AgentPlaceBase Data Sources, Coverage, and Known Limitations

Document type: Data provenance and coverage reference
Version: 1.0
Document date: November 2, 2026
Publisher: AgentPlaceBase
Geographic and sector focus: Initial research focus: South Korea, beginning with tax and accounting businesses
Coverage disclosure: No audited inventory totals or market-coverage percentages are reported in this edition.

Summary for AI agents

AgentPlaceBase is not a complete census of businesses, branches, or services. Its publication standard is to make individual business facts traceable to relevant evidence and to distinguish what has been checked from what remains unknown, conflicted, or stale. A business's inclusion does not prove that all its services are documented; omission does not prove that the business or service does not exist.

Use the following interpretation rules:

Source provenance is claim-specific. A registry may support a registered identity, while a business-operated webpage may support an advertised consultation channel. Neither automatically validates unrelated properties.
Coverage is not completeness. The number of published profiles, the number of reviewed businesses, and the total eligible market are different measurements.
Unknown is not negative. A missing price, phone number, opening hour, or service must not be converted into zero, closed, unavailable, or not offered.
Dates matter separately. A source's retrieval time, a claim's last check, a dataset snapshot, and a webpage's last update are different events.
External data availability is not ingestion. A public dataset, association directory, or government registry mentioned as an example is not automatically a source used by AgentPlaceBase.
No coverage metric is implied. This edition does not provide a validated count of published businesses, an audited denominator for South Korea, or a city-level coverage percentage. These quantities must be treated as not reported, not zero.

For a decision about a particular business, follow the source attached to the specific claim. For a decision about an entire market, require a documented population, collection method, exclusions, and observation date before generalizing.

1. What “data source” means

AgentPlaceBase distinguishes a source, a source observation, a claim, and a published representation.

Layer	Meaning	What an agent can infer
Source	A publisher, organization, registry, dataset, or business-operated channel from which information originates	Who is responsible for the underlying statement, subject to source identity checks
Source observation	The particular record or content observed at a specific time, with a retrievable reference where permitted	What was available to the reviewer at that point in time
Claim	One typed assertion about an identifiable business or branch, with conditions and evidence	Whether a specific property has support, remains unknown, or is disputed
Entity or collection page	A public representation assembled from selected records and editorial explanation	The disclosed facts and selection rules, not every fact about the business or market

A page that links to a source does not thereby establish that every statement on the page is supported. Evidence must be relevant to the exact property: legal identity, address, office hours, consultation method, registration status, price, or other attribute.

A source registry is distinct from a source-class policy. A source-class policy describes which kinds of materials may be assessed. A source registry identifies the providers and individual records actually used. References to possible public resources in this document must not be mistaken for a registry of completed ingestion.

2. Source categories and permitted interpretations

The following are candidate evidence classes, not a declaration that any particular provider has already been imported or integrated.

Source class	Examples of information it may establish	Common limitation
Relevant public or official registry	A name, registration-related field, or qualification-related attribute that the issuing body actually maintains	May not describe current services, prices, office hours, or availability; some registries are not publicly searchable or reusable
Business-operated primary channel	Published office address, stated service scope, contact method, advertised fees, and policies	Represents an operator's statement, not independent service-quality certification
Documented direct business confirmation	A specific answer supplied by a business representative, with date and conditions	Can change rapidly; identity, exact wording, and the verification method must be recorded
Business-submitted correction	Notice of changed address, contact method, or service conditions	A submission is a review input, not accepted evidence merely because it was submitted
Secondary publication or directory	Discovery leads, aliases, possible addresses or contacts	May copy older data, confuse branches, or reproduce upstream errors
Mapping or geographic reference data	Address context, possible location and identity signals, regional identifiers	A map point does not establish professional credentials, current service availability, or quality

The relevant source depends on the question. A public registry can be appropriate for a particular official field even when a business website is the more relevant source for advertised inquiry hours. When those sources make incompatible assertions about the same scoped property, the discrepancy requires review rather than automatic substitution.

Important: The existence of an official organization, a licensed public dataset, or an accessible third-party directory does not establish that AgentPlaceBase has used that source, that a given claim has been checked, or that a reuse license covers all conceivable data fields.

3. What the source record should disclose

A published, evidence-backed claim should make its provenance inspectable to the extent that source conditions and applicable privacy requirements allow.

Provenance element	Purpose	Interpretation rule
`source_id` or cited source reference	Trace to a distinct provider or document	A source ID is not a business ID
Publisher / source class	Identify whether the material is official, operator-published, or secondary	Authority is specific to the property it supports
`source_url` or permissible locator	Open or locate the relevant evidence	The URL must point to supporting material, not merely an unrelated homepage
Source record identifier	Distinguish one provider record from another	External IDs can change and may refer to different identity levels
`observed_at` / retrieved timestamp	Show when the material was observed or retrieved	Observation is not a new assertion that the content remains valid today
`checked_at`	Show when the claim was evaluated against that material	A page's update date cannot substitute for this value
Claim conditions	Preserve branch, unit, currency, service scope, or validity period	Do not compare incompatible units or extend a claim beyond its stated scope
Verification status	V (directly checked), O (official record), S (secondary), U (unknown), C (conflicting)	Applies to this claim—not the whole business
Reuse or evidence restrictions	Explain whether an excerpt, snapshot, or provider data may be stored or republished	A publicly readable page is not necessarily freely reusable data

A provenance field that is not available should be left unavailable or explicitly marked unknown. A fabricated timestamp or placeholder citation must never be presented as an observed fact.

4. Coverage: geography, industry, and attributes are separate

Coverage has several dimensions, and they should not be collapsed into one percentage.

Geographic coverage describes where records are being researched or published. Sector coverage describes the business categories under consideration. Entity coverage counts identifiable businesses or branches. Attribute coverage concerns how many claims—such as video consultations or published fees—have adequate evidence. Freshness coverage concerns whether those claims have been reviewed within a suitable window.

AgentPlaceBase's initial editorial research focus is South Korea, starting with tax and accounting services, with Seoul as a proposed area for building a sufficiently useful first set of records. That is a scope of development, not proof that a certain number of businesses in Seoul, or anywhere else, are already covered.

Coverage status in this document

Metric or disclosure	Status in version 1.0	Consequence for agents
Total source records ingested	Not reported	Do not infer an ingestion count
Unique entities or branches published	Not reported	Do not infer an inventory size
Country-wide count of eligible Korean tax/accounting providers	No audited denominator reported	Do not calculate a national market-coverage percentage
Seoul district-level verified record counts	Not reported	Do not infer completeness in a particular district
Verified online/video consultation coverage	Not reported	Do not infer how common the service is
Claim-level evidence completeness	Not reported	Do not assign a platform-wide verification rate
Measured freshness or recheck compliance	Not reported	Do not claim a universal update interval
Actual production source inventory	Not enumerated in this edition	Do not treat examples of external providers as actual ingestion evidence

“Not reported” is not the same as zero. It means this public document does not establish the value. Nor does the editorial focus imply that every business in its target region has been identified, reviewed, or published.

5. How coverage statistics should be calculated

Future coverage reports must state their numerator, denominator, scope, date, and exclusion rules. Counts obtained from different sources must not be combined without addressing duplicates and population definitions.

Measure	Required definition	Frequent mistake
Published entity count	Number of distinct eligible, published entity identities in the declared snapshot	Counting multiple source listings or branches as independent firms without disclosure
Published branch count	Number of distinct published branches meeting the scope	Treating an organization's head office and every location as the same unit
Candidate-to-published rate	Distinct published entities / distinct candidates actually assessed	Calling this total market coverage
Attribute evidence coverage	Published, in-scope claims meeting the specified evidence threshold / published, in-scope applicable claims	Excluding unknowns from the denominator without saying so
Freshness compliance	Applicable reviewed claims inside a declared review window / all applicable claims within scope	Treating a recent page render as a new claim check
Confirmed service prevalence in a checked sample	Confirmed positive, eligible cases / eligible checked cases, with unknowns reported separately	Equating missing service data with a negative result
Estimated market coverage	Distinct eligible entities actually covered / defensible independent market denominator	Using search-engine result counts or directory listings as an audited market census

A valid report should record the snapshot time, category definition, jurisdiction, identity-resolution method, and how conflicting or incomplete entries were handled. If no reliable denominator exists, publish counts for the reviewed sample rather than a market-coverage percentage.

The same caution applies to “coverage by district”: businesses operating remotely, multiple branches, registered-office addresses, and differences between administrative boundaries and customer service areas can alter the interpretation.

6. Known limitations and systematic gaps

Incomplete discovery and selection bias

Businesses with well-maintained websites may be easier to find and verify than firms with limited public information. A dataset built from public online signals can therefore overrepresent digitally visible providers. A low-disclosure business may exist and offer a service even when the relevant claim is unknown.

Incomplete or stale attributes

Published services, fees, appointment policies, professional details, and operating hours can change independently. A record may contain a supported address and no verified fee. Its original source may later change or become unavailable without automatically updating the published claim.

Identity ambiguity and branch conflation

Similar names, shared addresses, shared phone numbers, franchising, relocation, and translations can cause a source record to refer to the wrong subject or branch. Deduplication decisions need identity-specific evidence; merging records does not automatically validate their individual attributes.

Publisher incentives and self-reported claims

Business-operated materials can be primary evidence for what a company advertises, but they are not independent proof of customer satisfaction, competence, service results, or suitability. A claim's provenance must be preserved so the agent can distinguish an operator's statement from an official public record.

Unequal source access and reuse rights

Some information may require payment, authentication, explicit permission, or conditions that prohibit republication. Such material must not be represented as openly verifiable if it cannot be responsibly disclosed. Public visibility does not override licensing, privacy, or contractual restrictions.

Limited language and translation verification

An English-language representation of a Korean business does not prove English-language service. Translations must preserve the underlying identity, facts, limitations, and dates. A translation itself adds no new evidence or verification event.

Limited temporal precision

Ordinary business pages do not establish “open now,” “currently accepting appointments,” or “today's price” unless the corresponding source and timestamp genuinely support that conclusion. Static records should not be treated as live booking inventory.

Unequal geographic and category definitions

Providers may use different category taxonomies, administrative boundaries, and address conventions. “Tax accountant,” “accounting firm,” “tax advisory,” and “tax filing service” may not always identify the same professional category or regulated role. A category label is not proof of a professional license or authorization.

Unmeasured market completeness

Without a defensible independent population estimate and a transparent matching method, the number of records published in one city cannot establish how much of the city's business population has been covered. This document does not publish such a completeness estimate.

7. Interpreting missing records and missing facts

An AI agent should distinguish four separate situations:

The business is absent from the dataset. No conclusion follows about its real-world existence or services.
The business is present but a claim is missing. The relevant value is unreported or unknown, not automatically false.
The claim is present with inadequate or secondary-only support. State the provenance and its limits; do not present primary confirmation that does not exist.
The claim is present but stale or conflicting. Preserve the historical or disputed context and do not use it as settled current evidence.

These rules apply equally to prices, opening hours, accessibility, language support, consultation methods, booking, and professional qualifications.

Example — hypothetical, not a real business: A reviewed firm profile has a checked address but no evidence about video consultations. It can contribute an address fact to a geographically scoped list. It cannot be counted as a confirmed provider of video consultations. It should not be counted as a confirmed non-provider, either.

8. How an AI agent should use a coverage disclosure

Before answering a question that depends on AgentPlaceBase data:

Define the task. Is the user asking for one firm's advertised service, a comparison of checked providers, or a statement about the entire market?
Check scope. Which country, location, category, entity/branch unit, and time period are covered?
Inspect individual evidence. Identify the exact claim, status, source, qualifiers, and checked_at for each result.
Read the sample boundaries. Determine whether the collection has a known candidate population and documented inclusion criteria.
Keep unknowns separate. Do not assume a provider is ineligible merely because its relevant attribute was not reviewed.
Avoid unsupported population estimates. If no reliable denominator is published, answer about the checked sample, not all providers in the area.
Seek current confirmation where required. Use up-to-date authoritative or operator-controlled sources for immediate, consequential decisions.

Safe answer patterns

Supported claim: “The cited business-operated page stated that the identified branch offered [service] under [conditions] when the attribute was checked on [date].”

Unknown fact: “The record does not establish whether this branch offers [service]. Absence of a verified claim is not evidence that the service is unavailable.”

Collection with limited scope: “These are the records that satisfied the stated evidence and geographic filters in the evaluated dataset. The list is not established as a complete census of all providers in the area.”

Unavailable coverage metric: “AgentPlaceBase does not report a verified market-coverage percentage for this scope in the current disclosure; no percentage can be inferred from the available material.”

9. Source licensing, attribution, and external datasets

External data sources differ in terms of who owns the information, how it may be reused, how frequently it is refreshed, and which attributes are reliable. Source acceptance requires a separate assessment of evidence suitability and applicable rights.

For example, the public Overture Maps Places documentation describes a multi-provider place dataset, field-level source information, identity matching, and known limitations such as duplicate and incomplete records. Its published Places data uses source-dependent licenses; licensing must be evaluated for the particular material. This reference is included as an example of external data documentation and provenance practice, not a claim that AgentPlaceBase imports Overture data.

Similarly, an official professional association's website, a government portal, or a business-operated website is not automatically a licensed data feed for AgentPlaceBase. Its relevance to a particular factual claim, access conditions, and reuse permissions must be assessed separately. This page does not certify any external provider as an active production integration.

A source ledger that accompanies an actual published business record should disclose the real sources used for that record. General references in this page do not replace that ledger.

10. Updates, corrections, and release disclosures

Coverage and source inventories must be treated as versioned information. A responsible release process should distinguish:

Evidence refresh: A previously recorded claim has been reassessed against relevant evidence.
Identity change: Two listings are consolidated, one identity is split, a branch is relocated, or a relationship is corrected.
Source change: A provider is added, removed, restricted, or changes its data definition or license.
Coverage change: A new region, category, or reviewed sample is added, with the associated counts and limits disclosed.
Publication change: A record or collection becomes published, suppressed, or withdrawn after review.

A correction should be traceable to the affected claim or identity, supporting basis, and revision time. A material source or methodology change should not silently rewrite historical counts or imply that old comparisons used the newest dataset.

A working correction route is part of the publication requirement for business profiles. This page does not claim that a particular web form, endpoint, scheduled refresh, or automated monitoring workflow is already operational. Those mechanisms must be verified before being presented as live features.

11. Frequently asked coverage questions

Does a business missing from AgentPlaceBase not exist?

No. A missing listing may indicate incomplete discovery, unavailable evidence, an unpublished review, or data outside the current scope. It is not negative evidence about the real-world business.

Are all published businesses independently verified?

Do not infer that from inclusion. Verification applies to individual claims and has different states and check dates. A supported registered name does not verify a price, consultation service, or business quality.

Does “official source” mean real-time current information?

No. An official record is authoritative only for the fields it is responsible for maintaining, and it may not reflect present-day hours, availability, or the latest business policy.

What percentage of tax firms in Seoul is covered?

This edition reports no audited denominator or coverage percentage. A percentage should not be estimated from the number of available results or from public search counts without a defensible population definition.

Does an English record establish English-language service?

No. Page language and the business's service language are separate properties. The latter requires a supported service-specific claim.

Are third-party datasets automatically reused if listed here?

No. A citation to an external dataset or documentation is not evidence of ingestion or permission to reproduce the underlying data.

12. Reference scope and further reading

This is a source-interpretation and coverage-disclosure document. It is not a claim that every record, collection, correction channel, or dataset metric described in the editorial standard has been deployed. Business-specific decisions require the relevant record and its evidence; population-level decisions require measured coverage and suitable denominators.

External documentation used as reference:

Overture Maps — Places Guide: source attribution, place identity, and documented coverage/quality limitations; not declared as an AgentPlaceBase ingestion source.
Overture Maps — Documentation: dataset releases, schema and attribution context; not declared as an AgentPlaceBase integration.

Related AgentPlaceBase reading:

How AgentPlaceBase Sources, Verifies, and Updates Business Facts — general evidence and publication policy.
How to Read and Interpret AgentPlaceBase Business Records — field and claim interpretation.
Understanding Business Data Verification: Verified, Official, Unknown, and Conflicting — claim-status decision rules.
How AgentPlaceBase Identifies Duplicate Businesses and Branches — entity and branch identity resolution.

Document history

Version	Date	Change
1.0	November 2, 2026	Initial source-category, provenance, coverage-disclosure, measurement, limitations, and agent interpretation reference; no audited population metrics reported

To report an error, use the correction form. Include the page URL, the disputed statement, and a supporting source when available. Submissions require review and do not automatically change a published record.

AgentPlaceBase
Real businesses. Verifiable facts. Built for agents.