Decision-makers use aspiration levels rather than global optimization. When an alternative meets the aspiration level, search stops. This is rational given bounded cognitive and informational resources. "Click and hope" is satisficing under extreme cognitive load — rational behavior, not character failure.

CLT's three load types — intrinsic (task complexity — cannot be reduced without simplifying the task), extraneous (presentation burden — can and should be eliminated), germane (productive schema construction — should be maximized) — were formalized in subsequent work (Sweller, van Merrienboer, & Paas 1998), building on Sweller's (1988) foundational framework. When intrinsic + extraneous exceeds working memory capacity, germane processing collapses. The 2019 retrospective reconceptualized germane load not as a separate additive type but as working memory resources devoted to intrinsic load. Extended to collaborative settings (Kirschner et al. 2018: teams can share cognitive load across members). CLT remains entirely within instructional design — no published work applies it to governance, compliance, or organizational decision infrastructure. This is a confirmed gap.

Three independent traditions discover the same dual-process architecture:

Cognitive science (Kahneman): System 1 — fast, automatic, heuristic, bias-prone. System 2 — slow, effortful, analytical, capacity-limited. Transition: load/fatigue pushes 2→1.

AI research (LeCun, S8): System 1 — reactive, pattern-matched. System 2 — model-based planning, deliberative. Transition: amortized inference compiles 2→1.

Cybernetics (Beer, S9): Systems 1–3 — operations (here/now). System 4 — intelligence (there/then). Transition: variety engineering balances inside/outside.

Loss aversion (Kahneman & Tversky 1979) — the finding that losses loom larger than equivalent gains — explains why accountability is perceived as threat (potential loss) rather than support (potential gain). Governance infrastructure must trigger gain framing.

The Israeli parole board data pattern (65% favorable at session start → near 0% by session end → reset after break) is real. But the ego depletion mechanism is largely discredited: Hagger et al.'s 23-lab preregistered replication found d = 0.04 (near-zero effect). Glockner (2016) showed that case ordering artifacts could produce the observed pattern without any depletion mechanism. The general claim that decision quality may degrade with sequential decisions under load remains plausible but the specific mechanism is unresolved. No comprehensive meta-analysis on decision fatigue exists in the 2022-2026 window. The concept is in theoretical limbo.

The jam study: 24 options → 3% purchased; 6 options → 30% purchased. Schwartz distinguishes maximizers (consider all options, experience regret) from satisficers (seek "good enough," experience less regret). As choice sets grow, both satisfaction and confidence decline. Governance systems that present all business rules, precedents, constraints, and stakeholder concerns without structure overwhelm decision-makers — the governance parallel to choice overload.

Process accountability (accountable for HOW you decided) produces more systematic thinking, reduced cognitive bias, genuine consideration of alternatives, openness to disconfirming evidence. Outcome accountability (accountable for WHAT happened) produces defensive reasoning, premature closure, confirmation bias, impression management. The key moderator: accountability to an unknown audience → better thinking; accountability to an audience with known views → conformity. No published work applies this framework to AI governance or algorithmic accountability — the Raji et al. (2020) "Closing the AI Accountability Gap" framework (~830 citations) is the most influential algorithmic accountability framework but does not reference Lerner & Tetlock or use the process/outcome distinction. This is a genuine gap.

Chinese mobile phone factory: complete observability REDUCED performance; workers behind curtains performed better because privacy enabled experimentation and productive deviance. Not directly replicated but conceptually well-supported and extended to digital/algorithmic contexts (Kellogg et al. 2019). The transparency paradox distinguishes surveillance (triggers impression management) from process accountability (supports learning). Governance infrastructure must capture decision reasoning (process) without creating panoptic visibility into operational behavior (surveillance). The difference is structural.

Four independent justice dimensions: distributive (outcome fairness → satisfaction), procedural (process fairness → rule compliance and institutional trust), interpersonal (respect/dignity → supervisor attitudes), informational (transparency/explanation → voluntary helping behavior). Governance infrastructure should activate procedural justice (transparent, documented process) and informational justice (decision lineage showing why decisions were made). People comply with rules they perceive as procedurally fair.

Psychological safety — a shared belief that the team is safe for interpersonal risk-taking — predicts asking questions, admitting mistakes, reporting errors, seeking feedback. It is distinct from trust and cohesiveness. If governance infrastructure is perceived as surveillance, it will suppress the very behaviors it needs to capture. People will not document uncertainty, flag constraints, or admit process gaps if the record is used against them. Boundary condition (Edmondson 2018, 2023): PS is necessary but not sufficient — must be paired with high standards. Performative safety without genuine inclusion can be weaponized.

Silence is collective, driven by organizational structures and norms, not individual shyness. Contextual antecedents: hierarchical channels, punitive metrics, low participation. Implicit voice theories ("managers don't want bad news," "raising concerns signals disloyalty") suppress voice EVEN in psychologically safe environments with pro-voice leadership. Voice mechanisms must not just permit voice — they must actively disrupt implicit voice theories by making voice visibly rewarded, structurally protected, and normatively expected.

Near & Miceli (1985) provide the foundational definition of organizational dissidence. Building on their work, this report observes that whistleblowing occurs when internal voice channels are absent or ineffective. Hirschman's framework: Exit undercuts Voice (if unhappy members can leave, they don't advocate for change); Loyalty retards Exit, creating space for Voice. Organizations that enable Voice have higher adaptive capacity. Governed internal voice channels reduce pressure toward external disclosure while catching problems earlier.

The Recognition-Primed Decision (RPD) model: experts recognize patterns from experience and mentally simulate the first viable option. If the simulation reveals problems, they adapt; if not, they act. Decision support should surface relevant constraints, precedents, and evidence to enrich pattern recognition — not present exhaustive decision trees that increase cognitive load.

Intelligence emerges from coordination across individuals, technologies, and procedures. The cognitive unit is the system (people + tools + procedures), not any individual. Governance infrastructure that captures decision lineage, constraint histories, capacity patterns, and evidence bases composes a cognitive artifact that extends organizational decision-making capacity beyond what any individual holds.

Defaults matter enormously (organ donation opt-in vs. opt-out). Information structuring matters. Feedback timing matters. Governance infrastructure structures the decision environment (what's visible, what constraints are surfaced, what precedents shown) without restricting the decision-maker's choice. Arnott & Pervan (2005) found ~50% of DSS papers failed to ground design in decision-making research — a "crisis of relevance" that persists. No published work applies choice architecture specifically to internal organizational governance design — confirmed gap.

Frankfurt (2005): bullshit is truth-indifference (worse than lying — the liar respects truth by inverting it; the bullshitter disregards it). Spicer (2018): organizations are truth-indifference machines through incentives, hierarchy, and jargon. Ferreira et al. (2020/2022): three measurable dimensions — Regard for Truth, The Boss, Bullshit Language. Hannigan et al. (2024): "botshit" = AI-generated truth-indifferent content — AI cannot care whether its output is true (it has no epistemic stance). When humans uncritically use AI outputs, they become bullshit amplifiers. The truth-indifference chain: organizations produce truth-indifferent communication (OBPS) → AI amplifies it (botshit) → humans cannot filter it (cognitive BS sensitivity limits, per Pennycook et al. 2015) → and training itself breaks the invariances needed to detect it (Chlon, per S11). This chain feeds P2.

Graeber's five types of bullshit jobs include "box-tickers" — people who create the appearance of governance without substance. The critical warning for governance infrastructure: process accountability risks becoming box-ticker theater if captured information is merely filed rather than used. The distinction is operational: captured state transformations must flow into decision support, deviation measurement, and organizational learning — not into a compliance archive.