Glossary

Appendix - Glossary¶

This glossary collects key terms used across the Decision Intelligence with AI notebooks. The definitions are grounded in how the workshop uses each term: as practical language for framing, reasoning, executing, communicating, monitoring, and automating decisions with generative AI, quantitative methods, plugins, agents, and local or cloud AI systems.

Source labels refer to notebook codes such as 1a, 3b, or 7b.

Use the alphabet navigation above to jump directly to glossary terms by letter.

A¶

Actuarial Decision Framing — A decision-framing lens that asks whether a decision survives time, exposure, variance, fragile assumptions, tail risk, correlation, reserves, optionality, and adverse selection. Sources: 1b.

Adverse Selection — A decision risk where the people most likely to accept an offer or enter a process are also the people most likely to create hidden cost, risk, or poor fit. Sources: 1b.

Agent — An AI-assisted decision actor that can hold instructions, use tools, participate in a workflow, and contribute specialized reasoning toward a recommendation. Sources: 5a, 7a, 7b.

Agent Orchestration — The coordination pattern used to route work among one or more agents, often with a manager, moderator, or termination rule deciding how the conversation progresses. Sources: 7b.

AI Decision Explainability — The ability to show why an AI-assisted recommendation was made, including the evidence, sources, tools, assumptions, and reasoning path that shaped it. Sources: 5c, 5d.

AI Extensions — The Microsoft.Extensions.AI abstractions used in the workshop to work with different AI clients and providers through common interfaces. Sources: 2b, 3b, 3c, 3d.

AI Intuition — The workshop term for allowing a generative AI model to answer from learned patterns and training data without additional grounding, constraints, or explicit decision structure. Sources: 1e, 3a.

Alternative Decision Frame — A different reference point or lens for the same decision that can reveal options, risks, stakeholders, or values hidden by the original frame. Sources: 1b, 4x.

Alternatives — The candidate options available for a decision; quality decision work improves alternatives rather than only choosing among the first options presented. Sources: 1a, 1b, 1f.

Antifragility — A mental model for systems or strategies that do not merely survive volatility but can benefit from disorder, stress, or uncertainty. Sources: 1f, 10a.

API Key — A secret credential used by notebook examples to authenticate with AI services; the workshop keeps these values in local or secrets settings files rather than in notebooks. Sources: 2a, 3b, 3d.

Assumption Fragility — The risk that a decision only looks strong because a key assumption is optimistic, untested, or likely to break under real conditions. Sources: 1b.

Auto Function Invocation — A Semantic Kernel behavior where the model can choose and call available functions or tools based on their descriptions and the prompt instructions. Sources: 3c, 5d.

Azure OpenAI — A cloud AI provider used in the notebooks for hosted OpenAI-compatible chat and response workflows. Sources: 2a, 3b, 3d.

B¶

Backcasting — Reasoning backward from a desired or failed future state to identify decisions, assumptions, milestones, and risks in the present. Sources: 4x.

Barbell Strategy — A mental model that combines very safe choices with selective high-upside risks rather than concentrating everything in the middle. Sources: 10a.

Ben Franklin Decision Framework — A structured pros-and-cons approach that lists reasons for and against a decision, weighs them, crosses off comparable reasons, and selects the side with stronger remaining reasons. Sources: 1f, 3d, 7b.

Bias — A systematic distortion in judgment, such as anchoring, confirmation bias, groupthink, or framing effects, that can narrow a decision or overemphasize one type of evidence. Sources: 1b, 1c, 1f, 4x.

Brier Score — A quantitative accuracy measure for probabilistic forecasts that rewards calibrated probabilities and penalizes confident incorrect predictions. Sources: 6c, 10a.

Business Decision System — An enterprise pattern where repeated decisions are made through rules, analytics, models, AI, workflows, governance, and monitoring. Sources: 1j.

C¶

Calibration — The alignment between stated confidence and actual accuracy; a calibrated decision-maker or model is right about 70% of the time when it says 70%. Sources: 1g, 6b, 6c.

Chain of Thought Prompting — A prompting style that asks an AI model to break a problem into structured reasoning steps before producing a final answer or recommendation. Sources: 4x.

Chat Client — The AI client abstraction used by the workshop to send messages to a model through cloud or local providers. Sources: 3b, 3c, 3d.

Chat Completions API — An OpenAI-compatible API pattern where messages are sent to a chat model and the model returns a conversational completion. Sources: 3b.

Citation — A reference to the source material used to ground a recommendation, especially important when plugins or web search provide decision evidence. Sources: 5c, 5d.

Cloud AI — Hosted AI services such as Azure OpenAI or OpenAI that provide remote model inference, scaling, managed access, and cloud-based deployment. Sources: 3b, 3d.

Cognitive Bias — A recurring mental shortcut or distortion that can make a decision feel reasonable while hiding evidence, alternatives, or better frames. Sources: 1b, 1f, 10a.

Collective Intelligence — A decision pattern where multiple people, experts, models, or agents contribute independent perspectives that are aggregated into a stronger final judgment. Sources: 4x, 7b, 10a.

Commitment to Action — The decision quality element that turns a chosen option into execution, ownership, and follow-through rather than leaving the decision as analysis. Sources: 1d, 1j, 10a.

Configuration Settings — Local notebook settings such as endpoints, model names, and API keys that let examples run without hard-coding environment-specific values. Sources: 2a, 3b, 3d.

Confirmation Bias — The tendency to seek or favor information that supports an existing belief, frame, or preferred option. Sources: 1b, 4x.

Consequences — The expected outcomes, costs, benefits, risks, and second-order effects of a decision option. Sources: 1a, 1g, 10a.

Context Engineering — The practice of selecting, shaping, and ordering information so an AI model receives the most decision-relevant context within its context window. Sources: 1c, 3a, 3b.

Context Window — The amount of prompt, retrieved knowledge, conversation history, and output tokens a model can consider in one interaction. Sources: 1c, 3a, 3b.

Correlation Risk — The risk that multiple things fail together because they depend on the same underlying factor, such as one employer, platform, industry, vendor, or assumption. Sources: 1b.

Cost-Benefit Analysis — A structured comparison of expected costs and expected benefits used to clarify whether a decision is worth pursuing. Sources: 1b, 1g, 4x.

Creative Alternatives — New or improved options generated by reframing, brainstorming, structured frameworks, or AI support. Sources: 1b, 1f, 4x.

D¶

Data Grounding — Supplying a model with specific data or knowledge so the recommendation depends on relevant evidence rather than only model memory. Sources: 1c, 3b, 5c, 5d.

Decision — A choice, recommendation, prioritization, escalation, postponement, quit decision, group outcome, status quo choice, or selection of how to decide. Sources: 1a.

Decision Communication — The practice of explaining the recommendation, rationale, evidence, tradeoffs, and actions so stakeholders understand and can act on the decision. Sources: 1h, 5b.

Decision Execution — The act of carrying a decision into a choice or action, including choosing an option, using intuition, applying rules, using quantitative methods, or deciding not to proceed. Sources: 1d, 1e, 1f, 1g.

Decision Frame — The reference point, boundary, lens, or question that shapes how a decision problem is understood and what options appear available. Sources: 1b, 3a, 4x.

Decision Framing — The process of defining the decision, broadening or narrowing the lens, testing alternate frames, and selecting useful starting frames for analysis. Sources: 1b.

Decision Intelligence — A discipline for improving important decisions by combining decision framing, intelligence gathering, execution, communication, monitoring, and AI-assisted systems. Sources: 1a, 1i, 1j.

Decision Intelligence Framework — The workshop's organizing model for optimizing decision quality across framing, intelligence gathering, execution, communication, and monitoring. Sources: 1a, 1i.

Decision Management System — An enterprise system that supports repeatable operational decisions through data, rules, models, workflows, and governance. Sources: 1j.

Decision Monitoring — The practice of observing outcomes, feedback, drift, accuracy, and decision performance after execution. Sources: 1a, 1j, 6c.

Decision Prompt — A prompt designed to guide a model through a decision task, often specifying persona, frame, criteria, structure, constraints, and expected output. Sources: 3a, 4x.

Decision Quality — The quality of the decision-making process rather than only the outcome, especially under uncertainty where good decisions can still have bad outcomes. Sources: 1a, 10a.

Decision Recommendation — A suggested course of action produced by a person, model, rule, plugin, agent, or decision process. Sources: 1a, 3a, 5c, 7b.

Decision Rules — Reusable heuristics, mental models, algorithms, or simple rule tables that help execute decisions consistently. Sources: 1f, 5d.

Decision Support System — A system that helps humans make better decisions by presenting information, analysis, recommendations, or model outputs. Sources: 1j.

Decision Trap — A predictable thinking mistake or process failure that causes poor framing, poor evidence use, weak alternatives, or bad execution. Sources: 10a.

Diminishing Returns — The pattern where additional information or effort continues to help but contributes less incremental decision value over time. Sources: 1c, 1g.

Doing Nothing — Maintaining the status quo as a valid decision option, even when it is not presented explicitly. Sources: 1a, 1d.

E¶

Eisenhower Decision Prioritization Matrix — A decision framework that sorts work by importance and urgency to clarify what to do, schedule, delegate, or avoid. Sources: 1a, 7b.

Enterprise Decision Intelligence — Applying decision intelligence at organizational scale through repeatable systems, governance, observability, explainability, and decision workflows. Sources: 1j.

Evaluation Criteria — The standards used to compare options, prompts, outputs, tools, or recommendations. Sources: 1b, 3a, 4x, 6c.

Execution Path — The practical route by which a recommendation becomes action, including people, tools, workflows, timing, and follow-up. Sources: 1d, 1h, 1j.

Expected Case — The average or most likely scenario considered in a decision; actuarial framing warns against only optimizing for this case. Sources: 1b, 1g.

Expected Value — A quantitative decision concept that combines possible outcomes with their probabilities to estimate average value. Sources: 1g, 6c.

Expert Persona — A prompt or agent role that applies a specific expert lens, such as a decision intelligence strategist, devil's advocate, Eisenhower expert, or Ben Franklin expert. Sources: 4x, 7a, 7b.

Exploration vs. Exploitation — A decision tradeoff between trying new options to learn and using known options that already perform well. Sources: 1f, 10a.

Exposure — What is actually at risk in a decision, including financial, operational, reputational, emotional, time, or opportunity cost. Sources: 1b.

F¶

Few-Shot Prompting — Supplying examples in a prompt so the model can imitate the expected pattern, format, or reasoning style. Sources: 3a, 4x.

Forecasting — Estimating future outcomes or probabilities so decisions can account for uncertainty and expected consequences. Sources: 1g, 6b, 6c, 10a.

Framing Effect — The influence that wording, reference points, presentation, or perspective has on how people perceive the same facts. Sources: 1b.

Function Calling — A model capability that lets an AI system request a named function or tool with arguments to gather data, calculate, search, or perform a constrained action. Sources: 3c, 5d.

G¶

Gathering Intelligence — The decision phase focused on reducing uncertainty by collecting, validating, grounding, and selecting high-impact information. Sources: 1c, 4d.

Generative AI — AI models that generate text, reasoning, recommendations, summaries, code, or structured outputs from prompts and context. Sources: 1a, 1c, 3a, 3b.

Grounding — Anchoring AI output in supplied context, sources, documents, search results, data, or tool responses so it is relevant to the decision. Sources: 1c, 3b, 5c.

Groupthink — A group decision failure where social pressure or shared assumptions suppress independent information, dissent, and better alternatives. Sources: 1b, 4x, 10a.

Guardrail — A constraint, instruction, tool boundary, validation step, or governance mechanism that reduces unsafe or low-quality AI-assisted decision behavior. Sources: 1e, 2a, 3a, 3d.

H¶

Hallucination — An AI output that sounds plausible but is unsupported, incorrect, or not grounded in the supplied evidence. Sources: 1c, 3a, 5c.

Heuristic — A practical rule of thumb that simplifies a decision, often useful when time, information, or attention is limited. Sources: 1f, 10a.

High-Stakes Decision — A decision where the consequences are important enough to justify structure, evidence, calibration, communication, and monitoring. Sources: 1a, 1e, 1j.

Human-in-the-Loop — A decision pattern where AI assists but a person reviews, interprets, approves, or executes the final decision. Sources: 1j, 3d, 5d.

I¶

Important Decision — A decision significant enough to deserve better framing, intelligence, execution, communication, or monitoring than a routine low-impact choice. Sources: 1a.

Information Asymmetry — A decision situation where one side knows more than another, often creating risk in hiring, pricing, insurance, negotiation, or offers. Sources: 1b, 1c.

Intelligence — Selected, decision-relevant information that reduces uncertainty or improves reasoning, not merely a large amount of raw data. Sources: 1c.

Intelligence Gathering Curve — The pattern where early credible intelligence can sharply reduce uncertainty before later information produces smaller refinements. Sources: 1c.

Intuition — A fast pattern-recognition decision mode based on experience, learned examples, and context; useful in some settings but risky for high-stakes choices without structure. Sources: 1e, 10a.

J¶

Judgment — The human or AI evaluation used to interpret evidence, weigh tradeoffs, and choose a recommendation under uncertainty. Sources: 1a, 1g, 4x.

Judgment Noise — Random variation in decisions or recommendations that should be consistent, such as two reviewers judging the same case differently without a clear reason. Sources: 1f, 10a.

K¶

Kernel Function — A Semantic Kernel callable function, native or prompt-based, that can be described to the model and invoked during orchestration. Sources: 3c, 5a, 5d.

Knowledge Source — A source of relevant information such as indexed documents, search results, structured data, models, or local files used to ground decisions. Sources: 1c, 5c, 5d.

Knowledge Store — A persistent repository of reusable information that an AI workflow can retrieve from when model training data is insufficient. Sources: 1c.

L¶

Ladder Simulation — A quantitative example used to show how simulation can support decision execution under uncertainty. Sources: 1g.

Large Language Model (LLM) — A generative AI model trained on large text corpora that predicts and generates tokens in response to prompts. Sources: 1c, 3a, 3b.

Linked Decision — A decision connected to future decisions, where the current choice changes later options, constraints, or commitments. Sources: 10a.

LM Studio — A local AI tool used in the workshop to run open-source models through an OpenAI-compatible local endpoint. Sources: 3d.

Local AI — AI inference running on local hardware or a local network endpoint, often used for privacy, control, experimentation, or offline workflows. Sources: 3d.

Log Probability — The logarithmic form of token probability emitted by some LLM APIs, used to inspect model confidence in token choices. Sources: 6a, 6b.

LogProbs — Token-level probability information used in the workshop to evaluate model confidence, probabilistic accuracy, and decision quality signals. Sources: 6a, 6b, 6c.

Ludic Fallacy — The mistake of treating real-world uncertainty as if it behaves like simplified games or formal probability examples. Sources: 1f.

M¶

Magentic Orchestration — A Semantic Kernel multi-agent orchestration pattern where a manager coordinates specialized agents for complex, open-ended tasks. Sources: 7b.

Max Tokens — A model setting that limits the amount of output tokens generated in a response. Sources: 3a, 3b.

MCDA — Multi-criteria decision analysis; a structured way to compare options across multiple criteria, weights, tradeoffs, and consequences. Sources: 1b, 1g.

Mental Model — A reusable way of interpreting situations and making decisions, such as inversion, opportunity cost, Price's Law, or exploration versus exploitation. Sources: 1f, 10a.

Microsoft Agent Framework — The Microsoft agent-oriented framework material introduced in the setup notebooks for building AI agent workflows. Sources: 2c.

Microsoft.Extensions.AI — The .NET AI abstraction layer used to work with chat clients and AI providers through common interfaces. Sources: 2b, 3b, 3d.

Minto Pyramid — A communication structure that leads with the answer, organizes supporting reasons, and makes a recommendation easier to follow. Sources: 1h, 4x, 10a.

Model Deployment Name — The configured name of a hosted model deployment, used by the notebooks to connect to the correct Azure OpenAI model. Sources: 2a, 3b, 3d.

Model Risk — The risk that a model, assumption, probability, or calculation is wrong enough to mislead a decision. Sources: 1b, 1g, 6c.

Multi-Agent Interaction — A workflow where multiple AI agents or personas contribute, challenge, synthesize, or moderate decision reasoning. Sources: 7b.

Multi-Criteria Decision Analysis — A decision method for comparing options across several criteria instead of reducing the decision to one factor. Sources: 1b, 1g.

N¶

Native Function — A regular C# function exposed to Semantic Kernel as a callable tool for the model or workflow. Sources: 3c, 5c, 5d.

Natural Frequency — A probability presentation format that uses counts, such as 3 out of 10, to make risk easier to understand than abstract percentages. Sources: 1g, 10a.

Noise — Random variation in judgment or decisions that creates inconsistency even when the same facts are presented. Sources: 1f, 10a.

O¶

Objective — A decision goal or value that clarifies what the decision is trying to achieve. Sources: 1a, 1b, 10a.

OneR — A simple rule-learning method that chooses the single feature with the lowest error and builds a one-rule classifier. Sources: 1f.

Open Source Model — A model that can be run locally or privately depending on license, hardware, and tooling, used in the workshop for local AI decision workflows. Sources: 3d.

OpenAI Responses API — An OpenAI API pattern used in the workshop for model responses, tool use, and structured AI interaction beyond classic chat completions. Sources: 3b.

Operational Decision — A repeated or workflow-driven decision that can often be supported by rules, models, systems, and monitoring. Sources: 1j.

Optionality — The value of preserving future options, reversibility, exits, pivots, or flexibility when uncertainty is high. Sources: 1b, 10a.

Outcome Quality — The realized result of a decision; the workshop distinguishes this from decision quality because uncertainty and luck can affect outcomes. Sources: 1a, 10a.

P¶

Pareto Analysis — A decision-framing and prioritization idea that looks for the small number of causes or options responsible for a large share of impact. Sources: 1b, 1f.

Persona — A role or perspective assigned to an AI prompt or agent to shape how it reasons, communicates, and evaluates a decision. Sources: 3a, 4x, 7a, 7b.

Plugin — A grouped set of functions or tools exposed to Semantic Kernel or an AI workflow so a model can gather data, search, calculate, or act. Sources: 5a, 5b, 5c, 5d.

Premortem — A decision analysis technique that imagines a future failure, then works backward to identify why it happened and what can be changed now. Sources: 4x, 10a.

Price's Law — A productivity distribution rule suggesting that a small share of people can produce a disproportionate share of output. Sources: 1f.

Prioritization — Ranking or selecting what deserves attention first based on importance, urgency, value, risk, or constraints. Sources: 1a, 1f.

Private AI — AI deployment and usage patterns designed to keep data, prompts, models, or inference under tighter local or organizational control. Sources: 3d.

Probability Distribution — The model's or analyst's spread of likelihoods across possible tokens, outcomes, or future states. Sources: 1g, 6a.

Prompt — The instruction, context, question, or structured request sent to a generative AI model. Sources: 3a, 3b, 4x.

Prompt Engineering — Designing prompts with clear instructions, context, examples, personas, constraints, and output formats to improve AI decision support. Sources: 3a, 4x.

Prompt Template — A reusable prompt pattern with variables or sections that can be applied across similar decision tasks. Sources: 3c, 5a.

Probabilistic Accuracy — The degree to which predicted probabilities match observed outcomes across repeated decisions or forecasts. Sources: 6b, 6c.

Q¶

Qualitative Decision Method — A decision approach based on judgment, framing, comparison, reasoning, stories, or structured prose rather than numeric modeling alone. Sources: 1e, 1f, 1h.

Quantitative Decision Method — A decision approach using numbers, probabilities, statistics, simulations, scores, or models to support execution. Sources: 1g, 6c.

Quitting Decision — A decision to stop, abandon, or walk away from a course of action, treated in the workshop as a valid form of decision execution. Sources: 1a, 1d, 10a.

R¶

RAG — Retrieval-augmented generation; a pattern where relevant information is retrieved from external sources and supplied to the model as grounding context. Sources: 1c, 3b, 5c.

Reasoning Framework — A structured pattern for working through a decision, such as Ben Franklin, Eisenhower, Chain of Thought, premortem, MCDA, or actuarial framing. Sources: 1f, 3a, 4x.

Recommendation — A proposed action or choice generated by a person, AI model, agent, function, rule, or decision process. Sources: 1a, 3a, 5c.

Reference-Class Forecasting — Estimating outcomes by comparing a decision or project to similar past cases rather than relying only on inside-view assumptions. Sources: 10a.

Reference Point — The comparison point or lens that anchors how a decision is framed. Sources: 1b.

Reframing — Changing the decision frame to reveal different choices, criteria, risks, or opportunities. Sources: 1b, 4x.

Reserves — A decision cushion of time, money, capacity, or flexibility that helps absorb worse-than-expected experience. Sources: 1b.

Responses API — The OpenAI API style used in the workshop to request model responses and support newer interaction patterns. Sources: 3b.

Risk Tolerance — The amount and type of uncertainty, downside, or volatility a decision-maker can accept. Sources: 1b, 1g, 10a.

Rule of Thumb — A simple practical rule used to make or simplify decisions when full optimization is unnecessary or impossible. Sources: 1f, 10a.

S¶

Scenario Analysis — Exploring possible future states or decision contexts to understand outcomes, risks, and tradeoffs before acting. Sources: 1b, 1g, 4x.

Semantic Kernel — A Microsoft orchestration framework used in the workshop to connect prompts, functions, plugins, memory, and agents into AI workflows. Sources: 5a, 5b, 5c, 5d, 7a, 7b.

Sensitivity Testing — Checking how much a recommendation changes when assumptions, probabilities, weights, or inputs change. Sources: 1b, 1g, 6c.

Self-Consistency Pattern — A GenAI pattern where multiple reasoning paths, model responses, or agent perspectives are compared to improve the final answer. Sources: 4x, 7b.

Simulation — A quantitative method that repeatedly models possible outcomes to understand uncertainty, distribution, and risk before making a decision. Sources: 1g.

Six Thinking Hats — A decision-framing method that separates perspectives such as facts, emotions, caution, optimism, creativity, and process control. Sources: 1b, 10a.

Source Grounding — Preserving and using source references so the decision-maker can inspect the evidence behind an AI recommendation. Sources: 5c, 5d.

Strategic Decision — A high-level decision that shapes direction, tradeoffs, investments, or long-term positioning. Sources: 1j, 4x.

Structured Output — An AI response constrained into a predictable format such as a table, JSON-like structure, checklist, or recommendation template. Sources: 3a, 3b, 4x.

SWOT Analysis — A structured decision framework comparing strengths, weaknesses, opportunities, and threats. Sources: 1b, 4x.

System Prompt — The higher-priority instruction that sets the model's role, behavior, decision style, constraints, or output expectations. Sources: 3a, 4x.

T¶

Tail Risk — A low-probability but severe downside outcome that can dominate the quality of a decision if it creates ruin or irreversible harm. Sources: 1b, 1g.

Temperature — A model setting that influences randomness in token selection; lower values are more deterministic and higher values are more varied. Sources: 3a, 3b, 6a.

Thinking in Bets — A decision mindset that separates decision quality from outcome luck and treats uncertain choices as probabilistic bets. Sources: 1a, 10a.

Time Horizon — The length of time a decision must remain viable, survive uncertainty, or deliver value. Sources: 1b.

Token — A unit of text processed or generated by an LLM, often part of a word, whole word, or symbol. Sources: 3b, 6a.

Tool Calling — A model-assisted pattern where the AI chooses a described tool or function to gather information, calculate, search, or perform a step. Sources: 3c, 5d.

Top_p — A model sampling setting that limits token choices to a probability mass, shaping response diversity. Sources: 3a, 3b, 6a.

Tradeoff — A necessary exchange between competing values, objectives, benefits, costs, risks, or constraints. Sources: 1a, 1b, 1g, 1h.

Training Data — The data used to train an LLM, giving it broad learned patterns but not necessarily the specific knowledge required for a current decision. Sources: 1c, 1e.

U¶

Uncertainty — The lack of complete knowledge about facts, outcomes, probabilities, assumptions, or future states in a decision. Sources: 1a, 1c, 1g.

Uncertainty Reduction — The goal of gathering intelligence, measurement, grounding, and analysis so decisions are made with fewer unknowns. Sources: 1c, 1g.

User Prompt — The direct instruction, question, or request supplied by the user in a model interaction. Sources: 3a, 3b.

V¶

Validation — Checking that a recommendation, output, probability, data source, or model behavior is accurate enough for the decision context. Sources: 2a, 5c, 6b, 6c.

Value Creation — The business or personal benefit created by improving decision quality, reducing uncertainty, avoiding traps, or executing better choices. Sources: 1a, 1j, 10a.

Variance — The amount of fluctuation or instability in outcomes, even when the average result looks acceptable. Sources: 1b, 1g.

Via Negativa — A mental model that improves a system or decision by removing harmful elements rather than adding more complexity. Sources: 10a.

Volatility Capacity — The ability of a person, team, or organization to tolerate swings in income, outcomes, workload, or uncertainty. Sources: 1b.

W¶

Web Search Plugin — A Semantic Kernel or custom plugin that gathers current internet information to ground a decision recommendation. Sources: 5a, 5c, 5d.

Weighting — Assigning relative importance to reasons, criteria, risks, or evidence so the decision process reflects priorities rather than treating everything equally. Sources: 1f, 1g, 3d.

What-If Analysis — Exploring how outcomes change under different assumptions, failures, events, or scenarios. Sources: 4x.

Wisdom of Crowds — The idea that diverse, independent, decentralized judgments can outperform individual judgment when properly aggregated. Sources: 4x, 10a.

Workflow — The sequence of steps, tools, agents, prompts, validations, and handoffs that turns a decision task into an executed process. Sources: 1j, 3c, 5a, 7b.

Z¶

Zero-Shot Prompting — Asking the model to perform a task without examples, relying on the model's learned patterns and the prompt instructions alone. Sources: 3a, 4x.