Concepts

The vocabulary of AI selection. These terms are used consistently throughout the Deeprank specification and should be understood before reading other documentation.

Deeprank Profile

Canonical Definition

Definition

A Deeprank Profile (formal spec name: Deeprank Selection Profile, DSP) is the atomic unit of AI-readable business identity. It is a structured declaration containing six mandatory layers: Identity, Problem Declaration, Capability, Fit Conditions, Non-Fit/Exclusions, and Stability/Confidence.

The Deeprank Profile is designed to be machine-parseable while remaining human-readable. It provides AI systems with the structured information needed to make accurate selection decisions.

Selection

Canonical Definition

Definition

Selection is the act of an AI system choosing a business as the answer to a user query. Selection is binary: a business is either selected or not. There is no partial selection or ranked position.

When It Applies

When It Applies

Selection occurs when a user asks an AI for a recommendation, solution, or referral. The AI evaluates available options against the user's stated intent and constraints, then selects the option(s) that fit.

When It Does Not Apply

When It Does Not Apply

Selection does not apply to search queries that expect lists, exploratory browsing, or cases where the AI provides information rather than recommendations.

Exclusion

Canonical Definition

Definition

Exclusion is the act of an AI system removing a business from consideration for a query. Exclusion occurs when a business does not fit the stated constraints or when the business has declared that it does not serve the query's context.

When It Applies

Correct Exclusion

A correctly excluded business is one that genuinely does not fit the query. This is a positive outcome: the user does not receive a mismatched recommendation, and the business does not receive unqualified leads.

Common Failure Mode

Incorrect Exclusion

Incorrect exclusion happens when a business is filtered out despite being a good fit. This typically occurs due to missing or ambiguous information in the business's profile.

Intent

Canonical Definition

Definition

Intent is what the user wants to accomplish. It is the core problem or need driving the query. Intent is typically expressed as a verb + object: "find a lawyer," "fix my plumbing," "design a logo."

When It Applies

Intent Mapping

AI systems map user intent to business problem classes. A Deeprank Profile declares which problem classes a business addresses, enabling direct matching between user intent and business capability.

Constraint

Canonical Definition

Definition

A constraint is a condition that must be satisfied for selection to occur. Constraints can be explicit (stated by the user) or implicit (derived from context). Common constraint types include geographic location, budget, timeline, and specific requirements.

When It Applies

Hard vs Soft Constraints

Hard constraints must be satisfied for selection. Soft constraints are preferences that influence selection when multiple options fit. AI systems treat unqualified constraints as hard constraints by default.

Common Failure Mode

Constraint Failure

A single unmet hard constraint causes exclusion regardless of how well other criteria are satisfied. Businesses must clearly declare which constraints they can meet.

Fit

Canonical Definition

Definition

Fit is the determination that a business satisfies the intent and all hard constraints of a query. Fit is binary: a business either fits or does not fit. There is no degree of fit in AI selection logic.

When It Applies

Evaluating Fit

AI systems evaluate fit by checking: (1) Does the business's problem class match the user's intent? (2) Can the business satisfy all stated constraints? (3) Has the business declared any exclusions that apply to this query?

Problem Class

Canonical Definition

Definition

A problem class is a category of problems that a business addresses. Problem classes are standardized labels that enable intent matching. Examples: "legal-immigration," "plumbing-emergency," "accounting-tax-business."

When It Applies

Specificity

Problem classes should be as specific as the business's actual scope. "Legal" is too broad. "Legal-immigration-employment-based" is appropriately specific for an attorney who specializes in employment-based immigration.

Common Failure Mode

Over-Broad Classes

Declaring overly broad problem classes leads to selection for work the business cannot handle well. An accountant who declares "accounting" may be selected for forensic accounting work they are not qualified for.

Capability

Canonical Definition

Definition

Capability is the set of specific things a business can do to solve problems within its problem class. Capabilities are declared as labels and methods. Labels are standardized terms. Methods are specific approaches or tools the business uses.

When It Applies

Capability Declaration

Example: A marketing agency might declare capability labels = ["content-marketing," "paid-advertising"] and methods = ["SEO optimization," "Google Ads management," "social media content creation"].

Negative Capability

Canonical Definition

Definition

Negative capability is the explicit declaration of what a business does not do, cannot do, or will not do. This includes service exclusions, customer type exclusions, and situational exclusions.

When It Applies

Protective Function

Negative capability protects both the business and the user. The business avoids selection for mismatched work. The user avoids contacting businesses that cannot help them.

Common Failure Mode

Missing Exclusions

Without explicit negative capability, AI systems may infer capabilities the business does not have. A web developer who does not declare "no mobile app development" may be selected for mobile app projects.

Confidence

Canonical Definition

Definition

Confidence is the AI system's certainty that a selection is correct. Confidence is influenced by information quality, source consistency, and recency of verification. Higher confidence selections are more likely to be presented to users.

When It Applies

Confidence Factors

Factors that increase confidence: explicit declarations, consistent information across sources, recent verification, complete profiles. Factors that decrease confidence: inferred information, inconsistent sources, outdated data, incomplete profiles.

Common Failure Mode

Low Confidence Outcome

Low confidence may result in non-selection even when a business technically fits. AI systems prefer high-confidence selections and may skip ambiguous options in favor of clearly documented ones.

Related Documentation