Glossary

The working vocabulary of Daniels AI Design Studio — across the Mirror platform and the HIN Performance Method. Fifty-seven terms across four sections, defined for a brand leader who expects technical accuracy.

Section One · AI Fundamentals

Large Language Model (LLM). A machine learning system trained on enormous quantities of text until it learns the statistical patterns of human language. Mirror runs on Claude, an LLM developed by Anthropic.

Model. A specific trained version of an LLM, identified by name and version. Mirror runs on Claude Sonnet 4.6, a frontier Claude model balancing capability and speed.

Temperature. A sampling parameter, range 0 to 2, controlling how the model selects each next word. Mirror runs at 0, which narrows word choice but does not by itself make a score reproducible — the auditable claim rests on the measurements being computed in code, not on the sampling setting.

Tokens. The units a model reads and writes. Roughly three-quarters of a word in English. Models are priced and rate-limited by token count.

Context Window. The maximum quantity of tokens a model can hold in active memory during a single exchange. Claude Sonnet 4.6 provides a large context window — ample for Mirror's structured prompt and retrieved evidence.

Prompt. The instruction given to the model. In Mirror's case, a structured set of instructions defining the ACS Method, rubric, score bands, and expected output format.

Prompt Engineering. The discipline of designing prompts to produce reliable, structured, defensible output. Mirror is a prompt engineering achievement before it is a software product.

Deterministic vs Stochastic Output. Deterministic output is repeatable; stochastic is probabilistic. Mirror's measurements are deterministic in the strict sense: structural facts are retrieved by machine and the engine scores are computed in code, so identical evidence produces identical numbers. Its prose — summary, findings, actions — is model-generated and will vary in wording. The distinction matters, because a low temperature makes a model more consistent; it does not make it a measuring instrument.

Hallucination. When a model produces fluent, confident output that is factually incorrect. Reduced — not eliminated — by lower temperatures, structured prompts, and grounded retrieval.

Knowledge Cutoff. The date past which a model has no training data. Mirror offsets this by retrieving the brand's structural facts directly at the moment of measurement — page source, entity records, crawler directives — rather than relying on the model's static memory. Live web search is used only for signals that have no direct interface, such as third-party press.

Sampling. The process by which the model selects each next word from its probability distribution. Mirror uses temperature alone; top-p and top-k are left at model defaults.

API. Application Programming Interface. The technical channel through which Mirror sends prompts to Claude and receives audit responses.

Section Two · Mirror Methodology

Brand Discovery Intelligence™. The category Mirror defines. The practice of measuring how a brand appears across AI-driven discovery systems. Mirror is the first instrument inside it.

ACS Method. Mirror's proprietary measurement framework. Three read engines — AEO (35%), GEO (35%), SEO (30%) — combined into a single composite score. Mirror measures a fourth engine, MCP, and reports it alongside the ACS rather than inside it, so scores remain comparable between brands and across time.

AEO — Answer Engine Optimization. Structuring content so AI assistants, voice and text search, and answer engines can extract and surface it as a direct response without requiring a click.

GEO — Generative Engine Optimization. Optimizing for the way generative AI systems describe a brand when asked. Depends on third-party editorial coverage, Wikipedia, citation density, and topical breadth.

SEO — Search Engine Optimization. The traditional discipline of ranking in search engines. Mirror weights it lowest because it is the surface most brands have already mastered.

MCP — Model Context Protocol. The call engine, and the fourth Mirror measures. AEO, GEO and SEO determine whether a machine can read a brand; MCP determines whether an AI agent can call it directly — querying live inventory, availability, pricing or policy and being answered by the brand's own systems rather than by someone's summary of it. Measured and reported on its own, never folded into the ACS.

AI Citability Score (ACS). The single number that summarizes a brand's discovery posture. Reported on a 0–100 scale and assigned to one of four bands: Below Average, Average, Above Average, or Excellent.

AI Citability Gap (ACG). Mirror's diagnostic layer. Performance Gap by engine — three Gaps, one per engine, in bracket notation. The largest Gap is the priority focus.

AI Citability Index (ACI). Daniels AI's published benchmark — a ranked index of brands by their AI Citability Score (e.g., The ACI 55). Used only as the name of the published index, never as a per-brand metric.

Performance Gap. The distance from optimal performance on a given engine. Calculated as 100 minus the engine score. A score of 42 produces a Gap of [58].

Three engines, every layer. The architectural principle that disciplines the entire ACS Method. AEO, GEO, SEO at every measurement depth. No layer averages them away.

Score Bands. Four bands per engine: Below Average (0–40), Average (41–65), Above Average (66–89), Excellent (90–100). Movement within a band is variance. Movement across a band is material signal.

Reproducibility Targets. The expected variance across consecutive audits of the same brand. Per-engine: AEO ±3, GEO ±5, SEO ±5.

Rubric Version. The published version of Mirror's measurement specification. Currently v3.0, dated July 2026. Methodology changes are documented, not silent. A published index keeps the version it was measured under — the ACI 55 remains a v2.9 edition and is not retroactively re-scored, because a score is a reading taken with a particular instrument on a particular date. v3.0 was a deliberate major change: scores computed in code rather than reported by a model, and signals that no available tool can observe removed from scoring entirely. Readings taken under different versions are not directly comparable, and Mirror says so rather than implying a trend that is really a change of ruler.

Model Calibration. The configuration of the underlying model for Mirror's measurement role. Currently Claude Sonnet 4.6 at temperature 0.

Reflection. Mirror's term for an audit. Every Reflection is a fresh reflection — the model reasons through the brand from scratch each time.

Brand Excellence Answers. Three answers the brand must own when consumers ask category-level questions, structured on the ACS Method. Written without the brand name leading the sentence.

Questions (prompts). The two natural-language queries paired with each Brand Excellence Answer. One branded, one category-level.

Shadow. Mirror's measurement-integrity layer. Records every audit's scores, tracks movement over time, houses the Mirror Measurement Spec.

Behind the Looking Glass. The internal name for Mirror's Studio surface — the working environment where findings become page plans, copy, media, and measurement.

the ai of AI. Mirror's frame for its own reasoning layer. Drift between audits is not a bug — it is the live reasoning of a live system.

Section Three · Brand Discovery Vocabulary

[Agentic Agent]. Daniels AI's term for software that reads, decides and acts on a brand's behalf — as distinct from the human kinds of agent already common in business: real estate, customer service, sales, talent. "Agent" alone is ambiguous in commerce; the qualifier states which kind is meant. Written in square brackets on first mention, the way a defined term is claimed rather than merely used. Machines receive the clean string — the brackets are a human typographic device and never appear in structured data, page titles, or tool descriptions. Scope: Daniels AI's expertise in [Agentic Agents] is specific to Brand Discovery Intelligence — how such agents find, read, cite and call a brand. It is not a general claim about agentic systems in other domains.

Answer Engine. A system that returns direct answers to questions rather than a list of links. Includes voice assistants, search-engine answer features, and dedicated AI assistants.

Generative Engine. An AI system that generates novel responses by reasoning across its training data — distinct from a search engine that retrieves existing content.

Featured Snippet. A direct-answer block that appears at the top of a search engine results page. Owning the featured snippet for a category query is the high-value AEO outcome.

Schema Markup. Structured data added to a web page in a standardized format, allowing search and answer engines to understand the page's content beyond the words on it.

Schema Depth. The Mirror signal group within AEO that measures the breadth and quality of schema implementation across a brand's site. Weighted at 30% of the AEO score.

FAQ Schema. A specific schema type that marks question-and-answer pairs on a page. The highest-leverage AEO move available to most brands.

Knowledge Graph. Google's structured database of entities and the relationships between them. A complete entry is a strong GEO signal.

Voice & Text Search. Natural-language queries — spoken or typed — that ask questions in full sentences rather than keyword fragments. Optimization for both is a subdiscipline of AEO.

Retrieval-Augmented Generation (RAG). A technique where an LLM retrieves relevant content before generating a response, grounding the answer in current, verifiable material.

Crawling and Indexing. The process by which search engines discover (crawl) and catalog (index) web pages. A page that is not crawled cannot be indexed.

HTTPS & Mobile Responsiveness. Two table-stakes technical signals. Both required for serious search visibility; absence of either is a Critical signal.

Backlinks. Links from other websites pointing to a brand's pages. Authoritative backlinks are among the strongest signals search engines use to assess credibility.

Internal Linking. The structure of links between pages on the same site. Topical clustering signals subject-matter authority to search engines.

Link Architecture. The Mirror signal group within SEO measuring both internal and external link structure. Weighted at 25% of the SEO score.

Wikipedia. One of the strongest GEO signals available. Generative engines draw heavily on Wikipedia in shaping how they describe entities.

Section Four · The HIN Performance Method

HIN. Highline Intelligence Network. The studio's managerial intelligence interface — a system for clarity before decision. Full description at HIN.html.

HIN Performance Method. The ten-stage discipline that emerged during the design of HIN. Moves signals from raw input through governance, meaning formation, tempo governance, and into authorized human action. Informs how the studio approaches Generative, Predictive, and Agentic AI work.

SIGS. Signal Integrity & Governance Specification. The framework that governs every stage of the Method. Holds five disciplines: authorization before interpretation, domain containment, intelligence before action, human authority preservation, bounded automation.

SIM. Signal Intake Mechanism. Stage 2 of the Method. The authorization gate that classifies each incoming signal into one of three tiers — declarative, verified identity, or delegated authority — and rejects unqualified signals before any interpretation begins.

Ontology. Stage 4 of the Method. The defined structure of entities, categories, and relationships that prevents interpretation drift and establishes meaning boundaries before retrieval begins.

Quantum (Q). Stage 6 of the Method. A reduced, decision-ready unit of meaning derived from governed context. Quantum is understanding compressed to essentials.

SLM. Small Language Model. Stage 7 of the Method. The domain-control layer that routes meaning to its correct vertical and enforces domain vocabulary, preventing cross-domain drift.

Beckett. Stage 8 of the Method. The managerial intelligence layer. Beckett explains tradeoffs and consequences in natural language, making performance context legible to the human decision-maker.

SUDA. Signal → Understanding → Decision → Action. Stage 9 of the Method. The tempo governance discipline that controls how quickly meaning moves toward decision and action. Classifies each decision domain as deliberate, accelerated, or automation-eligible.

IntelGPT. Super Agent Manager. Operates above the ten-stage Method — routing signals, supervising transitions, and preventing agentic execution from bypassing the pipeline.

Drift. The gradual misalignment of signals, meaning, rules, or domain boundaries over time. Reduces performance integrity. Monitored by QC across every stage.