Use these forms to connect claims to evidence, decisions to responsible actors, and proposed repairs to observed results. Keep reasons, uncertainties and outcomes visible so that another reader can assess the judgment and later findings can correct it. Fit the record to the task.
A local claim record
- Claim and scope. Write one proposition that could be denied. Name subject, domain, discipline, scale, population or corpus, place, and time.
- Competent warrant. Give exact sources and loci, date or version, method, source role, uncertainty, and the premises connecting them to the claim.
- Dependencies. Name what must hold for the conclusion to follow. Identify doctrinal, empirical, historical, normative, and analogical steps, the explanatory work each connection adds, and any dependence among source sets. Give rational order, personal source, creatorhood, moral standing and Christological restoration their own premises.
- Assessment. Record inferential status, confidence, maturity, and the separate assessments of confessional adequacy, public explanatory fit, and embodied fruit.
- Rival and adverse case. State the strongest alternative, the comparison criterion, the cost of the added premise, and a case that would expose the proposed account's weakness.
- Action and revision. Name the decision, affected subjects, expected benefit, costs, power and consent, protection, review method, and a result requiring change.
For a concrete policy or treatment, supply the relevant study, diagnosis, duty and facts about the affected person. Make each general commitment answerable to the particular decision.
A domain map
A local claim record states a proposition and its warrant. A domain map shows how such claims belong to the larger subject being explained. Begin with the inquiry's scope, its actual recipients, and the goods at stake. Record six connected entries:
- The warranted local claims, their evidence, and their limits.
- The mechanisms and channels through which the good reaches its recipients.
- The forms of damage and the routes by which their effects enter neighboring lives and systems.
- The agents, their available powers, their responsibilities, and the authority required for action.
- The protection or repair available now, evidence of its receipt, and the goods left unmet.
- The limits of the domain's own fulfillment, its relation to resurrection and judgment, and the rival account or evidence that could change the treatment.
For each connection, identify the kind of relation and mark its status as established, disputed, or open. The theological horizon gives the domain its larger place; local empirical warrant remains with the evidence that establishes the claim. Retain the underlying source records so that repeated use remains traceable to the same evidence.
An open entry identifies further work. A conclusion depending on it retains that condition, while warranted findings guide present action. Integration adds explanatory content where the map identifies actual dependencies and shows what changes when one fails. The record directs the next inquiry, action, or revision.
Matched workflow comparison form
Use this form to test whether requiring an explicit record of dependencies improves decisions. The ordinary comparator includes causal debugging, rights, accountability and competent specialist practice.
- Question and estimand. State whether the added DDF record changes
causal diagnosis, repair selection, received benefit, downstream persistence, or the burdens imposed by the decision. Define the primary outcome and the effect to be estimated.
- Case construction. State the hidden causal construction, case family,
eligible recipient, domain, time, and authority. Include false and true positive cases, coexisting causes, already adequate cases, and cases in which intervention would create a burden. Freeze the generator and preserve a custodied final set before assignment.
- Assignment and parity. Record the unit randomized or matched, team or
practitioner experience, case-family balance, shared data, tools, technical assistance, time, privacy limits, repair menu, and decision authority. Do not count cases or training seeds as independent workflow assignments when they share a team, generator, or learned answer.
- Workflow record. Before choosing a repair, both teams use the same
ordered fields for the failed good, actual bearer, carrying conditions, good to preserve, predicted repair effect, and evidence that would overturn the diagnosis. The DDF format requires explicit references to the entries on which a judgment depends; ordinary formulation may connect them freely. Count training, writing, review, and maintenance for both.
- Repair and follow-up. Record the proposed change, its location in the
record, input, data, model, rule, office, payment, route, or deployment; the person with authority to make it; the immediate and delayed checks; and the conditions under which the repair is stopped or reversed. Verify receipt by the affected person or group, not only a changed file or dashboard.
- Outcomes and harms. Record diagnosis accuracy, actual receipt,
downstream persistence, privacy, access, displaced burdens, staff time, computational cost, and explanation burden separately. Preserve continuous measurements beneath any threshold. State which outcome is primary and which tradeoffs cannot be exchanged for it.
- Decision rule. Apply the joint criteria below, with numerical margins and an analysis fixed before assignment. Assess fidelity, leakage, lost capability, unequal resources and adverse effects separately from the primary outcome.
- Scope of conclusion. State the tested tasks, population and period.
The assignment effect includes effort induced by the record; a claim about the mental mechanism requires a further comparison. Keep evidence of using the record separate from evidence of better decisions.
Freeze the sample rationale, margins, subgroups and analysis before access to final cases or outcomes. Obtain variance information for the intended workflow outcome and submit a registered report where available. The completed classification run in Chapter 43 supplies a working technical task; this form specifies the investigator comparison.
The joint decision rule
For the household comparison, use the durable correct resolution defined in the household protocol below as the primary recipient outcome. Let Δ_B be the DDF arm's resolution rate minus the active comparator's, and let Δ_(H_j) be its increase in each prespecified harm or burden. Choose a smallest worthwhile benefit δ_B>0, a permissible increase m_j for every protected outcome, a cost ceiling, and the protected case families with affected people and domain specialists. The quantities need numerical values, units and reasons before data collection.
A superiority claim requires the lower confidence bound for Δ_B to exceed δ_B, the upper bound for every Δ_(H_j) to lie below m_j, every protected family to pass its separately specified noninferiority limit, and the cost requirement to be met. Preserve each outcome rather than averaging an unpaid benefit, privacy loss or unnecessary intervention into a favorable total. State whether the cost requirement concerns a mean, a tail risk or an absolute resource limit, and assess that quantity.
A lower-cost claim requires the entire confidence interval for Δ_B to fall inside a justified equivalence interval (-ε_B,ε_B), a prespecified worthwhile cost reduction, and the same harm and family protections. Choose a primary success route before collection, or prespecify error control for claiming either route. Requiring every protection within one route differs from choosing among alternative success claims; additional confirmatory comparisons also need appropriate error control.
Match inference to assignment: team allocation requires uncertainty valid for teams, even when outcomes concern many household episodes. Retain case-family dependence, and prespecify treatment of missing or unverified receipt with sensitivity analysis. A bound crossing a required limit leaves that criterion unresolved. Equivalence requires the stated interval test; a favorable secondary outcome cannot replace a failed primary one. Informative evidence ruling out a worthwhile added benefit retires that practical claim.
The proposed advantage on failures spanning several connections requires a direct comparison of workflow effects across those cases and equally demanding local failures. Significance in one case type and nonsignificance in another does not establish a difference between their effects. Prespecify whether this interaction is confirmatory or exploratory, and include cases where a second intervention is unnecessary or harmful.
Classification task and repair protocol
Generate circles and triangles against horizontal or vertical striped backgrounds. Shape determines the label; location, size, rotation and brightness vary while leaving the full outline visible. Begin with equal shape counts, but pair 95 percent of circles with horizontal stripes and 95 percent of triangles with vertical stripes. The remaining examples reverse the pairings. Class balance therefore leaves four unequally frequent groups.
An intended larger recipe uses 64-pixel-square images, a small network with three convolutional blocks and a final classification layer: 20,000 training images, 4,000 fresh group-balanced repair images, 2,000 diagnostic images, 2,000 tuning images and 8,000 final images per evaluation condition. Fix architecture, optimization, initialization procedure and training budget during feasibility. These quantities are starting choices, not a power analysis. The completed preliminary run in Chapter 43 uses an explicitly smaller architecture and sample.
Keep every variation of one foreground instance within its split. File names, borders and rendering artifacts must supply no unintended answer key. If the network already solves the task, report that result rather than searching secretly for an unusually weak model to rescue. Freeze the case-generating procedure before comparison.
Competent alternatives
An untreated average-error model can exhibit the problem but cannot serve as the sole opponent. Include balanced-data training, group distributionally robust optimization and deep feature reweighting, implemented and tuned competently.
Group DRO trains against specified groups' largest loss. Sagawa and colleagues found that fitting every training group could coexist with poor rare-group generalization; regularization or early stopping materially improved the results. DFR instead retrains the final layer on balanced data while preserving the feature extractor, where useful target features remain available. Le and colleagues' restricted image analysis found remaining dependence after DFR, with limitations in activation-map evidence. Improved accuracy is not proof that every unwanted dependence disappeared.
Applying the matched workflow form
Apply the matched workflow form above to a comparison of how teams select repairs, separately from blinded assessment of the saved interventions. Both groups may inspect preprocessing, vary backgrounds, probe representations, and check deployed versions. The only required-format difference is the explicit dependency record.
Household case families and received outcomes
A household-assistance case tests the relation developed in Chapters 31 and 36. Two households use the same temporary-housing address; falsely treating them as one creates a fraud finding and blocks a heating credit. Repair may also depend on payment, delivery, or a usable route. Both workflows receive the records, decision path, repair menu, and authority limits, without wording that reveals the hidden causal answer.
Prespecified case families include false household merges, true duplicate claims, stale payment records, unusable routes, failed delivery after approval, combined causes, and already adequate cases. Assign distinct teams or matched team bundles to workflows, balancing experience and access before assignment. Account for dependence among decisions made by one team or generated from one case family.
Test the selective prediction by prespecifying cases whose operative repair crosses records, authority, resources, or access, together with equally difficult cases whose relevant evidence and remedy lie within one responsible setting. Independent reviewers classify these relations from the facts before seeing either workflow's results and set the task weights before assignment. Compare the workflow difference between those strata as well as the primary outcome across the registered case mix. A larger gain in fragmented cases supports the narrower prediction; it cannot replace the joint benefit, harm, and cost requirements. Matching difficulty and retaining local cases prevent the comparison from selecting only problems built to reward the extra record.
Register the causal question, eligibility, assignment unit, workflow, follow-up, outcomes, and decision rule before collecting results. Randomization is preferred; a controlled nonrandomized design must state its remaining exchangeability assumptions. A target-trial description fixes which intervention and outcome are being compared.
The primary recipient outcome is the proportion of assigned cases correctly resolved and still correct at the prespecified follow-up. For an entitled household, resolution requires the correct benefit actually received and a correct continuing status. For an already correct case or justified denial, it requires preserving that status without inventing a defect. The hidden case construction supplies the reference; independent assessment checks the actual route and downstream result. Report each case family separately so that preserving easy cases cannot conceal failure to restore wrongfully denied assistance.
Apply the joint decision rule above to durable resolution, reporting each harm, cost, and protected family separately. Better paperwork cannot compensate for an unpaid claim or a displaced loss.
The household workflow study needs numerical margins, sample size, and variance information for its own clustered outcomes. The classification feasibility run supplies none of those estimates.
The shared questions may themselves explain any benefit. Required documentation earns additional practical credit only through better recipient outcomes or lower cost under the decision rule. Assess its use separately: a complete record can accompany ineffective repair, while ordinary reasoning can make the right connections without that record.
The sharpest prediction concerns broken connections concealed by accurate local records. Required dependency recording should help most when entitlement, authority, delivery, and continued access come apart. Compare these cases with equally demanding local faults, including cases where a needless second intervention would delay receipt or expose private information.
Prespecify the contrast across these case types and retain case-family dependence. Finding a necessary second repair on unfamiliar cases would support the prediction; recalling a taught label would not. Decision traces can show which links were used, and a controlled component comparison can test their contribution.
For the classifier study, reproduce the failure and identify the model actually loaded, transformed input and any display override. Change backgrounds while holding foreground fixed, then foreground while holding background fixed. Valid matched images test irrelevant dependence. Cutting out a shape can alter contrast or borders and introduce another cause; a heat map suggests investigation rather than replacing intervention.
Admit incorrect labels, missing triangle points, normalization faults and stale weights as competing or coexisting causes. If the pipeline is correct and backgrounds drive prediction, probe the fixed representation with a new readout trained on separate balanced repair data. Success shows that this representation and newly learned readout support the task, not that the original classifier used that rule. Failure may concern the probe or its optimization rather than prove absence of shape information.
Select the restricted repair if it meets tuning criteria; otherwise assess full retraining or another justified intervention within budget. A pipeline fault should be repaired before reassessing the model. Identify whether the change lives in a record, input, weights, data, objective or deployment control. Durable service reliability does not require every support to become internal: maintained human review or a reference store may be appropriate. Test removal of a protective control offline, and count its continuing operational burden.
Untouched evaluation evidence
Reserve final collections preserving, balancing and reversing the original association, plus unfamiliar stripe details and new valid combinations of shape position, size and rotation. Missing outlines would test a different task. A custodian should withhold final examples until diagnosis, intervention and claimed operating conditions are frozen. A subsequent change needs new untouched confirmatory evidence.
Specify worst-group error on the balanced final collection as the primary case outcome, with useful original-distribution performance retained within a prespecified margin. Recompute the worst group after each repair; report every group's counts and error, stated-frequency averages and separate stress tests. Define the worthwhile improvement before viewing results.
Background-change consistency is secondary: always answering circle would be consistent but useless. Report correctness on both members of matched pairs. Assess supplied probabilities against outcomes within relevant groups and distributions. If the service abstains, report coverage, deferred cases, errors and review burden; refusing everything cannot count as a successful classifier.
Diagnosis needs predicted effects beyond plausible narration. In constructed cases the injected fault is known, but its actual operation still needs confirmation. “Not yet distinguished” should outperform a confident false explanation. Count annotation, preparation, failed repairs, specialist help, computation, time and continuing controls. Retain continuous measurements beneath pass marks; a threshold can make gradual improvement appear discontinuous.
Assignment and negative results
A single model establishes feasibility, not workflow effectiveness. Freeze multiple cases involving shortcut dependence, pipeline faults, loading errors and already adequate performance. Include the last category to detect invented problems and needless intervention. Assign practitioners with balanced expertise to matched case bundles, avoiding repeat exposure to a case whose answer they learned under the other workflow.
Images within a model, training seeds within a case, and cases sharing a practitioner or generating family are not independent workflow assignments. Specify case and practitioner counts, meaningful effect, intended precision and uncertainty analysis after feasibility but before final evaluation. A wide interval containing worthwhile benefit and no benefit is inconclusive, not equivalence.
Assess workflow use separately from blinded assessment of the saved interventions. If the package helps, a component comparison can distinguish useful causal questions, failure predictions, attention to goods, and the effect of writing a decision down.
Lost capability and additional effort count alongside any accuracy gain. Leakage, invalid transformations, or poorly implemented alternatives defeat the comparison. Informative evidence excluding a worthwhile gain retires that practical claim; a different educational claim requires its own study.
The first environment remains offline. A service assessment must also test users' actual input and output paths, permissions, consequences, maintenance, usable review, stopping authority, and whether corrections reach affected decisions. An unread alert or powerless appeal does not provide accountability.
Classroom transfer study
Chapter 43 describes the initial staff simulation and its unrehearsed transfer task.
Success justifies a distinct learning study of pupil understanding. That study needs prior, immediate, delayed and transfer assessments, accessible instruction and essential support in every group. Teacher or classroom assignments determine the analysis: pupils sharing instruction are not independent method assignments. Permission, learner participation, privacy and prompt correction of discovered disadvantage belong in that study. Include absences, increased staff work and burdens displaced onto families.
Formation over time
The analytic state is
Record given nature and kinds of powers N; durable dispositions Θ_t; active context C_t; source trust T_t; valuation V_t; embodied habit H_t; relational openness R_t; and material, social, liturgical, and institutional environment E_t. Present ability depends on their interaction. The notation specifies questions rather than numerical coordinates.
Compare the same person at relevant times. What changed quickly? What persisted? What support or constraint changed? What acquisition history is known? Which powers were available at the earlier act? How freely a pattern was acquired is a question for its history. Distinguish intended change, observed change, attributed cause, and the subject's own account.
CRM: a record for a correction moment
- Name. What occurred? Separate observation, report, interpretation, and uncertainty. Preserve the original expectation before hindsight changes it.
- Sort. Which prediction failed? Which meaning, obligation, identity, hope, or allegiance was pressured? The two axes are distinct: a true obligation can survive a failed expectation.
- Check. What warrants source trust? What do safety, health, time, attention, and support permit now? What truthful action lies within this person's agency?
- Repair. Identify the actual problem: mistaken description, unmet practical need, injury, chosen wrong, unreliable channel, institutional concealment, or a mixture. Record whose action is required and whose good it serves.
- Review. Set an appropriate occasion for checking facts, access, capacity, decisions, relationships, adverse effects, and unresolved loss. Confirm the outcome by these checks rather than by immediate calm.
Manageable pressure permits reflection; overwhelming distress calls for support and pacing; danger or acute clinical concern calls for protection and qualified help. Give care freely; never make disclosure, confrontation, reconciliation, or religious interpretation its price.
Check four false resolutions: facts denied; identity protected against evidence; spiritual closure announced before lament or repair; and institutional reputation protected from correction. Another person's agreement, a completed form, and reduced discomfort can each coexist with these failures.
Testing CRM
Test whether prediction error (expectation mismatch) and meaning pressure can be distinguished reliably, whether they add information beyond existing measures, and whether the linked record improves selection of a warranted action. The coding study and the workflow comparison answer these different questions. Specify participants, follow-up, sample rationale, missing-data treatment and adverse-effect recording before collection; report null, mixed and harmful findings alongside favorable ones.
Materials and procedure
The companion ddf-crm-pilot-2026-10-12 supplies twelve public construct cases, five matched families with two factual states and two fact orders each, provisional keys, matched workflow instructions, blank rating and independent-review forms, and a facts-only packet exporter. It contains no participant observations. Use these materials for development, then have a custodian construct and withhold new final families; changing public cases' names or numbers does not create a held-out test. The reading PDF includes ddf-crm-pilot-companion.zip as an attachment, also supplied beside the PDF.
Park's distinction between effort to make meaning and meaning actually made gives the measurements their order. She also separates appraisal from the larger beliefs and goals against which an event is assessed. In a one-year observational study of 172 cancer survivors, Park and colleagues associated meaning-making efforts with adjustment through measured meanings made. Record expectation, meaning pressure, proposed action and later outcome separately, so that interpretive effort is not counted as achieved benefit.
First, have independent readers code facts-only cases before seeing their keys. Cross expected and unexpected events with high and low meaning pressure: anticipated bereavement can be deeply disruptive, while a harmless measurement surprise may require little revision of identity. Each axis permits low, high or insufficient-information judgments supported by a case statement. This constructed crossing specifies the distinction; empirical separability requires observations. Record source reliability, available capacity and actual authority separately. Include calm but false settled beliefs. Assess interrater agreement, retain disagreements and revise unclear items before freezing the manual. Check translation and measurement comparability when extending to another setting.
Second, test whether making a dependency visible helps a practitioner notice a missing factual warrant or responsible actor. Compare two equally structured Park-informed formulations using the same five ordered rows: event and expectation, meaning and appraisal, conditions and authority, response and alternative, and review and burden. Match evidence fields, questions, training, response space and resource limits. The ordinary format permits connections freely; the experimental format requires an explicit cross-reference whenever another entry is necessary to a judgment. Allow the answer that no dependency is established, and count the added effort in the intervention's cost.
Use neutral names and the same truthful introduction. Check perceived credibility and expected benefit in a separate design sample before freezing the formats, since active controls can generate different expectations. Analyze by original assignment, retaining ordinary participants who spontaneously link and experimental participants who do not comply. Leave postassignment attention, expectancy and adherence unconditioned in the primary analysis. The effect being estimated includes what the required record demands of its users.
The matched families ask whether judgment tracks the fact that matters. A working access route should be preserved; an obsolete operating identifier requires correction. A communication aid can carry a learner's own explanation or supply an answer the learner could not inspect. A tutor can hold correction authority or need another office's action. Specify which action or evidential claim changes and which commitments remain warranted, then compare those state differences directly between formats. The order variants preserve identical statements and test order robustness. To test the proposed larger benefit on interdependent failures, obtain new families with independently matched difficulty, justified task weights and adequate precision.
Before assignment, independent domain reviewers judge facts-only cases, examine the proposed contrasts, and see the author key last. Freeze acceptable action sets, factual warrants, prohibited interventions, residual uncertainty and whose interests each judgment protects. Retain multiple defensible actions, disagreements and indeterminate cases. New authors draw cases from a justified task-sampling frame without selecting for expected CRM advantage. Outcome raters receive the common action response with its reasons and case citations; withhold planning worksheets and format-specific codes. Score against evidence and justified responsibilities. Treat a material invented warrant as a factual error even when the proposed action is plausible.
The primary adverse outcome is a case-level occurrence of any intervention aimed at an unsupported object: for example, requiring a recipient to revise a sound commitment when an office must correct its record. Give both arms one first action or justified nonintervention and at most two necessary subsequent steps, with reasons and review. Rate every action within a coordinated step; safe actions cannot dilute a wrong intervention's rate. Before assignment, define family and state weights and the treatment of missing or indeterminate ratings. Improvement must pass the omission, factual-accuracy, coercion, burden and time protections as well as reduce wrong-target interventions; this prevents inaction from winning by default.
Use the scenario pilot to estimate completion time, rater disagreement, missingness and variability for the later comparison. Assign practitioners or teams and retain case-family dependence; set meaningful margins and precision before confirmation. A separate recipient study establishes whether corrected records, accessible care or agreed repair reach people and persist. Keep these outcomes distinct from immediate relief: unresolved grief can accompany a better decision, and relief can accompany an uncorrected injury.
Transfer between fields
Name the source relation and its evidence, target question, proposed correspondence, points of failure, extra premises, source dependence and target-domain test. State what the transfer explains and what would expose a merely verbal resemblance. Identify its intended use as hypothesis, illustration, explanation, prediction or decision aid, with the warrant that use requires.
For an emergence claim, add base, macrovariable, coarse-graining, timescale, intervention, mechanism of constraint, comparison with a lower-level model, and negative cases. For historical reconstruction, add source dependence and competing reconstructions. For a clinical application, add qualified assessment and relevant evidence. For a public intervention, add distribution of costs, implementation, appeal, and effects on neighboring systems.
From intended repair to observed repair
Record the damaged good, actual bearer, responsible office or agent, remedy, resources, timeline and evidence of receipt. Check restitution, corrected records, restored access and changed rules separately. Trace any residual loss or cost displaced onto another person. Reopen the decision when observed effects depart from the promised benefit.
Related research
Compare this chapter with source documents or study its biblical references.
Chapter comments