it. Morgan does not use the error as a reason to rewrite the whole page. The completed recommendation says:

For six training headsets, Model A totals £132 including delivery and fits the £180 limit, leaving £48. Model B totals £183 and exceeds the limit by £3. We recommend A for the next compatibility check. This is not an order request until compatibility and purchasing approval are complete. If A does not meet the required connection and usage needs, we will return with a revised comparison rather than buy B under this budget.

Morgan notices a long opening sentence elsewhere on the page and wants to rewrite it. It does not alter the facts or hide a qualification. Morgan leaves it in Dee’s wording, as planned, and records the urge to intervene. This is the specific changed behavior; necessary oversight remains present.

Capture the actual response and role decision

At the 14 September review, Esther says the recommendation is understandable and should proceed to the compatibility check. She also asks Dee to split the long opening sentence before circulating the page more widely. Dee accepts and makes that edit. Esther does not approve a purchase and makes no statement that Morgan’s oversight is inadequate.

Morgan initially hears the edit request as confirmation of the whole fear. The written observation is narrower: one stylistic correction was requested, the recommendation proceeded conditionally, and no adverse oversight statement was made in that meeting. The absence of a statement is not proof that nobody had a private criticism.

The invitation for 21 September names Morgan as chair. Morgan actually chairs that routine review; responsibility has not been removed in the way predicted for the next meeting. They do not claim that all future roles are secure. Compatibility and purchasing remain separate operational work and have not been completed by the coaching review.

Check validity before revising the assumption

On 22 September, Morgan compares the actual action with the brief. The fact review occurred, Dee corrected the budget error, Morgan left the remaining style alone, and the leadership group saw that version. The test therefore addressed the agreed stylistic prediction under retained oversight. It did not test abandoning technical review, a higher-value purchase or a different audience.

The result is mixed. The predicted need for a stylistic correction occurred. The predicted explicit criticism of oversight did not occur in the 14 September test meeting, and removal from the next chairing role did not occur. Other private or later judgments remain unknown. Morgan still reports discomfort and a wish to control wording. They choose the conditional revision:

For a routine internal recommendation in this group, I can retain the required factual review and let a lead own the style. A request for a wording edit can happen without the adverse oversight statement or immediate removal I predicted. I do not yet know how this applies to more consequential recommendations.

Morgan chooses another comparable test only if a suitable ordinary recommendation arises. No larger risk is automatically authorized, and no permanent belief change is declared. Dee’s factual mistake remains visible; the result does not prove that every colleague’s work can proceed unchecked.

Dee records 45 minutes of drafting and five of correcting the budget and sentence, using 50 of the allocated 60 minutes. Morgan records 20 minutes of test preparation, ten of observation notes and invitation checking, 15 of factual review and two 15-minute review meetings: 75 minutes. The additional test planning and notes use Morgan’s existing 30-minute development allocation; the other 45 minutes are ordinary review work. Coaching sessions, Esther’s time, other participants’ time and later procurement work are outside that account. No headsets have been bought, and no productivity saving is claimed.

Use a six-step practical process

1. Agree the client’s learning question. Check that the assumption and desired improvement are recognizable and chosen, rather than supplied as a diagnosis.
2. Specify the prediction and alternatives. Define what could support, challenge or qualify it, and which private judgments remain inaccessible.
3. Choose an appropriate ordinary action. Retain duties, clarify authority and obtain affected people’s real agreement and resources.
4. Set stop and redesign conditions. Explain what would make the action unsuitable, including new risk, missing support or changed delivery requirements.
5. Record what actually happened. Include the client’s actual behavior, relevant responses, feelings and differences from the plan.
6. Review validity, then meaning. Decide what the test can reveal before choosing a qualified conclusion or a further action.

These steps are original practical guidance, not a copied ITC worksheet or research protocol. The model’s language of testing does not make a workplace interaction a controlled experiment. Keep the scope of the conclusion proportional to the action and evidence.

Use Coachful to hold the test and its limitations

Create an original reflection worksheet for the agreed brief and a client-owned goal for the learning purpose. Add only accepted actions, with clear boundaries. A completed task for “send the page” does not establish that the budget was met, the manager accepted it or the assumption was disproved.

For submitted forms and worksheets, Coach review is private to the coach team and is not shown to the client or included in the response PDF. Deliver agreed feedback through the client’s private conversation with the authorized coach-team audience or during a session. Marking an internal review does not send the client a comment.

If summarized sessions are used, recaps send automatically by default. Check Sessions > settings gear > Session AI > Session recaps, choose Approve before sending when review is needed and check any session-type override. Review lets the coach edit client-facing content before Send recap or Keep private; assigning tasks is separate. Keep confidential assumptions within the agreed audience and do not send a sponsor a private psychological interpretation as if it were fact.

The screenshots are illustrative demonstrations, not Morgan’s test or a formal ITC assessment. Keep employer and purchasing records in their proper systems. For broader review distinctions, see progress and accountability.

Troubleshoot before increasing the stakes

The action is too risky. Do not run it to demonstrate commitment. Retain the responsibility, clarify the situation or use a less consequential source of information. The client’s development does not override another person’s safety or rights.

The client quietly changed the test. Compare plan and action. If Morgan had rewritten every sentence, a favorable response would not test leaving style alone. Record the limitation and decide whether another suitable opportunity is worth using.

Every response appears to confirm the assumption. Revisit the prediction and the evidence rules. If no possible observation could add information, the proposed test is not yet useful. Direct clarification may fit better.

The result reveals a real adverse expectation. Take it seriously. The client may need a standards conversation, appropriate support or a different decision. Do not insist that a genuine organizational constraint is merely a belief to overcome.

The work becomes distressing or moves beyond coaching competence. Pause and revisit scope. The 2025 ICF competencies distinguish coaching from other support professions and include appropriate referral. An ITC test is not a substitute for clinical care or specialized risk advice.

Frequently asked questions

Does safe mean the client feels comfortable?

No. Unfamiliar behavior can feel uncomfortable, but that does not make every risk appropriate. Examine actual duties, consequences and affected people separately from the client’s feelings.

Must the test disprove the assumption?

No. It can support part of the concern, reveal a boundary or remain inconclusive. The objective is relevant information, not a required success story.

Does SMART mean the usual goal-setting criteria here?

No. In Kegan and Lahey’s test-design discussion it concerns safe, modest, actionable testing with a research stance. Do not replace careful design with a generic measurable-goal checklist.

Should the next test always be more ambitious?

No. A comparable repeat or a direct conversation may be more informative. A larger consequence is not evidence of deeper learning, and a client may choose to stop.

What if the work succeeds but the test was not valid?

Keep the successful work result and the test limitation separate. A completed recommendation cannot establish that an assumption was examined if the relevant changed behavior never occurred.