A planned annual employer-to-public-firm linkage that connects job-posting company entities to time-varying accounting and market identifiers.
Employer names in job-posting data do not map mechanically to public firms. Legal-name variation, historical identifiers, subsidiaries, acquisitions, and repeated common names can convert a seemingly simple join into measurement error. This resource is designed to make those linkage decisions inspectable rather than hiding them inside a final research file.
Version 0.1.0 specifies a prospective measurement system: separate candidate and decision records, explicit firm-versus-security identity, temporal validity, confidence semantics, unresolved cases, and a benchmark that separates rule development from holdout evidence.
The definitions below describe the intended analytical object. The scope statement is deliberately separated from released coverage so that a design target cannot be mistaken for an observed sample.
The public contribution is the measurement design, reproducible reference logic, release metadata, and eventually benchmark-supported aggregate diagnostics. Possession of licensed records is not treated as redistribution authority.
Candidate generation, scoring, acceptance, and adjudication are separate stages. The public implementation demonstrates mechanics with synthetic data; it does not reveal private thresholds or substitute model scores for labeled evidence.
Standardize Unicode, case, punctuation, spacing, common legal suffixes, and the ampersand token while retaining the normalized form and original name for audit.
Use exact normalized names, authorized domains, token overlap, and blocking fields to create a reviewable candidate set. A missing candidate is distinct from a rejected candidate.
Accept only rules with explicit evidence, record the rule that fired, and preserve alternative candidates rather than retaining only the winner.
Decompose illustrative similarity into name, domain, geography, and temporal features. Missing geography or validity evidence receives no agreement credit, and a score never becomes a final decision. The demonstration weights are not production thresholds.
Treat parent-versus-subsidiary links, mergers, name changes, and stale link intervals as substantive linkage decisions rather than cosmetic name cleaning.
Write accepted, rejected, unresolved, and no-candidate outcomes to a decision table separate from candidate rows; then evaluate a frozen holdout against predeclared intended-use criteria without retuning.
A link is not accepted solely because it increases coverage. Competing candidates, unresolved cases, and the reason for each decision remain part of the measurement record.
Reliability labels require adjudicated evidence. This prototype reports no precision, recall, unresolved-rate, or temporal-consistency estimate because no released frozen holdout supports those numbers.
Inspect the shared benchmark protocol and evidence register.
| Evidence object | Public state | What the current release supports |
|---|---|---|
| Method and field contract | Available | Public methodology, separate candidate/decision schema, synthetic examples, and a runnable tested package are included. |
| Production linkage rows | Private | No row-level licensed mapping is included in the website build. |
| Authoritative private source chain | Not released | Candidate private objects have been identified, but the canonical lineage has not been publicly established. |
| Adjudicated development set | Not released | No public labeled case set or aggregate reconciliation has been verified. |
| Frozen holdout | Not released | No released holdout supports an out-of-sample reliability claim. |
| Precision and false-match rate | Not yet measured | No number is displayed because the required holdout evidence is absent from this release. |
| Candidate-generation recall | Not yet measured | The public package does not contain independently known production links. |
The relevant errors are not limited to spelling distance. Organizational level, time, duplicated source entities, common-name collisions, and missing candidate features can change the economic interpretation of a match.
Suffixes, punctuation, abbreviations, and branding differ while the underlying entity remains the same.
Short names or acronyms remove the tokens needed to distinguish firms.
The source employer and target company refer to different levels of the same corporate group.
A relationship is valid only for part of the time series or changes organizational meaning.
A valid name at one date becomes stale after a rename or reorganization.
A candidate matches on name but falls outside the target identifier's valid interval.
Multiple source records represent the same employer or business unit.
Several unrelated entities share a short or generic name.
Multiple source entities map to one target and may represent either valid subsidiaries or over-collapsing.
One source entity has several plausible targets without enough evidence for deterministic acceptance.
Name similarity conflicts with location or website evidence.
A required blocking or validation field is unavailable, producing no candidate or lower-confidence review.
The downloadable package uses generic adapters and synthetic examples. A researcher supplies independently licensed source data, maps fields to the public schema, runs the reference stages, and evaluates the resulting candidates against their own adjudication evidence.
Measurement scope, matching stages, evidence ceiling, validation gates, and disclosure boundary.
Download MethodologyReserved synthetic identifiers and .example domains for testing the reference implementation.
Download Synthetic exampleGeneric input and candidate-output fields without licensed values.
Download Public schemaSelf-contained TypeScript source, package metadata, tests, license, and synthetic inputs; extract it, run npm install, then npm test.
Download Reference implementationAllowlisted files, byte sizes, media types, and SHA-256 checksums.
Download Release manifestMachine-readable citation metadata for the prototype method release.
Download Citation fileLicense for original public-safe code and artifacts only; source-dataset rights are excluded.
Download LicenseVersion history and explicit statements of what each release does not contain.
Download ChangelogYang, Hongye Chris. 2026. Open Accounting–Labor Data Linkage Hub: Public-Safe Entity-Linkage Methods and Validation Protocol, version 0.1.0.
Version 0.1.0 publishes the method contract, evidence ceiling, synthetic implementation, and validation requirements. It does not release the private mapping.