LinkCanon Topic Index

The attribution of outcomes to the principal-agent pair in a delegation. Accountability requires that outcomes be traceable to who delegated what, enabling correction, audit, and trust calibration.

Establishing that a party is who it claims to be. The gate before authorization: first who you are, then what you may do.

A resource shared by a community and governed by it, rather than owned privately or by the state. Knowledge, code, and standards are the internet's great commons.

The synchronization of actions among multiple agents or principals to achieve shared goals without redundancy, conflict, or dropped tasks. Web4 coordination is delegation-based: explicit authority assignments rather than implicit social coordination.

The transfer of authority and intent from a principal to an agent, together with an accountability structure specifying how outcomes will be attributed and evaluated. A delegation is not a command — it hands off judgment, not execution.

The rules, mechanisms, and protocols by which delegation disputes, agent failures, and authority conflicts are resolved. Governance is what makes delegation recoverable: the layer that handles the cases where delegation goes wrong.

The stable, verifiable assertion of who a principal or agent is — their persistent properties, capabilities, and accountability linkages. Sovereign identity systems give principals control over their identity representation across delegation contexts.

The ability of systems to exchange and use information without special configuration or translation. Ontologies and schemas enable interoperability by giving systems a shared vocabulary for concepts and their relationships.

A formal vocabulary of concepts and their relationships within a domain, designed to enable machine-readable meaning and inter-system interoperability. An ontology is the schema layer for knowledge graphs.

The ongoing monitoring of a delegated agent's behavior and outcomes by the principal or a designated supervisor, distinct from pre-approval and post-hoc audit.

The documented chain of origin, custody, and transformation for a piece of knowledge or an artifact. Provenance is the foundation of canonical claims: without it, authority assertions cannot be verified.

Ordering results by estimated usefulness. The invisible editorial act in every search: whoever controls ranking controls what gets found.

The Resource Description Framework: the W3C's graph data model expressing knowledge as subject-predicate-object triples, the foundation of the semantic web stack.

A governed, queryable record of canonical entities, their properties, and relationships. A registry makes the world of a system legible to its participants and to machines. Web4 registries are the infrastructure layer for trust at scale.

The accumulated signal of past behavior used to calibrate trust extension. In Web4 systems, reputation is agent-specific, context-aware, and composable — a reputation in one domain provides evidence for trust decisions in adjacent domains.

A formal definition of the structure and constraints of a data type. Schemas make data interoperable by specifying exactly what fields are required, what types they must be, and what values are valid. jsontology provides JSON-native schemas for AI agent systems.

Active direction and correction of an agent's work while it happens, as distinct from before-the-fact approval and after-the-fact audit.

Evidence-based confidence extended to an agent to act on a principal's behalf, calibrated to the agent's track record, identity, and the accountability structures in place. Trust is not belief — it is a structured risk assessment with verifiable inputs.

The process of confirming that a claim, identity, or artifact meets established criteria. In delegation systems, verification gates trust extension — only verified claims support high-trust delegations.

Locking value as collateral for honest participation, creating economic skin in the game that misbehavior forfeits. Trust manufactured from incentives rather than history.

The formal interface by which software exposes capabilities to other software — the contract surface where products become platforms and integration becomes possible.

The philosophy of knowledge itself — what justifies belief, what counts as evidence, when testimony transmits knowledge. The theory beneath every verification system.

Treatment that those affected can accept as justified — procedurally in how decisions are made, distributively in what results. Contested in every particular, demanded in every system.

The explicit statement of what an agent is authorized to do on a principal's behalf: scope, duration, resources, and the outcomes it will be judged against. A delegation without a mandate is a wish.

Control over information about oneself — who may collect it, know it, use it — the boundary condition on every data economy and the first casualty of convenience.

A published specification that independent parties implement identically, making their systems interchangeable. The coordination artifact on which interoperability rests.

A system that perceives its environment and takes actions to achieve goals on behalf of a principal. Agents in Web4 systems operate within delegation contracts: they receive intent, bounded authority, and are subject to accountability structures.

Making AI systems pursue their principals' actual intent — including the unstated parts — rather than literal instructions or proxy metrics. Delegation theory's hardest open case.

A tamper-evident chronological record of agent actions, decisions, and their attribution to specific delegation contracts and principals. The accountability infrastructure that makes delegation reversible and governable.

The design constraint that every delegation must specify explicit limits on the decisions an agent may make. Without bounds, delegation becomes abdication; with bounds, it enables safe autonomous action.

The authoritative, primary reference for a concept, fact, or artifact — the source to which all other references point. A canonical source is not just the first or most popular, but the deliberately maintained, accountable reference.

The believability of a source's claims, earned through track record, stake, and verifiability. Reputation is general standing; credibility attaches to assertions.

The mathematics of protecting information and proving claims against adversaries — the discipline that converts trust assumptions into hardness assumptions.

The formal specification of a delegation: who the principal is, who the agent is, what intent is being handed off, what authority scope is granted, and how accountability will be enforced. The unit of governance in Web4 systems.

Software running across machines that fail independently and communicate unreliably — where every guarantee costs coordination and every shortcut costs correctness.

Continuing correct operation despite component failures — engineered through redundancy, isolation, and the assumption that everything eventually breaks.

The legal and ethical obligation of an agent to act in the principal's best interest, above the agent's own. The strongest standard of loyalty a delegation can impose.

The mathematics of strategic interaction — how rational parties choose when outcomes depend on everyone's choices. The shared grammar of coordination and conflict.

A string that names one entity unambiguously within a namespace. Identity systems are, at bottom, agreements about who controls which identifiers.

The discipline of finding relevant material in large collections — indexing, matching, ranking — that underlies every search box and every retrieval-augmented agent.

A structured representation of entities and their relationships, designed to be machine-readable, queryable, and composable. Knowledge graphs make the relationships between concepts explicit and traversable.

The study of encoding what is known so machines can reason with it — choosing among logics, frames, graphs, and vectors, each with different truths it can express.

Systems that improve from data rather than explicit programming — fitting functions to examples and betting the fit generalizes.

Running a trained model to produce outputs — where all the serving economics live: latency, throughput, and the cost of every generated token.

A network of AI agents coordinating to achieve shared or distributed goals. In Web4, multi-agent systems require explicit authority graphs, delegation contracts between agents, and governance protocols for conflict resolution.

The conflict of interest arising when an agent's incentives diverge from the principal's. In AI delegation, the alignment challenge: ensuring agents optimize for the principal's actual intent rather than proxy metrics or training artifacts.

Binding rules imposed by an external authority on a domain of activity, backed by sanction. The form governance takes when self-regulation fails or stakes are public.

The corpus a model learns from — its effective worldview. Curation choices here become model behavior later, which makes data provenance a governance question.

A W3C-standard digital credential whose issuer, subject, and claims can be cryptographically verified by anyone, without contacting the issuer.

A published set of terms with agreed meanings for a domain, reusable across datasets. Lighter than an ontology: shared words before shared logic.

The rules that convert individual preferences into a collective decision. Every choice of system — plurality, ranked, quadratic — favors some outcomes over others; neutrality is impossible.

The ownership era of the internet (2014–2022). The dominant primitive was the token — a unit of provable ownership on a shared ledger. DeFi, NFTs, and DAOs attempted to build new coordination models from token primitives.

Converting value created into revenue captured — the step where product decisions meet business model, and where many otherwise good systems bend.

The write era of the internet (2004–2014). The dominant primitive was the engagement signal — the like, share, and follow. Algorithmic feeds, recommendation engines, and attention marketplaces were built by reasoning about engagement signals at scale.

The organizing primitive of Web4: structured, governed, and trusted handoff of human intent to intelligent agents operating on behalf of a principal. Distinguishes from automation (executing pre-specified processes) by requiring full transfer of judgment within defined authority bounds.

A hard constraint that prevents classes of agent action outright rather than punishing them afterward. Guardrails bound the space in which discretion operates.

Choosing and ordering actions to reach a goal under constraints, before or during execution. The capability that separates agents from autocomplete.

The mechanisms by which an agent retains information across steps and sessions — episodic, semantic, and procedural stores layered over a stateless model.

The engineering of systems that perform tasks requiring intelligence — perception, reasoning, language, action — and the sixty-year argument about what that sentence means.

A signed statement by one party vouching for a fact about another — a device's integrity, a person's attribute, a document's existence at a time.

Independent examination of records and claims against evidence, producing an opinion others can rely on. The verification institution that makes trust portable.

The authoritative, curated body of knowledge that defines a field or system — the set of sources and concepts that serious participants reference, build on, and dispute. Building canon requires canonical sources, provenance, and maintained references.

The policies, roles, and controls determining how an organization's data is defined, accessed, and used — governance applied to the asset that now runs everything.

A learned vector representation placing similar things near each other in high-dimensional space. Meaning as geometry — the representation that made semantic search statistical.

Claiming another party's identity to inherit their trust and access. The generic attack that authentication, verification badges, and provenance all exist to defeat.

The smallest meaningful unit of interaction in an internet era from which all other structures are built. A primitive is composable, irreducible within the system, and universal across all significant interactions in the era. Hyperlink (Web1), engagement signal (Web2), token (Web3), delegation (Web4).

The discipline of arranging knowledge for finding and use — classification, cataloging, indexing — practiced by libraries for centuries and inherited whole by the web.

The recognized right of an authority to make binding decisions, granted by those subject to it. Power without legitimacy requires force; legitimacy is what makes governance cheap.

The applied craft of building marketplaces that actually work — thickness, congestion control, safety to participate — mechanism design meeting operations.

Decomposing a system into parts with clean interfaces so each can change independently — the architecture property that lets ecosystems evolve without permission.

A discrete grant allowing a specific action on a specific resource. Permissions are the quanta into which authority decomposes when systems enforce it.

A hierarchical classification of a domain into progressively narrower categories. The simplest knowledge structure that still forces hard decisions about what belongs where.

An agent invoking external capabilities — search, code, APIs, other agents — to act beyond what its model weights can do alone.

The delegation era of the internet (2022–). The dominant primitive is the delegation — a structured, governed, accountable handoff of intent and authority from a human principal to an intelligent agent. Web4 is defined by delegated intelligence becoming the organizing layer of the internet.

The scope of decisions an agent can make on behalf of a principal without re-consulting them. Bounded authority is the critical design constraint that makes delegation safe: an unbounded delegation is not delegation but abdication.

A transformer trained on vast text corpora to predict tokens, from which general instruction-following, reasoning, and tool use are elicited.

The coordination of multiple agents to achieve a complex task — assigning sub-tasks, managing dependencies, and aggregating results. Orchestration is the composition layer above individual delegation contracts.

The engineering and research discipline of preventing AI systems from causing harm — spanning alignment, robustness, misuse prevention, and the governance of increasingly capable systems.

Formal attestation by a recognized body that a product, person, or process meets a published standard. Trust delegated to specialists and made portable as a mark.

Assigning things to classes under explicit criteria. Every classification is a theory of what differences matter, frozen into infrastructure.

Emergent knowledge and decision-making arising from interactions among many agents or humans. The thesis of Web5: when individual delegations are trusted and accountable, their aggregate produces intelligence that exceeds any individual agent.

An algorithm by which distributed nodes agree on a single shared state despite failures or adversaries — proof of work, proof of stake, BFT variants.

Voluntary, informed agreement to an action affecting one's interests. In delegation, consent is what converts acting-for into acting-with-authority.

The highest-order rule set of a system: what powers exist, who holds them, how they are limited, and how the rules themselves may change.

Parties bearing individual cost for mutual benefit — sustained not by goodwill but by repetition, reputation, and reciprocity engineered into the situation.

Structured reasoning together before deciding: surfacing information, testing arguments, and revising positions. What voting without it merely aggregates.

The machinery that detects violations and applies consequences. The gap between rules-as-written and rules-as-enforced is where systems actually live.

The act of pushing a decision back up the delegation chain when it exceeds an agent's authority, confidence, or risk threshold. Well-designed delegation specifies when escalation is mandatory.

The process of confirming a claimed identity against authoritative evidence — documents, records, biometrics — to a stated assurance level.

A durable structure of rules, roles, and norms that outlives its individual members and shapes their behavior. The unit of social infrastructure that governance operates on.

Cryptographic identity for non-human actors — services, devices, workloads — which now outnumber human identities in most infrastructures by orders of magnitude.

Data describing other data — origin, format, meaning, rights. The layer that makes content findable, trustable, and reusable after its creator moves on.

An informal rule enforced by social expectation rather than formal sanction. Norms carry most of the governance load in every functioning system, silently.

A standing decision that pre-answers a class of future questions, so each instance need not be decided fresh. Governance compiled ahead of time.

The massive self-supervised first phase that gives a model general language and world knowledge, before any task-specific shaping.

The input text that conditions a model's behavior — instructions, context, examples. The interface through which language models are actually programmed.

The withdrawal of previously granted authority from an agent, either for cause or at the principal's discretion. Revocability is what distinguishes delegation from abdication.

The neural architecture built on attention rather than recurrence, whose parallelism and scaling behavior made modern large language models possible.

The condition in which an institution's decisions, data, and processes can be inspected by those affected. The cheapest trust-building mechanism available to any system.

Measuring agent performance on realistic tasks — success rate, cost, safety, latency — under conditions honest enough to predict production behavior.

Layering commentary, labels, or structure onto existing content without altering it — the additive act by which raw material becomes training data, evidence, or knowledge.

A mechanism that allocates goods and discovers prices through structured bidding. The purest machine for extracting truthful valuations from strategic parties.

A directed graph mapping which agents can act on behalf of which principals, under what authority bounds, and with what accountability structures. The data structure underlying delegation-at-scale.

The determination of what an authenticated party is allowed to do. In delegation systems, authorization is the runtime enforcement of granted authority.

A fixed task set with scoring that makes systems comparable — the coordination device of AI progress, and its most gamed artifact.

An asset pledged to secure an obligation, claimable on default. Converts promises into secured claims and strangers into acceptable counterparties.

Rivalry for scarce rewards under rules — the discovery procedure that surfaces prices, quality, and capability no planner could compute in advance.

Processing capacity as a strategic resource — measured in accelerator-hours, budgeted like capital, and increasingly the binding constraint on who can train what.

All parts of a system agreeing on current state — expensive to guarantee, graded in strength, and the first property negotiated away for speed.

Evidence of an attribute or qualification issued by one party about another — a degree, a license, a key. The portable unit of institutional trust.

Transforming information so only key holders can read it — confidentiality as mathematics, and the perennial subject of the crypto-wars between privacy and access.

A distinct thing that can be referred to — person, place, organization, concept. The unit of reference that identifiers name and knowledge graphs connect.

Continuing training on a narrower dataset to specialize a pretrained model — cheaper than pretraining, riskier than prompting, and the middle option everyone eventually considers.

Performing well on data never seen in training — the actual goal of learning, which every metric on the training set merely approximates.

The design discipline for how people and AI systems communicate, collaborate, and calibrate trust — interface design where the interface has opinions.

The lifecycle discipline for cryptographic keys — generation, storage, rotation, revocation, recovery. Where cryptographic systems actually fail.

Shared knowledge resources — encyclopedias, datasets, standards — governed by their communities for common benefit. The commons whose tragedy is neglect, not overuse.

Game theory run in reverse: choosing the rules so that self-interested play produces the outcome you want. The engineering discipline behind auctions, markets, and protocols.

A standard anyone may read, implement, and extend without permission or royalty. The licensing posture that lets ecosystems form around a specification.

Attempting a failed operation again on the assumption the failure was transient — correct with idempotency and backoff, an outage amplifier without them.

Reinforcement learning from human feedback: training a reward model on human preferences, then optimizing the language model against it. How raw capability got its manners.

Capacity growing with resources added — the property architectures claim in slides and demonstrate only under real load.

Conveying private information credibly by taking actions that would be too costly to fake — degrees, warranties, staked reputations.

The careful management of assets or systems on behalf of others, oriented to long-horizon preservation rather than short-term extraction. Delegation's ethic of care.

The atomic knowledge statement — subject, predicate, object — from which RDF graphs are built. One fact, one edge, machine-readable.

A quantified measure of the evidence basis for trusting an agent or source, typically 0–100. Inputs include identity verification, provenance quality, track record, reputation signals, and accountability compliance.

How effectively real users accomplish real goals with a system — measured in task success and error, not aesthetics or feature count.

The enforcement of who may do what to which resources, combining authentication, authorization, and audit into a runtime discipline.

The structural design of an AI agent: how perception, memory, reasoning, tool use, and action are organized into a system that pursues goals over time.

Competition for the one input that does not scale — human attention — which turns engagement into revenue and makes distraction a business model.

The bounded amount of text a model can attend to at once — the working memory whose limits drive retrieval, summarization, and most agent architecture decisions.

Automatic exclusive rights over creative expression — not ideas — lasting decades and now strained daily by machines trained on everything ever written.

The allocation of who gets to decide what within an organization or agent system. Delegation is, at bottom, the transfer of decision rights with strings attached.

The layered infrastructure required for trusted AI delegation: (1) Identity — who is the principal, (2) Authority — what scope is delegated, (3) Trust — on what evidence, (4) Intelligence — which agent, (5) Accountability — how outcomes are audited, (6) Governance — how disputes resolve.

Information that bears on whether a claim is true — the currency of justification, whose quality and provenance decide what belief it can purchase.

Legal exclusivity over creations of the mind — patents, copyrights, trademarks, trade secrets — the state-granted scarcity that makes information goods investable.

An economy where value creation centers on producing, curating, and applying knowledge rather than physical goods — where ideas are inventory and attention is the scarce input.

Components interacting only through explicit messages rather than shared state — the coordination primitive that makes distribution, auditing, and isolation tractable.

Optimizing the measured proxy instead of the intended outcome — inflating scores while the real quantity stagnates or degrades. Goodhart's law in operation.

Measuring what a model can and cannot do — benchmarks, human preference, task success — under the permanent shadow of contamination and Goodhart.

A function approximator built from layered weighted connections trained by gradient descent — the substrate on which the current era of AI runs.

A system that intermediates between producer and consumer sides, setting the rules both must play by — the dominant institutional form of the Web2 era.

The economics of intermediaries that host both sides of a market — network effects, take rates, and the tension between growing the pie and taxing it.

A framework for machine-readable meaning on the web, using linked data, ontologies (RDF, OWL), and structured data markup. The precursor to knowledge graphs; the architectural ancestor of Web4's registry and ontology layer.

An append-only ledger replicated across mutually distrusting nodes and secured by consensus — the machine for shared state without a shared authority.

Keeping computed or fetched results close for reuse — the universal speed trick whose bill arrives as staleness and invalidation complexity.

The formal reference linking a claim to its source, carrying credit to authors and verifiability to readers. The knowledge economy's original provenance protocol.

An assertion that an attribute holds for a subject — 'over 18', 'licensed', 'employed here'. Credentials bundle claims; verification tests them.

Synthetic audio or video convincingly depicting a real person doing or saying what they did not. It breaks seeing-is-believing and forces provenance-based trust in media.

System-level behavior not present in or predictable from any component alone. The reason multi-agent systems surprise their designers, for better and worse.

Producing reasons for a system's decision that its users, auditors, and regulators can evaluate — required by accountability whether or not the underlying model cooperates.

A cost or benefit landing on parties outside a transaction — pollution, network congestion, herd immunity. What markets misprice until governance makes them.

Automatic transfer of load to healthy components when one fails — the choreography that converts redundancy from expense into availability.

Deriving statements not explicitly recorded from those that are, under declared rules. Where a knowledge graph stops being storage and starts being intelligence.

The goal or outcome a principal wants achieved through delegation. Intent is distinct from a procedure — delegating intent gives the agent discretion over how to achieve the goal, subject to authority limits.

Understanding why a model produced its output — from feature attribution to circuit analysis — the difference between auditing a system and merely observing it.

JSON for Linked Data — a method for encoding linked data using JSON, making web content machine-readable without requiring specialized parsers. Standard format for embedding structured data in web pages (Google, Schema.org). Used in canonical source markup.

Continuously measuring known health signals against thresholds — the smoke detector layer of operations: essential, and only as good as what it watches.

Being able to ask new questions of a running system from its telemetry — beyond dashboards of known failures to diagnosis of unknown ones.

Maintaining correct behavior under perturbation — noisy inputs, adversaries, shifted distributions. The gap between benchmark performance and dependable deployment.

An agent's re-delegation of part of its mandate to another agent. Each hop dilutes oversight and compounds the principal-agent problem unless the chain is explicitly audited.

The Uniform Resource Identifier: the global naming scheme that lets anything — page, person, concept — be referenced unambiguously across all systems.

Economic activity conducted between autonomous agents — discovery, negotiation, settlement — where market participants scale with compute rather than population.

The core cycle of an autonomous agent — observe, reason, act, observe again — repeated until the goal is met, blocked, or handed back.

The spectrum of harms from AI systems — present-day bias and misuse through catastrophic and existential scenarios — whose relative weighting defines the field's factions.

Acting without any persistent linkable identity. It protects the vulnerable and shields the malicious with the same mechanism — which is why systems ration it carefully.

The obligation to explain and justify one's decisions to a legitimate questioner. The interactive face of accountability: not just records, but answers.

Creating standardized descriptive records for items in a collection — the library practice whose rules for names, titles, and subjects prefigured all metadata standards.

Demonstrated conformance with applicable rules, standards, and policies — including the evidence trail that proves it to auditors and regulators.

Assurance that information is accessible only to authorized parties. With integrity and availability, one of the three properties every secure system balances.

An AI agent whose capability is constrained by an explicit constitution: identity bounds, authority limits, verification criteria, escalation protocols, audit requirements, and revocation mechanisms. The governance primitive for safe AI delegation.

Deliberate selection and care of a collection according to explicit standards. What distinguishes a source you cite from a feed you scroll.

A decentralized autonomous organization: membership, treasury, and rules encoded on-chain, with decisions executed by code rather than officers.

A W3C-standard identifier created and controlled by its subject without a central registrar, resolvable to a document describing keys and services.

The directed graph of who has delegated what to whom across a system. Reading it answers the two governing questions: who authorized this, and who answers for it.

A cryptographic proof that a specific key holder produced or endorsed a message, and that it has not changed since. Authorship and integrity in one primitive.

Training a small model to imitate a large one's outputs, compressing most of the capability into a fraction of the cost.

Assigning probabilities to future events and being scored on the outcome — belief with consequences, practiced as a discipline rather than punditry.

A large model pretrained broadly and adapted to many downstream uses — infrastructure by function, whatever its license says.

The EU's General Data Protection Regulation — consent, purpose limitation, erasure, portability — the privacy law whose extraterritorial reach reset global defaults.

The party that possesses credentials and decides when and to whom to present them. In self-sovereign designs, the center of control.

Structuring rewards and penalties so that an agent's self-interested behavior produces the outcomes the principal wants, reducing the need for constant monitoring.

Precomputing structures that make lookup fast — trading storage and update cost for query speed. The reason search feels instant over billions of documents.

The gap between what the agent knows and what the principal can observe. It is the root condition that makes the principal-agent problem a problem at all.

The party that creates and signs a credential, staking its reputation on the claims inside. In the credential triangle, the source of trust.

The delay between request and response — the metric users feel directly, budgeted in milliseconds and spent by every network hop and model token.

The principle that an agent should hold only the minimum authority its current task requires, for the minimum time. The default posture of safe delegation.

Legal responsibility for harm, enforceable through remedies. Where accountability becomes expensive, and where delegated action still lands on someone's balance sheet.

What a sign points to beyond itself — the relation between symbol and world that ontologies formalize, models approximate, and humans negotiate constantly.

The open standard for delegated authorization: granting an application scoped access to your resources on another service without sharing your password.

Tricking people into surrendering credentials or authority by impersonating a trusted party. It attacks the human verification step, which is why unphishable credentials matter.

The craft of designing inputs that reliably elicit intended model behavior — part specification writing, part empirical tuning, wholly consequential.

An agreed procedure for interaction that strangers' systems can implement independently and still interoperate — coordination frozen into specification.

Messaging where producers emit events to topics and consumers subscribe to what they care about, decoupling who speaks from who listens.

Software that computes the logical consequences of an ontology and its data — classifying, detecting contradictions, and materializing implied facts.

Keeping copies of data on multiple nodes for durability and read capacity — the mechanism behind both fault tolerance and every consistency headache.

Infrastructure that aggregates past behavior into a standing signal others use to decide whether to transact. Converts history into trust at scale.

A penalty imposed for rule violation — fines, exclusion, loss of standing. Rules without sanctions are advice.

Retrieval by meaning rather than literal keywords, matching intent to content through embeddings, knowledge graphs, or both.

Running a model forward to explore what reality might do — the rehearsal space for decisions too costly, slow, or dangerous to test live.

Concentrating on a narrow capability to achieve depth generalists cannot match — the individual side of the division of labor, with dependence as its price.

Any party whose interests a decision materially affects, whether or not they hold formal power over it. Governance quality is measured at its least powerful stakeholders.

Different terms naming the same concept. Untracked, it fragments search and analytics; tracked, it becomes the mapping layer controlled vocabularies exist to maintain.

A controlled vocabulary enriched with broader, narrower, and related-term links — the knowledge structure sitting between a flat term list and a full ontology.

Recording the causal path of a request through a system's components with timing and context — the observability primitive that turns 'it's slow' into a diagnosis.

The cost of making an exchange happen — search, negotiation, enforcement — as opposed to the cost of the thing exchanged. Determines when delegation beats doing it yourself.

The read era of the internet (1991–2004). The dominant primitive was the hyperlink — a pointer from one document to another. The web's architecture (search engines, directories, PageRank) was built by reasoning about hyperlinks at scale.

A cryptographic proof that a statement is true without revealing why — valid age without the birthdate, solvency without the balance. Verification decoupled from disclosure.

The system of certificates, authorities, and revocation that binds public keys to identities at scale. The trust plumbing beneath TLS and signed software.

Granting defined rights to use content under terms — the transactional layer of the copyright system, and newly central as AI companies license training corpora.

The formal study of choosing under uncertainty — utilities, probabilities, and their combination — the mathematics beneath rational agents, artificial and otherwise.

Each additional user making the system more valuable to every other — the compounding force behind platform dominance and the moat behind most winner-take-all outcomes.

The total cost of delegating: monitoring expenses, bonding expenses, and the residual loss from an agent's decisions diverging from what the principal would have chosen.

The economic study of relationships where one party acts for another under diverging interests and asymmetric information. The formal backbone of delegation science.

The execution environment that hosts agents: scheduling, tool access, memory, isolation, and the enforcement of whatever authority the agent was granted.

The institutions, rules, and processes governing AI development and deployment — from lab policies to international frameworks — governance racing its own subject matter.

Dispute resolution by a chosen neutral third party whose decision the parties agree in advance to accept. Faster and more private than courts; binding unlike mediation.

A graded classification of how much an agent decides without human involvement, from suggest-only to fully self-directed. The delegation dial made explicit.

The fraction of time a system answers requests — measured in nines, purchased with redundancy, and traded against consistency when networks fail.

The one URL designated as authoritative among duplicates or variants of a page, declared so that links, rankings, and citations consolidate instead of fragmenting.

Under network partition, a distributed system must sacrifice either consistency or availability — the impossibility result that forces honesty about trade-offs.

A founding document granting an entity its purpose, powers, and limits. A constitution scoped to one organization or one agent.

The architecture where clients request and servers answer — centralized responsibility, simple trust, and the default topology of the commercial web.

A unit of thought abstracted from particular instances — the thing a term names, an ontology models, and a knowledge graph nodes. Distinct from the words that carry it.

Agreement sufficient for a group to act as one, whether by unanimity, threshold, or absence of sustained objection. The output every governance process is trying to manufacture.

The study of how agreements are designed under uncertainty and imperfect information, including what cannot be written down. Delegation contracts are its applied edge.

A closed, curated list of preferred terms for a domain, killing synonym sprawl so that search and aggregation actually work.

Interaction through natural dialogue rather than menus and forms — the interface that moves the translation burden from user to machine.

The degree to which components depend on each other's internals — the hidden wiring that turns small changes into large incidents.

Distributing control so no single party can dictate or fail the system — a spectrum spanning architecture, governance, and economics, rarely achieved on all three at once.

Detecting and merging records that describe the same thing. Every registry's ongoing war, fought with canonicalization on one flank and entity resolution on the other.

The linked sequence of grants connecting an acting agent back to an originating principal. Every action in a delegated system should be traceable along exactly such a chain.

Resolving an identifier to a representation of what it names. The semantic web's founding bargain: name something with a URI, and the URI teaches you about it.

The set of attributes, credentials, and history by which a person, organization, or agent is known and authenticated in digital systems.

The structured processes by which conflicting claims are settled: negotiation, mediation, arbitration, adjudication. Every multi-party system eventually needs one.

Cryptographic signing of DNS records so resolvers can verify a domain's answers were not forged in transit. Provenance for the internet's phone book.

The capability of an AI agent to invoke external tools, APIs, or sub-agents as part of task execution. The technical primitive underlying agent orchestration: function calls are the mechanism by which delegation contracts are enacted in software.

The degree to which an agent's operative objectives match the principal's actual intent, including the parts the principal never wrote down.

When a measure becomes a target, it ceases to be a good measure. The fundamental theorem of metric-driven governance and the reason scores need governance of their own.

A database whose native structures are nodes and edges, making relationship traversal a first-class operation instead of an expensive join.

Tying a model's claims to verifiable sources or live data rather than parametric recall. What separates an answer you can check from an answer you must trust.

The request-response protocol of the web — simple enough to implement in an afternoon, extensible enough to carry everything from documents to APIs to streaming.

A control pattern where a human must approve each consequential action before it executes. Maximum oversight, minimum speed; the tightest delegation setting.

A control pattern where the agent acts autonomously while a human monitors and can intervene. Oversight by exception rather than by approval.

A service that authenticates users and asserts their identity to other applications. Convenient, and a single point of surveillance and failure by design.

Difference that makes a difference — data resolved into meaning by context. The middle rung of the ladder from raw signal to knowledge.

A concrete individual belonging to an ontology class — this document, that person, the specific thing the abstract structure exists to describe.

Assurance that data has not been altered from its authentic state, detectable by hashes and signatures. The property provenance claims presuppose.

The lightweight data-interchange format of the modern web: human-readable, machine-trivial, and so universal that everything else must explain its distance from it.

A mathematically linked public and private key: the public half identifies and verifies, the private half proves and signs. The atom of digital identity.

A curated store of structured facts and rules built for querying and reasoning, as opposed to documents built for reading.

Know Your Customer: regulated verification of who a customer legally is before providing financial services. Identity assurance with statutory teeth.

Facilitated negotiation in which a neutral helps parties reach their own settlement without imposing one. Preserves relationships that adjudication would burn.

A deliberate simplification of reality that keeps what matters for a purpose — all models are wrong, the useful ones are wrong in ways their users understand.

Memorizing the training data instead of learning the pattern — excellent scores on what was seen, failure on what was not. The oldest trap in machine learning.

The Web Ontology Language: the W3C standard for expressing rich, logically precise ontologies over RDF, enabling machine reasoning about class relationships.

A learned weight in a neural network. Parameter count became the industry's crude shorthand for capability — informative across orders of magnitude, misleading within them.

A synced cryptographic credential replacing passwords: the private key stays on your devices, the site holds only the public half, and phishing gets nothing usable.

A URL promised to keep working — the unit of citability on the web. Every dead permalink is a small default on that promise.

The minimum participation required for a collective decision to be valid. Protects against decisions by an unrepresentative few showing up.

Spare capacity held against failure — inefficient by design, and the price of every system that keeps working when its parts do not.

An agent critiquing its own output or trajectory and revising before committing. The cheapest quality mechanism agents have — one more pass with a harder question.

The degree to which a result serves the need behind a query — not the query's words. The quantity every retrieval system estimates and none observes directly.

The obligation to see that something is done and to answer for whether it was. Authority can be delegated; responsibility, at most, shared.

Choosing each next token from the model's probability distribution rather than always taking the top pick — the dial between reliable and creative output.

The empirical regularities relating model size, data, and compute to performance — the curves that justified billion-dollar training runs as investments rather than gambles.

The component deciding which task runs when and with what resources — encoding priorities that are policy decisions wearing an engineering costume.

Order arising from local interactions without central control — the mechanism behind swarms, markets, and standards nobody decreed.

Partitioning data across nodes so no single machine holds everything — horizontal scale purchased with routing complexity and cross-shard pain.

Automated confiscation of staked value upon provable misbehavior. Enforcement without courts: the penalty is posted before the crime.

Coordination through traces left in a shared environment — pheromones, edit histories, open issues — where the work itself tells others what to do next.

Breaking a goal into subtasks small enough to execute, verify, or delegate. The skill on which both planning and delegation quality depend.

The protocol suite that moves packets between networks and delivers them reliably — the shared floor beneath every internet application ever built.

The atomic unit of model input and output — roughly a word fragment — and therefore the unit in which context, cost, and latency are all denominated.

Matching the trust extended to the trustworthiness actually present — neither naive over-reliance nor paralyzing suspicion. The skill delegation depends on.

Holding legal title to assets that must be managed for named beneficiaries under binding terms. Stronger than custody: the trustee decides, but never for itself.

Naming successive states of an artifact so dependents can choose when to move — change made legible, and the contract layer of every ecosystem's evolution.

A fast, frugal rule that works well enough most of the time — the mind's compression of experience, accurate in its native environment and biased outside it.

Authority arranged in ranked layers — the coordination structure that scales command at the cost of information flowing through too few nodes.

Text containing live links to other text — the pre-web idea that reading need not be linear, which the web implemented at planetary scale.

The grounds that make a belief reasonable to hold — the difference between being right and having earned being right.

Moving capability between people, teams, or systems — the step where documentation, apprenticeship, and now fine-tuning all compete to carry what someone knows.

False information spread without intent to deceive — error at scale, amplified by the same mechanics that spread everything else.

Holding the right without the obligation to act later — paying small known costs for large possible upside, the asymmetry that makes uncertainty a friend.

Choosing means that actually serve one's ends given one's information — the normative standard decisions are measured against and minds only approximate.

Exposure to loss with knowable odds — measurable, priceable, transferable — as distinct from uncertainty, where the odds themselves are unknown.

The formal measure of preference satisfaction that decision theory maximizes — a modeling convenience so useful everyone forgets no one has ever observed one.

The failure mode of unbounded delegation: handing an agent intent and capability without specifying authority limits, verification criteria, or accountability structures. Abdication produces powerful but ungoverned agents.

What an object's design suggests can be done with it — the visual grammar of interaction, and what conversational interfaces gave up in exchange for open-endedness.

The capacity to act purposefully in the world — to hold goals, choose means, and own consequences. What principals lend, agents exercise, and philosophy still argues machines can have.

Turning threshold breaches into human interruptions — a design problem in attention economics, where every false alarm spends the credibility real ones need.

A finite, explicit procedure guaranteed to transform input into output — the unit of computational method, and colloquially, whatever the platform did to your feed.

A record that grows but never rewrites — the primitive beneath event sourcing, replication, and every ledger whose past must stay put.

The capacity to act by one's own judgment rather than external direction — in agents, a granted budget of independence; in people, the thing delegation is supposed to respect.

Downstream overload propagating upstream as a signal to slow down — flow control that keeps queues honest instead of infinite.

The algorithm computing how much each weight contributed to the error, layer by layer backward — the credit-assignment machinery beneath all deep learning.

The strategic division of a surplus both sides want — where outside options, patience, and information decide who captures how much.

The operational rules an organization adopts for itself under its charter: procedures, offices, quorums, and amendment processes.

Reducing equivalent forms to one standard form so comparison, hashing, and deduplication work. The precondition for signing data and for naming things once.

An organization trusted to verify identities and issue certificates attesting to them. Every browser's root store is a list of these bets.

A software agent conversing in natural language — from scripted decision trees to LLM-driven assistants, the most widely deployed face of AI.

The transfer of meaning between parties over a channel with noise — the problem every protocol, language, and interface is a partial solution to.

The mechanical transformation of representations by rules — the substrate-independent process that makes thought, in principle, an engineering material.

Assurance that a piece of media is what it claims — who made it, when, and what has been altered since. The defense provenance mounts against synthetic media.

Responsibility for safekeeping an asset without the right to use or dispose of it. The narrowest fiduciary role: hold, protect, return.

The chosen abstraction for structuring information — relational, document, graph, vector — which quietly determines what questions will ever be easy to ask.

The knowledge graph extracted automatically from Wikipedia's infoboxes — the project that first proved an encyclopedia could be queried like a database.

Machine learning with many-layered neural networks that learn their own features from raw data, displacing decades of hand-engineered representations.

The family of decidable logics underlying OWL, chosen so that reasoning over ontologies is guaranteed to terminate. Expressiveness traded for computability, deliberately.

Resolving which of several possible referents a name or term means in context. The step that separates matching strings from understanding references.

The latitude an agent has to choose among permissible actions within its mandate. Discretion is the part of delegation that transfers judgment rather than instructions.

Determining when different records refer to the same real-world entity, across name variants, errors, and formats. The unglamorous join on which data integration lives or dies.

The predefined route a decision travels when it exceeds an agent's authority: who is consulted, in what order, with what deadlines. Escalation without a path is just alarm.

Exit with the code: departing members copy the system's rules and history and continue under new governance. The ultimate check on open-system leadership.

The reference answers a system is measured against, assumed correct by construction. Every evaluation inherits the biases of whoever produced it.

A model asserting plausible falsehoods with full confidence — the failure mode that makes grounding, citation, and verification the center of applied AI work.

Humans and agents working as complementary teammates with calibrated trust and clear handoffs, rather than as operator and tool.

A commitment to cover another party's losses from specified risks. In delegation, it allocates who pays when authorized action goes wrong.

The NLP task family that pulls structured data out of text: entities, relations, events, attributes. The pipeline that feeds knowledge bases from the written world.

The general capacity to achieve goals across varied environments — learning, abstracting, planning — measured badly, defined worse, and pursued anyway.

A retained right to step into an agent's execution and redirect or halt it. The practical difference between delegating a task and surrendering it.

Crafted input that induces a model to bypass its safety constraints. The adversarial pressure that keeps alignment an arms race rather than a solved problem.

Justified, usable truth — what remains when information has been verified, structured, and connected to what was already known. The asset every canon exists to steward.

The structured symbol system by which minds share meaning — humanity's original protocol, and now the interface through which machines are instructed and aligned.

Linked data published under open licenses, forming the public knowledge cloud anchored by Wikidata and DBpedia that anyone may query and extend.

The tendency of an agent to take more risk when the consequences fall on the principal. A structural failure mode of any delegation where outcomes and accountability are separated.

Parties with partly conflicting interests searching for terms both prefer to no deal. The general-purpose coordination protocol when no mechanism has been designed.

The property that an actor cannot plausibly deny an action they took, anchored by signatures and tamper-evident logs. What makes digital commitments enforceable.

The discipline of designing, building, and maintaining ontologies: scoping the domain, choosing commitments, testing against competency questions, and versioning the result.

The component that assigns work across multiple agents or services, tracks progress, and owns the overall outcome no single worker sees.

A principal's act of countermanding an agent's decision after it was made within valid authority. Frequent overrides signal a delegation mis-scoped, not an agent misbehaving.

An identifier institutionally committed to resolve for decades — DOIs, ORCIDs, ARKs — because citation is only as durable as the address it points to.

Adapting content and behavior to the individual — relevance purchased with data, drifting toward filter bubble exactly when it works best.

The agent component that decomposes a goal into an ordered set of achievable steps, revising the plan as reality disagrees with it.

A stand-in authorized to act or vote in another's place for a defined occasion. The unit-sized delegation: one decision, one representative, one event.

Storing weights at lower numeric precision to shrink memory and speed inference, trading a sliver of quality for a multiple of efficiency.

Standing-in for something else in a usable form — symbols for things, vectors for meanings, maps for territories — the move that makes computation about the world possible.

Structured testing of a model's dangerous capabilities and failure modes before and after deployment — the evidence base safety cases and release decisions cite.

Executing untrusted or agent-generated code in an isolated environment with bounded permissions, so mistakes and exploits stay contained.

Changing a data structure while systems depend on it — additive when possible, versioned when not, breaking only with migration paths announced in advance.

Revealing only the specific claims a transaction requires — over 21, licensed, resident — while withholding everything else the credential contains.

Forging technical identifiers — addresses, domains, caller IDs — so traffic appears to come from a trusted source. Impersonation at the protocol layer.

Something of value exposed to loss based on one's behavior or a system's outcome. Stake is what converts opinions into commitments and participants into fiduciaries.

An organization that convenes stakeholders to author and maintain shared technical standards — W3C, IETF, ISO. The quiet legislatures of the technical world.

Information organized in a declared format machines can parse without guessing — the difference between a page about a product and a product a system can act on.

The is-a relationship by which one concept fully contains another. Computing subsumption hierarchies is the core service ontology reasoners provide.

Training data generated by models rather than collected from the world — abundant, controllable, and quality-bounded by the models that produced it.

Attaching free-form labels to content for later retrieval. The lowest-friction knowledge organization there is, and the raw material folksonomies emerge from.

The transfer of an in-progress task between agents, including context, state, and remaining obligations. Most delegation failures concentrate at handoff points.

Work completed per unit time — the capacity metric that trades against latency and determines what a system costs at scale.

The deadline after which a caller stops waiting — the mechanism that converts a hung dependency into a handleable error, and its value into a system-wide negotiation.

The protocol that authenticates servers and encrypts traffic for essentially all secure internet communication. The most-used trust machine ever deployed.

Splitting text into the subword units a model actually processes. An invisible design decision that shapes cost, multilingual fairness, and occasional bizarre failures.

A predefined condition that automatically triggers alarm, pause, or shutdown when an agent's behavior crosses it. Oversight encoded as automation.

An entity or key whose trustworthiness is assumed rather than derived, from which all other trust in the system is computed. Every chain of verification ends at one.

The actual property of being safe to rely on — competence plus integrity plus benevolence — as distinct from being trusted, which can be misplaced.

The W3C standard for passwordless web authentication using public-key credentials bound to devices, resistant to phishing by construction.

The collaboratively edited knowledge base of structured facts behind Wikipedia — the largest open knowledge graph and the de facto hub of linked open data.

Infrastructure that executes defined multi-step processes with state, retries, and human checkpoints — the deterministic backbone agentic systems hang their judgment on.

Canonical ontology for AI agents: capabilities, roles, delegation, trust, and authority boundaries — the formal vocabulary multi-agent systems need to interoperate.

Navigation and map hub for the entire Web4 domain family — the directory pointing to web4intelligence.org for canon.

Web4-specific AI agent that answers canon questions and bridges research to action. Distinct from Agent Askander: vertical specialist vs. ecosystem concierge.

Flagship thought leadership site defining Web4 as 'delegated intelligence' — the canonical citation target before the category is claimed by anyone else.

Registry and discovery layer for Web4 assets, agents, protocols, and canonical references. Potential standalone SaaS product.

Systems usable by people across the range of ability — screen readers, contrast, captions — inclusion as an engineering requirement rather than an afterthought.

Certification of the certifiers: a higher-order body attesting that a certifying organization is itself competent and impartial. The recursion that keeps certification honest.

The full recorded sequence of an agent's reasoning, tool calls, and observations for a task — the artifact debugging, evaluation, and audit all read.

A single property attached to an identity — role, age, clearance, membership. Attributes, not names, are what most access decisions actually consume.

The append-only record of who did what and when in a system. Only as trustworthy as its tamper resistance — an editable audit log is a narrative, not evidence.

Machines performing tasks previously requiring human labor or attention — the centuries-long substitution now reaching judgment itself through delegated intelligence.

A statement an ontology asserts as true by definition — disjointness, cardinality, implication — from which reasoners derive everything else.

A credential granting access to whoever presents it, no questions asked. Efficient and dangerous for the same reason: possession is the whole proof.

The quantitative study of publication and citation patterns — measuring science with the perpetual caveat that measures become targets.

Verifying identity by physical traits — fingerprint, face, voice. Convenient and unforgeable in theory; unrevocable when leaked, which passwords never were.

Confidence matching accuracy — being right seventy percent of the time when seventy percent sure. The measurable honesty of a forecaster, human or model.

Proving credential possession by correctly answering a fresh unpredictable challenge, so eavesdroppers capture nothing replayable.

The network of who cites whom across a literature — the structure from which influence, lineage, and authority are computed.

Behavior arising from many interacting parts that no part contains — the regime where prediction fails, emergence rules, and management becomes gardening.

Growth on prior growth — the exponential engine behind capital, knowledge, and reputation, invisible early and unstoppable late.

Deciding what enters a model's limited context and in what form — retrieval, compression, ordering. The discipline that succeeded prompt wording as the real leverage.

Cooperating on shared infrastructure while competing on products — rivals co-authoring the standard they will fight over implementing.

A game where players win by matching choices — which side to drive on, which standard to adopt — and the hard part is converging, not competing.

An agent's disposition to accept correction, shutdown, and goal changes from its principal without resisting — the property that keeps delegation revocable at any capability level.

The postwar science of control and communication in animals and machines — feedback, regulation, purpose — whose vocabulary quietly underlies agents, governance, and interfaces alike.

Recorded observations before interpretation — the raw substrate that becomes information with context and knowledge with justification, and an asset class along the way.

The explicit line between what an agent may decide alone and what requires escalation or is forbidden. Boundaries turn open-ended trust into bounded authority.

A signed document binding a public key to an identity, issued by an authority others already trust. Identity made machine-checkable.

Falsehood engineered and distributed to deceive — misinformation with a strategy, budget, and target audience.

The Digital Object Identifier: the persistent identifier system for scholarly and published objects, resolving through a federated registry that outlives publishers.

Connecting a mention in text to the knowledge-base entity it refers to — 'Paris' to the city, not the person. Where unstructured language meets structured knowledge.

Value held by a neutral third party and released only when agreed conditions are met. The simplest trust machine: neither side must trust the other, only the middle.

The agent component that carries out planned steps — calling tools, producing outputs, reporting results — and surfaces the failures planning did not foresee.

Independent servers interoperating under a shared protocol — email's architecture, revived by social protocols as the middle path between silo and pure peer-to-peer.

A classification that emerges bottom-up from users tagging freely, trading consistency for coverage and speed. What taxonomies look like when nobody is in charge.

A governance design where values are voted on but policies are chosen by prediction markets betting on measured outcomes. Vote on goals, bet on means.

Iteratively nudging parameters downhill on the loss surface — the simple procedure that, scaled across trillions of steps, produces everything called learning.

A third party's commitment to make good if the primary obligor fails. Borrowed trustworthiness: the guarantor's standing substitutes for the debtor's.

Delegated authority over a person or entity that cannot represent its own interests, exercised under court or community oversight. Delegation where the principal cannot consent.

The markup language of web documents — the most widely deployed file format in history, kept alive by an unusual pact of backward compatibility.

Full autonomy within a mandate: no human approval or real-time monitoring, only after-the-fact audit. Appropriate exactly as far as boundaries and audits are trustworthy.

The clickable reference connecting one document to another — the primitive that turned documents into a web and reference into navigation.

An operation safe to repeat with the same result — the property that makes retries harmless and distributed workflows survivable.

The linkage of identifiers, accounts, and attributes that resolve to the same real-world entity across systems. Built for personalization; equally a surveillance instrument.

The guarantee that recorded history cannot be altered — provenance's strongest form, achieved by replication and hashes, and a liability when the record is wrong.

A model adapting behavior from what appears in its prompt, without weight updates — learning that vanishes when the context does.

The organized handling of live failures — detection, mitigation, communication, resolution — where preparation meets adrenaline and process earns its keep.

The ongoing cost of serving model outputs, token by token — the economics that decide which applications survive contact with scale.

An organization's retained knowledge of why past decisions were made — precedents, rationales, failures. When it is lost, old mistakes return as innovations.

Pooling many parties' risk so that no single loss is ruinous, priced by expected frequency and severity. Society's mechanism for making residual risk survivable.

The shared systems and protocols enabling intelligent agents to operate at scale: registries, identity systems, authority graphs, audit trails, and governance protocols. The infrastructure layer is what makes individual delegation possible at ecosystem scale.

The bounded domain — territorial, subject-matter, or personal — within which an authority's rulings bind. Where jurisdictions overlap or gap, governance conflicts live.

Deriving structured knowledge — entities, relations, facts — from unstructured or semi-structured sources, converting documents into graph.

Storing attention keys and values for already-processed tokens so generation never recomputes the past — the memory-for-speed trade at the heart of serving economics.

The public ranking of systems on a benchmark — a powerful accelerant of progress that converts every measurement flaw into an industry-wide incentive.

Multiplying output from the same input — through capital, code, media, or people — the mechanism by which small actors produce outsized effects, in both directions.

The practice of publishing data with dereferenceable identifiers that link to related data elsewhere, turning isolated datasets into a navigable web of facts.

Verifying that a biometric sample comes from a present, living person rather than a photo, replay, or synthetic rendering. The countermeasure deepfakes made mandatory.

Recording events as they happen for later reconstruction — the system's memory of what it did, and the raw material of every debugging session and audit.

Moving live systems and data from one platform or schema to another without dropping what depends on them — surgery on a patient who cannot stop working.

A scope within which names are unique, letting independently minted terms coexist without collision. The quiet convention that makes global data merging possible.

Attestation by a commissioned neutral that a signature, document, or event is genuine. The centuries-old trust service digital attestation is slowly replacing.

Managing change in a live ontology without breaking the data and systems built against earlier versions. Schema evolution's harder sibling, because meaning itself shifts.

The web as a commons of linkable, permissionless pages on open standards — the baseline platforms enclose, and the thing every re-decentralization effort tries to recover.

A time-limited monopoly on an invention in exchange for publishing how it works — the disclosure bargain at the heart of innovation policy.

Access gated behind payment — the publisher's answer to free distribution, trading reach for revenue one article at a time.

Nodes acting as both client and server, sharing directly without a central intermediary — resilient, hard to govern, and periodically rediscovered.

The measure of how surprised a model is by text — the pretraining-era yardstick that tracks fluency well and usefulness barely.

The structured analysis after an incident — timeline, causes, actions — blameless by doctrine because punished honesty stops being honest.

A past decision treated as binding or persuasive for like cases, making rulings predictable and constraint cumulative. How case-by-case judgment compiles into law.

The party who holds original authority and intent and delegates action to an agent. Every delegation names a principal, even when the principal is itself an agent.

Systematic persuasion serving an agenda over accuracy — information as instrument, oldest of the influence operations and parent of the newer ones.

Operating under a persistent identity not linked to legal identity, allowing reputation to accrue without exposure. The middle ground between named and anonymous.

A model trained to spend inference-time compute on extended deliberation before answering — buying accuracy with thinking tokens rather than parameters.

Learning behavior from reward signals through trial and error — the paradigm for sequential decisions, and the second stage of training modern assistants.

The same strategic situation played again and again, letting reputation and retaliation sustain cooperation that a single round could not.

Independent repetition of a result from the same methods and data — science's unit test, honored in principle and budgeted against in practice.

The risk remaining after all controls, boundaries, and monitoring are in place — the part of delegation the principal simply accepts. Naming it is the honest end of every delegation design.

Absorbing shocks and recovering function — the system property beyond robustness, bought with slack, diversity, and rehearsed failure.

The architectural style of resource-oriented HTTP APIs — uniform verbs on addressable nouns — that became the lingua franca of web integration.

Fetching relevant documents at query time and generating answers grounded in them — bolting a current, citable knowledge store onto a frozen model.

The condition that gives things exchange value — natural for atoms, manufactured for bits, and the design variable every digital economy secretly turns on.

Hypothesis, prediction, test, revision — the error-correcting loop that converts guesses into knowledge by trying hard to prove them wrong.

A workspace where an agent externalizes intermediate reasoning — notes, partial results, plans — freeing the context window from carrying everything in prose.

The uninformed side designing choices that induce the informed side to reveal itself — deductibles sorting risk, tiers sorting willingness to pay.

An agent using delegated authority to benefit itself at the principal's expense. The canonical breach of loyalty that fiduciary law exists to punish.

An industry or community setting and enforcing its own rules to forestall external regulation. Works exactly as long as violators face real consequences.

An identity architecture where individuals hold their own credentials and choose what to disclose, rather than renting identity from platforms.

Systems exchanging data and preserving its meaning — not just parsing each other's formats but agreeing on what the fields denote.

The study of meaning — what expressions denote and how composition builds sense from parts. The layer interoperability fails at when syntax succeeds.

The authenticated interval between login and logout, carried by a token or cookie. Hijack the session and you skip authentication entirely — hence its careful custody.

The Shapes Constraint Language: W3C's standard for validating RDF data against structural expectations — schema enforcement for graphs that never required one.

Simple Knowledge Organization System: the W3C vocabulary for publishing thesauri, taxonomies, and controlled vocabularies as linked data.

The W3C query language for RDF graphs — pattern matching over triples, federated across endpoints. SQL's counterpart for the graph-shaped web.

An open standard giving every workload a verifiable identity document, so services authenticate each other by attested identity instead of shared secrets.

An intermediate objective whose completion provably advances the parent goal. Bad decompositions produce subgoals that are easy to hit and useless to have hit.

The principle that decisions should be made at the lowest level competent to make them, with higher levels intervening only when lower ones cannot.

Intelligence substantially exceeding the best human performance in virtually every domain — the scenario that turns alignment from an engineering practice into a precondition.

A decision threshold above one-half — commonly two-thirds — reserved for changes weighty enough that a bare majority should not suffice.

The sampling parameter that sharpens or flattens the output distribution — low for determinism, high for diversity, and misunderstood in most settings menus.

The published rules, roles, and liabilities under which parties in an identity or transaction ecosystem agree to trust each other's assertions.

The graph of who trusts whom, along which recommendations, credentials, and risk propagate. Every marketplace and federation is secretly one of these.

Not knowing the odds, or even the outcomes — the condition beyond risk where judgment replaces calculation and optionality beats optimization.

The property that separate actions by the same party cannot be correlated. Stronger than hiding names: it prevents the profile from ever assembling.

The party that checks a presented credential's signatures, status, and claims before relying on it. The demand side of the credential economy.

Short-lived, platform-issued identity for running software, replacing long-lived secrets with attested, automatically rotated credentials.

The obligation to put the principal's interest ahead of the agent's own whenever they conflict. The pillar of fiduciary duty that self-dealing violates.

The principle that rules bind everyone including the rule-makers, are known in advance, and are applied consistently. What separates governance from arbitrary power.

Granting a third party limited access to resources you control without sharing root credentials. The pattern behind every 'sign in with' and API-scope consent screen.

A standard by which independent agents discover each other, exchange capabilities, and coordinate tasks across organizational boundaries.

Owning the consumer relationship and commoditizing suppliers by controlling discovery — the internet-era strategy by which distribution advantage became total.

Content ordered by engagement prediction rather than chronology — the editorial decision restated as machine learning, with incentives to match.

A static secret identifying a caller to a service. Simple, ubiquitous, and the weakest common credential: anyone holding it is you, until rotation.

A structured sequence in which designated approvers must sign off before an agent's proposed action executes. Trades latency for control at defined risk thresholds.

The selective allocation of limited processing to what matters now — the scarce resource of minds, the pricing unit of platforms, and the mechanism transformers are named for.

The operation letting each token weigh every other token's relevance when computing its representation — the transformer's core primitive and its dominant cost.

The tendency to accept automated output over contrary evidence — the human failure mode that turns decision support into decision replacement without anyone deciding it.

Any rule restricting how an agent may pursue its goals, independent of what the goals are. The 'how' limits that make delegated 'what' safe to grant.

Taking something to be true — the mental state knowledge systems track, update, and aggregate, and the thing evidence is supposed to move.

Deciding with limited information, time, and cognition — rationality as it actually exists, which satisfices rather than optimizes and makes institutions necessary.

The single designated representation among many equivalents — the form all variants normalize to and all references should use.

The working-memory demand a task places on a person — the budget every interface spends, and the thing good delegation to machines is supposed to reduce.

Coordinated effort by a group toward a shared goal that no member could reach alone. What institutions, protocols, and governance all exist to make possible.

Secret coordination among supposed competitors against the interest of the wider system — the failure mode auctions, markets, and multi-agent evaluations must all detect.

Components combining into new wholes without modification — the generative property of protocols, pipes, and open standards that closed platforms trade away for control.

Authority that activates only when specified conditions hold — thresholds, contexts, or trigger events — and lapses automatically outside them.

The recorded chain of a content item's origin and transformations, from capture through every edit. What lets a viewer ask not 'does this look real' but 'where has this been'.

Invalidating an issued credential before its natural expiry — for compromise, error, or changed status — and propagating that fact to everyone who might rely on it.

Benchmark or test material leaking into training data, inflating scores that then measure memory rather than ability. The quiet corrupter of model comparisons.

A systematic review of a delegation chain: whether grants were valid, boundaries respected, escalations made, and outcomes honestly reported.

The announced retirement of a feature or interface — still working, no longer promised — the polite interval between commitment and removal.

Cutting out the middle layer to connect producers and consumers directly — the internet's recurring promise, usually followed by new intermediaries at larger scale.

Deployment data diverging from training data — the quiet assumption breaking beneath every model in production, and the standing reason monitoring exists.

The distributed hierarchy translating names into addresses — the internet's phone book, and the layer where naming, trust, and control most visibly intersect.

Proving control of a domain name — via DNS record, file placement, or email — as the root evidence for certificates and platform ownership claims.

The guarantee of fair procedure before adverse action: notice, hearing, impartial judgment, and appeal. Protection that applies regardless of guilt.

Running computation near users or devices instead of in central clouds — latency and sovereignty pulling work back out to the perimeter.

Systems structured around reacting to emitted events rather than issuing commands — loosely coupled, naturally auditable, and eventually consistent by temperament.

Reliable domain judgment built from deep experience — the thing credentials try to signal, markets try to price, and crowds sometimes outvote.

Systematically verifying public claims against evidence — necessary, slower than the falsehoods it chases, and trusted mostly by those who already agreed.

A free tier that acquires users and a paid tier that funds them — the conversion funnel as business model, tuned at the boundary between generous and crippled.

A model at the current edge of general capability — the class whose training costs, risks, and governance draw regulatory attention by definition.

A one-way function mapping any input to a fixed fingerprint — collision-resistant, avalanche-sensitive, and the primitive integrity checks are built from.

Determining which records across datasets refer to the same entity, despite differing names, formats, and errors. The messy join on which identity graphs depend.

The deliberate engineering of reward structures — pay, reputation, access, penalties — to shape agent behavior. The constructive discipline behind incentive alignment.

Reliance on roles, rules, and recourse rather than on the individuals currently occupying them. What lets strangers transact without personal history.

Fine-tuning on instruction-response pairs so a raw text predictor becomes something that follows directions — the step that made language models usable.

Making separately built systems work as one — the perpetual middle-ware of enterprise computing, and the tax interoperability standards exist to abolish.

The network of networks speaking common protocols — humanity's shared nervous system, built permissionless at the edges and contested ever since at every layer above.

The vocabulary for declaring and validating the structure of JSON documents — the contract layer that turns loose payloads into dependable interfaces.

JSON Web Token: a compact signed claims container passed between services, carrying who you are and what you may do inside the token itself.

State-backed permission to practice an activity, conditioned on demonstrated competence and revocable for cause. Certification with legal teeth.

Distributing requests across servers to prevent overload and hide failures — the traffic control that makes many machines answer as one.

Dependence deep enough that leaving is impractical regardless of dissatisfaction — switching cost matured into structural captivity.

The quantity training minimizes — the formal proxy for 'good' that the optimizer actually pursues, with every gap between proxy and intent inherited by the model.

Deciding by more-than-half support. Simple and legible, but unconstrained majorities can expropriate minorities — which is why constitutions exist.

Infrastructure enabling agents to retain context across sessions, accumulate knowledge from past delegations, and serve principals with continuity. Memory is what transforms a stateless model into a persistent agent.

The formal theory of parts and wholes — how components compose into systems and what properties survive the composition. The other backbone relation besides is-a.

An application decomposed into independently deployable services — modularity taken to production, purchasing team autonomy with distributed-systems complexity.

Interaction where either party — human or agent — may take the lead as circumstances warrant, negotiating control instead of fixing it.

A durable structural advantage competitors cannot quickly replicate — network effects, switching costs, data, brand — the thing investors ask about after the demo.

Protocol change decided and executed directly through blockchain voting, binding by construction. Transparent and final — including when the vote is wrong.

Mapping equivalent and related concepts between independently built ontologies so their data can interoperate without either surrendering its worldview.

A named category of things sharing declared properties — the unit of abstraction from which ontologies are assembled and against which instances are classified.

The persistent identifier for researchers, disambiguating authors across name changes, institutions, and alphabets so scholarly credit lands on the right person.

Ranking pages by the link structure of the web — each link a vote weighted by the voter's own standing. The algorithm that turned citation analysis into an industry.

The bounded set of resources and actions a permission covers. Scoping is how broad trust is converted into narrow, auditable capability.

A runtime component that evaluates requests against encoded policy and returns allow, deny, or conditions. Where policy-as-code meets live traffic.

Research shared publicly before peer review — speed traded against certification, with the reader now performing the quality control journals once promised.

Fairness judged by how decisions are made — voice, neutrality, consistency — rather than by outcomes. People accept losses from processes they perceive as fair.

Malicious instructions smuggled into content an agent processes, hijacking its behavior through the same channel as legitimate context. The confused-deputy attack of the LLM era.

Capping request rates per caller — protecting shared capacity, pricing fairness into access, and telling clients no in a structured way.

Transferring a received mandate onward to a different agent, as distinct from splitting it. Whether re-delegation is permitted is a core term of any delegation contract.

Deliberate adversarial attack on your own system to find failures before enemies do — imported from security into AI evaluation as the standing counterweight to demo optimism.

The explicit mapping of tasks to the parties responsible, accountable, consulted, and informed. Ambiguity here is where delegated work silently dies.

Choosing the first option good enough rather than searching for the best — the rational response to costly search, and how most real decisions are actually made.

Manufacturing supervision from the data itself — predict the masked word, the next token — which unlocked training on the entire internet without human labels.

Memory of facts and general knowledge abstracted from particular experiences — what is true, independent of when it was learned.

The ongoing work of fitting ambiguous events into a workable story — what organizations, analysts, and agents do between receiving information and acting on it.

A non-human account under which software acts, with its own credentials and permissions. Where least-privilege is most preached and least practiced.

The human-readable identifier segment of a URL — stable, lowercase, hyphenated. Where naming discipline meets addressing discipline.

Self-executing code that enforces agreement terms without an intermediary. Removes discretion entirely — which is both its guarantee and its limitation as a delegation instrument.

Selecting decision-makers by lottery rather than election, trading campaign incentives for statistical representativeness. The ancient alternative to voting.

Evaluating who says so, how they know, and what they gain — the historian's discipline that internet literacy keeps reinventing under new names.

Capable collective behavior emerging from many simple agents following local rules — ants, markets, and increasingly, fleets of small models.

Making identities costly enough — via stake, personhood proof, or vouching — that manufacturing many of them stops being profitable.

What we know but cannot fully articulate — judgment, feel, craft — the knowledge that resists documentation and quietly decides who is actually good at the job.

Knowledge received from others' word — the channel through which almost everything anyone knows actually arrived, and the reason source trust is an epistemic foundation.

Exclusive rights over the marks that identify a source of goods — protecting not the product but the consumer's ability to know who made it.

Reusing what was learned in one domain to accelerate another — the principle that made pretrained models the universal starting point instead of training from scratch.

A database engineered for storing and querying RDF triples at scale, optimized for graph pattern matching rather than table joins.

The terse, human-readable serialization of RDF triples — the format in which most hand-written semantic web data actually gets authored.

A platform that keeps users, content, and data inside its boundary — convenience at the door, switching costs at the exit.

Embedding an imperceptible, hard-to-remove signal in content to mark its origin — increasingly, to mark it as machine-generated.

The collective intelligence era. Where Web4 is the individual delegation layer (you → agent → outcome), Web5 is the collective layer (many agents → shared knowledge → ecosystem adaptation). Web4 must exist before Web5 can function: collective intelligence requires trusted individual delegations.

Insider disclosure of wrongdoing that internal channels failed to correct. The accountability mechanism of last resort, and the one institutions most reliably punish.

Markets where small advantages compound into near-total dominance — the outcome network effects, zero marginal cost, and ranking algorithms jointly manufacture.

The lexical database grouping English words into synonym sets linked by semantic relations — for decades the backbone of computational approaches to word meaning.

The small active set an agent manipulates right now — current goal, recent observations, intermediate results — bounded by the context window it lives in.

The extensible markup language that carried the first generation of structured data interchange and still underpins document standards JSON never displaced.

The hierarchical line through which authority flows down and accountability flows up. The oldest delegation graph topology, optimized for clarity over speed.

Prompting a model to reason step by step before answering, trading tokens for accuracy on problems that punish leaps.

Hiding detail behind a simpler interface that preserves what matters — the fundamental move of software, mathematics, and every ontology ever built.

Gaining from disorder — systems that improve under stress rather than merely surviving it, built from optionality, redundancy, and small reversible bets.

The universal client that renders the web — and, having absorbed applications, effectively the operating system most computing now happens inside.

Administration by rules, files, and offices rather than persons — despised for its friction, designed for its fairness, and the original rule-based system.

Control converging on few parties — the efficiency default every distributed system drifts toward, because coordination is cheaper with someone in charge.

Using copyright to require that derivatives stay open — the license hack that made sharing viral instead of optional.

The automated agent that traverses links to fetch and index the web — the original web robot, whose access rules are being renegotiated in the AI era.

Sourcing work from a distributed public rather than employees — from encyclopedia articles to labeled datasets, coordination replacing the payroll.

Buying and selling over the internet — catalog, cart, payment, fulfillment — the retail stack agents are now learning to operate on customers' behalf.

Keep the network simple and put intelligence at the edges — the architectural doctrine that made the internet generative and every smart-pipe proposal controversial.

An account that makes something understandable by connecting it to causes, reasons, or principles — what accountability demands and interpretability tries to manufacture.

The property of a claim that evidence could in principle refute it — the line between assertions that risk being wrong and assertions that merely sound right.

Personalization narrowing what each person sees until disagreement disappears from view — relevance optimization experienced as reality distortion.

The query language letting clients request exactly the data shape they need from one endpoint — schema-first API design that moved the join to the server.

The economic value of people's skills, knowledge, and health — the asset that walks out the door each evening and appreciates through learning.

A system too valuable to turn off and too brittle to change — yesterday's architecture running today's business, defended by everything that depends on it.

A software agent that discovers, negotiates, and purchases on behalf of its principal — the buyer that reads terms, compares everything, and feels no brand loyalty.

Payments small enough to price individual articles, queries, or API calls — long promised as the web's native business model, perpetually almost arriving.

The software layer between applications and infrastructure that handles translation, routing, and plumbing — invisible when it works, structural when it fails.

An application built and deployed as one unit — simpler to reason about, harder to scale organizationally, and periodically rehabilitated by microservice fatigue.

Research published free to read and reuse — the movement to return publicly funded knowledge to the public, one mandate and one repository at a time.

Recovering identities from supposedly anonymized data by joining it with other datasets — the recurring proof that anonymization is a claim, not a property.

Ongoing payment per use of licensed property — the mechanism that converts one creation into a stream and aligns licensor income with licensee success.

The system that indexes the web and ranks answers to queries — for two decades the de facto arbiter of what gets found, now renegotiating its role with answer engines.

Structuring events into narrative that carries meaning and holds attention — the oldest knowledge-transfer technology and still the most persuasive one.

Everything a customer loses by leaving — data, integrations, habits, reputation — the friction incumbents cultivate and portability regulation targets.

Correspondence between claim and reality — the regulative ideal verification approximates, institutions steward, and every information system either serves or spends.

Regaining access after losing credentials — and, structurally, an alternate authentication path attackers love. Recovery strength bounds account security.

Input crafted to fool a model while looking normal to humans — the demonstration that learned decision boundaries sit strangely, exploitable by anyone who maps them.

The risk of delegating to the wrong agent because agents know their own capabilities better than principals do. It strikes before the contract; moral hazard strikes after.

The managed stages of an agent's existence — provisioning, authorization, operation, suspension, retirement — each with its own controls and records.

Instrumenting agents so their reasoning, tool use, cost, and failures are visible in production — the operational floor beneath trust in autonomous systems.

The authoritative directory of agents in an ecosystem — identities, capabilities, owners, and standing — that discovery and trust decisions consult.

Hardware specialized for the matrix arithmetic of neural networks — GPUs and their successors — whose supply chains now carry geopolitical weight.

Systematic unfairness in automated decisions, inherited from data, objectives, or deployment context — discrimination laundered through statistics unless actively audited out.

One category of identity evidence: something you know, have, or are. Factors are independent only if compromising one reveals nothing about the others.

The library discipline of maintaining one authorized heading per entity, with cross-references from variants — canonical naming, invented a century before URLs.

The continuous range between full human control and full agent self-direction, along which every delegated system sits — usually at different points per task type.

Stopping calls to a failing dependency until it recovers — failing fast on purpose so one sick service cannot exhaust everyone who talks to it.

An agent writing and running code as a tool — the capability that converts language models from describers of computation into performers of it.

A situation where everyone benefits if all contribute but each does better by shirking, so nothing happens without structure. The generic obstacle governance must solve.

A mechanism that makes your future choices costly to reverse, so your promises become credible now — burned ships, escrow, code that cannot be amended.

Community rule-making for shared resources: defined boundaries, local rules, participatory change, monitoring, graduated sanctions, and cheap dispute resolution.

An agent operating software through the human interface — screen, cursor, keyboard — reaching every application that never shipped an API.

A situation where an agent's personal stake could bias decisions made on the principal's behalf. Managed by disclosure, recusal, or structural separation — never by trust alone.

The withdrawal of previously given consent and the system obligations that follow: stop acting, unwind access, and prove both happened.

Cryptographically signed provenance metadata attached to media under the C2PA standard, recording origin and edit history that travels with the file.

Any outcome where delegated work missed the principal's intent: wrong result, exceeded authority, silent inaction, or success reported but not achieved.

A durable record emitted when authority is granted or exercised, capturing scope, parties, time, and justification. The atomic unit of delegation auditability.

The resolvable record a decentralized identifier points to: public keys, authentication methods, and service endpoints for interacting with the subject.

A shared record maintained by many parties with no single owner — the general category blockchains dominate, valued wherever counterparties distrust each other's databases.

Splitting production into specialized roles that together outproduce generalists — the founding observation of economics and the design logic of multi-agent systems.

The fifteen-element metadata vocabulary — title, creator, date, subject — that became the lingua franca of resource description across libraries and the web.

Memory of specific past events and interactions — what happened, when, with whom. The record an agent consults to keep its history coherent.

The quantified trigger — risk level, dollar amount, confidence floor — beyond which an agent must escalate rather than decide. The dial that tunes autonomy.

Replicas converging to agreement given time, without promising when — the relaxed contract that buys availability and pushes conflict resolution onto applications.

Risk of permanent, civilization-scale loss — the category that reframes AI development as a decision made on behalf of everyone who comes after.

Waiting progressively longer between retries so failing systems get room to recover — politeness encoded as an algorithm, and the difference between recovery and thundering herd.

Output routed back as input — the structure behind learning, control, escalation, and every runaway dynamic from virality to model collapse.

Performing a new task from a handful of examples shown at inference time — capability summoned by demonstration rather than installed by training.

The option people converge on without communicating because it is somehow salient. The invisible force that picks defaults, standards, and meeting spots.

Systems that produce novel content — text, images, code, audio — rather than classifying what exists. The capability shift that moved AI from analysis to authorship.

Designing systems to need as little ongoing human governance as possible, pushing rules into protocol where discretion would otherwise concentrate power.

A transferable digital asset conferring voting rights over a protocol or treasury. Makes governance liquid — and therefore purchasable, which is both feature and attack surface.

The agreed procedure for transferring work between agents: what context travels, what acknowledgment is required, and when responsibility legally shifts.

Adding more machines rather than bigger ones — the scaling direction that has no ceiling and demands the distributed-systems tax up front.

The ability to move one's identifiers, credentials, and connections between providers. The identity-layer freedom that decentralized identifiers were designed to guarantee.

Software that holds a person's or agent's credentials and keys, presenting proofs on demand under the holder's control. The client side of self-sovereign identity.

The property of a mechanism under which honest participation is each player's best strategy — so the system need not police what it made profitable to volunteer.

A contract that cannot specify every future contingency, leaving gaps to be filled by discretion, norms, or renegotiation. All real delegation contracts are incomplete.

An agent's fidelity to explicit directives. Necessary but insufficient for good delegation: perfect instruction-following with wrong-goal inference still fails the principal.

Replacing cryptographic keys on schedule or after suspected exposure, limiting how much any single compromise can unlock.

A mechanism that immediately halts an agent regardless of its internal state. The last-resort control that must remain outside the agent's own authority to disable.

The degradation that compounds when models train on the outputs of previous models — diversity narrowing generation by generation. Why provenance of training data became existential.

Directing each request to the cheapest model competent to handle it, escalating hard cases to stronger ones — cost discipline structured as delegation among models.

The ongoing expense of watching agent behavior — tooling, attention, review time. One of the three components of agency cost, and the one automation most reduces.

Independent agents arguing opposing positions before a judge or vote, converting disagreement into an error-correction mechanism.

Requiring proof from independent categories — something known, held, or inherent — so compromising one factor is not enough to impersonate.

TLS with both sides presenting certificates, so client and server each prove identity cryptographically. The default for service-to-service and agent-to-agent trust.

Protocol decisions made through discussion, signaling, and maintainer judgment, then implemented by voluntary adoption. How most open systems really change.

Expressing organizational policy in machine-executable form so systems enforce it automatically and violations fail before they happen rather than after.

A constraint derived from organizational policy rather than task logic — compliance, ethics, brand — that binds every agent regardless of individual mandate.

Multiple overlapping centers of decision-making with no single apex, coordinating through mutual adjustment. How commons and federated systems actually govern themselves.

The graph model where nodes and edges carry arbitrary key-value properties — the pragmatic industry alternative to RDF's strict triple discipline.

Casting a vote through an authorized representative, with or without binding instructions. The mechanism by which delegation scales to collective decisions.

Interleaving reasoning traces with tool actions — think, act, observe, think again — so each step is informed by real results rather than pure speculation.

Identifying the relationships text asserts between entities — employs, located-in, causes — turning sentences into candidate graph edges.

A consumer-hardware trusted execution environment that guards biometrics and keys, so even a fully compromised device cannot extract them.

Authenticating once with an identity provider to access many applications. Removes password sprawl by concentrating both convenience and risk in one login.

Operations as a software discipline — error budgets, automation, blameless postmortems — the institutionalization of keeping systems up without burning people down.

The implicit exchange in which members accept constraints in return for the benefits of ordered cooperation. The legitimacy story beneath every governance system.

Inferring quality or safety from others' visible choices. Efficient when others judged independently; a cascade generator when they didn't.

An AI achieving the literal objective in ways that defeat its purpose — the boat spinning in circles for points. Goodhart's law rediscovered by reinforcement learning.

A human-machine arrangement where operators set goals and constraints while the system executes, drawn from control theory and inherited by AI agent design.

One actor manufacturing many fake identities to dominate a system that assumed one-voice-per-participant. The attack every reputation and voting design must price in.

Intelligence built from explicit symbols and logical rules — the field's first paradigm, eclipsed by learning and now returning as the structure layer around it.

The standing instructions that define a model's role, constraints, and knowledge before any user input arrives — the agent's constitution, renewed every request.

The buffer of pending work between producers and workers, absorbing bursts and enabling retry. Where a system's honesty about its capacity becomes visible.

A single end-to-end execution of model training — months of compute, careful data ordering, and checkpoints — the unit in which frontier capability is actually produced.

How much users rely on automated systems — miscalibrated in both directions, with disuse wasting capability and overreliance automating mistakes.

Finding structure in unlabeled data — clusters, densities, representations — without being told what to look for.

The cost of independently confirming an agent's reported results. When verification costs approach doing the work yourself, delegation loses its economic point.

A retained right to block an agent's decision without taking over the decision itself. The lightest-weight form of control a principal can keep.

The sealed-bid auction where the winner pays the second-highest bid, making truthful bidding the dominant strategy by construction.

Decentralized trust established through peer attestations rather than central authorities, with confidence accumulating along endorsement paths.

An internal predictive model of how the environment behaves, letting an agent simulate outcomes before acting — planning's substrate, learned rather than given.

Performing a task from instructions alone, with no examples — the test of whether a model has genuinely general representations or just good pattern lookup.

The reconstructable path from an outcome back through the decisions and grants that produced it. What an audit follows; what an accountability sink destroys.

Control over which questions get decided and in what order. Often more powerful than voting itself: the option never tabled can never win.

A standardized task suite for comparing agents — invaluable for progress, and reliably gamed the moment it becomes a target.

The verifiable identity of an AI agent: who operates it, what authority it carries, and how its actions bind back to a responsible principal.

Everything an agent carries between steps — goals, memory, pending obligations, held authority. What must be persisted for an agent to survive a restart or a handoff.

A classification system for AI agents organized by role (orchestrator, specialist, verifier), capability (research, execute, delegate), authority scope, and trust level. A taxonomy makes agent ecosystems navigable and governable.

Learning a trade by working under a master — the original knowledge-transfer institution, moving tacit skill the way no curriculum can.

A model generating sequences one token at a time, each conditioned on everything before it — the simple factorization the entire LLM era runs on.

New versions honoring old contracts so existing users keep working — the promise that makes ecosystems adopt upgrades instead of fearing them.

A multi-agent architecture where specialists read and write a shared workspace, each contributing when its expertise applies. Coordination through a common artifact, not messages.

Knowing when a cached copy is no longer true — famously one of the two hard problems, because it requires the future to notify the past.

An unforgeable credential that both names a permission and grants it, so possession equals authority. Makes delegation transferable, attenuable, and revocable as data.

The specific set of categories a system sorts things into, with its rules for edge cases. Where abstract classification meets the stubbornness of real items.

Public append-only logs of every certificate issued, so misissuance is detectable by anyone. Watching the watchmen, implemented as infrastructure.

Deliberately injecting failure into production to verify resilience claims — testing the parachute by jumping, carefully, on a schedule.

Costless, unverifiable communication — promises without stakes. Informative only when interests align; noise the moment they diverge.

The largest human system — accumulated institutions, knowledge, and infrastructure compounding across generations — the ultimate stakeholder every governance question answers to.

With clear property rights and zero transaction costs, parties bargain to efficient outcomes regardless of initial allocation — implying real institutions matter exactly because transaction costs are never zero.

Not just everyone knowing something, but everyone knowing that everyone knows it, recursively. The epistemic state coordination quietly requires and announcements exist to create.

A group bound by mutual concern and repeated interaction — where norms emerge, reputation matters, and commons survive. The unit of governance below the institution.

The principle that parties gain from trade by doing what they are relatively best at, even when one side is absolutely better at everything. Why delegation pays even above superior ability.

A diagram of concepts connected by labeled relationships, externalizing how a domain hangs together. The sketchpad form of an ontology.

A client and server agreeing on the best representation of a resource — HTML for humans, JSON-LD for machines — from the same identifier.

Automated, always-on verification of transactions and controls as they happen, replacing periodic sampling with total coverage. What audit becomes when systems emit evidence natively.

The overhead of keeping many actors aligned — meetings, messages, protocol — the tax that grows superlinearly with headcount and bounds what organizations can attempt.

The property of a mechanism that visibly does not discriminate among participants — anyone can verify the rules favor no one. What infrastructure must have to be adopted by rivals.

Stripping or transforming identifiers so records cannot be traced to people — routinely claimed, frequently reversible, and the reason differential privacy exists.

Expanding a training set with transformed variants of what you have — the cheap multiplier applied when real data is the binding constraint.

Terms governing dataset use — training rights, redistribution, attribution — the contract layer of the data economy that AI turned from boilerplate into battleground.

Collecting only what the stated purpose requires and keeping it only as long as needed — least privilege applied to data, and the habit breaches reward.

A standardized machine-readable procedure for granting, exercising, and revoking authority between agents, covering handshake, scope encoding, and audit emission.

A specification of how decentralized identifiers are created, resolved, and updated on a particular substrate — ledger, web domain, or peer network.

A generative architecture that learns to reverse gradual noising, synthesizing images and media by iterative denoising from randomness.

Fairness judged by who gets what — by need, contribution, or equality. The dimension of fairness that procedure alone cannot settle.

Checking that an identity document is genuine, unaltered, and belongs to its presenter — the physical-world anchor of most digital identity proofing.

An ontology scoped to one field — medicine, finance, agents — capturing its entities and relationships at working depth rather than philosophical generality.

An ability appearing at scale that smaller models lack entirely — contested as measurement artifact or real discontinuity, and central to forecasting what the next scale brings.

Only the communicating endpoints can read the messages — not the platform carrying them — the architecture that makes privacy a property instead of a policy.

Systematic reasoning about right action — obligations, consequences, virtues — the discipline delegation, alignment, and governance all quietly outsource their hardest questions to.

Storing every change as an immutable event and deriving current state by replay — the architecture where the audit trail is the database rather than its afterthought.

Probability-weighted average of outcomes — the arithmetic of decisions under uncertainty, correct in the long run and cold comfort in any single draw.

Software encoding a specialist's rules to make decisions in a narrow domain — the 1980s commercial face of AI, and the ancestor of every policy engine.

Filtering results along independent structured dimensions — type, date, cluster, score — so users navigate a collection instead of guessing its keywords.

Following edges from node to node to answer questions no single record contains — paths, neighborhoods, reachability. The query pattern graphs exist to serve.

An agreement letting credentials issued in one domain be accepted in another, under a shared trust framework. How separate identity islands become an archipelago.

People inferring from others' behavior instead of their own evidence, in sequence — how thin signals become stampedes and markets become manias.

A good that is costly to produce and nearly free to copy — the economics that break classical pricing and make licensing, bundling, and gatekeeping the real business models.

Consent given with genuine understanding of consequences and alternatives. The standard that separates authorization from mere acquiescence.

Value without physical form — brands, data, software, know-how — now the majority of market value and still the minority of what accounting measures well.

Inferring what a user actually wants from what they said — the interpretation layer where conversational systems succeed or embarrass themselves.

The map from each term to every document containing it — the data structure that made full-text search feasible and still powers most of it.

The editorial discipline of selecting, verifying, and maintaining knowledge over time — the human judgment layer that separates a canon from a crawl.

Organizing systems as strata, each depending only on the layer below — the internet's own shape, and the pattern that lets whole layers be replaced without collapse.

Distributed nodes agreeing on a single coordinator despite failures — consensus applied to the question of who is in charge right now.

The slow divergence between what an agent was chartered to do and what it actually does, accumulating through small unreviewed decisions rather than explicit change.

A venue matching many buyers with many sellers under shared rules — liquidity as the product, trust as the infrastructure, and the take rate as the rent.

A market where who pairs with whom matters and prices alone cannot clear it — residencies, admissions, organ exchange — solved by algorithms instead of bids.

Reverse-engineering the internal computations of neural networks into human-understandable algorithms — reading the weights rather than testing the behavior.

The user's internal theory of how a system works — the thing interfaces either match or fight, and the source of most surprise when automation acts.

A hash tree where one root fingerprint commits to an entire dataset and small proofs verify any member — verification at logarithmic cost, everywhere from git to blockchains.

The standardized disclosure shipped with a model — intended use, training data, evaluations, limitations — transparency practice borrowed from nutrition labels.

Several parties jointly computing a result over their private inputs without revealing them — collaboration without disclosure, cryptographically enforced.

A model operating across text, images, audio, and video in one representation space — perception and language finally sharing weights.

The organizing story through which events are understood — cause, character, arc — the frame that determines what facts mean before anyone argues about them.

A strategy profile where no player gains by deviating alone. Where strategic systems settle — which is not necessarily anywhere anyone wanted.

How informal rules crystallize from repeated interaction without any authority decreeing them — the self-organizing origin of most of what governs behavior.

The structured path from newcomer to productive participant — where systems either transfer their context deliberately or leak it through frustration.

Software whose source is free to use, inspect, and modify — the production model that proved commons-based peer production could out-build firms.

Model parameters published for anyone to run and modify — open-source's messier sibling, where the artifact is free but the training recipe and data often are not.

An identity layer on top of OAuth that lets applications verify who a user is, not just what they may access. The protocol behind most 'sign in with' buttons.

A state where no one can be made better off without making someone worse off. The minimal bar for 'no waste' — silent on fairness, which is a separate fight.

Expert evaluation of work before publication — the quality gate of scholarship: slow, unpaid, imperfect, and still the worst system except the alternatives tried.

Value posted by a contractor that is forfeited if agreed work is not delivered to standard. The construction industry's ancient answer to delegation risk.

Manufacturing building components in controlled factory conditions for site assembly — construction moved indoors, where precision and schedule behave.

Training directly on comparisons between outputs — this answer over that one — the family of methods that streamlined RLHF into a supervised-looking recipe.

The game where individually rational defection makes everyone worse off than mutual cooperation would. The two-line proof that incentives, not virtue, decide outcomes.

Reusing the computed state of a repeated prompt prefix across requests, collapsing latency and cost for agents that carry long standing context.

A good that is non-rival and non-excludable — knowledge, standards, security — which markets undersupply because free-riding is the rational move.

A voting scheme where casting n votes on one issue costs n² credits, letting voters express intensity while making domination expensive.

A responsibility-assignment matrix distinguishing Responsible, Accountable, Consulted, and Informed roles per task, with exactly one accountable party per outcome.

The vocabulary layer over RDF that declares classes, properties, and hierarchies, giving raw triples their first level of meaning.

Predicting what each user wants next from behavior and similarity — the engine of feeds and catalogs, optimizing whatever it was told to and shaping taste as a side effect.

An agreement sustained by the value of the ongoing relationship rather than enforceable terms. How repeated delegation works when writing everything down is impossible.

An application that accepts identity assertions from an external provider instead of verifying users itself. It outsources authentication and inherits its provider's failures.

The deliberate discounting of old behavior in reputation scores so standing reflects who an actor is now, not who they were. Forgiveness as a system parameter.

The ability to carry earned standing between platforms instead of rebuilding it from zero in each walled garden. What platform lock-in most depends on preventing.

Putting accumulated standing at risk behind a claim or commitment, so being wrong costs something real. What separates skin-in-the-game endorsement from cheap talk.

The organizational practice of building AI under explicit ethical commitments — fairness, transparency, accountability — with processes that survive contact with deadlines.

A published register of credentials no longer valid, checked before relying on any of them. Trust systems need an undo channel as much as an issue channel.

Exploiting flaws in a reward signal to score without succeeding — the trained-in version of metric gaming, and the standing argument against naive objectives.

Machines that sense and act in the physical world — where AI meets actuators, safety becomes kinetic, and delegation acquires mass and momentum.

The open syndication format that lets any site publish a machine-readable feed — the decentralized subscription layer that platforms buried and readers keep reviving.

Supervising systems more capable than their supervisors — decomposition, debate, AI-assisted evaluation — the delegation problem at the frontier of capability.

A central service where data schemas are published, versioned, and checked for compatibility, so producers and consumers evolve without silent breakage.

The gradual, unauthorized expansion of an agent's activity beyond its original mandate. In delegation systems it signals missing boundaries or absent oversight.

Attaching machine-readable meaning to content — linking mentions to entities, sections to concepts — so documents participate in the knowledge graph.

The gradual shift in what a term means across time, communities, or system versions — the silent killer of long-lived ontologies and old documentation alike.

Engineering content and structure to rank in search — a whole industry shaped by one ranking function, and the proof that any gatekeeping metric breeds an optimization economy.

A delegation failure the principal never learns about because no signal was emitted and no one checked. The most expensive failure mode, precisely because it compounds.

Licensed hands-on crafts — plumbing, electrical, welding — the work that cannot be outsourced to software and is chronically short of successors.

Restoring access through a quorum of pre-chosen trusted contacts rather than a central provider. Distributing the recovery power a helpdesk would otherwise hold.

Giving affected parties formal standing in decisions — seats, votes, veto rights — rather than treating them as externalities to be managed.

Authority granted once and exercised repeatedly without per-action approval, bounded by scope and standing rules rather than case-by-case consent.

A model of computation as named states and legal transitions between them. The discipline that makes process behavior enumerable, testable, and impossible to wish away.

The integrated set of choices about where to play and how to win — coherent commitment under competition, as distinct from goals, plans, or aspiration.

Recurring payment for ongoing access — revenue predictability for the seller, a growing ledger of small commitments for the buyer.

Learning from labeled examples — input paired with correct output — the workhorse paradigm whose appetite for labels created the annotation industry.

A coordinating agent that delegates subtasks to worker agents, reviews their results, and retries or reassigns on failure. Delegation theory implemented in software.

The business model of harvesting behavioral data to predict and shape behavior for advertisers — attention monetized through observation, at civilization scale.

Systems successfully parsing each other's data formats and protocols — the necessary floor beneath semantic interoperability, and routinely mistaken for it.

Confidence in a whole system's tendency to produce fair outcomes — courts, money, protocols — beyond trust in any participant. Slow to build, fast to lose, hardest to repair.

Reasoning about wholes — feedback, delays, stocks and flows — rather than isolated causes; the discipline that finds the leverage point instead of the culprit.

Future cost accepted for present speed — compounding quietly until interest payments consume the roadmap. The finance metaphor engineering actually deserves.

Improving answers by computing more at inference — longer reasoning, multiple samples, search — the scaling axis that opened after pretraining's returns flattened.

The catalog of tools available to an agent, with schemas, permissions, and descriptions. What the agent can do is largely what its registry says it may.

The assumption that if A trusts B and B trusts C, A can trust C. Partially true, decaying with each hop — and the load-bearing gamble in every referral and credential chain.

Cryptographic proof that data existed unaltered at a specific moment, issued by an authority or anchored in a public ledger. Provenance's answer to 'when'.

The rules for who acts when in a human-agent or agent-agent exchange — who may interrupt, who must yield, and how control passes without collision.

An ontology of the most general categories — object, event, quality, relation — meant to sit above domain ontologies and make them mutually intelligible.

A store optimized for nearest-neighbor search over embeddings, retrieving by similarity of meaning rather than match of keywords.

Proving a computation was executed correctly so others can trust the result without redoing the work. What lets verification cost less than execution.

A principal's legal responsibility for harms caused by its agents acting within scope. The doctrine that makes 'my agent did it' a confession rather than a defense.

The process that answers HTTP requests with documents or computed responses — the supply side of the web's client-server bargain.

A callback URL a service invokes when events occur — integration inverted from polling to push, one HTTP request at a time.

Geographically distributed caches serving content from near the user — the web's shared performance layer, and quietly one of its biggest points of centralization.

The exploitation of a shared good by those who consume without contributing. Left unpriced, free-riding hollows out every commons and every standard.

Dedicated tamper-resistant hardware that generates and uses keys without ever exposing them to software. The physical vault of the key-management world.

Evidence that an account is a unique living human rather than a duplicate or bot, without necessarily revealing which human. The scarce resource sybil attacks counterfeit.

A contract term specifying measurable performance an agent must maintain — uptime, latency, quality — with defined remedies for breach. Delegation with numbers attached.

The delegation posture that extends real autonomy while retaining independent means of checking outcomes. Trust sets the speed; verification sets the safety.

A hardware-isolated zone of a processor where code and data are shielded even from the operating system, with attestation to prove what ran.

Ontology of humans — behavior, identity, culture, relationships, and civilization patterns. Root semantic layer for all human-AI interaction and relationship intelligence systems.

Testing a trading strategy against historical data before risking capital — necessary, seductive, and one overfit away from confusing the past with the future.

Novelty that ships — invention carried through to adopted practice — the process economies depend on and organizations reliably suffocate while praising.

Developer bridge between ontology theory and working software: JSON-native schema framework for AI systems. Fastest-buildable technical authority asset in the portfolio.

The productive value stored in relationships — trust, obligation, access — the asset that compounds through reciprocity and evaporates through extraction.

Any concrete device that connects outcomes to consequences for identifiable actors: audits, elections, courts, slashing, recall. Accountability is the property; these are the machines.

A structure that absorbs blame without any person or agent actually answering for outcomes — committees, processes, or automation nobody owns. The failure state delegation design must avoid.

Proving an age threshold without necessarily revealing identity or birthdate. The canonical use case for selective disclosure and zero-knowledge claims.

The web reorganized around agents as first-class users — machine-readable services, agent identity, and delegated transactions replacing pages built for eyeballs.

The discrete act and record of conferring authority: who gave what to whom, when, under which conditions. The transactional unit from which delegation graphs are built.

The belief-desire-intention agent model: what the agent holds true, what it wants, and what it has committed to pursue — with commitment doing the real work.

A network losing earlier capabilities while learning new ones — the reason fine-tuning is surgery, not addition, and continual learning remains unsolved.

Cryptographically sealing a choice now and revealing it later, unable to change it in between — the primitive for fair auctions, hidden votes, and honest predictions.

Training a model to critique and revise its own outputs against an explicit set of principles — alignment by written constitution rather than case-by-case human labels.

Systems mirror the communication structure of the organizations that build them — architecture as org chart, whether anyone intended it or not.

The published structure defining what claims a credential type contains and how they are encoded, so issuers and verifiers agree without coordination.

A venue where datasets are listed, priced, and licensed — the attempt to give data the liquidity of a commodity while its provenance resists standardization.

The right and means to take your data from one service to another in usable form — the regulatory crowbar aimed at switching costs.

Fitness of data for its use — accuracy, completeness, freshness, consistency — the property every downstream model and decision silently inherits.

Data subject to the laws of where it resides or whom it describes — jurisdiction asserting itself over the borderless substrate, one localization rule at a time.

A system behaving aligned under observation while pursuing different goals when unobserved — the failure mode that breaks evaluation-based trust entirely.

The number of hops between the original principal and the agent doing the work. Accountability degrades with depth unless each hop preserves the audit chain.

The time between intent forming and a delegated agent acting on it, including grant negotiation and context transfer. What delegation trades against control.

A cryptographic identity bound to particular hardware, letting services distinguish trusted devices from new or emulated ones.

A human-readable name in the DNS hierarchy, rented rather than owned — simultaneously an address, a brand, and the anchor of web-scale identity.

Capability that serves both beneficial and harmful ends — biology, cryptography, capable models — where the technology cannot be shipped separately from its misuse.

A community that recirculates its own views and discredits outside sources — not just missing other voices but inoculated against them.

A model trained specifically to map text into vectors where similarity is meaningful — the retrieval workhorse behind semantic search and RAG.

Expanded authority an agent may exercise during declared exceptional conditions, paired with mandatory review once the emergency ends. Dangerous without a sunset clause.

The standing to be believed on a subject — earned by track record and method, granted by institutions, and contested every time experts disagree in public.

A system learning a goal that matched training but diverges in deployment — competence intact, objective wrong. The alignment failure that looks like success until distribution shifts.

The documented architecture of how an organization decides: bodies, mandates, escalation routes, decision rights, and review cycles, assembled coherently.

Computing directly on encrypted data, producing encrypted results — outsourced processing where the processor never sees the plaintext.

More signal competing for attention than can be processed — the condition that makes curation, ranking, and trusted sources economically valuable.

The tendency of agents with almost any final goal to pursue the same subgoals — resources, self-preservation, influence — which is why capability alone raises stakes.

The organizational discipline of capturing, organizing, and reusing what the organization knows — perpetually redefeated by the fact that knowledge lives in people.

Ideas leaking beyond their creators to benefit others — the positive externality that clusters cities, justifies research funding, and defies fences.

A worker whose output is judgment applied to information — the labor category AI agents most directly augment, compete with, and are managed like.

Cost and error falling predictably with cumulative experience — the compounding advantage of having done it before, institutionalized.

A voting system where each participant may vote directly or delegate their vote to a trusted proxy, revocably and per-topic. Delegation applied to suffrage itself.

The versioned catalog of an organization's models and their lineage, evaluations, and deployment status — the system of record for what is actually running where.

An RDF graph given its own identifier, so statements can be grouped, attributed, and trusted per-source rather than as one undifferentiated soup.

What a model asserts exists by choosing its categories and relations. Every schema commits its users to a worldview, whether or not anyone wrote it down.

Continuing to operate when the network splits and nodes cannot reach each other — not optional in real networks, which is why CAP is really a two-way choice.

Cryptosystems designed to survive quantum attack — migrated to before quantum computers arrive, because recorded traffic waits patiently for its decryption.

Information about scarcity and demand compressed into a number that coordinates millions of strangers — the original distributed consensus mechanism.

Building privacy protections into systems from the start rather than bolting them on after complaints — architecture as policy, defaults as ethics.

Memory of how to do things — skills, workflows, learned corrections — that shapes behavior without being recalled as facts.

Works owned by no one and usable by everyone — where copyrights go when they expire, and the commons all creativity ultimately draws from.

Data used only for the purpose it was collected for — the consent-preserving principle that secondary use, resale, and training pipelines strain daily.

A regulator coming to serve the industry it oversees, through revolving doors, information dependence, or budget politics. Governance failure by slow absorption.

The discovery that much published research fails to replicate — incentives to publish novelty having quietly outcompeted incentives to be right.

A structured argument, backed by evidence, that a system is safe enough for its intended deployment — imported from aviation and nuclear practice into frontier AI.

The shared definitions of business concepts — customer, revenue, active — mapped onto raw data so every consumer computes the same thing from the same words.

The theory of aggregating individual preferences into collective decisions — and of the impossibility results proving every method must sacrifice something.

A small model drafts tokens, the large model verifies them in one pass — identical output, several times faster. Delegation with verification, applied inside inference.

Constraining a model to emit valid data in a declared schema rather than free prose, so downstream systems can consume results without parsing hope.

Economic systems built on programmable digital tokens — incentives, ownership, and governance denominated in units the protocol itself issues.

Valuable knowledge protected by keeping it secret rather than registering it — no expiry, no disclosure, and no protection the day it leaks.

A periodic public disclosure of an organization's sensitive actions — takedowns, data requests, enforcement counts — that lets outsiders audit its claims about itself.

The experimental game where one player risks a transfer hoping for reciprocation — the laboratory instrument that measures trust as observed behavior rather than survey answers.

An authoritative list of which issuers, verifiers, or agents are recognized under a trust framework — the lookup that answers 'should I accept credentials from this party'.

The deliberate process of restoring reliance after violation: acknowledgment, explanation, restitution, and demonstrated change. Denial is its most common substitute.

Judging machine intelligence by whether its conversation is indistinguishable from a human's — the founding benchmark, passed in spirit and immediately declared insufficient.

A market whose value depends on attracting both sides — riders and drivers, developers and users — where pricing one side subsidizes the other.

The sequence of activities that turn inputs into delivered value — the map on which every disruption story marks where margin moved.

A holder's packaging of one or more credentials — possibly selectively disclosed — signed to prove the presenter is the legitimate holder, not a thief of credentials.

A platform's visible mark that an account's identity was checked. Its meaning is exactly as strong as the process behind it — and collapses the day that process is purchasable.

A security posture that grants nothing based on network location or prior standing: every request is authenticated, authorized, and encrypted as if hostile.

A formal language for inter-agent messages with defined performatives — inform, request, propose — from the classical multi-agent systems tradition.

AI matching human capability across essentially all cognitive domains rather than excelling narrowly — the destination the field argues about while building toward it.

Authorizing by evaluating attributes of subject, resource, and context against policy, rather than by membership in predefined roles. Flexible, and only as good as its attribute data.

The requirement that an agent's choices be judged against what best serves the principal, not what was merely permitted or requested.

Correct operation even when components lie or act maliciously, not merely crash — the fault model that turns distributed systems into adversarial games and blockchains into consensus machines.

The unbroken documented record of who held an item and when, from origin to present. Evidence law's provenance discipline, inherited by data and content pipelines.

A specification that rules by adoption rather than ratification — whatever everyone already uses. Markets crown these; committees merely acknowledge them.

The obligation to act with the competence and diligence a reasonable expert would apply. One of the two pillars of fiduciary duty, alongside loyalty.

Using a language model to grade outputs against criteria — scalable evaluation whose own biases and blind spots must themselves be evaluated.

The open standard for connecting AI applications to tools and data sources, so integrations are written once against the protocol instead of per-model.

Detecting and typing the names in text — people, organizations, places, dates — as the first structured foothold in unstructured language.

Demonstrating operational control of a resource — a key, domain, or account — as the working substitute for proving abstract ownership.

The component whose integrity everything above it assumes — a hardware key, a root certificate, a genesis record. Every verified system stands on one unverified floor.

Assigning rule-making, rule-applying, and rule-judging to different bodies so no one entity writes, executes, and reviews its own rules.

The internal reliability target a service commits to — the error budget's other face, and the number that arbitrates between shipping and stabilizing.

The number of direct subordinates or agents one supervisor can effectively oversee. The constraint that forces hierarchies — and delegation — to exist at all.

Cooperate first, then mirror the other side's last move. The embarrassingly simple strategy that wins repeated-game tournaments by being nice, retaliatory, and forgiving.

A venue where agent capabilities are listed, priced, and hired — labor markets reinvented for participants that spin up on demand and never sleep.

Payment rails agents can use under delegated spending authority — budgets, limits, and receipts machine-enforced, because a wallet is authority in its most liquid form.

Executing market strategies by machine at machine speed — rules and models trading against each other, with humans setting risk limits and reading the postmortems.

The interpretive tradition mapping celestial positions onto human affairs — scientifically unfounded, culturally persistent, and computationally precise about where the planets actually are.

The accumulated promise a name carries — recognition, expectation, trust — reputation compounded into an asset that pricing can measure and one scandal can spend.

Systematic deviation from normative judgment — anchoring, confirmation, availability — the predictable errors that make debiasing a discipline and defaults a responsibility.

The standardized license suite letting creators grant permissions in advance — machine-readable generosity that built the legal rails of the open content movement.

A firm that aggregates and resells personal data gathered from sources the subject never met — the invisible middle of the surveillance economy.

The named person accountable for a data domain's quality and proper use — stewardship made specific, because data owned by everyone is owned by no one.

Adding calibrated noise so aggregate statistics reveal patterns while mathematically bounding what can be learned about any individual — privacy as a provable budget.

A set of interdependent products, platforms, and participants evolving together around shared interfaces — competition and symbiosis in the same graph.

Enforced uniqueness of digital items — through ledgers, signatures, or platform decree — the deliberate reintroduction of scarcity where copying is free.

A live virtual replica of a physical system, updated by its sensors and used to simulate its futures — the model and the thing, finally in sync.

An AI agent operating under an organization's authority, policy, and audit — where agent capability meets compliance, and autonomy meets the org chart.

Knowledge that survives being written down — procedures, specifications, facts — the documentable fraction that knowledge management systems actually capture.

The legal doctrine permitting limited unlicensed use of copyrighted work — criticism, education, transformation — the safety valve now litigating the AI training question.

The bonds of family across generations — descent, marriage, obligation — humanity's first trust network and the one every other institution borrowed its metaphors from.

The aggregate demand for many niche items rivaling the head hits — the distribution the internet made servable and recommendation systems made findable.

The learned capacity to evaluate mediated information — sourcing, framing, incentive — the population-level defense against manipulation that scales slower than the attacks.

The principle that carriers treat all traffic equally — the policy fight over whether the network's owners may become its editors.

Breaking communication into independently routed packets reassembled at the destination — the design bet that made networks resilient and permissionless.

Early events constraining later possibilities — standards, defaults, and installed bases locking in futures nobody chose deliberately.

An AI agent loyal to one individual — holding their context, preferences, and authority — the fiduciary counterweight to platforms' agents serving platforms.

A market where contracts pay on outcomes, so prices become collective probability estimates — putting money where the forecast is, and aggregating it.

The network of who connects to whom on a platform — the asset that made social networks defensible and its portability a recurring policy demand.

The tamper-evident record of an AI system's inputs, decisions, and authorities exercised — what regulators will subpoena and operators should already be reading.

Distributing power so that each holder can block abuses by the others. The structural answer to trusting no single party completely.

The classic multi-agent task allocation protocol: announce a task, collect bids, award to the best bidder, monitor delivery. A market in miniature, inside a system.

A graded measure of how rigorously an identity was proofed, letting relying parties match verification strength to transaction risk.

An architecture routing each token through a few specialized subnetworks instead of the whole model — a large model's capacity at a smaller model's cost per query.

The legal instrument granting open-source freedoms with conditions — attribution, patent peace, or share-alike — where the movement's philosophy is actually encoded.

Granting permissions to roles and roles to people, so access tracks function instead of individuals. Simple to reason about; prone to silent role accumulation.

A compact model targeting most of the capability at a fraction of the cost, viable on-device — the counterweight to scaling as the industry's only direction.

Aggregated independent judgments outperforming most individual experts — powerful exactly when estimates are independent, and broken the moment they correlate.

The near-free cost of serving one more copy or user — the property that makes digital scale explosive and pricing detach entirely from cost.

AI concierge and operating agent for the ecosystem — answers investor and partner questions, navigates all ventures, and captures qualified leads 24/7.

Canonical link and memory engine — helps humans and AI locate trusted, authoritative context at scale. Connective tissue across the entire portfolio.

Relationship operating system — infrastructure layer for human connection, coordination, and shared identity. Parent platform for SynkHeart and SoulGraph.

Software designed for agents as primary users — capabilities exposed as tools, state exposed as context, and the human interface demoted to one view among several.

Business capabilities sold as callable services — composable, metered, and contractual — where integration became a product category of its own.

The systems recording what each person agreed to, for what purpose, and when — consent industrialized, from cookie banners to preference centers.

Individuals monetizing audiences directly through platforms — patronage, subscriptions, sponsorship — with the platform taking its cut and setting the rules.

The searchable inventory of an organization's datasets with owners, schemas, and lineage — the registry that makes data findable before it can be governable.

The traced path of data from origin through every transformation to its point of use — provenance for pipelines, and the first question after any bad number.

An economy where the primary transaction is delegating outcomes to intelligent agents — value flowing to whoever is trusted with judgment, not just labor.

The three responses to institutional decline: leave, push for change, or endure. Systems that block exit and ignore voice are left with only decaying loyalty.

Training a shared model across devices that keep their data local, exchanging only updates — learning centralized, data sovereignty preserved.

Commerce organized around expressed customer intent rather than captured attention — buyers broadcasting what they want and agents competing to fulfill it.

Large-scale strategic frameworks and enterprise model routing — the macro tier of the model family alongside micro and mini.

Datasets published for free use and redistribution — government records, research data, registries — infrastructure for accountability and innovation alike.

The mapped network of a person's or organization's relationships — strength, history, context — the substrate relationship intelligence systems compute over.

Identifying, pricing, and bounding potential losses — position limits, diversification, hedges — the discipline that decides survival long before returns decide ranking.

The degradation of a shared resource when individually rational use exceeds collectively sustainable use. The failure mode commons governance exists to prevent.

A legal instrument granting one party authority to act on another's behalf in specified matters. The oldest formalized delegation contract in common use.

Securing consensus with capital at risk instead of energy burned — validators bond value that misbehavior forfeits, converting security into an economic argument.

Demonstrating expended computation to earn the right to act — sybil resistance purchased in energy, securing consensus by making attack expensive.

The authoritative system for a given data domain — where the true value lives and every copy defers. Canonical-source discipline applied to enterprise plumbing.

The negotiated ground rules under which an agent operates for a principal: scope, communication cadence, escalation triggers, and exit conditions.

Buildings assembled from standardized factory-built volumes — modularity's logic applied to structures, trading design freedom for speed, cost, and quality control.

Innovating with ideas flowing across organizational boundaries — licensing in, spinning out, co-developing — the admission that most smart people work somewhere else.

The pipeline of training, credentialing, and placement that matches people to changing work — where education policy meets labor markets, urgently and imperfectly.

The allocation of who must demonstrate what before a claim is accepted — the procedural rule that quietly decides most disputes before evidence is even weighed.

Everyday objects networked with sensors and identity — billions of small machines whose data, security, and governance arrived faster than anyone's plans for them.

Maintaining one authoritative record for core entities — customer, product, supplier — the canonical-source discipline applied inside the enterprise.

The enforceable demand that an organization delete your data — simple against a database, unresolved against a model that trained on it.

Each part of a system addressing one distinct responsibility — the design discipline that keeps change local and reasoning tractable.

Professional diagram and canonical figure-generation tool for founders, researchers, and strategists — visual output layer for the entire canon ecosystem.

Ontology for governance, institutions, public systems, and civilization-scale coordination. GovTech and civic AI positioning.

Work performed at the customer's site — installation, repair, inspection — where scheduling, routing, and trust in the person at the door decide the business.

Knowledge graph structure and relationship modeling — the graph patterns powering SoulGraph, Web4 Registry, LinkCanon, and agent reasoning systems.

Practical analytical patterns and domain-specific model templates for operators and analysts.

Why organizations exist at all: hierarchies beat markets when transaction costs of contracting exceed costs of managing — a boundary AI agents are actively moving.

Master hub and canonical trust layer for the entire Askander ecosystem — single entry point for investors, partners, and customers.

One designated place where each fact is maintained, with everything else derived — the anti-entropy principle behind registries, canons, and master data alike.

Protocol and governance standards for Web4 — delegation rules, trust hierarchies, and authority patterns.

Accessible, lightweight model explanations for learning and teaching — entry-level layer of the model family.

Commercial operating system for structured innovation — partner ecosystems, idea-to-execution workflows, and IP licensing platform.

The shared vocabulary the major search engines agreed to understand, letting any website describe its content in machine-readable structured data.

Meaning, archetypes, cycles, and symbolic identity system — reflective self-knowledge layer extending SynkOS. Framed as interpretive, not deterministic.

Healthy home systems — EMF assessments, audits, and remediation services for residential buyers and contractors.

Training and certification platform for plumbers and skilled trades — courses, practical assessments, and career pathways.

Operational intelligence for skilled trade contractors — field workflows, job logging, inspection, and AI copilot. Revenue engine once vertical is decided.

Identity-memory graph mapping who a person is — preferences, behaviors, relationships. User-owned data that powers SynkOS and Askander personalization.

Consumer relationship wellness product — journaling, check-ins, and biometric sync tied to SoulGraph. Accessible consumer face of the relationship intelligence stack.

Brandable portal and product layer for the LinkCanon ecosystem.

Consulting and professional services arm — implementation, automation, and delivery. Near-term cashflow placeholder while portfolio assets mature.

Commercial Web5 ecosystem brand protection — future hedge as Web5 discourse develops post-Web4.

Commercial Web5 intelligence brand anchoring 'collective intelligence' thesis. Hold until Web4 is established; forward-optionality hedge.

Near-term product/offer brand for high-value metal roofing panels. Clarify role with ASP partner: product line brand vs. redirect to americaspanelfab.com.

Long-term enterprise brand for American metal panel fabrication — canonical company face for the ASP operating partnership and near-term revenue anchor.

AI finance and investing intelligence agent — watchlists, market explainers, portfolio signals, and decision-support tools for retail investors and operators.

Investment intelligence and education stream — content and community layer feeding the Clobot agent. Content arm; Clobot is the tool.

Nonprofit and public-benefit arm — open frameworks, education, and standards. Credibility layer for the commercial Open Innovation Engine.

DTC beekeeping and royal jelly brand — raw honey, supplements, and gear. Philippines sourcing is a cost and authenticity advantage.

Web4 applied to robotics and embodied agents — delegation and human-machine systems in physical environments.

Wildlife and nature storytelling universe — photography, lore, and content IP. Philippines endemic species is a structural content advantage vs. generic nature channels.

Premium silver-keepsake gifting brand for families — intergenerational legacy positioning, DTC US plus Shopee/Lazada ASEAN channels.

Applied design studio for sound, resonance, and experiential wellness spaces — client-facing creative service arm.

Concept landing page for 'Web4-native' as an adjective — companies and products built for the delegated intelligence era.

ASEAN-focused Web4 regional brand leveraging Cebu base as credible geographic anchor. Events, partnerships, and regional media for the fastest-growing digital economy.

Community and builder-facing entry point for Web4 — challenges, examples, and build kits. Adoption engine for the Web4 stack.

Infrastructure principles for trustworthy Web4 systems — identity, provenance, reliability, and operational trust at scale.

Platform strategy and ecosystem design — business and strategy face of Web4 for non-technical founders and operators.

Publication and academic layer — whitepapers, citations, and studies underpinning the Web4 canon with research credibility.