Decision making, judgment, and bounded rationality

A photograph became a choice among dangerous futures

U-2 photographs taken over Cuba on October 14, 1962, were analyzed and brought to President John F. Kennedy on October 16 as evidence of Soviet missile sites under construction. The images did not contain a decision: analysts inferred capability and timing, military staffs supplied executable plans, diplomats anticipated reactions, and the president had to act while intentions and consequences remained unsettled.¹

The Executive Committee of the National Security Council, or ExComm, considered inaction, diplomatic pressure, a missile trade, a naval quarantine, air strikes, and invasion. The October 21 minutes show participants revising and sequencing alternatives as they argued about military feasibility, alliance politics, surprise, and escalation.¹ The group constructed a choice under a deadline; the record does not show discovery of a complete menu or a provably optimal answer.

The interpretive proposition is that organizations shape a decision before the authorized person commits. They decide which cues become facts, which plans are ready, who may frame an alternative, what remains secret, and which consequences enter the room. The crisis is revealing because the possible effects were global while the recorded U.S. forum was narrow. Its roster contains U.S. officials and officers, not Cuban or Soviet representatives. Translated Cuban-Soviet records and a later Cuban state account describe mobilization under fear of invasion, Cuban objections to exclusion from the settlement, and demands absent from the superpower agreement.²

Capabilities and moral argument constructed the alternatives

An air strike looked concrete because military organizations possessed targets, plans, forces, and estimates. Officials could not be certain that it would destroy every missile, and they anticipated possible retaliation and invasion. A quarantine looked slower and less decisive but preserved time for observation and bargaining. Calling it a “quarantine” rather than a blockade was itself decision work tied to the legal and political position the United States would present.¹

Participants also disputed what kind of act an unannounced strike would be. Some invoked Pearl Harbor; others treated delay as an intolerable military risk. CIA Director John McCone's October 17 memorandum warned against a lasting “Pearl Harbor indictment” even while proposing a warning followed by a massive strike if the missiles were not removed.¹ Values were not appended after technical analysis. They altered which alternatives could be described as responsible.

Kennedy held final U.S. authority, but reconnaissance crews, analysts, service chiefs, diplomats, lawyers, and advisers constructed premises from which he chose. Secrecy gave officials room to change positions without immediate public commitment. It also excluded wider scrutiny and people exposed to the plans. Those observations support an organizational reading; they do not prove that one routine, office, or participant independently caused the quarantine.¹

The quarantine changed the next decision

Kennedy announced the quarantine on October 22. U.S. forces prepared for further action while Washington and Moscow exchanged public and private messages. On October 27, after a U-2 was shot down, Robert Kennedy warned Ambassador Anatoly Dobrynin that further incidents could lead to retaliation and pressed for a Soviet commitment within a day.¹ Preserving time made another round of bargaining possible without removing escalation risk.

The settlement joined public Soviet missile withdrawal and a U.S. pledge not to invade Cuba with a private understanding that U.S. Jupiter missiles would leave Turkey. Khrushchev's October 28 confidential letter recorded a four-to-five-month timetable and Kennedy's wish that the exchange remain private.¹ A public story of resolve can therefore obscure reciprocal concession.

Cuban leaders had not been consulted about missile removal and contested the settlement, inspection, continuing U.S. pressure, and the status of Guantánamo. The Wilson Center documents reproduce translated participant records; the later Granma account preserves the Cuban government's official retrospective position.² Both enlarge a U.S.-centered record. Neither represents Cuban civilians, political opponents, or the full range of Cuban judgments.

Records and models answer different questions

Minutes, memoranda, letters, and option papers establish what particular officials recorded and did. They do not automatically reveal private motives, unrecorded exchanges, or the complete Cuban and Soviet situation. U.S. repositories preserve the most accessible record used here; later translated Cuban and Soviet materials change the perspective without producing a representative account of everyone exposed to the crisis.¹²

Graham Allison's Essence of Decision uses the crisis to compare three explanatory frames: rational-actor choice, organizational process, and governmental bargaining. The revised edition incorporates evidence unavailable to the first edition and explicitly draws on organizational theory associated with March and Simon.³ The models are interpretive devices, not neutral transcriptions.

Barton Bernstein credits the book's influence while arguing that the revised analysis neither explains the crisis nor resolves important conceptual problems.³ That disagreement is a substantive limit. The same outcome can look different when the unit of analysis is a state, an organization, or a coalition, but comparing models does not identify a settled causal account.

The early genealogy is partly historical and partly editorial

Node I06-N01, “Classical prudence and statecraft,” and node I06-N02, “Probability, statistics, and military staff work,” are broad organizing labels. No reviewed source establishes a single tradition or direct passage between them. They preserve an imported periodization while marking the absence of a defensible lineage.

Node I06-N03 has firmer historical footing. Official U.S. Army history and the professional history maintained by INFORMS describe British and U.S. wartime operations-research groups that joined civilian scientists to operational problems, often through empirical and statistical analysis.⁴ This supports wartime institutionalization, not every claimed intellectual ancestor of operations research.

The movement from I06-N03 to Simon at I06-N04 is also not a simple revolution against wartime optimization. Independent history traces later interaction among operations research, computation, management science, and Simon's procedural rationality, but Simon's administrative research began before or during the war, and the preliminary edition reviewed here is dated 1945.⁴⁵ Chronology and conceptual contrast are supportable; a direct wartime-OR-to-Simon cause is not.

Simon replaces the fictional complete chooser

Herbert A. Simon made limits to organizational choice central. In Administrative Behavior, decision makers do not survey every alternative and maximize a complete utility function. Limited attention, information, time, representation, and computational capacity bound rational action.⁵

Satisficing is one response to those limits. Search stops when an alternative crosses an aspiration level rather than when an unknowable global optimum has been proven. Simon's 1956 paper and 1978 lecture connect bounded rationality to search, targets, learning, and adaptation.⁵ A deadline, threshold, or escalation condition is therefore not clerical detail; it helps decide when inquiry becomes commitment.

Simon also describes organizations as suppliers of decision premises. Objectives, routines, categories, authority, training, and communication narrow the space in which participants choose.⁵ This coordination is necessary for scale and is also a form of power: a nominally decentralized choice can inherit centrally constructed categories and values.

The ExComm application is editorial. Intelligence categories made some observations salient, military organizations supplied ready plans, legal language changed the diplomatic meaning of an action, and presidential authority set a stopping point.¹ Simon did not design or study ExComm in the sources reviewed here, so his theory is a lens rather than a historical influence claim.

March adds routines, coalitions, and learning

James March and Simon's Organizations describes performance programs, search, and organizational conflict. March's later model distinguishes exploration of uncertain alternatives from exploitation of known capabilities and argues that adaptive processes may refine exploitation faster, creating short-run effectiveness and long-run vulnerability.⁶

Allison's organizational-process model is a documented use of the March-and-Simon tradition. It asks how reconnaissance, air-strike, naval, and diplomatic routines shaped available action; the governmental-politics model asks how offices and bargaining positions mattered.³ Neither model proves that routines or bargaining alone caused the quarantine.

Routines store experience and lower search costs, but those are analytical functions rather than guarantees of benefit. A routine can remain reliable, become obsolete, or distribute costs outside the measures used to evaluate it. March's exploration model is a formal argument, not a general empirical estimate that one fixed exploration ratio improves every organization.⁶

Behavioral research contests a different ideal of rationality

Daniel Kahneman and Amos Tversky's experiments identify recurring effects of availability, representativeness, anchoring, framing, reference points, and losses in specified judgment and choice tasks. Prospect theory models reference-dependent choice under risk rather than every form of organizational decision.⁷

A multinational study of 4,098 participants in 19 countries reproduced most of the prospect-theory contrasts it tested, with attenuation and substantial heterogeneity in some effects. Gigerenzer and Gaissmaier's review argues that simple heuristics can be accurate in environments whose structure fits them.⁷ The evidence therefore does not warrant treating every heuristic as error or exporting one laboratory contrast to every institution.

Thinking, Fast and Slow is a reading route into this tradition, not evidence that Kahneman and Tversky analyzed ExComm. Framing and loss aversion can generate later questions about surprise attack, inaction, and reputational loss, but the reviewed sources do not establish those mechanisms as causes of any participant's position.⁷

Behavioral findings can support tests of defaults, forecasts, and review. They can also be used to manipulate people or to attribute structural failures to individual cognition. The latter propositions are ethical cautions, not findings from the 1962 record or from the cited experiments.

Sensemaking asks how a photograph became a crisis

Karl Weick directs attention before alternative selection. Sensemaking in Organizations describes identity, retrospection, enactment, social context, ongoing activity, extracted cues, and plausibility as properties of organizational sensemaking.⁸ The framework asks how participants come to inhabit a situation in which particular choices appear possible.

Dissent may therefore signal more than another preference inside a shared frame; it may expose a disputed frame. Critical-sensemaking scholarship argues that Weick gives insufficient attention to power and context, including why some identities and voices become plausible while others do not.⁸ That limitation matters in a room whose membership and authority were already restricted.

Weick did not analyze ExComm in the sources used here. Applying sensemaking to the movement from aerial image, to “offensive missile,” to intolerable threat, and then to response is an analytical comparison. Cuban and Soviet records show different interpretations, but no reviewed source reconstructs all of those frames symmetrically.²

AI expands search without escaping constructed premises

AI-assisted systems can support search, classification, simulation, synthesis, and alternative generation. A 2026 systematic review maps a large management-focused literature but does not establish uniform decision improvement. A field experiment with consultants found performance gains for some tasks and deterioration for tasks outside the tested capability frontier, making task fit a central boundary rather than a minor exception.⁹

Generative systems also introduce confabulation, attribution error, bias, homogenization, automation bias, and persuasive fluency. NIST's risk profile recommends documenting knowledge limits and checking sources and citations. Semantic-entropy research detects one subset of arbitrary wrong answers while explicitly not solving consistent or systematic errors.⁹ None of these sources establishes that an AI system can hold legitimate authority or bear the consequences of a value tradeoff.

Applying those findings to the 1962 case is a design thought experiment, not historical evidence. A system trained mainly on U.S. records could reproduce the U.S. room's categories while making Cuban experience less visible. Search, recommendation, commitment, and execution remain different grants of authority; source provenance, uncertainty, and dissent must survive synthesis.

The reviewed AI studies do not estimate effects across displaced workers, people represented in training data, data annotators, communities bearing energy and water demand, nonhuman life, or future generations. Those absences prevent a general claim that increased decision capacity creates net benefit.⁹

Four product hypotheses remain untested

A decision record can preserve a question, owner, deadline, premises, alternatives, evidence, uncertainty, dissent, grant, and expected result. That is a local design proposition. The ExComm records show the value of inspectable premises and changing alternatives, but they do not validate a particular software object or workflow.¹

• I06-P01 proposes a structured decision object. Its status remains research-hypothesis; evidence must show that people consult it and that it improves review or learning without becoming form-filling theater.
• I06-P02 proposes AI assistance for search and critique while keeping authority explicit. Its status remains research-hypothesis; comparative trials must test outcomes, automation bias, labor displacement, resource use, and whether nominal human approval is substantive.⁹
• I06-P03 proposes provenance for inherited, generated, and unsupported premises. Its status remains research-hypothesis; audits must test whether sources and corrections can be reconstructed without exposing sensitive people or dissolving responsibility.
• I06-P04 proposes later prediction review. Its status remains research-hypothesis; longitudinal evidence must test calibration, hindsight bias, gaming, punishment for honest uncertainty, and selection of only easy-to-score decisions.

The evaluation rule remains an editorial caution: a good process can produce a bad outcome under genuine uncertainty, and a reckless process can be lucky. Confidence, speed, documentation, narrative polish, and outcome are therefore insufficient alone. Whether prediction review improves decisions is an empirical question, not a result established by the historical case.

Eight nodes and seven edges make the map auditable

The sequence combines historical developments and theoretical comparisons. I06-N01 and I06-N02 are broad, provisional prehistory labels; I06-N03 marks wartime operations research; I06-N04 marks Simon; I06-N05 marks March; I06-N06 marks Kahneman and Tversky; I06-N07 marks Weick; and I06-N08 marks contemporary AI-assisted decision systems.⁴⁵⁶⁷⁸⁹

The structured edges have different evidentiary standing:

• I06-E01 is a grade-C provisional editorial periodization from practical judgment to formal uncertainty. It has no source IDs because no reviewed work supports the broad transmission claim.
• I06-E02 is a grade-B historical-institutionalization relation supported through Allied and U.S. World War II mobilization and the operations-research histories. Its support is limited to wartime groups and methods, not all of I06-N02's prehistory.⁴
• I06-E03 is a grade-C comparative periodization. Its source IDs, Allied/U.S. mobilization and Administrative Behavior, support the two endpoints; they do not establish that Simon reacted to wartime operations research.⁴⁵
• I06-E04 is a grade-B organizational-extension relation sourced to Administrative Behavior and Organizations. Shared authorship and textual continuity support the Simon-to-March connection, but the edge compresses several later developments.⁵⁶
• I06-E05 is a grade-C editorial counterpoint sourced to Organizations and Thinking, Fast and Slow. The works support the endpoints, not direct transmission or one unified behavioral theory.
• I06-E06 is a grade-C editorial counterpoint sourced to Thinking, Fast and Slow and Sensemaking in Organizations. It contrasts choice within a frame with construction of a frame; it is not a generalization from individual experiments to organizations.
• I06-E07 is a grade-C editorial extension. Sensemaking in Organizations supports the Weick endpoint; recent external AI research supports a bounded comparison, and no local AI work record supports the target node.⁹

Evidence grades describe support for bounded relations, not their importance or the size of an effect. These relations are editorial comparisons, not evidence of causal influence or institutional descent.

Related records are reading routes, not prerequisites

The six direct relations serve different purposes:

• Administrative Behavior is the primary theoretical route for organizational decision premises and bounded choice;
• Organizations is the primary-work route for programs, routines, search, and conflict;
• Sensemaking in Organizations is a theoretical counterpoint about constructing situations before choosing;
• Thinking, Fast and Slow is a later synthesis route into heuristics-and-biases research;
• organizational ignorance is an adjacent problem route for unknowns produced or protected by institutions; and
• benefit for all life is an ethical route for asking whose consequences a decision frame excludes.

No structured reading dependency is recorded. None of the six records is a formal prerequisite for another, and a related link does not establish influence, agreement, or shared empirical findings.

Established institutions are comparison paths, not descendants

The institution links are editorial test routes. Their internal records contain their own sourcing. Grouping them here does not assert adoption of a theory, common performance, equal moral status, or historical descent.

Administrative, military, scientific, and corporate comparison paths:

• Prussian–German General Staff, Allied and U.S. World War II mobilization, NASA Apollo program.

Participatory, community, local-knowledge, and ecological comparison paths:

• Botswana's kgotla assemblies, Cairo's Zabaleen recycling system, CECOSESOLA, Madagascar's GELOSE resource-management transfers, Navajo Nation Peacemaking, Niger's farmer-managed natural regeneration, Peru's rondas campesinas, Porto Alegre participatory budgeting, Somaliland's peace conferences, and vTaiwan and g0v.

Public-system, emergency, oversight, and transitional comparison paths:

• Chile's individual-account pension system, ECOWAS and The Gambia's 2016–2017 transition, Nigeria's Ebola Emergency Operations Center, Project Cybersyn, PROMESA oversight board, Rwanda's Gacaca courts, South Africa's Truth and Reconciliation Commission, and Tunisia's National Dialogue Quartet.

Coercion, exclusion, institutional failure, and harm comparison paths:

• Dawes Act allotment administration, Ethiopia's Gambella villagization program, Fujimori's reproductive-health program and forced sterilization, MINUSTAH and Haiti's cholera response, Navajo livestock reduction, PDVSA's institutional hollowing, U.S. federal Indian termination policy, and Vale and the Brumadinho dam disaster.

The comparison question is limited: how did each institution construct facts, alternatives, authority, dissent, stopping rules, and review? The links add no new assertion about a case's history or effects.

Impact remains unclassified

No structured impact record is assigned. Bounded rationality, judgment, sensemaking, and decision architecture are analytical categories rather than one intervention with a common beneficiary or outcome. The crisis record supports a bounded account of superpower decisions and Cuban leadership objections; it does not support a net classification across Cuban civilians, Soviet personnel, U.S. service members and communities, Turkish people affected by the private missile bargain, political opponents, future generations, nonhuman life, or ecosystems.¹²

The absence of an impact record means impact is unclassified, not absent. It prevents preservation of time, avoidance of immediate war, or expanded analytic capacity from silently standing in for every affected party's experience. The AI evidence is likewise task-bounded and cannot support a general net-impact judgment.⁹

Evidence still needed

• Cuban civilian, dissident, military, household, and local records that do not treat government leaders as proxies for a whole population; comparable Soviet, U.S. military-family, Turkish, and other affected-party records.
• A documented history capable of separating the broad I06-N01 and I06-N02 traditions, or evidence strong enough to replace those aggregates with narrower nodes and testable edges.
• Process evidence that reconstructs how particular ExComm participants changed premises and positions without inferring private motives from recorded speech alone.
• Comparative organizational studies testing when routines, dissent, wider representation, secrecy, and delay improve decisions and when they create paralysis, coercion, or displaced harm.
• Direct evidence connecting individual behavioral effects to organizational decisions, rather than assuming that experimental findings scale unchanged.
• AI-assisted decision trials across task types, populations, model versions, and authority arrangements, including failures, labor effects, privacy, energy and water use, and consequences for people unable to appeal.
• Longitudinal tests of I06-P01 through I06-P04, including whether decision records improve learning, whether premise provenance remains usable, and whether prediction review changes later action without punishing uncertainty.

Paths into deeper study

• Begin with the primary crisis records in the source notes and trace how the option menu changes before adopting a retrospective success narrative.
• Read Administrative Behavior for bounded choice and Organizations for routines and search, then inspect Allison's use of that tradition and Bernstein's disagreement.
• Read Sensemaking in Organizations for the construction of situations before choice.
• Compare the decision problem with executive attention and organizational sensing: one asks how to construct choice under limitation; the other asks how material reality reaches the lowest authority able to respond.

Additional reciprocal institution comparisons

Newly developed institutional records add these reciprocal comparison paths:

• Apple — institution-comparison
• Athens, 508/7–403 BCE — institution-comparison
• Intel — institution-comparison
• Meta / Facebook — institution-comparison

Each path identifies a sourced case where this idea is a defining emphasis. The relation is editorial comparison, not evidence of direct influence, shared terminology, or equivalent outcomes.