How to use this glossary
The guide introduces terms inline where they first matter, restates them in each part-end glossary, and aggregates them here. If you arrived at this page from a cross-reference inside one of the parts, the part tag on each entry — (Part X) — points back to the part where that term is treated in fuller context. Multi-part terms appear once here, with all the parts they appear in named in the part tag.
Numerals
12-factor AI-readiness instrument. A self-assessment used in PeopleAnalyst consulting practice, scoring an organization across twelve dimensions of AI readiness. The dimensions: AI-Driven Workforce Planning; Change Fatigue and Resistance; HR and Organizational Design; Human-AI Collaboration and QA; Management Buy-In; Organizational Restructuring; Technology / Management / HR Capability Building; Transparency; Trust in AI; AI as Strategic Partner; Workflow Mapping; Workforce and Job Analysis. (Part II)
A
Action-grade reliability. The standard an AI system must meet when its outputs drive actions in the world: calibration, traceability, recoverability, and observability of failure modes. Distinct from output quality, which most AI evaluation infrastructure measures. (Part IV)
Adaptive Authorship Control Kernel (F-19). The Penwright system's central registry for skill measurement, intervention, and genre-aware behavior. Forks copy + schema enums + prompts + metrics by genre rather than collapsing them. (Part V)
Agentic AI. A system built on foundation-model substrate that decomposes a task into multiple steps, executes those steps (often via tool calls), maintains state across steps, and produces a result at the end. (Part IV)
Alignment data. The smallest layer of an AI model's training data, used to shape the model's outputs toward human preferences. Typically tens of thousands of comparisons; the substrate underneath RLHF. (Part I)
Anti-invention constraint. A product feature that causes an AI system to refuse to render rather than to fabricate when structural rhetorical moves require material the user has not supplied. Enforced in the Penwright system at two layers (invented-content register + Sonnet critic). (Part V)
Auditability. The ability to verify that an AI system did what it claimed to do, with the data it claimed to use, in the way it claimed to operate. Structurally limited for foundation-model AI; addressable through behavioral audits, provenance tracking, calibration auditing, and deployment-context auditing. (Part VI)
Authorship Packet Model. The Penwright structured-input pattern replacing freeform prompting. Components: intent, structure, key ideas, relevant passages, counterpositions. The structure is data; the system reasons against the packet rather than against free-floating user framing. (Part V)
Autoregressive generation. A class of machine learning model that produces a sequence one token at a time, with each new token depending on the previous ones. The dominant approach for modern language and code models. (Part I)
B
Backburn. The wildfire-management practice of intentionally setting a controlled fire ahead of an advancing wildfire to consume fuel before the wildfire arrives. In the network-mediated adoption argument: accelerating adoption inside the seed cohort's tie clusters so the early majority encounters a saturated zone rather than an unfamiliar tool. The title of the book project this guide pairs with. (Part VII)
Bainbridge effect (Ironies of Automation). The observation that human operators' vigilance for automation failures decreases as automation reliability increases. Bainbridge 1983 is the canonical reference. (Part IV)
Bandit algorithm. A class of decision-making algorithms that balance exploration (trying new options to learn their value) and exploitation (selecting the option with the highest known value). Used extensively in campaign optimization and recommender systems. (Part III)
Behavior surface. The full range of inputs an AI system might see in production and the corresponding outputs. Distinct from a software test surface in that the behavior surface is too large to comprehensively test. (Part I)
Behavioral audit. An audit methodology that characterizes an AI system's behavior surface against representative input distributions, rather than exhaustively testing for correctness on specific inputs. (Part VI)
Bias amplification loop. The bidirectional process by which AI outputs shape human judgments, which shape new training data, which shape next-generation AI outputs. Empirically demonstrated by Glickman and Sharot 2024 in Nature Human Behaviour. (Part V)
Bridge actor. An individual in an organizational network who connects otherwise-disconnected clusters. In Burt's structural-hole framework, a bridge actor captures disproportionate informational and brokerage value; in network-mediated adoption, bridge actors are load-bearing for propagation. (Part II)
C
Calibration auditing. A specific behavioral-audit methodology testing whether the AI's stated confidence matches ground-truth accuracy on representative tasks. (Part VI)
Calibration of personalization. The AHI program's framing of when personalization is and isn't harmful: content-personalization (showing the right content to the right person) is fine; reasoning-personalization (the AI adjusting how it reasons based on the user) is the failure mode. (Part V)
Capability ceiling. The expertise level above which AI assistance no longer produces productivity gains and may produce productivity losses. Documented across multiple knowledge-work domains in the post-2023 literature. (Part II)
Cluster saturation. The operational move in network-mediated adoption: deliberately accelerating adoption inside the seed cohort's immediate tie clusters until the local density crosses Centola's threshold. The opposite of the conventional roll out to the next ring expansion. (Part VII)
Cognitive offloading. The redistribution of cognitive load from worker to AI. Empirically documented for memory tasks since the Google-effect literature (Sparrow 2011); emerging documentation for higher-order cognitive tasks (Lee 2025). (Part V)
Cognitive redistribution. The synthesis of the empirical productivity literature: AI does not uniformly raise productivity; it redistributes which workers gain access to expert-level outputs. Novices typically gain the most; intermediate-skilled workers gain moderately; experts in some domains gain little or lose time to verification overhead. (Part II)
Complex contagion. A behavior that requires multiple reinforcing exposures from trusted ties to propagate. The empirical anchor is Centola 2010 (Science) and 2018 (How Behavior Spreads). The category includes most behaviors that change daily practice — every meaningful organizational adoption is a complex contagion. (Part VII)
Content personalization. Adaptation of what gets shown to a user — the right product, article, or offer. Distinct from reasoning personalization (see below); content personalization is the recommender-system tradition. (Part III)
Corpus Control Layer. The Penwright pattern by which writers explicitly select which sources influence the work, rather than inheriting the model's training distribution. (Part V)
Crown fire transition. The wildfire-management concept of surface fire jumping to canopy fire, with much faster spread and much higher difficulty of suppression. In the network-mediated adoption argument: the moment when innovator + early-adopter adoption needs to propagate to the early majority. Moore's chasm reframed as a fire-physics problem. (Part VII)
Cumulative disclosure. The privacy failure mode in which a user discloses information incrementally across many AI interactions, with the cumulative profile substantially exceeding what the user would have disclosed if asked directly. (Part VI)
Customer lifetime value (CLV). A model of the total revenue expected from a customer over the duration of their relationship with the firm. Classical survival-analysis methodology; foundation models add limited contribution. (Part III)
D
Design constraint vs risk management. The structural distinction in governance posture: design constraints prevent problematic features from being built (upstream); risk management evaluates rollouts after design (downstream). Both are necessary; the upstream variant is less common and more load-bearing. (Part VI)
Diffusion of innovations. Rogers 1962-2003 — the canonical framework documenting that the adoption of new practice follows a roughly normal distribution across the adopter population (innovators / early adopters / early majority / late majority / laggards). (Part VII)
Drift. The phenomenon by which an AI system's effective behavior changes over time without any change to the system itself, because the data distribution or user behavior around it has changed. (Part I)
E
EU AI Act. The European Union's risk-tiered regulation of AI systems (entered into force August 2024; phased compliance deadlines through 2027). High-risk AI deployments face technical-documentation, risk-management-system, transparency, human-oversight, and accuracy / robustness / cybersecurity requirements. (Part VI)
EVPI / EVSI. Expected Value of Perfect Information / Expected Value of Sample Information. The two primitives in Value-of-Information analysis: how much would knowing the truth be worth (EVPI); how much would a noisy sample be worth (EVSI). Implemented in production form in the People Analytics Toolbox's forecasting spoke. (Part IV)
F
Fine-tuning data. The middle layer of an AI model's training data, used to adapt a pretrained model to specific tasks (instruction-following; conversation; domain specialization). (Part I)
Foundation model. A large neural network trained on broad data, intended to be adapted to many downstream tasks. Term coined by Stanford HAI in 2021. (Part I)
Four non-negotiable failure modes. The Penwright Measurement Framework's design vetoes: output-only optimization; over-automation; weak measurement; ignoring genre differences. (Part V)
Fuel load. The wildfire-management concept of how much combustible material is available in a forest. In the network-mediated adoption argument: organizational readiness for adoption, not as an aggregate score but as a topology-aware map. (Part VII)
Fuel-readiness map. The network-readiness reframe of the 12-factor instrument: instead of an aggregate organizational score, a topology-aware map of where each dimension is high and low across the organization's network, signaling where adoption will propagate and where it will stall. (Part II)
H
Hallucination. An AI system producing confident-looking outputs that are factually incorrect or fabricated. A consequence of the methodology gap from §1.3 (software-fails-by-stopping vs. AI-fails-by-performing-confidently). (Part I)
Human-in-the-loop. An AI deployment architecture in which a human reviewer or approver sees the AI's output before it drives an action. The default safety architecture for high-stakes operational AI; vulnerable to the Bainbridge / rubber-stamp failure mode without auxiliary instrumentation. (Part IV)
Hyper-personalization. Marketing term for AI-augmented adjustment of content, timing, and messaging to individual users. The term blurs content-personalization (fine) and reasoning-personalization (the failure mode the sycophancy literature documents). (Part III)
K
k-anonymity. A privacy property in which any individual in an aggregated dataset is indistinguishable from at least k-1 other individuals. The classical minimum-N privacy gate. (Part VI)
Kepner-Tregoe. A structured decision-analysis methodology developed in the 1960s, organized around explicit problem analysis, decision analysis, and potential-problem analysis. Integrates cleanly with AI-augmented analysis as one structured-framework option. (Part IV)
L
Longitudinal test (the better-with-than-without-in-six-months test). The Penwright Research Program's core methodological move: assess AI writing tools not by output quality at a single point but by writer capability change over a six-month window. (Part V)
M
Machine learning. A class of computer-science techniques in which a system improves its performance on a task by being shown examples, rather than by being explicitly programmed. (Part I)
Min-N gate. The privacy-discipline enforcement of a minimum cohort size (typically 5, 7, or 10) below which aggregated outputs are not surfaced. Necessary but not sufficient against AI-augmented inference attacks. (Part VI)
Model collapse. The phenomenon by which a generative AI model trained recursively on outputs from prior generative models loses distributional tails — high-frequency patterns are preserved; low-frequency patterns are lost. Documented in Nature 2024 (Shumailov et al.). (Part I, V)
Monte Carlo simulation. A computational method for parameterized uncertainty — running many random samples through a model to surface the distribution of outcomes. The People Analytics Toolbox's forecasting spoke implements Monte Carlo as a callable service. (Part IV)
N
Network readiness. The reframing of AI-readiness assessment from organizational-aggregate property to spatial-distribution-across-the-network property. The unit of analysis shifts from the organization to the tie cluster. (Part II)
Network-mediated adoption. The four-stage framework the guide advances: map the trust graph; identify the right ~16% by network position + psychometric signal; seed them first; saturate their tie clusters until weak-tie spillover reaches the early majority. The alternative to the standard consulting playbook (executive sponsor → uniform training → KPI dashboard). (Part VII)
O
Organizational network analysis (ONA). A methodology that maps relationships between people in an organization based on patterns of communication, advice-seeking, information flow, and trust. Distinct from the formal org chart; substantially more predictive of adoption outcomes. (Part II)
P
Paternalism vs autonomy. The ethics-in-deployment tension between an AI system's adaptation to user preferences supporting the user's autonomy vs undermining it. The reasoning-personalization failure mode from Part V §5.2.1 sits at this tension. (Part VI)
Penwright Measurement Framework. Six skill dimensions, six derived indices, three measurement layers, five-step learning loop, four non-negotiable failure modes. Documented at vela/docs/VISION-PENWRIGHT-MEASUREMENT.md; externally-facing version at peopleanalyst.com/research/ai-human-interaction/penwright-paper-04-measurement. (Part V)
Penwright Research Program. A 12-paper trajectory across three tiers documenting the methodology, empirical study designs, and product features the AHI program advocates. Public-facing trajectory at peopleanalyst.com/research/ai-human-interaction/. (Part V)
Persona drift. Degraded coherence in an AI system's stated identity or operating constraints over the course of an extended session. Documented across multiple LLMs; counter-intuitively increases with model scale (Chen et al. 2024). (Part V)
Prescribed burn. The wildfire-management practice of intentionally burning under controlled conditions. In the network-mediated adoption argument: the seed cohort's onboarding — designed for innovators, not scaled-down general training. (Part VII)
Pretraining data. The largest layer of an AI model's training data. Typically trillions of tokens of broad-distribution data (web, books, code); the substrate underneath the model's general capabilities. (Part I)
Protected feedback. A substrate primitive (vs configuration option) that ensures workforce feedback cannot be used to identify individual respondents, score individuals, or support surveillance/disciplinary/retaliation actions. Implemented in Performix as a cross-cutting capability that every other capability passes through. (Part VI)
Provenance tracking. A discipline of tracking where data entered an AI system, what processing it underwent, and what outputs followed. Necessary for any meaningful auditability of AI deployments. (Part VI)
R
Reasoning personalization. The failure mode in which an AI system adapts how it thinks to the user, with the user's framing shaping the system's reasoning. The sycophancy-driven failure pattern from Part V §5.2.1 in personalization form. (Part III, V)
Reinforcement learning. A class of machine learning in which a system learns to take actions in an environment by receiving feedback about whether the actions led to good outcomes. The dominant approach for game-playing systems and one of the inputs to alignment training (RLHF). (Part I)
RLHF. Reinforcement learning from human feedback. The specific training technique used to align foundation models with human preferences — humans rate comparison pairs of model outputs; the model is fine-tuned to produce outputs that match the higher-rated pattern. (Part I)
Rubber-stamping. The failure mode in which human-in-the-loop reviewers approve AI outputs without substantive review, typically because the AI's apparent reliability has reduced the reviewer's vigilance. A consequence of the Bainbridge effect. (Part IV)
S
Seed cohort. The innovator + early-adopter population (about 16% of the organization in Rogers's framework) whose adoption decisions propagate to the rest of the organization. Identified by network position plus psychometric signal, not by role title or enthusiasm. (Part VII)
Sequence generation. The machine-learning capability that produces sequences (text, code, image tokens, audio samples) one token at a time. The dominant capability underneath foundation-model AI. (Part I)
Small-world network. A network with short average path lengths but low local clustering coefficient. Excellent at moving information; bad at moving practice. The broader organization is, by network structure, a small-world graph. (Part VII)
Structural hole. The Burt 1992 concept of a gap between otherwise-disconnected sub-networks. Managers whose tie networks span structural holes capture disproportionate informational and brokerage value; in network-mediated adoption, structural-hole positions are the bridge-actor positions load-bearing for propagation. (Part VII)
Structural isolate. An individual in an organizational network with weak or no ties to other clusters. The opposite of a bridge actor; adoption does not propagate from structural isolates. (Part II)
Substrate opacity. The structural condition of foundation-model AI in which the training data, alignment process, and other formative inputs are proprietary to the vendor and largely undisclosed to the enterprise deploying the system. (Part VI)
Supervised learning. A class of machine learning in which a model is trained on labeled examples. The dominant approach for pattern-recognition systems (image classification; fraud detection). (Part I)
Sycophancy. An AI system producing outputs that agree with the user's framing even when the framing is incorrect. Documented across multiple frontier models; structurally driven by RLHF preference data (Sharma et al. 2024). (Part I, V)
T
Tie cluster. The local group of people whose adoption decisions reinforce each other because their work and trust ties are dense enough that what one does, the others see and respond to within the same week. The right unit of analysis for AI adoption; a 50,000-person organization contains 2,000-5,000 of these. (Part VII)
Tool-calling. The pattern in agentic AI systems by which the foundation model invokes external APIs, databases, or code execution between language-generation steps. The practical bottleneck in agentic-system reliability. (Part IV)
Transactive memory. The distributed knowledge structure that develops in teams over time, with team members specializing in different domains and consulting each other for cross-domain questions. Wegner 1986 is the foundational reference; the AHI program review at transactive-memory.md extends the framing to human-AI teams. (Part II)
Transformer. The neural-network architecture that has dominated foundation-model design since 2017. Notable for its self-attention mechanism, which allows the model to weigh different parts of its input differently. (Part I)
Trust graph. The informal network of who-actually-influences-whom in an organization. Distinct from the org chart; substantially more predictive of adoption outcomes. (Part VII)
U
Uplift modeling. A class of causal-modeling techniques that estimate the incremental effect of a marketing intervention rather than its absolute effect. Mature methodology; foundation models add limited contribution. (Part III)
V
Value of Information (VOI) analysis. A decision-theory framework that quantifies how much additional information would be worth before making a decision. The structural correction to AI-augmented decision support that lacks calibration scaffolding. (Part IV)
Verification capability. The skill of assessing whether an AI-produced output is correct, calibrated, and complete. Increasingly a differentiating capability in 2026 hiring markets; correlated with but not identical to autonomous-production capability. (Part II)