Researchers, methodologists, and tool builders. move from conceptual clarity to coding reliability, validation, proposed studies, ethics, and revision.
Operational models need disciplined testing. The goal is not to claim clinical authority, but to make the account clear enough to be tested, criticized, and improved.
Operational Constructs
- Construct | Operational handle
- Incoming reality | Event, testimony, observation, diagnosis, result, repeated pattern, failure, or discovery.
- Factual model | What the person or group expected reality to do.
- Meaning frame | Identity, faith, values, mission, loyalty, hope, worldview, or assumed meaning.
- Prediction error | Degree of mismatch between event and factual model.
- Meaning gap | Degree of pressure on identity, duty, faith, hope, loyalty, or another meaning-bearing commitment. Assess its truth separately.
- Source trust | Perceived and assessed reliability of the channel carrying the signal.
- Capacity | Embodied, relational, clinical, communal, and temporal ability to process the pressure.
- AG response | Proposed warranted action, matched to evidence, actual powers, authority, supports, and constraints. Record actual action separately.
- Resonance quality | Outcome fruit: reality-contact, humility, repair, confession, protection, wise action, communion, or persistent distortion.
Coding Manual for Case Studies
The manual below is for research coding, supervised training, and institutional review. It is not a diagnostic instrument. Coders should code only what is present in the case material, mark uncertainty explicitly, and keep the five fields independent. A severe meaning gap does not automatically make the source trustworthy. Low capacity does not make a signal false. A strong action preference does not decide the prediction-error or meaning-gap score.
Coding unit. Code one pressure episode at a time: one incoming signal, one primary expected model, one primary meaning frame, one source channel, one capacity state, and one available action set. If a case contains several distinct signals, split it into separate episodes before assigning codes.
Coding Order
Coders should work in this order before comparing notes:
- Write the incoming signal in one sentence.
- Identify the failed expectation, if one is present.
- Identify the threatened meaning frame, if one is present.
- Code prediction error and meaning gap separately.
- Code source trust from the channel evidence, not from agreement with the claim.
- Code capacity from safety, embodiment, support, coercion, and clinical risk.
- Code agency as the best available action class under current evidence and capacity.
- Add one edge-case note if the coding was difficult.
Prediction-Error Anchors
Prediction error (PE) codes the mismatch between the incoming signal and the factual model. It is about expectation failure, not emotional pain by itself.
- Code | Anchor | Code when | Do not code from
- PE0 | No mismatch established | A sufficiently specified expectation fits the evidence. | Distress, dislike, disagreement, or vague unease.
- PE1 | Mild breach | A small expectation fails, but the model can absorb it with ordinary adjustment. | A large emotion attached to a small factual mismatch.
- PE2 | Material breach | The signal requires real updating, verification, explanation, or correction of the model. | Mere identity threat without a clear factual mismatch.
- PE3 | Severe breach | The signal overturns a central expectation, exposes serious risk, reveals a pattern, or invalidates the working model. | Catastrophic language unsupported by the case evidence.
Meaning-Gap Anchors
Meaning gap (MG) codes pressure on identity, faith, loyalty, mission, duty, hope, justice, calling, safety, or love. An expected event can create severe pressure while the commitment remains true. Code the pressure separately from surprise and from the commitment's truth or moral quality.
- Code | Anchor | Code when | Do not code from
- MG0 | No clear breach | The event has no evident pressure on identity, faith, loyalty, mission, hope, or moral meaning. | Technical correction or ordinary uncertainty.
- MG1 | Mild breach | The meaning frame is unsettled but still intact; the person can name tension without major rupture. | Momentary embarrassment or annoyance alone.
- MG2 | Material pressure | A commitment faces substantial pressure requiring attention to grief, protection, counsel, an external remedy, or warranted interpretation. | A factual update that does not touch the meaning frame.
- MG3 | Severe pressure | Identity, duty, faith, hope, or belonging is deeply disrupted, even if the commitment remains true. | Strong doctrine, ideology, or group loyalty unless the case shows meaning-level pressure.
Source-Trust Coding
Source trust (ST) codes the channel carrying the signal. Coders should look for access, competence, method, corroboration, incentives, records, accountability, and exposure to correction.
- Code | Anchor | Typical indicators
- ST-high | Strong channel trust | Firsthand access or reliable record; competent method; independent corroboration; clear chain of custody; incentives named; open to correction.
- ST-mixed | Partial or contested channel trust | Some access or corroboration, but gaps, conflicts, incomplete records, memory limits, role pressure, or unresolved incentives remain.
- ST-low | Weak channel trust | Hearsay, anonymous claim without support, manipulated context, no method, obvious incentive distortion, AI output without verification, or protected rumor.
- ST-unknown | Insufficient channel data | The case does not give enough information to assess the source responsibly.
Capacity Coding
Capacity (CAP) codes the receiver's present ability to process the pressure truthfully. It concerns timing, support, safety, and care. It does not decide whether the signal is true.
- Code | Anchor | Typical indicators
- CAP-green | Ordinary reflection is available | Basic safety, enough sleep and attention, no immediate coercion, manageable emotion, and enough support for ordinary processing.
- CAP-yellow | Support or slower timing is needed | High stakes, fatigue, trauma activation, relational pressure, grief, shame, institutional pressure, or limited support.
- CAP-red | Interpretation must yield to safety or qualified help | Active danger, abuse, coercion, self-harm risk, psychosis, medical emergency, severe dissociation, credible threat, or active cover-up.
Agency Coding
The AG field codes a proposed warranted action, not the whole power of agency and not a verdict on guilt. Record what the person actually did separately. AG codes the truthful action available now. It should be scaled to source trust and capacity. Code available action, not merely the action the person prefers.
- Code | Use when the next truthful action is | Common confusion
- AG-adjust | Update an ordinary plan, expectation, schedule, or practice without a major repair process. | Treating ordinary adjustment as if it required deep crisis interpretation.
- AG-verify | Check source, records, testimony, measurement, or context. | Endless checking to avoid action.
- AG-wait | Delay interpretation until capacity, evidence, or timing improves. | Passive avoidance.
- AG-document | Preserve records, dates, messages, observations, or decisions. | Weaponized documentation without care for truth.
- AG-protect | Establish safety, boundaries, separation, reporting, or safeguarding. | Calling protection unforgiveness.
- AG-refer | Involve clinician, supervisor, safeguarding officer, elder board, legal counsel, mediator, or qualified helper. | Dumping responsibility without handoff.
- AG-practice | Repeat a concrete skill, habit, discipline, or repair behavior under feedback. | Consuming more information instead of training.
- AG-negotiate | Seek a truthful agreement about duty, boundary, timing, role, resource, or shared expectation. | Using compromise language to avoid truth, protection, or responsibility.
- AG-lament | Grieve, pray, name loss, and refuse forced meaning. | Treating continuing grief as evidence of evasion or failed repair.
- AG-confess / repent | Own wrongdoing and turn toward repair. | Shame collapse without concrete change.
- AG-repair | Apologize, restore, compensate, correct, reconcile where safe, or change practice. | Performative repair that protects image.
- AG-correct / disclose | Publicly correct falsehood, policy, teaching, record, or institutional communication. | Over-disclosure before verification.
- AG-escalate | Move to emergency, formal, disciplinary, protective, or legal process. | Escalating to punish when lower warranted action is enough.
Examples and Edge Cases
- Case sketch | Likely profile | Coding note
- Weather forecast fails and plans change. | PE1--2 / MG0 / ST-high or mixed / CAP-green / AG-adjust + AG-verify | Factual model pressure without major meaning pressure.
- Medical diagnosis contradicts a belief that faithfulness guarantees health. | PE2 / MG3 / ST-high if medically supported / CAP-yellow / AG-refer + AG-lament + AG-adjust | Same signal creates factual and meaning pressure.
- Anonymous accusation against a leader with no records yet. | PE-unknown / MG2 / ST-low or unknown / CAP-yellow / AG-document + AG-verify + AG-protect if risk exists | Do not let low source trust erase protection duties when possible risk is serious.
- Multiple independent abuse reports with records. | PE3 / MG3 / ST-high / CAP-red or yellow / AG-protect + AG-refer + AG-escalate + AG-correct/disclose when verified | Institutional meaning pressure must not override evidence or safety.
- AI summary sounds confident but gives no sources. | PE0--1 / MG0--1 / ST-low / CAP-green / AG-verify | Fluency is not source trust.
- Person feels intense shame after a minor correction. | PE1 / MG2--3 / ST-high or mixed / CAP-yellow / AG-lament + AG-refer + AG-repair if needed | Emotional intensity may indicate meaning gap, not severe factual breach.
- Group stays calm after repeated near misses. | PE2--3 / MG1--2 / ST-high / CAP-green or yellow / AG-correct/disclose + AG-adjust | Calm can be false resonance when repeated signals are normalized.
Edge-case rules. Code unknown rather than guessing. Missing information is not a zero score. Record experienced surprise separately from evidenced expectation mismatch, and meaning-pressure intensity separately from the truth or moral quality of the commitment. Split compound cases into episodes. Record the strongest warranted code, not the loudest emotion. Escalate capacity to red when safety requires it, even if prediction error or meaning gap is still uncertain. In institutional cases, code the vulnerable person's capacity separately from the institution's capacity to process the signal.
Interrater Reliability Goals
Reliability is not agreement for its own sake. It shows whether coders can apply the model without turning it into private interpretation. Use training vignettes, independent coding, blinded comparison where possible, adjudication notes, and revision of anchors before outcome claims are made.
For nominal fields such as source-trust category, capacity category, and agency class, report percent agreement and a chance-corrected statistic such as Cohen's kappa for two coders or Fleiss' kappa / Krippendorff's alpha for more coders. For ordered PE and MG levels, report weighted kappa or ordinal alpha. Krippendorff's common rule of thumb treats α ≥ 0.80 as a strong target and α ≥ 0.667 as the lowest range for tentative conclusions; CRM studies should aim for 0.80 or higher before making strong claims, and should revise the manual if core fields repeatedly fall below 0.70.
Minimum reporting should include:
- coder training procedure and number of practice cases;
- whether coders saw full cases, excerpts, or structured vignettes;
- reliability for PE, MG, ST, CAP, and AG separately;
- common disagreement patterns and how they changed the manual;
- whether final codes came from independent ratings or adjudicated consensus;
- examples of low-agreement edge cases retained for future testing.
Validation Ladder
The model becomes stronger through staged validation, not by being defended as finished. If it is used only as a reflective map, the standard is clarity and usefulness. If it becomes a questionnaire, training protocol, pastoral workflow, clinical-adjacent aid, or digital tool, the standard rises. MRC guidance distinguishes development, feasibility, evaluation, and implementation as interacting phases. Theory, context, stakeholder experience, uncertainty, and cost need attention throughout. Choose the next study by the unresolved question; return to earlier work when evidence requires it.
- Stage | Question | Evidence needed
- Conceptual clarity | Do users understand the difference between prediction error, meaning gap, source trust, capacity, and agency? | Interviews, cognitive interviewing, inter-rater discussion, failure-case review.
- Coding reliability | Can trained users code the same case similarly without forcing agreement? | Case vignettes, independent coding, agreement metrics, adjudication notes.
- Construct validity | Do the constructs relate to adjacent measures without collapsing into them? | Factor analysis, convergent and discriminant validity, test-retest patterns.
- Usefulness | Does CRM improve clarity, source handling, help-seeking, truthful action, or repair? | Pilot studies, comparison conditions, follow-up outcomes.
- Safety | Does it reduce false closure without increasing rumination, shame, coercion, or delayed care? | Adverse-event tracking, referral audits, red-pressure review, participant feedback.
- Implementation | Can it be used responsibly across cultures, institutions, and digital tools? | Translation studies, process evaluation, stakeholder review, context adaptation.
Instrument, Translation, and Release Gates
The word validated should never be used as a single all-purpose stamp. The Standards for Educational and Psychological Testing requires evidence for the interpretation and use being claimed, including validity, reliability or precision, fairness, administration, scoring, and the consequences of use. COSMIN likewise places content validity first: the proposed content must be relevant, comprehensive, and comprehensible for the construct, target population, and context before internal structure or reliability can make the instrument meaningful.
For CRM this produces a release sequence.
- Proposed use | Minimum gate before release | Permitted label before the gate
- Personal reflection map | Clear instructions, user comprehension, red-pressure stop rules, privacy expectations, and adverse-use feedback. | Proposed field guide for voluntary low-risk reflection.
- Facilitator or pastoral training | Defined competence, supervised practice, case-based reliability, power and consent rules, referral pathways, and safeguarding review. | Training draft; not certification or authorization to practice outside one's role.
- Research case coding | Stable coding unit, trained independent coders, field-specific agreement, uncertainty rules, and transparent adjudication. | Proposed coding manual.
- Questionnaire or score | Target construct and population, content validity, cognitive interviews, structural validity, reliability and measurement error, hypotheses testing, cross-cultural validity, fairness, and score-interpretation evidence. | Candidate items; not the CRM Inventory as an established measure.
- Intervention or decision support | Feasibility, comparison condition, benefit and harm outcomes, qualified governance, implementation evidence, and a stopping rule. | Experimental protocol; not treatment, diagnosis, or clinical decision rule.
- AI or digital implementation | All relevant gates above plus privacy, audit logs, hallucination testing, crisis routing, human review, bias testing, dependency monitoring, and defined accountability. | Prototype reflection aid.
Translation is adaptation, not word replacement. A meaning-frame question that sounds private and individual in one language may carry household, honor, caste, communal, religious, or political meanings in another. International Test Commission guidance therefore requires attention to cultural context, construct overlap, item development, administration, scoring, interpretation, equivalence, and documentation. CRM translation work should use bilingual and cultural experts, target-user cognitive interviews, independent review, pilot data, and explicit tests of whether PE, MG, ST, CAP, and AG still distinguish what the local setting needs them to distinguish. Back-translation alone is not enough.
No score without a use claim. Before reporting any total or subscale score, state who will use it, with whom, for what decision, under what conditions, and with what consequence if it is wrong. If those fields cannot be stated, keep the output as ordinary-language reflection rather than measurement.
Predictions That Can Fail
CRM proposes a better way to select and review a response, not a new general mechanism of cognition. Its central practical claim is narrower than the claim that structured reflection helps: requiring a reasoned connection between evidence, the object of repair, and the responsible actor should reduce interventions aimed at the wrong problem. That claim must earn support against competent practice using the same questions.
- Trained readers should distinguish expectation mismatch from meaning pressure when cases vary them separately. Anticipated bereavement can carry high meaning pressure; a harmless surprise can carry little.
- When a relevant fact changes, the warranted response should change with it. When only fact order or emotional wording changes, a judgment whose warrant is unchanged should remain stable.
- Explicit dependency recording should reduce wrong-target interventions without increasing missed duties, invented facts, coercion, or burden. A candidate benefit is noticing that an office must correct its record while the recipient's sound commitment should be preserved.
- A useful response should reach its recipient and persist under the conditions claimed. Better forms, greater confidence, and immediate relief are separate outcomes rather than proxies for that result.
Set the worthwhile effect, allowable harms, and intended precision before examining confirmatory results. A wide interval compatible with benefit and no benefit is inconclusive. Adequately informative evidence excluding a worthwhile advantage retires that use claim; a harmful use stops even if users like it.
Proposed Study Portfolio
The next studies answer different questions. Agreement on codes does not establish that using a worksheet improves decisions, and a better decision does not establish that help reached the person. No participant or comparative outcome results for CRM are reported in this book.
Study 1: Distinguish the Constructs
Develop facts-only cases crossing expected and unexpected events with high and low meaning pressure. Include missing information, calm but false settled beliefs, reliable and unreliable reports, and different levels of available support. Use separate judgments for PE, MG, source, capacity, and actual authority. Each judgment needs a case citation; unknown is a permitted answer.
Independent readers code before seeing author keys. Independent domain reviewers examine the cases and proposed warrants before seeing those keys too. Retain disagreement and multiple defensible readings. Revise unclear instructions during development, then freeze the manual and test it on new cases. Report agreement by field, uncertainty, and the errors the total agreement rate might conceal. Constructed contrasts establish the intended distinction; observations must establish empirical separability and added information beyond existing measures.
Study 2: Compare Equally Supported Formulations
Use two Park-informed formats with the same five ordered rows: event and expectation; meaning and appraisal; conditions and authority; response and alternative; review and burden. Give both arms the same facts, questions, evidence fields, training, response space, and resource limits. The ordinary format permits connections freely. The experimental format requires an explicit cross-reference when a judgment depends on another entry. Allow participants to say no dependency is established. This isolates a demand to record connections rather than an advantage in information, structure, or teaching time.
Park distinguishes efforts to make meaning from meanings actually made. Record initial expectation, meaning pressure, proposed action, and later outcome separately, so that completing an interpretation does not count as achieved benefit.
Use neutral format names and the same truthful introduction. In a separate design sample, check whether credibility and expected benefit differ before freezing materials. Analyze by original assignment, including ordinary-format users who spontaneously connect entries and experimental-format users who do not. Do not adjust the primary estimate for attention, expectations, or adherence produced after assignment.
Create matched case families with two factual states. One may establish that an access route works; the other that an obsolete operating record blocks it. Another may distinguish a communication aid carrying a learner's own explanation from an aid supplying an uninspected answer. Specify which warranted action changes and which commitment remains. Vary fact order separately while preserving the statements. Rewriting public training examples with new names does not produce held-out tests.
Before assignment, independent reviewers freeze acceptable action sets, factual warrants, prohibited actions, and residual uncertainty. Keep indeterminate cases and reasonable alternatives. Use new case authors and a justified task-sampling frame, including already adequate cases; selecting only cases built to favor CRM would defeat the comparison. Final case families stay with a custodian until instructions, analysis, and stopping rules are frozen.
Both formats yield the same response: one first action or justified nonintervention, at most two necessary subsequent steps, reasons, and a review point. Blind outcome raters to format by withholding planning sheets and format-specific codes. Rate against case evidence and justified responsibility, not CRM vocabulary. A material invented warrant is a factual error even when the action sounds sensible.
The primary adverse outcome is a case containing any intervention aimed at an unsupported object. Rate every action within a coordinated step; safe actions cannot dilute a wrong intervention. Prespecify case-family and state weights, missing-data treatment, and handling of indeterminate ratings. Improvement must also pass protections for omitted duties, factual accuracy, coercion, burden, and time. Doing nothing cannot win by default. A claimed larger benefit on interdependent failures needs new cases matched for difficulty and enough precision to test that contrast.
Study 3: Follow the Remedy to the Recipient
Only after feasibility and decision studies justify the next step, test whether a corrected record changes access, an agreed referral reaches care, or restitution reaches the person owed it. Prespecify the primary recipient outcome and follow-up. Retain those unreached, implementation failures, continuing support costs, and harms displaced to another person. Measure relief separately. Grief can remain after a good decision; confidence can grow around an ineffective remedy.
Compare the same person's later understanding and available action under the conditions claimed. Distinguish a temporary change of frame, learned skill, maintained external support, and an altered institutional rule. All can help; each supports a different persistence claim. Never remove essential care or safety to manufacture a test.
Later Studies Need Their Own Claim
Journaling, misinformation prompts, pastoral training, after-action reviews, and digital aids remain separate applications. Compare each with a credible active alternative, match attention and resources, and identify the benefit the actual recipient needs. A study of source prompts must preserve response to reliable warnings; a study of pastoral training must count missed protection as well as unnecessary escalation. Evidence for one application does not validate another or certify a helper's professional competence.
A Feasible Pilot Before Confirmation
Begin with offline scenarios rather than exposing people in crisis to a new intervention. Estimate comprehension, completion time, missingness, rater disagreement, variability, and the cost of required documentation. Use development cases openly; reserve independently authored final cases for later confirmation. Do not turn a successful demonstration or a small favorable pilot into an effectiveness claim.
Plan case and practitioner numbers from the intended contrast, smallest worthwhile effect, desired precision, and feasible dependence structure. Cases sharing an authoring family or practitioner are not independent workflow assignments. Team assignments need team-level analysis. A feasibility sample supplies information for design, not an automatic sample size for the later recipient study.
Publish the frozen instructions, task-sampling rationale, analysis plan, deidentified usable data where ethical, and null, mixed, and harmful results. If the comparison fails, correct the recommendation in training and digital tools as well as the research record.
Research Ethics
Consent, participant protection, privacy, fair selection, referral, and adverse-event review govern the work. Research participation must be separate from access to pastoral care, employment, education, benefits, and institutional approval. Obtain the ethics review and local expertise appropriate to the actual study.
Do not recruit people in acute danger into ordinary reflective pilots. Where a later study involves clinical risk or safeguarding, it needs qualified governance, a usable response pathway, clear confidentiality limits, and stopping authority. Care cannot depend on finishing a worksheet, disclosing injury, accepting a religious explanation, or reconciling with someone. Review harms including rumination, shame, reassurance dependence, coerced forgiveness, delayed care or reporting, false accusations, and institutional closure. A translated study also needs local comprehension, cultural review, and evidence for any proposed cross-language score comparison.
Example Inventory Items
These are not validated items. They show the kind of measurement work validation would require.
- This event violated what I expected would happen.
- I can name the specific expectation that failed.
- This event threatens the meaning frame I use to understand my life.
- I know which source or channel I am trusting.
- I have enough support and stability to process this honestly.
- I can name one truthful action available to me now.
- My current explanation helps me face reality rather than avoid it.
- My current explanation protects someone or something from needed correction.
Related research
Compare this chapter with source documents or study its biblical references.
Chapter comments