---
title: "Altss Entity Resolution Methodology | Altss Frameworks"
description: "The Altss methodology for deciding whether two records describe the same entity: identity, relation and role kept separate; organisations merged only on…"
canonical: "https://altss.com/knowledge-center/frameworks/entity-resolution-methodology"
---

Methodology · Version 1.2 · Last reviewed 2026-10-01

# Altss Entity Resolution Methodology

The Altss methodology for deciding whether two records describe the same entity: identity, relation and role kept separate; organisations merged only on registry identifiers or a filing naming both as one; similarity creates relation candidates, never merges; no metrics summed across entity families.

Publisher: Altss LLCContent modified 2026-10-01

LIB-04

Image: Abstract illustration: a wire mesh of connected points rising across a dark navy field.

Altss editorial illustration. AI-generated and decorative; it does not show data.

## 1. Purpose

Private-markets organisations come in families that look alike. A state can sponsor several retirement systems; a university has an endowment, campus foundations and a hospital system; a manager has a management company, a general partner entity per fund, the funds themselves, parallel and feeder vehicles; a family has an office, a foundation and holding companies. Members of a family often share a name stem, a web domain, an address and some people, and they transact differently. Treating them as one record destroys the distinctions a reader needs; treating one organisation as several hides its history.

This methodology sets out how Altss decides whether two records describe the same real-world entity, how it records the connections between entities that are distinct, and how it keeps both separate from the question of what an entity does. Its governing rule for organisations is short: only a registry identifier, or a filing that names both as one entity, merges two organisation records. Everything else relates them. Records of natural persons follow a separate conservative rule (5.12). Nothing infers a role from a name.

## 2. Scope

**In scope**

- Organisations: companies, funds and other vehicles, managers, general partner entities, allocators, intermediaries, public bodies and their agencies.

- Identity decisions (merge or keep separate), relation records, and role assignment as it bears on identity.

- Matching methods used to find candidate pairs, and the limits on what they may decide.

- Entity families and the rules for metrics across them.

- Natural persons, in outline (section 5.12).

**Out of scope**

- The classification rules for specific allocator and manager types: see the Family Office Classification Standard and the Institutional Allocator Classification Standard.

- The evidence that must be recorded for each claim: see the [Evidence & Provenance Standard](https://altss.com/knowledge-center/frameworks/evidence-and-provenance-standard).

- Relationship strength and other Altss-defined relationship measures: see the [LP/GP Relationship Intelligence Methodology](https://altss.com/knowledge-center/frameworks/gp-lp-relationship-mapping).

- Identity verification for anti-money-laundering or know-your-customer purposes. This methodology resolves records for research; it is not a regulated verification process.

## 3. Definitions

- **Entity.** A real-world organisation or person.

- **Record.** A set of claims held about what is believed to be one entity.

- **Identity.** The question "is this the same entity?". Answered by a registry identifier or by a filing naming both as one.

- **Relation.** A typed, directed, dated and sourced connection between two distinct entities. See [relationship graph](https://altss.com/glossary/relationship-graph).

- **Role.** What an entity does (allocates capital, manages funds, advises, operates a business, regulates). Multi-label, with one primary, each with its own evidence. See role history.

- **Registry identifier.** An identifier assigned by a statutory register, a regulator or a tax authority to the entity or filer it registers, under rules that make it unique within its scheme and jurisdiction.

- **Identification scope.** The kind of thing an identifier scheme identifies (a legal entity, an establishment, a filer, a tax account, a benefit plan). An identifier merges only things of that kind.

- **Merge.** Combining two records into one canonical record because they describe one entity. See canonical entity.

- **Unmerge.** Reversing a merge, restoring both records with their claims.

- **Relation candidate.** A pair of records that share signals suggesting a connection, held for review. Not a relation and not a duplicate.

- **Entity family.** A set of distinct entities connected by relations such as sponsorship, ownership or household. A family is a view over relations, not an entity.

- **Deterministic matching.** Matching on exact agreement of specified fields. See deterministic matching.

- **Probabilistic matching.** Scoring candidate pairs by how likely their pattern of agreement is among true matches compared with non-matches. See probabilistic matching.

- **Disambiguation.** Telling apart distinct entities with similar or identical names. See [disambiguation](https://altss.com/glossary/identity-disambiguation).

## 4. Inputs and source types

**4.1 Registry identifiers that establish identity.** Each is used only within its identification scope and is stored with its scheme and, where the scheme is national, its jurisdiction. The facts in this table come from the issuing body's own publications (section 12).

| Identifier | Assigned by | Identifies | Format and scope | Where it is checked |
| --- | --- | --- | --- | --- |

| LEI | LEI issuers in the Global LEI System, whose network GLEIF manages, under ISO 17442 | Legal entities | 20-character alphanumeric code. Each legal entity is assigned a single LEI, and each LEI represents only one entity. Level 1 data record who is who; Level 2 data record who owns whom. Level 2 parent data are relations, not identity. See legal entity identifier. | Global LEI Index, free of charge |

| CIK | US Securities and Exchange Commission, through EDGAR, when a filer signs up to file | EDGAR filers: corporations and individual people who have filed with the SEC | Unique to the filer and never recycled. Identifies the filer, not entities related to it. One filer can hold more than one CIK: Form ADV asks an adviser for all of its CIK numbers. | SEC CIK lookup and EDGAR |

| CRD number | FINRA's Central Registration Depository (CRD), or the Investment Adviser Registration Depository (IARD), which FINRA operates for the SEC and NASAA | Broker-dealer firms and their associated individuals; investment adviser firms; investment adviser representatives | A firm's number identifies the firm: Form ADV asks for the firm's own number, not that of an officer, employee or affiliate, and reports each private fund under the fund's own private fund identification number. A firm can hold additional CRD numbers. | BrokerCheck; Investment Adviser Public Disclosure (IAPD) |

| EIN | US Internal Revenue Service | Tax accounts of employers and of certain others with no employees | Nine digits, in the form XX-XXXXXXX. Tax-account scope. A benefit plan is identified by its sponsor's EIN plus a three-digit plan number, together the unique 12-digit number the IRS, the Department of Labor and the PBGC use, so one EIN can sit behind several distinct plans. An entity should hold one EIN, but duplicates are issued. | No public IRS register among the sources cited; taken from a filing that states it |

| Company number (UK: registered number) | The registrar of companies (UK: Companies House) | Companies on that register, including registered overseas companies | UK: allocated to every company under the Companies Act 2006, s.1066, in a form of figures or letters the registrar determines. The registrar may change existing numbers when it adopts a new form. A number is compared only within its own register. | The register; Companies House makes registered company information available to the public |

| SIREN | INSEE, which manages France's SIRENE register | Legal units (legal persons and natural persons) | Nine digits, the ninth a check digit. Carries no meaning, is allocated once and is never reused. The SIRET (14 digits: the SIREN followed by the NIC, a four-digit establishment number and a check digit) identifies an establishment; establishments of one legal unit are one entity. | SIRENE, managed by INSEE |

Other identifiers assigned by a statutory register on equivalent terms qualify in the same way, once the issuing body's own statement of the identifier's scope has been recorded.

**4.2 Filings that establish identity.** A filing or registry record that names two descriptions as one entity establishes identity: a change-of-name record, a registration that lists a former name or an alternative business name of the registrant, a conversion or re-domiciliation record that states the entity continues; a filing in which the filer lists two identifiers of one scheme as its own (Form ADV Item 1.D asks for all of an adviser's CIK numbers, Item 1.E for any additional CRD numbers). A filing describing a merger of one entity into another does not establish identity; it establishes succession (section 9).

**4.3 Signals that do not establish the identity of an organisation, alone or together.** Web domain or email domain; name, name stem or acronym; street address, telephone or postcode; shared directors, trustees, staff or advisers; shared sponsor or parent; the same auditor, custodian or consultant; any string-similarity or model score at any threshold.

**4.4 How identifier evidence is classified.** Identifier evidence is described on the three separate dimensions of the [Evidence & Provenance Standard](https://altss.com/knowledge-center/frameworks/evidence-and-provenance-standard): evidence origin, on each evidence item (what kind of source stated it); derivation status, on the value (how it was produced); and validation status, on the claim with its date (what checking it has passed).

- A registry identifier retrieved from the issuing body's own system (the Global LEI Index, EDGAR, BrokerCheck or IAPD, SIRENE, the Companies House register), or from a filing on a regulator's system that states it, has origin REGULATORY_PUBLIC_RECORD and derivation OBSERVED.

- The same identifier quoted on the subject's own website has origin DISCLOSED; quoted by another party, OSINT_SOURCED; supplied in a licensed dataset, LICENSED_THIRD_PARTY. None of these merges records until the identifier is found in the issuing register (5.6, step 5).

- REGULATORY_PUBLIC_RECORD says where a record is kept, not that its content is true. Registers contain errors (section 7).

- A match score is a DERIVED value used to rank candidates. It is never evidence of identity.

- Validation status (UNVERIFIED, CORROBORATED, RESEARCH_VALIDATED, CONFLICTING) is recorded per claim with its date. It is never collapsed into origin or derivation.

## 5. Method

**5.1 Ask three questions separately.** For any pair of records: Are they the same entity (identity)? If not, how are they connected (relation)? What does each do (role)? An answer to one never answers another.

**5.2 The merge rule.** Two organisation records are merged only when:

- both carry the same registry identifier of the same scheme, within that scheme's identification scope and jurisdiction; or

- a filing or registry record names both as one entity (4.2).

No other evidence merges organisation records, however strong it looks. Records of natural persons are resolved only under 5.12.

**5.3 Everything else relates.** Signals in 4.3 create a relation candidate. A relation candidate is reviewed against evidence. It is not published as a relation, not counted, and not used to combine data. If evidence supports a connection, a relation is recorded; if not, the candidate is closed.

**5.4 A relation is typed, directed, dated and sourced.** Every relation records its type, its direction, the counterparty, its start and end dates (or that they are unknown), its source, its confidence and its status. A connection missing its type, date or source is not published as a relation. Relation types include:

- sponsored by (an entity whose board or funding a sponsor controls), and sibling under a sponsor (derived from two sponsored-by relations);

- administers / administered by;

- invests for (an investment office or company investing on behalf of an institution);

- grants to;

- regulated by;

- subsidiary of / parent of, and other ownership relations;

- general partner of; advised by or managed by (a fund and its manager);

- advises (a consultant, OCIO or other adviser and its client);

- employs (an organisation and a person);

- succeeded by (a predecessor and its successor after a merger or transfer);

- household or family relation (members of one family's group of entities).

**5.5 Similarity writes nothing to identity.** A similarity score, from string comparison or from a model, is a ranking hint. It orders candidate pairs for a registry lookup or a reviewer. No threshold turns a score into identity.

**5.6 Resolution steps.**

- **Normalise for comparison.** Strip legal-form suffixes, punctuation and case; transliterate scripts; expand common abbreviations. Normalised strings are used only to compare. The names as stated in sources are kept. See [data normalization](https://altss.com/glossary/data-normalization).

- **Generate candidates.** Restrict comparisons to plausible pairs (blocking) by keys such as normalised name tokens, domain, jurisdiction and identifiers.

- **Compare.** Record field-by-field agreement, partial agreement or disagreement.

- **Score.** Apply deterministic rules and, where useful, a probabilistic model (5.8).

- **Look up identifiers.** For candidate pairs, look up registry identifiers in the issuing register. An identifier found anywhere else is a lead until the register confirms it.

- **Decide.** Merge under 5.2; record a relation under 5.4 if evidence supports one; otherwise keep separate and close the candidate.

- **Record.** Write the decision with its evidence, actor and date.

**5.7 Deterministic matching.** Exact agreement on a registry identifier of the same scheme, scope and jurisdiction is the deterministic match that establishes identity. Exact agreement on other fields, even several at once (identical name and identical address), is not identity: it is a strong relation candidate.

**5.8 Probabilistic matching.** The Fellegi–Sunter model scores a candidate pair by comparing, for each field, the probability of the observed agreement among true matches (m) and among non-matches (u). Agreement on a field adds a weight of log2(m/u); disagreement adds log2((1−m)/(1−u)). The total weight places the pair in one of three bands. In classic record linkage the upper band is accepted as a link, the middle band goes to clerical review and the lower band is rejected. In this methodology, no band merges: the upper band is prioritised for registry lookup and relation review, the middle band is reviewed if resources allow, and the lower band is closed.

**5.9 Merges are recorded and reversible.** Every merge records the surviving record, the absorbed record, the identifier or filing that justified it, the actor and the date. The absorbed record's identifier is kept and resolves to the surviving record. A merge with no recorded justifying identifier or filing is a defect. Any merge can be undone: the unmerge restores both records, and each claim returns to the record its evidence supports.

**5.10 Roles are separate from identity.** Role is multi-label with one primary, and each role has its own evidence and confidence. Role is never inferred from a name or a free-text type description: an organisation named "Police Fund" is not thereby an allocator. "Does not allocate" is a stated value, displayed with a link to the entity that manages the capital where one exists. Two records with different roles can still be one entity (an insurer that allocates and also operates a business), and two records with the same role can be different entities.

**5.11 Never sum across a family.** No metric (assets under management, commitments, headcount, deal counts) is summed across the members of an entity family. A family total invents a quantity that no entity has. Each figure is shown with the entity it belongs to. A consolidated figure that an entity itself reports (for example in group financial statements) is a claim about that reporting entity, attributed to it with the scope it states. It is not recomputed by adding members.

**5.12 Natural persons.** Identity, relation and role stay separate for people as for organisations. A person's position at an organisation is a dated relation (employs), not part of either party's identity. Two person records are resolved to one person automatically only on one of two bases:

- **Deterministic identity evidence.** A unique identifier that clearly refers to the same individual in both records: for example an individual's CRD number (the CRD system holds the registration records of individuals associated with broker-dealers, and investment adviser representatives are registered through it), or a unique canonical identifier issued by the source for that individual.

- **Several independent strong signals that agree.** Compatible full names or name variants together with further independent evidence, such as the same organisation, a compatible role, a consistent employment history, consistent geography, the same unique professional-profile URL, the same verified professional contact identifier, or other independent corroborating sources. The signals must come from evidence chains that do not share an originating source.

No single weak attribute is sufficient. A name alone, an employer alone, a location alone, or an assertion by a language model alone never establishes that two records are one person.

Hard evidence conflicts block automatic resolution: incompatible unique identifiers, clearly distinct professional profiles, independent evidence that establishes two separate individuals, or identity histories that cannot reasonably belong to one person.

Where the evidence is insufficient or conflicting, the records stay separate. Records are never merged to make a profile look more complete. Ambiguous cases with a high impact, such as a senior decision-maker at an allocator, go to research review before any merge, and the review is recorded under the validation rules of the [Evidence & Provenance Standard](https://altss.com/knowledge-center/frameworks/evidence-and-provenance-standard). This methodology does not publish numeric matching thresholds: the rule above states which evidence is admissible, not a calibrated score. Collection and publication of personal data are additionally limited by privacy law.

## 6. Classification rules

**6.1 Verdict by signal (organisation records).**

| Signal | Verdict |
| --- | --- |

| Same registry identifier, same scheme, scope and jurisdiction | IDENTITY: merge |

| Filing or registry record naming both as one entity, or listing both identifiers as one filer's own | IDENTITY: merge |

| Different identifiers in a scheme that gives each entity a single identifier (LEI) | DISTINCT: never merge |

| Different identifiers in a scheme where one entity can hold several (CIK, CRD number, EIN) | Kept separate, not merged; identity only if a register or filing links both to one entity |

| Same domain | Relation candidate |

| Same name or name stem | Relation candidate |

| Same address, phone or postcode | Relation candidate |

| Same person on both boards or staff lists | Relation (if sourced), never a duplicate signal |

| Same sponsor or parent | Relation (sponsored by, subsidiary of) |

| High similarity score at any threshold | Nothing: a ranking hint only |

**6.2 Duplicate vs sibling.** A duplicate is a second record of one entity; it is merged when the merge rule is met and labelled as a possible duplicate until then. A sibling is a distinct entity in the same family; it is related, never merged. The two are opposite claims and are never shown in the same way. Where a pair cannot yet be classified as either, it remains a relation candidate and is not shown as either.

**6.3 Fund, general partner entity and manager.** A fund, its general partner entity and its management company are distinct entities with distinct identifiers, connected by "general partner of" and "advised by" relations. A shared name does not merge them (see rule 6.9 and 6.10 of the [Taxonomy & Classification Standard](https://altss.com/knowledge-center/frameworks/private-markets-taxonomy-classification-standard)).

**6.4 Parallel, feeder and successor funds.** A parallel fund, a feeder fund, an alternative investment vehicle and a continuation vehicle are distinct entities related to the main fund. A successor fund (Fund IV after Fund III) is a distinct entity related to the same manager.

**6.5 Asset owner and investment office.** Where an institution owns capital and a separate entity invests it (an investment company owned by a university, an OCIO), the two are distinct entities connected by "invests for" or "advises". The allocating role is recorded on the entity whose capital it is; the managing entity carries its own role.

## 7. Conflict handling

**7.1 Identifiers disagree with signals.** Two records with the same name and website and different LEIs are distinct legal entities. The identifiers win, always. In schemes where one entity can hold more than one number (CIK, CRD number, EIN; see 4.1), different numbers keep the records apart but do not by themselves prove two entities. The records stay separate until a filing settles it; a filing that lists both numbers as the filer's own establishes identity (4.2).

**7.2 One record, two identifiers of one scheme.** A record carrying two different identifiers of one scheme is investigated; it is not resolved by choosing one identifier. If a register record or a filing by the entity links both to it (an adviser's Form ADV listing two CIK numbers, or an additional CRD number), the record is one entity and both identifiers are kept. In a scheme that gives each entity a single identifier, such as the LEI, or where nothing links the two, the record is suspected of combining two entities and is split or investigated. Where the issuing register itself marks one identifier as a duplicate of another, the register's resolution is followed and recorded.

**7.3 A merge is challenged.** If evidence shows a merge was wrong (for example a filing showing two distinct entities behind one identifier assignment), the merge is reversed under 5.9 and the reason is recorded.

**7.4 Registers disagree.** If two registers disagree about an entity (a name or status mismatch between a national register and an LEI record) and the pre-checks of the Evidence & Provenance Standard (7.1) show a real conflict, both values are kept and the claim's validation status is CONFLICTING until a review resolves it under the preference order of the [Source Reliability & Confidence Methodology](https://altss.com/knowledge-center/frameworks/source-reliability-and-confidence-methodology). The identity decision rests on the identifier, not on the disputed attribute.

## 8. Confidence and limitations

- **Precision over recall, deliberately.** Wrongly merging two entities is costly and hard to detect: claims from both are mixed and the distinction is lost. A missed merge leaves a duplicate that can be labelled and merged later. This methodology accepts more duplicates in exchange for fewer false merges. Measures of resolution quality therefore report false merges and missed merges separately; see [precision and recall](https://altss.com/glossary/precision-and-recall).

- **Many entities have no public registry identifier.** Private family entities, foreign entities outside accessible registers and informal groups may carry none. Their records are kept separate, related where evidence supports it, and merged only when an identifier or naming filing appears. The correct number of merges without such evidence is zero.

- **Registers contain errors.** A registry identifier assigned in error, or a register entry with a wrong name, is handled under section 7.

- **Relations depend on evidence.** A relation is only as good as its source. Unsourced connections are candidates, not relations.

- **Disambiguation is not identification.** Showing that two similarly named entities are different does not establish anything else about either.

## 9. Temporal rules

- **Identity persists through change.** A renamed entity, a re-domiciled entity that the register treats as continuing, or an entity that changes its legal form while keeping its registry identifier remains one entity. Its earlier names become aliases with their validity dates.

- **Mergers and acquisitions are relations.** When one entity merges into another and ceases to exist, the two records are not merged. The ceased entity keeps its record with an end date, and a "succeeded by" relation points to the survivor from the effective date. When one entity acquires another that continues to exist, a "subsidiary of" relation begins on the completion date.

- **Identifiers have validity periods.** Where a scheme allows a number to be changed or retired, the identifier is matched together with its validity period. The UK registrar, for example, may change existing registered numbers when it adopts a new form of number, with effect from the date the company is notified (Companies Act 2006, s.1066(3)-(4)). Where the issuer states that numbers are never reused (CIK, SIREN, Form 5500 plan numbers), a number keeps denoting the same entity or plan after it ceases to exist.

- **Relations and roles are dated.** A relation or role with no known start or end date records the dates as unknown, and shows when it was last observed. See the [Temporal Data & Freshness Methodology](https://altss.com/knowledge-center/frameworks/temporal-data-and-freshness-methodology).

- **Point-in-time views use dated relations.** A view of a family or a list "as of" a date includes only the entities, relations and roles valid at that date.

- **Dissolved entities are retained.** A dissolved or deregistered entity keeps its record and its history, with an end date.

## 10. Edge cases

- **Benefit plans and their sponsor.** Several plans can share a sponsor's EIN; each plan is a distinct entity identified by the EIN together with its three-digit plan number, the 12-digit combination the IRS, the Department of Labor and the PBGC use. The sponsor is related to each plan by "sponsored by".

- **Identifiers of tax-account or filer scope.** Where an identifier identifies a tax account or a filer rather than a legal entity, a shared identifier establishes that the records share that account or filer. Identity of the legal entities then needs an identifier of legal-entity scope or a naming filing.

- **Establishment and legal-entity numbers.** Where a register numbers establishments (as SIRET does, by adding an establishment number to the SIREN), two establishment numbers under one legal-unit number are one entity. An establishment is a place of business, not a separate entity.

- **Umbrella funds and sub-funds.** Where sub-funds carry their own legal-entity identifiers, each is a distinct entity related to the umbrella. Where they do not, the umbrella is the entity and the sub-funds are recorded as compartments of it.

- **Generic or shared domains.** Free email domains, shared hosting and platform sub-domains create no relation candidate on their own.

- **Spin-outs.** A team that leaves to form a new manager creates a new entity. People relations link it to the former firm; the new manager does not inherit the former firm's identity or track record.

- **Name collisions across jurisdictions.** Identical names in different jurisdictions are distinct unless the merge rule is met; company numbers are compared only within one register.

- **Translated and transliterated names.** Name variants in other scripts or languages are aliases of one record only once identity is established; until then they are separate records with a relation candidate.

- **Rebrands without a filing.** A new trading name used on a website without any registry change is recorded as a self-described alternative name with its source. It is an alias only if the merge rule is otherwise met.

## 11. Worked examples

**Example 1: two pension systems sharing a domain family (illustrative).** "State X Public Employees' Retirement System" and "State X Teachers' Retirement System" both use sub-domains of the state government's web domain, share the name stem "State X" and the state as sponsor, and use the same custodian. Each was established by its own statute, has its own board, staff, investment consultant, asset allocation policy and registry identifiers. Verdict: two entities, each with the role "allocates". Relations: each "sponsored by" State X, so they are siblings under a sponsor. Their assets are never added together. A manager that wins a mandate from one has a relationship with that one only.

**Example 2: a fund and its adviser.** "Example Capital Partners III, L.P." files Form D under its own CIK. "Example Capital Management, LLC" files Form ADV under its own CRD number and lists the fund as a private fund it advises, under the fund's own private fund identification number. Verdict: two entities. Relation: the fund is advised by the management company, from the date the filing supports. The general partner entity named in the fund's documents is a third entity with a "general partner of" relation.

**Example 3: a rename.** One record holds "Alpha Holdings Ltd" and another "Beta Group Ltd". The LEI record for one shows the same LEI with a legal-name history from Alpha to Beta. Verdict: identity; the records are merged with the LEI as the justifying identifier. "Alpha Holdings Ltd" becomes an alias valid up to the date of the name change.

**Example 4: an acquisition.** A manager is acquired by a larger firm and continues to file under its own registration. Verdict: two entities; a "subsidiary of" relation begins at completion. If the acquired manager later withdraws its registration and its business is transferred, its record ends at that date and a "succeeded by" relation points to the acquirer. At no point are the records merged, and the acquired manager's historical funds stay attributed to it.

**Example 5: a family group.** A family's single-family office, a private foundation, two holding companies and the family's principal share a surname, an office address and a web domain. Verdict: five entities (four organisations and one person). Relations: the office "invests for" the family; the foundation "grants to" its recipients and is a family member entity; the holding companies are owned by the principal or the office as filings show. Each entity's figures are shown separately. A family total of assets is not computed.

**Example 6: a high similarity score.** "Acme Holdings Ltd" (England and Wales) and "Acme Holdings LLC" (Delaware) agree on name (m = 0.90, u = 0.01, weight log2(90) = 6.49), web domain (m = 0.80, u = 0.005, weight 7.32) and city (m = 0.85, u = 0.05, weight 4.09): a total of 17.90. Band cut-offs are set for each application; a weight this high would usually fall in the upper band. Had the city disagreed (weight log2(0.15/0.95) = −2.66), the total would have been 11.15. Under this methodology neither score merges. The registry lookup finds two different identifiers in two jurisdictions. Verdict: two entities. If a filing shows that one owns the other, a "subsidiary of" relation is recorded; if not, the candidate is closed.

**Example 7: duplicate vs sibling.** Two records each carry firm CRD number 999999 (illustrative): one entity, merged. Two other records share a website and a name stem but carry different LEIs: two legal entities, related if evidence shows how, never merged. Had they carried different CRD numbers instead, they would stay separate but the question would stay open, because a firm can hold additional CRD numbers. The adviser's Form ADV settles it: if it lists both numbers as the firm's own, the records are one entity (4.2).

## 12. External references

- **Fellegi and Sunter (1969), "A Theory for Record Linkage", Journal of the American Statistical Association.** The foundation of probabilistic record linkage: m and u probabilities, agreement weights and the three decision bands. This methodology uses the model to rank candidates and differs in not accepting any band as a merge.

- **ISO 17442 and the Global LEI System (GLEIF).** The LEI is a 20-character code under ISO 17442; each legal entity is assigned a single LEI and each LEI represents only one entity. Level 1 data records who is who; Level 2 data records who owns whom (direct and ultimate parents). This methodology treats Level 1 as identity evidence and Level 2 as relation evidence.

- **Issuing bodies' definitions of each identifier.** CIK: SEC, *EDGAR Glossary*, *Accessing EDGAR Data* and *CIK Lookup*. CRD number: FINRA, *Central Registration Depository*; FINRA's IARD site; SEC Form ADV Part 1A (Items 1.D and 1.E, Schedules A and D). EIN: IRS Publication 1635 and the IRS EIN page; *2025 Instructions for Form 5500*, line 1b (plan numbers). SIREN and SIRET: INSEE definitions and *Connaître et comprendre le répertoire Sirene*. UK company number: Companies Act 2006, section 1066; Companies House. Each body defines the scope of its own identifier; this methodology uses each only within that scope.

- **Where this methodology differs from common practice.** Classic probabilistic record linkage accepts pairs in the upper weight band as links (Fellegi and Sunter, 1969). This methodology does not: it merges only on registry identifiers or naming filings, keeps every merge reversible, and records all other connections as typed, dated and sourced relations.

## 13. Version and change log

- **Version 1.0 (2026-10-01).** First publication. Establishes the separation of identity, relation and role; the merge rule; the treatment of signals as relation candidates; the limits on deterministic and probabilistic matching; reversible merges; the no-family-sum rule; temporal rules for renames, mergers and acquisitions; and worked examples.

- **Version 1.1 (2026-10-01).** 5.12 natural persons rewritten to the conservative identity rule (deterministic identifier, or several independent strong signals; no single weak attribute; hard conflicts block; insufficient evidence keeps records separate; high-impact ambiguity goes to research review; no published numeric thresholds).

- **Version 1.2 (2026-10-01).** Independent review. The merge rule of 5.2 (and the purpose, 4.3 and 6.1) now states that it governs organisation records and that natural persons are resolved only under 5.12, removing a contradiction between 5.2 and 5.12. Section 4.4 attributes the three evidence dimensions to evidence item, value and claim. Register-access rights removed from the identifier table. Section 7.4 applies the conflict pre-checks first. The comparison with common practice in section 12 is restated from the record-linkage literature.

**Change control.** Adding an identifier scheme that qualifies under 4.1 is a minor change. Any change to the merge rule (5.2) or to the treatment of similarity scores (5.5) is a major change.

## Concepts used

- [Entity Resolution](https://altss.com/glossary/entity-resolution)

- Canonical Entity

- Deterministic Matching

- Probabilistic Matching

- [Disambiguation](https://altss.com/glossary/identity-disambiguation)

- Legal Entity Identifier

- [Legal Entity Hierarchy](https://altss.com/glossary/legal-entity-hierarchy)

- [Relationship Graph](https://altss.com/glossary/relationship-graph)

- Role History and Role Validation

- [Precision and Recall](https://altss.com/glossary/precision-and-recall)

## Sources

- [A Theory for Record Linkage](https://doi.org/10.1080/01621459.1969.10501049). Ivan P. Fellegi; Alan B. Sunter, Journal of the American Statistical Association, Vol. 64(328), pp. 1183-1210. Status: Published (paywalled) (checked 2026-10-01). Vol. 64(328), pp. 1183-1210

- [Introducing the Legal Entity Identifier (LEI)](https://www.gleif.org/en/about-lei/introducing-the-legal-entity-identifier-lei). GLEIF, Global Legal Entity Identifier Foundation, ISO 17442 identifier; web page accessed 2026-10-01. Status: Current (checked 2026-10-01). Introducing the LEI page (re-opened 2026-10-01)

- [Form ADV (Uniform Application for Investment Adviser Registration and Report by Exempt Reporting Advisers) - SEC form cover and index](https://www.sec.gov/about/forms/formadv.pdf). U.S. Securities and Exchange Commission, OMB No. 3235-0049; expires 2027-07-31. Status: in force (checked 2026-10-01). Part 1A and Schedule D (private funds advised)

- [Investment Adviser Public Disclosure (IAPD)](https://adviserinfo.sec.gov/). U.S. Securities and Exchange Commission, Live database (accessed 2026-10-01). Status: current (checked 2026-10-01). Firm and individual search pages

- [EDGAR Glossary](https://www.sec.gov/submit-filings/filer-support-resources/edgar-glossary). U.S. Securities and Exchange Commission, Page last reviewed or updated 2025-01-16. Status: current (checked 2026-10-01). Entry 'Central Index Key (CIK)'

- [Accessing EDGAR Data](https://www.sec.gov/search-filings/edgar-search-assistance/accessing-edgar-data). U.S. Securities and Exchange Commission, Page last reviewed or updated 2024-06-26. Status: current (checked 2026-10-01). CIK paragraph

- [CIK Lookup](https://www.sec.gov/search-filings/cik-lookup). U.S. Securities and Exchange Commission, Page last reviewed or updated 2024-11-01. Status: current (checked 2026-10-01). Page text

- [Form ADV Part 1A (paper version) - Uniform Application for Investment Adviser Registration and Report by Exempt Reporting Advisers](https://www.sec.gov/about/forms/formadv-part1a.pdf). U.S. Securities and Exchange Commission, SEC 1707 (07-24). Status: in force (checked 2026-10-01). Items 1.D(3), 1.E; Schedule A; Schedule D Sections 7.A and 7.B.(1)

- [Central Registration Depository (CRD)](https://www.finra.org/registration-exams-ce/classic-crd). FINRA, Web page accessed 2026-10-01 (no date shown). Status: current (checked 2026-10-01). Page text

- [What Is IARD? (Investment Adviser Registration Depository)](https://iard.com/what-iard). FINRA, as developer and operator of IARD for the SEC and NASAA, Web page accessed 2026-10-01 (no date shown). Status: current (checked 2026-10-01). Background information

- [Publication 1635, Employer Identification Number: Understanding Your EIN](https://www.irs.gov/pub/irs-pdf/p1635.pdf). Internal Revenue Service, Rev. 2-2014 (the revision hosted at irs.gov/pub/irs-pdf and linked from the IRS EIN page, reviewed 2026-07-17). Status: current (checked 2026-10-01). p.2 'What is an EIN?'; p.15

- [Employer identification number (IRS web page)](https://www.irs.gov/businesses/small-businesses-self-employed/employer-identification-number). Internal Revenue Service, Page last reviewed or updated 2026-07-17. Status: current (checked 2026-10-01). Page text (reviewed 2026-07-17)

- [Instructions for Form 5500, Annual Return/Report of Employee Benefit Plan (2025)](https://www.dol.gov/sites/dolgov/files/EBSA/employers-and-advisers/plan-administration-and-compliance/reporting-and-filing/form-5500/2025-instructions.pdf). U.S. Department of Labor (EBSA), Internal Revenue Service and Pension Benefit Guaranty Corporation, 2025 Form 5500 instructions (EFAST2). Status: current (checked 2026-10-01). Part II, line 1b

- [Siren number (definition)](https://www.insee.fr/en/metadonnees/definition/c2047). INSEE (Institut national de la statistique et des études économiques), Last update 2025-10-21. Status: current (checked 2026-10-01). Definition and side note

- [Numéro Siret (définition)](https://www.insee.fr/fr/metadonnees/definition/c1841). INSEE (Institut national de la statistique et des études économiques), Dernière mise à jour 2019-12-04. Status: current (checked 2026-10-01). Definition

- [Connaître et comprendre le répertoire Sirene](https://www.insee.fr/fr/information/1972043). INSEE (Institut national de la statistique et des études économiques), Mise à jour 2023-07-03. Status: current (checked 2026-10-01). Page text

- [Companies Act 2006, section 1066 - Company's registered numbers](https://www.legislation.gov.uk/ukpga/2006/46/section/1066). UK Parliament (legislation.gov.uk, The National Archives), Latest available (revised), up to date with changes known to be in force on or before 2026-10-01. Status: in force (checked 2026-10-01). s.1066(1)-(4), (6)

- [Companies House (GOV.UK organisation page)](https://www.gov.uk/government/organisations/companies-house). Companies House (GOV.UK), Web page accessed 2026-10-01. Status: current (checked 2026-10-01). Organisation description

## Related frameworks

- Standard · Version 1.2

### [Altss Evidence & Provenance Standard](https://altss.com/knowledge-center/frameworks/evidence-and-provenance-standard)

The Altss standard for recording evidence behind private-markets facts: claim-level provenance, evidence chains, three separate evidence dimensions (evidence origin, derivation status, validation status), distinct date types, null states such as unknown vs not disclosed, negative evidence, conflicts and claim citation.

- Methodology · Version 1.2

### [Altss Source Reliability & Confidence Methodology](https://altss.com/knowledge-center/frameworks/source-reliability-and-confidence-methodology)

The Altss methodology for grading sources and assigning confidence labels to private-markets claims: source authority classes, reliability and credibility grades adapted from the NATO Admiralty system, independence, use of validation status, conflict handling, label meaning and limits, calibration and decay.

- Methodology · Version 1.2

### [Altss Temporal Data & Freshness Methodology](https://altss.com/knowledge-center/frameworks/temporal-data-and-freshness-methodology)

The Altss methodology for time in private-markets data: valid time vs transaction time, point-in-time reconstruction, as-of reporting, freshness as an attribute of each claim rather than a page, decay, and why a content edit never implies that a fact was verified.

- Standard · Version 1.1

### [Altss Private Markets Taxonomy & Classification Standard](https://altss.com/knowledge-center/frameworks/private-markets-taxonomy-classification-standard)

The Altss standard for classifying private-markets concepts: 33 facets with one primary facet per concept, permanent concept IDs, one canonical page per concept, rules against common category errors, and crosswalks to external classifications.

## Canonical URL

https://altss.com/knowledge-center/frameworks/entity-resolution-methodology