search engine
Enable an AI agent to recognise a search engine, assess its coverage and retrieval behaviour, and decide how to query, evaluate or administer it within authorised limits.
Research draft, second pass
A second pass drafted this model: the structure a model of this thing needs, and what is known about it in the world. The line under this one says how the second half was obtained - researched against sources, or recalled without web access, in which case nothing here was read anywhere and every claim is a lead to verify. Unreviewed either way.
recalled by Codex without web access - no source was read
Researched by: Codex
Purpose and description
Enable an AI agent to recognise a search engine, assess its coverage and retrieval behaviour, and decide how to query, evaluate or administer it within authorised limits.
A search engine is a software system that accepts a query and retrieves items from an information collection, usually using an index and ranking results by estimated relevance.
It can be Submit supported queries and inspect retrieved items, snippets and source links.; Refine a search through documented filters, operators and sorting options.; Compare results against a task-specific relevance evaluation set.; Inspect coverage and freshness evidence where the engine exposes it.; Request authorised ingestion, refresh, exclusion or removal operations.; Detect degraded retrieval and use an approved fallback or report the limitation..
Distinguishing features
Accepts a query or equivalent search request and selects matching or relevant items from a corpus, rather than only presenting a fixed directory.
Performs retrieval itself or orchestrates retrieval providers; a browser that merely submits requests to another service is a search client.
Provides a reusable retrieval capability across a corpus, rather than only navigating to an already specified address.
Identifies retrieved items or passages, potentially alongside generated answers; answer generation alone does not establish a search engine.
May use lexical, semantic or other matching methods and need not crawl the public web.
Scope
+ Searchable corpus, acquisition boundaries and coverage
+ Indexing or equivalent searchable representations and their freshness
+ Query interpretation, matching, ranking and result presentation
+ Retrieval quality, service availability and observable limitations
+ Access controls, result exclusions and authorised search operations
- Physical servers, storage devices and network infrastructure
- The full content and lifecycle of indexed documents
- Web browsers and general navigation interfaces
- General-purpose database management beyond search capabilities
- Advertising marketplaces and the operator's wider business
- Standalone answer generators that do not retrieve from a corpus
Characteristics
- Search domain
- Public web, enterprise, site, collection, specialised domain, or mixed Determines which information needs the engine can plausibly satisfy.
- Retrieval architecture
- Own index, direct corpus search, federated retrieval, or hybrid Locates responsibility for coverage, freshness and retrieval failures.
- Searchable corpus
- Collections, repositories, source systems or upstream search providers Defines the information boundary behind returned results.
- Supported query capabilities
- Keywords, phrases, Boolean operators, field filters, natural language, image, audio, or other documented inputs Allows an agent to formulate requests the engine can interpret.
- Searchable item count
- Items, with item definition, corpus boundary and observation time Indicates retrieval scale without treating item count as proof of completeness.
- Source-to-searchable delay
- Seconds, minutes, hours or days, with event definition and percentile Indicates how quickly source changes become discoverable.
- Query latency
- Milliseconds, with percentile, workload and measurement boundary Supports interactive use and timeout decisions.
- Retrieval effectiveness
- Precision@k, recall@k, nDCG@k or other declared metric, with evaluation set and judgement method Makes quality claims interpretable for a stated search task.
- Service state
- Available, degraded, unavailable, or unknown Guides retry, fallback and escalation decisions.
- Visibility context
- User identity, tenant, permissions, locale and applicable search policies Explains why different users may legitimately receive different results.
Also called
Where this came from
wikidata · CC0 1.0
Drafted structure
Bundle to layer to finding to question, as the second pass will find it: 6 bundles · 11 layers · 18 findings · 28 questions.
Search identity and boundaries Establishes the retrieval system's role, search domain and relationship to adjacent systems.
An agent must distinguish the engine from its client interface, operator, hardware and optional answer-generation features.
Retrieval role
Identifies the capability that makes the system a search engine.
Engine versus client
Record whether the system performs or orchestrates retrieval, and which interfaces only submit requests.
- Which component selects corpus items in response to a search request? definition
- Which browser, search box or answer interface is a client of this engine? boundary
Corpus boundary
Defines what can be searched and the sources supplying that content.
Search domain and dependencies
Record intended collections, supported content types and reliance on external retrieval providers.
- Which collections and content types are within the engine's intended search domain? boundary
- Which results come from the engine's own retrieval resources and which come from upstream providers? provenance
Acquisition and searchable state Tracks how content becomes searchable, remains current and leaves the searchable corpus.
A missing or stale result can originate before query processing, so acquisition and representation require separate evidence.
Content acquisition
Describes discovery, submission and source access mechanisms.
Ingestion paths and gaps
Record crawlers, connectors, feeds or direct access paths and the content they cannot acquire.
- How are new and changed items discovered or supplied to the engine? provenance
- Which source restrictions, unsupported formats or acquisition failures prevent intended content from becoming searchable? boundary
Representation lifecycle
Tracks searchable units, updates, duplicates and deletion propagation.
Freshness and removal
Record how source items map to searchable units and how changes propagate to results.
- What is a searchable unit, and how are versions, duplicate items and canonical sources represented? definition
- How long do additions, updates and deletions take to appear in search, and how is completion verified? measurement
Query and result behaviour Describes how requests become candidate matches and how those matches are ordered and exposed.
An agent needs to predict query semantics and interpret results without assuming undocumented operators or ranking behaviour.
Query interpretation
Captures accepted input and transformations affecting retrieval.
Query semantics
Record supported syntax, matching methods and query rewriting where known.
- Which query forms, operators and filters are supported, and what do they mean? definition
- How can an agent detect or control spelling correction, expansion, translation or semantic interpretation? action
Ranking and presentation
Captures result ordering, grouping and evidence visible to the searcher.
Result interpretation
Record the meaning of result fields and distinctions among retrieved evidence, promotions and generated text.
- Which documented factors, filters or user context affect result order and inclusion? definition
- How are source items, extracted snippets, sponsored placements and generated answers distinguished and traced? provenance
Retrieval quality and operation Assesses search effectiveness and whether the retrieval service is usable under stated conditions.
Fast responses can still be irrelevant or incomplete, while service failures can resemble legitimate empty results.
Effectiveness evidence
Defines task-specific evaluations and limits on coverage claims.
Relevance and coverage evaluation
Record evaluation queries, relevance judgements and the evidence supporting reported metrics.
- Which query set, user tasks and relevance judgements support the engine's quality measurements? provenance
- What precision, recall or ranking metric is measured, at what cutoff, and against which known reference set? measurement
Runtime condition
Captures latency, limits and partial failure behaviour.
Degradation and recovery
Record observable distinctions between successful empty searches, partial retrieval and unavailable service.
- Which signals distinguish zero matches from timeouts, stale indexes, unavailable partitions or failed upstream providers? definition
- Which retries, request reductions or fallback routes are permitted under the engine's limits? action
Visibility and search governance Defines who may discover content and which controls govern search data and result exclusions.
Search can expose information through titles, snippets and aggregates even when opening the source is restricted.
Permission-aware retrieval
Examines enforcement of source and tenant access boundaries throughout retrieval.
Result access boundaries
Record how permissions apply to matches, snippets, suggestions, counts and cached representations.
- How are source permissions enforced for results, snippets, suggestions and result counts? boundary
- How are permission changes propagated and checked before restricted content remains discoverable? action
Search data and exclusions
Captures query-data handling and authorised controls over discoverability.
Logging and removal controls
Record query retention, personalisation inputs and the scope of exclusion or removal mechanisms.
- Which query, identity and interaction data are retained or used for personalisation, and under whose control? provenance
- Who may request exclusion or removal, and does each mechanism affect acquisition, indexed results, cached content or all three? action
Evidence and external alignment What the world already says about this thing, gathered so the model can be checked against it.
A model that cannot be lined up against existing standards, identifiers and practice cannot be adopted by anyone who already uses them.
Reported evidence
Findings from the breadth pass, kept separate from the structural claims.
Check these first
Recalled without web access and unsourced; every item is a lead to verify.
- This describes the general software-system sense, including web search; the PHY.OBJ domain assignment should be checked because a search engine is not inherently a physical device.
- The kinds overlap: academic search is domain-specific, while metasearch describes an architecture that queries other search services.
- These statements are recalled knowledge, not source-verified research; no universal capacity, latency range or certification requirement is asserted.
- Which of these check these first hold for the sense of search engine this model covers, and on what evidence? provenance
Kinds and varieties
Recalled without web access and unsourced; every item is a lead to verify.
- Web search engine
- Enterprise search engine
- Desktop search engine
- Domain-specific search engine
- Metasearch engine
- Academic search engine
- Which of these kinds and varieties hold for the sense of search engine this model covers, and on what evidence? provenance
Standards and regulation
Recalled without web access and unsourced; every item is a lead to verify.
- RFC 9309, Robots Exclusion Protocol, published by the IETF, standardizes robots.txt instructions for crawlers; these instructions are not access authorization.
- Unicode Standard, maintained by the Unicode Consortium, provides character encoding used in multilingual queries and indexed text.
- Which of these standards and regulation hold for the sense of search engine this model covers, and on what evidence? provenance
Real-world use
Recalled without web access and unsourced; every item is a lead to verify.
- Finding public web pages and other online resources
- Retrieving documents across an organization's information systems
- Locating files and their contents on a personal computer
- Discovering scholarly publications
- Finding products within a retail catalogue
- Which of these real-world use hold for the sense of search engine this model covers, and on what evidence? provenance
Typical measurements
Recalled without web access and unsourced; every item is a lead to verify.
- Precision at a specified result cutoff - 0-1 by definition; observed values depend on the query set and relevance judgments - proportion
- Recall against a defined set of relevant items - 0-1 by definition; open-web recall is generally difficult to establish - proportion
- Which of these typical measurements hold for the sense of search engine this model covers, and on what evidence? provenance
Failure modes and hazards
Recalled without web access and unsourced; every item is a lead to verify.
- Incomplete or stale indexes omit useful items or return outdated information.
- Ambiguous queries and inadequate language handling produce irrelevant results.
- Spam and ranking manipulation promote misleading or malicious content.
- Incorrect access-control enforcement exposes restricted documents or snippets.
- Query logging and personalization can expose sensitive interests or reinforce biased rankings.
- Which of these failure modes and hazards hold for the sense of search engine this model covers, and on what evidence? provenance
Regional variation
Recalled without web access and unsourced; every item is a lead to verify.
- Languages and writing systems require different tokenization, normalization and morphological processing.
- Ranking and available results can vary with location, language preferences and local content coverage.
- Privacy requirements and content-removal obligations vary by jurisdiction.
- Which of these regional variation hold for the sense of search engine this model covers, and on what evidence? provenance
Neighbouring kinds and how to tell them apart
Recalled without web access and unsourced; every item is a lead to verify.
- Web browser - A browser displays and interacts with web resources; a search engine retrieves resources matching a query.
- Web crawler - A crawler discovers and fetches resources; it may supply a search engine but does not itself provide the complete query-and-retrieval system.
- Search index - An index is a data structure supporting retrieval; a search engine includes query processing and result delivery.
- Database management system - A database management system manages stored data and structured operations; a search engine specializes in information retrieval, often with relevance ranking, although products can combine both.
- Recommendation system - A recommendation system suggests items from contextual or preference signals, often without an explicit query; search centers on a submitted information request.
- Which of these neighbouring kinds and how to tell them apart hold for the sense of search engine this model covers, and on what evidence? provenance
What the second pass must settle
- Does the registry intend the ordinary software retrieval sense of search engine, and how should that sense relate to its assigned PHY / PHY.OBJ classification?
- Does an existing Vercy world model already own this concept or its core retrieval capabilities?
- Should metasearch services and retrieval systems that return only generated answers be included as variants or linked neighbouring kinds?
- Which authoritative sources establish the query semantics, quality measures and operational controls appropriate to this kind?
- Which deployment-specific interoperability, accessibility, privacy or other requirements should be recorded without implying a universal search-engine certification?