Bounded Rationality Design

Match decision method, search depth, sufficiency threshold, and escalation to the real limits and stakes of the choice.

1. Overview

Bounded-Rationality Decision Design creates decision processes that work under real limits instead of assuming perfect information, unlimited computation, exhaustive search, or inexhaustible attention. It does not celebrate shortcuts. It makes every important bound explicit, allocates scarce rigor to the choices where it matters, and gives simple methods declared validity limits, escalation paths, and feedback.

The pattern starts from a practical claim: a method is rational only relative to its cost, available resources, consequence profile, reversibility, and accountability obligations. A low-stakes recurring choice can often use a vetted default. A consequential irreversible choice may require independent review even when time is scarce. The architecture therefore separates decision need from method habit.

The pattern integrates five moves: expose limits, classify consequence, define sufficiency, select and bound a method, and learn from outcomes. Each move produces an auditable artifact. Together they prevent both analysis paralysis and reckless simplification.

2. Why This Pattern Exists

Classical optimization is useful when objectives, options, probabilities, constraints, and computation are sufficiently available. Many human and organizational decisions meet none of those conditions. Information arrives late or is expensive; objectives conflict; attention is interrupted; expertise is distributed; coordination takes time; and consequences fall unevenly across people. Pretending these limits away merely hides where judgment enters.

Boundedness also changes across the lifecycle. A crisis compresses time. A recurring process generates data and supports automation. A high-stakes exception may deserve more resources than routine cases. A new participant may need accommodation or explanation that an expert does not. A mature design treats these changes as parameters, not noise.

The pattern exists to allocate decision effort deliberately. It turns unspoken shortcuts into governed methods, turns stopping into an explicit criterion, turns escalation into a designed transition, and turns observed outcomes into calibration evidence. The objective is not maximal analysis; it is reliable and legitimate decision performance within defensible limits.

Design principles

• Declare limits before evaluating whether performance is adequate.
• Spend rigor in proportion to consequence, uncertainty, and irreversibility.
• Prefer reversible provisional actions when delay is costly.
• Preserve minimum safety, rights, and fairness regardless of resource pressure.
• Treat every heuristic, default, algorithm, and delegation as context-bounded.
• Measure the burden shifted to affected parties, not only the burden saved for decision makers.

3. Problem Signature

The structural problem appears when the decision process requires more search, evidence, comparison, memory, calculation, or coordination than the actors can supply, or when it silently substitutes an unreviewed shortcut. Symptoms often alternate: exhaustive review for trivial choices, followed by depleted attention and careless treatment of consequential cases.

Diagnostic questions

1. What decision must close, by whom, and by when?
2. Which information is available now, obtainable later, unreliable, or unknowable?
3. What time, attention, expertise, computation, and coordination budget exists?
4. What is the harm of delay, error, inconsistency, exclusion, or irreversibility?
5. Which options or affected parties are easiest to overlook?
6. What method is currently used, and where does its validity end?
7. What counts as good enough, and who authorized that threshold?
8. What anomaly, uncertainty, disagreement, or impact should trigger escalation?
9. Can the decision be reversed, repaired, appealed, or revisited?
10. What outcome evidence will update the design?

Severity levels

A mild case produces inconvenience and avoidable delay. A moderate case produces inconsistent treatment, fatigue, rework, and missed options. A severe case affects rights, safety, livelihood, or irreversible commitments through opaque shortcuts or false precision. Severity must drive process allocation; it must not be inferred only from how difficult the analysis feels.

4. Intervention Signature

The intervention is a tiered architecture, not a single rule. First define the decision boundary and minimum obligations. Next expose resource and information limits. Then classify stakes, reversibility, recurrence, and error asymmetry. Establish a sufficiency criterion before option search. Choose the least burdensome method that can meet the obligation. Bound its search and execution cost. Add defaults, delegation, accommodation, exceptions, and escalation. Finally compare outcomes with expectations and revise the architecture.

End-to-end workflow

1. Frame: identify the decision, owner, affected parties, deadline, and exclusions.
2. Bound: inventory information, time, attention, computation, expertise, and coordination.
3. Protect: declare non-negotiable safety, legality, rights, fairness, and access requirements.
4. Classify: score stakes, reversibility, recurrence, uncertainty, and error asymmetry.
5. Define sufficiency: specify the minimum acceptable outcome and prohibited failure.
6. Select a method: choose a heuristic, algorithm, expert judgment, default, delegation, or hybrid.
7. Limit search: set evidence, option, time, and comparison budgets with reentry conditions.
8. Instrument escalation: define thresholds for anomaly, uncertainty, disagreement, and impact.
9. Execute and record: capture method, evidence, assumptions, exceptions, confidence, and residual risk.
10. Calibrate: compare forecast and outcome, then update thresholds, methods, and defaults.

The intervention succeeds when method selection is explainable before the result is known. Post-hoc rationalization is a failure even when an individual outcome happens to be favorable.

5. Components

1. Decision Context Boundary

Defines the decision class, actor, affected parties, time horizon, recurrence, and conditions outside the design. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

2. Resource Limit Profile

Makes available time, staffing, computation, money, expertise, and coordination capacity explicit. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

3. Information Availability Map

Separates known, observable, obtainable, delayed, unreliable, and irreducibly unavailable information. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

4. Attention and Cognitive Load Budget

Allocates scarce attention and working-memory demand across the decision sequence. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

5. Stakes and Reversibility Profile

Classifies consequence magnitude, recoverability, lock-in, and affected-party exposure. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

6. Aspiration and Sufficiency Criterion

Defines what counts as good enough without pretending that exhaustive optimization is possible. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

7. Method Portfolio

Lists admissible heuristics, algorithms, expert judgment, defaults, delegation paths, and hybrid methods. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

8. Search and Option Budget

Limits option generation, evidence gathering, comparison depth, and stopping cost. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

9. Escalation Threshold

Specifies when stakes, uncertainty, anomaly, or disagreement require more rigorous review. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

10. Default and Delegation Rule

Assigns repeatable low-stakes choices to safe defaults, automation, or qualified owners. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

11. Uncertainty and Error Budget

Records tolerated uncertainty, false-positive and false-negative costs, and residual decision risk. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

12. Accountability and Transparency Contract

Defines what reasoning, evidence, exceptions, and affected-party impacts must remain inspectable. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

13. Outcome Feedback and Calibration Loop

Compares outcomes with expectations and updates methods, thresholds, and limits. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

14. Accessibility and Accommodation Path

Adapts the decision environment for differing cognitive, language, sensory, and executive-function needs. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

15. Environment Simplification Plan

Removes avoidable choices, interruptions, and interface burden before asking people to reason harder. This component must state its owner, evidence source, update cadence, and relationship to escalation. It is incomplete if it only names a limit without changing the decision process. Review it for distributional burden: a process that saves time centrally by transferring confusing work, delay, or risk to affected people has not actually reduced total decision cost.

Minimum contents: scope; measurable or inspectable indicator; validity boundary; exception path; and a trigger for revision. Validation: test one routine case, one ambiguous case, one high-consequence case, and one accessibility or affected-party edge case.

6. Mechanisms

1. Decision Method Triage Matrix

Scores stakes, latency, uncertainty, recurrence, and reversibility to choose a heuristic, algorithm, expert review, or hybrid path. It implements one part of the parent architecture and must not be mistaken for the entire archetype. Configure it with an explicit decision class, resource limit, sufficiency or success criterion, exception path, and review owner.

Operating sequence: declare input and stakes; apply the rule or tool; record eliminated options or delegated work; test escalation conditions; preserve override and appeal where appropriate; and log evidence for calibration. Misuse risk: the mechanism can create false objectivity or premature closure when its context boundary is omitted.

2. Satisficing Threshold Rule

Closes search when a candidate meets explicit minimum criteria and continued search has lower expected value. It implements one part of the parent architecture and must not be mistaken for the entire archetype. Configure it with an explicit decision class, resource limit, sufficiency or success criterion, exception path, and review owner.

Operating sequence: declare input and stakes; apply the rule or tool; record eliminated options or delegated work; test escalation conditions; preserve override and appeal where appropriate; and log evidence for calibration. Misuse risk: the mechanism can create false objectivity or premature closure when its context boundary is omitted.

3. Timeboxed Search

Allocates a bounded search interval with a fallback, checkpoint, and escalation path. It implements one part of the parent architecture and must not be mistaken for the entire archetype. Configure it with an explicit decision class, resource limit, sufficiency or success criterion, exception path, and review owner.

Operating sequence: declare input and stakes; apply the rule or tool; record eliminated options or delegated work; test escalation conditions; preserve override and appeal where appropriate; and log evidence for calibration. Misuse risk: the mechanism can create false objectivity or premature closure when its context boundary is omitted.

4. Progressive Option Screening

Uses cheap broad filters before costly deep evaluation while retaining false-negative review and reentry. It implements one part of the parent architecture and must not be mistaken for the entire archetype. Configure it with an explicit decision class, resource limit, sufficiency or success criterion, exception path, and review owner.

Operating sequence: declare input and stakes; apply the rule or tool; record eliminated options or delegated work; test escalation conditions; preserve override and appeal where appropriate; and log evidence for calibration. Misuse risk: the mechanism can create false objectivity or premature closure when its context boundary is omitted.

5. Default and Delegation Protocol

Routes routine choices to vetted defaults, automation, or qualified owners with override and exception logging. It implements one part of the parent architecture and must not be mistaken for the entire archetype. Configure it with an explicit decision class, resource limit, sufficiency or success criterion, exception path, and review owner.

Operating sequence: declare input and stakes; apply the rule or tool; record eliminated options or delegated work; test escalation conditions; preserve override and appeal where appropriate; and log evidence for calibration. Misuse risk: the mechanism can create false objectivity or premature closure when its context boundary is omitted.

6. Cognitive Offloading Aid

Moves memory, comparison, sequencing, or calculation burden into checklists, tables, prompts, or tools. It implements one part of the parent architecture and must not be mistaken for the entire archetype. Configure it with an explicit decision class, resource limit, sufficiency or success criterion, exception path, and review owner.

Operating sequence: declare input and stakes; apply the rule or tool; record eliminated options or delegated work; test escalation conditions; preserve override and appeal where appropriate; and log evidence for calibration. Misuse risk: the mechanism can create false objectivity or premature closure when its context boundary is omitted.

7. Algorithmic Escalation Gate

Escalates from a simple rule to formal analysis when threshold conditions or anomaly signals are met. It implements one part of the parent architecture and must not be mistaken for the entire archetype. Configure it with an explicit decision class, resource limit, sufficiency or success criterion, exception path, and review owner.

Operating sequence: declare input and stakes; apply the rule or tool; record eliminated options or delegated work; test escalation conditions; preserve override and appeal where appropriate; and log evidence for calibration. Misuse risk: the mechanism can create false objectivity or premature closure when its context boundary is omitted.

8. Two-Stage Review

Separates rapid provisional action from later verification, correction, or ratification. It implements one part of the parent architecture and must not be mistaken for the entire archetype. Configure it with an explicit decision class, resource limit, sufficiency or success criterion, exception path, and review owner.

Operating sequence: declare input and stakes; apply the rule or tool; record eliminated options or delegated work; test escalation conditions; preserve override and appeal where appropriate; and log evidence for calibration. Misuse risk: the mechanism can create false objectivity or premature closure when its context boundary is omitted.

9. Choice Architecture Simplification

Reduces irrelevant options and presentation burden while preserving meaningful alternatives and agency. It implements one part of the parent architecture and must not be mistaken for the entire archetype. Configure it with an explicit decision class, resource limit, sufficiency or success criterion, exception path, and review owner.

Operating sequence: declare input and stakes; apply the rule or tool; record eliminated options or delegated work; test escalation conditions; preserve override and appeal where appropriate; and log evidence for calibration. Misuse risk: the mechanism can create false objectivity or premature closure when its context boundary is omitted.

10. Post-Decision Calibration Review

Compares forecast, confidence, process, and outcome to recalibrate future thresholds and methods. It implements one part of the parent architecture and must not be mistaken for the entire archetype. Configure it with an explicit decision class, resource limit, sufficiency or success criterion, exception path, and review owner.

Operating sequence: declare input and stakes; apply the rule or tool; record eliminated options or delegated work; test escalation conditions; preserve override and appeal where appropriate; and log evidence for calibration. Misuse risk: the mechanism can create false objectivity or premature closure when its context boundary is omitted.

7. Parameters

The design must parameterize rather than merely narrate its limits. A time budget can be fixed or stage-specific. Search can stop at a candidate count, evidence-cost ceiling, confidence band, or marginal-value rule. Stakes combine magnitude, probability, distribution, rights impact, and reversibility. Sufficiency can be conjunctive—every minimum constraint must pass—or tiered, with preferred and fallback criteria. Escalation thresholds should combine quantitative signals with qualitative red flags and affected-party challenges.

Parameter interaction

A shorter time budget does not automatically justify a lower safety threshold; it may instead require a reversible action and immediate escalation. Higher stakes can justify more costly information gathering, independent review, or algorithmic support. Greater recurrence can justify investment in defaults, automation, and calibration. Lower reversibility demands stronger pre-decision scrutiny. Higher uncertainty can warrant robust or minimax choices instead of false expected-value precision.

Parameter changes must be versioned. A process tuned during stable conditions can fail during crisis, growth, staffing change, or population shift. Every parameter therefore needs an owner, rationale, effective date, monitoring signal, and rollback rule.

8. Invariants

Minimum safety, legality, rights, procedural fairness, and accessibility remain protected even when time or information is scarce. The design must preserve meaningful escalation and exception access, visibility of residual uncertainty, traceability for consequential eliminations, and ownership of final decisions.

The method may change while these invariants remain. A heuristic can replace an algorithm for a low-stakes case, or a provisional action can precede full review during an emergency, but neither transition may erase accountability. A default can reduce burden, but it may not silently remove a meaningful alternative. Delegation can move work to expertise, but not responsibility into a void.

A further invariant is epistemic honesty: uncertainty must not be converted into arbitrary certainty merely to close the decision. The process can choose under uncertainty, but it must label what is known, assumed, estimated, and unresolved.

9. Outcomes

Expected outcomes include shorter and more predictable decision latency, fewer fatigue-driven errors, explicit good-enough stopping, better allocation of expert review, and more consistent escalation. Quality should improve not because every choice receives more analysis, but because consequential and uncertain choices receive the right analysis while routine choices consume less avoidable attention.

Measurement portfolio

• latency and work cost by stakes tier;
• decision reversals, rework, incidents, and appeals;
• missed-escalation and unnecessary-escalation rates;
• confidence calibration and forecast error;
• false-negative rates in pruning or screening;
• exception use and override outcomes;
• burden, comprehension, and accessibility by affected group;
• default drift and context-change response time;
• consistency across comparable cases; and
• residual-risk acceptance and ownership.

No single metric is sufficient. Faster closure can hide worse exclusion; fewer escalations can hide weak detection; higher consistency can reproduce a biased default. Use a balanced measurement portfolio and inspect subgroup impacts.

10. Tradeoffs

Speed competes with rigor, simplicity with agency, consistency with contextual judgment, automation with accountability, and search breadth with evaluation cost. These tensions cannot be eliminated. The pattern makes them governable.

Use stakes and reversibility to manage speed versus rigor. Use progressive disclosure and meaningful override to manage simplicity versus agency. Use validity boundaries and exceptions to manage consistency versus context. Use decision ownership, explanation, monitoring, and appeal to manage automation versus accountability. Use staged screening, diversity preservation, holdouts, and reentry to manage breadth versus cost.

There is also an investment tradeoff. Building method portfolios, defaults, and calibration systems costs more upfront than improvisation. Recurring decisions often repay this cost; one-off choices may need a lighter artifact. The decision architecture itself should be bounded and proportionate.

11. Failure Modes

Boundedness as excuse

+Resource limits become a rhetorical cover for negligence or discrimination. Preserve non-negotiable standards and explicit residual-risk ownership.

False sufficiency

+The convenient option is called good enough without a prior criterion. Set criteria before search and validate them against outcomes.

Premature pruning

+Cheap filters remove viable or underrepresented options. Audit exclusions, preserve diversity, sample rejected cases, and allow reentry.

Escalation failure

+A shortcut continues after its validity boundary is exceeded. Instrument anomalies, disagreement, uncertainty, and impact.

Default lock-in

+A once-safe default persists after the environment changes. Set expiry, monitoring, exception, and rollback.

Burden shifting

+Decision-maker effort falls while applicant, customer, patient, worker, or citizen burden rises. Measure end-to-end burden and distribution.

Automation reification

+Tool output is treated as objective. Preserve model limits, explanation, override, independent review, and accountability.

Accessibility failure

+The process assumes one cognitive, language, sensory, or executive-function profile. Provide accommodations and alternate paths without penalizing use.

12. Variants

1. Time-Critical Bounded Decision Design

Use precommitted rules, rapid triage, and later verification when delay is itself dangerous. Use it when the decision window is shorter than full analysis and a reversible provisional action exists. Its distinctive feature is that time and latency dominate the resource-limit profile. It remains the same parent because it retains explicit limits, method selection, sufficiency, escalation, and calibration.

Key components: resource_limit_profile, stakes_and_reversibility_profile, escalation_threshold. Key mechanisms: decision_method_triage_matrix, two_stage_review. Main failure mode: Treating urgency as permission to skip accountability.

2. High-Stakes Escalation Design

Use bounded preliminary screening but require independent, explainable escalation for consequential or irreversible choices. Use it when errors can materially affect rights, safety, livelihood, or irreversible commitments. Its distinctive feature is that stakes and affected-party exposure dominate method selection. It remains the same parent because it still designs around finite information, time, cognition, and computation.

Key components: stakes_and_reversibility_profile, accountability_and_transparency_contract, uncertainty_and_error_budget. Key mechanisms: algorithmic_escalation_gate, post_decision_calibration_review. Main failure mode: Using boundedness to excuse arbitrary or discriminatory shortcuts.

3. Recurring Operational Choice Design

Standardize repeated decisions with defaults, delegation, exception paths, and outcome calibration. Use it when a similar decision recurs often enough for method learning and operational automation. Its distinctive feature is that recurrence enables defaults, delegation, monitoring, and continuous calibration. It remains the same parent because the intervention still allocates finite attention and rigor to real decision limits.

Key components: default_and_delegation_rule, outcome_feedback_and_calibration_loop, environment_simplification_plan. Key mechanisms: default_and_delegation_protocol, choice_architecture_simplification. Main failure mode: Freezing an outdated default after context changes.

Variants should collapse into the parent when the only difference is terminology, industry, or tool choice. A new archetype is justified only if consequence structure, governance, protected invariants, or method-selection logic changes enough to require a separate review contract.

13. Boundaries and Neighbor Distinctions

Bias-Specific Decision Audit checks named bias vulnerabilities; it does not design the complete bounded process. Decision Load Management manages volume, timing, and fatigue; it does not by itself specify information bounds, sufficiency, method portfolios, or escalation. Heuristic vs. Algorithm Tradeoff and Selection is a strong method-selection neighbor, but the parent also governs search, defaults, delegation, accommodation, accountability, and calibration.

Cognitive Load Reduction removes avoidable mental burden. Attention Budgeting allocates focus. Search Space Pruning narrows candidates. Minimum Sufficient Solution limits solution scope. Satisficing Threshold Design defines good-enough closure. Each can instantiate part of the parent without replacing its integrated decision architecture.

The frozen proposal to merge this candidate into Self-Efficacy Scaffolding is rejected on semantic grounds. Self-efficacy concerns capability belief built through mastery, modeling, feedback, and graduated challenge. Bounded-rationality design concerns the objective and institutional limits of decision processes. Confidence support can be useful within the parent, but it is not its canonical owner.

14. Examples

Emergency triage

+A service cannot gather complete information before acting. It uses a rapid screen, explicit high-risk triggers, a reversible provisional disposition, documented uncertainty, and second-stage verification. The design is bounded but not careless because safety invariants and escalation are preserved.

Benefits administration

+Routine complete applications follow transparent defaults. Ambiguous, high-impact, inaccessible, or contested cases receive human review and an appeal path. Outcome monitoring checks error and burden by group. The design saves attention while protecting procedural fairness.

Vendor selection

+A team sets minimum criteria, timeboxes broad search, preserves a diverse holdout, deep-reviews finalists, records conflicts, and reopens search when every finalist fails a critical constraint. The stopping rule is explicit rather than an exhausted team’s intuition.

Product configuration

+Most users receive safe defaults and progressive disclosure; experts can reveal additional controls. Changes are reversible and the interface records when defaults cause repeated overrides. Simplification reduces burden without erasing agency.

Repeated operational approvals

+Low-stakes cases route to qualified owners under a delegation rule. Anomalies, novel conditions, and high-value cases cross an escalation threshold. Quarterly calibration updates the default and examines exceptions.

15. Non-Examples

A careless guess followed by the claim that people are bounded is not the archetype. Neither is a checklist with no validity boundary, a default with no override, an algorithm with no escalation, or a time limit that ignores consequence. Exhaustive analysis of a trivial reversible choice is also a non-example because it misallocates scarce rigor.

Other non-examples include hiding a discriminatory rule behind efficiency, removing meaningful options to benefit only the decision owner, calling the first feasible option sufficient without criteria, delegating work without authority or accountability, and simplifying an interface by shifting confusion into inaccessible documentation.

16. Review and Open Questions

The candidate is recommended for use after human review of family placement, rights impact, accessibility, and the rejected frozen merge proposal. Reviewers should confirm that it becomes the general q48 owner while bias_specific_decision_audit and decision_load_management remain accepted specialists.

Open questions include how to measure total decision burden across actors, when independent rather than internal escalation is required, how sufficiency criteria should be governed in contested domains, and which aliases best distinguish the intervention from the descriptive theory. Domain pilots should report latency, outcome quality, missed escalation, exception use, burden distribution, and calibration.

The q40 collapse stop, replan finding, checkpoint, authorization, and retained debt remain unchanged. This q48 full draft does not authorize q49, the q50 checkpoint, or finalization.

Common Mechanisms

• Algorithmic Escalation Gate
• Choice Architecture Simplification
• Cognitive Offloading Aid
• Decision Method Triage Matrix
• Default and Delegation Protocol
• Post-Decision Calibration Review
• Progressive Option Screening
• Satisficing Threshold Rule
• Timeboxed Search
• Two-Stage Review

Compression statement

Make resource and information bounds explicit; classify stakes and reversibility; set a good-enough criterion; choose a heuristic, algorithm, expert, default, delegation, or hybrid method; bound search; preserve meaningful options; escalate anomalies and consequential cases; record uncertainty and accountability; and recalibrate from outcomes.

Canonical formula: decision_process_fit = alignment(method_cost, available_resources, stakes, uncertainty, reversibility, accountability); stop when sufficiency is met or marginal search value falls below search cost, unless escalation conditions apply.

Related Abstractions

Abstractions this archetype builds on — directly (a source ingredient) or as a related pattern. Links follow the typed catalog namespace.

Built directly on (3)

• Bounded Rationality: Limited decision capacity.
• Constraint: Limits possibilities to guide outcomes.
• Decision: Committing to one alternative from a set under uncertainty and trade-off, collapsing open deliberation into a chosen path and foreclosing the others.

Also references 12 related abstractions

• Algorithm: Step-by-step problem-solving procedure.
• Boundedness: Values remain within limits.
• Cognitive Load: Mental effort.
• Computability: The in-principle boundary between problems an effective procedure can solve and those none can.
• Decision Fatigue: Reduced decision quality over time.
• Feedback: Outputs influence inputs.
• Heuristic: Mental shortcuts.
• Opportunity Cost: Value of best alternative.
• Resource Management: Allocation of finite assets.
• Satisficing: Accept good-enough solution.

▸ Show 2 more

Variants

Narrower or domain-specific specializations that share this archetype's core structure. Recognized variants are established; candidate variants are provisional.

Time-Critical Bounded Decision Design · temporal variant · recognized

Use precommitted rules, rapid triage, and later verification when delay is itself dangerous.

• Distinct from parent: Requires stronger precommitment, fallback, and post-action review.
• Use when: The decision window is shorter than full analysis and a reversible provisional action exists.
• Typical domains: emergency operations, incident response
• Common mechanisms: decision method triage matrix, two stage review

High-Stakes Escalation Design · risk or failure variant · recognized

Use bounded preliminary screening but require independent, explainable escalation for consequential or irreversible choices.

• Distinct from parent: Requires independence, stronger evidence, and explicit residual-risk acceptance.
• Use when: Errors can materially affect rights, safety, livelihood, or irreversible commitments.
• Typical domains: healthcare, public policy, finance
• Common mechanisms: algorithmic escalation gate, post decision calibration review

Recurring Operational Choice Design · implementation variant · recognized

Standardize repeated decisions with defaults, delegation, exception paths, and outcome calibration.

• Distinct from parent: Emphasizes operational cadence and method maintenance.
• Use when: A similar decision recurs often enough for method learning and operational automation.
• Typical domains: operations, service delivery
• Common mechanisms: default and delegation protocol, choice architecture simplification

Near names: Capacity-Aware Decision Design, Finite-Resource Decision Architecture, Realistic Decision Process Design, Bounded Decision Method Design.