Entity SEO
Entity Provenance Chain: Trace Sources, Changes and Decisions
An entity provenance chain records how a source statement became an accepted assertion and where that assertion was used. It connects original evidence, extraction, normalization, identity and relationship decisions, approval, publication and later correction. A final source URL alone cannot explain changes made along the way.
Follow the annotated evidence chain
An address assertion from source to publication
Illustrative scenario: a directory lists an address for Northbank Workshop. An extraction tool extracts it, an editor standardizes its format, and the address is proposed for an organization page and Organization markup. Without a chain, the final address looks authoritative even if nobody checked which organization, location or period it describes.
- Save the original evidenceSave the source location, exact observed statement, observation date and relevant context. Record the directory as the source of this statement; do not silently relabel it as confirmation from the organization itself.
- Extract without changing the claimGive the extraction its own event ID. Record the tool or person, input source revision and output fields. Preserve qualifiers such as “former address” or “correspondence only”.
- Normalize the representationLog formatting changes and the rule used. Expanding an abbreviation may change presentation; adding a missing unit, address component or inferred location changes meaning and needs evidence.
- Resolve the entity and locationCompare the entity ID, organization name, branch and applicable period. Check the organization’s own source where available. A similar name or same street does not justify a merge.
- Decide and approveRecord the accepted address, source comparison, unresolved conflict and reviewer. If the directory and organization page disagree, hold the disputed current address until scope or timing explains the difference.
- Link the outputsConnect the accepted assertion version to the visible organization page and structured data candidate. Record their placement identifiers and release state; approval of a record is separate from evidence that it was published.
- Track later correctionIf the organization moves, create a new assertion and effective period. Link it to the superseded record and identify every dependent placement needing change. Preserve the prior record as historical evidence.
The output is a traceable derivation record, not proof that the address is correct simply because a log exists. Evidence quality and the validity of each decision still need review.
Give every assertion an identity of its own
Record the subject entity, attribute or predicate, recorded assertion, scope, version and effective period. Separate an organization from its branch, a product from its family, and a person from the company they represent. Each distinction affects which source can support the assertion.
Retain the original statement and the accepted form side by side. Add source authority, observation date, uncertainty and any limits on use. A source snapshot or exact extract can preserve what was observed when a page later changes, subject to the rights and handling rules that apply to the material.
Use stable identifiers for assertions and events so a replacement does not erase the earlier decision. A source URL identifies a location; an assertion ID identifies what your system accepted from it. A new source version can support the same assertion, and one source can support different assertions.
Distinguish formatting from a new inference
| Operation | What to record | Mistake to avoid |
|---|---|---|
| Extraction | Input source, observed span or field, tool or person, method and output. | Treating extracted text as already verified or approved. |
| Normalization | Original entry, changed representation, rule, units or vocabulary and resulting entry. | Adding facts under the label of cleanup. |
| Identity resolution | Input entity records, matching evidence, merge or split decision and retained IDs. | Merging records because names or domains look alike. |
| Label or predicate mapping | Source wording, chosen label or relation, direction and type constraints. | Strengthening “mentioned with” into “owns”, “employs” or another unsupported relationship. |
| Inference | Input assertions, stated reasoning, uncertainty and separate derived assertion. | Presenting an editorial inference as a direct source quotation. |
| Conflict resolution | Competing assertions, scope/date comparison, accepted or held outcome and reviewer. | Deleting the losing source so nobody can inspect the disagreement. |
For a merge or split, retain links to the predecessor records and review their dependent assertions. A correct identity repair can still leave incorrect attributes on the surviving entity if propagation is ignored.
Record the agent, decision and release separately
An agent can be a person, tool or process. Log its name or stable identifier, method or version where relevant, the time of the operation, and the input and output IDs. This makes automated extraction distinguishable from human acceptance.
A reviewer should be able to see the evidence, changes to the record, confidence decision, conflict handling and reason for approval. Name the accountable owner and the next review trigger. Missing authority, unclear identity or unresolved scope should leave the affected assertion held, even when the surrounding batch is ready.
Record output destinations precisely: page URL or ID, content placement, graph record, table or structured data candidate, and version. Drafted, approved and published are separate states. Do not mark a dependent page corrected merely because its upstream assertion has been repaired.
Use the chain to find the smallest reliable repair
Start with the changed source or incorrect decision, then follow dependency links forward. If an abbreviation expansion was wrong, repair that change to the record and inspect every output derived from it. If identity resolution was wrong, review the attributes and predicates attached after the merge, not just the displayed name.
Retain the old event and create a superseding decision with its reason, reviewer and effective date. Connect the new output to the prior version so historical pages can be interpreted correctly. A current correction does not make an accurate statement about an earlier period false.
When the original source cannot be recovered, mark the gap. Seek a replacement source or hold the unsupported assertion; reconstructing a plausible origin is not a substitute for evidence. If independent evidence supports the same assertion, record a new derivation from that evidence and keep the missing part of the old chain explicit.
Finish by checking the visible text, tables, internal routes and machine readable assertions that depend on the decision. Store the remaining unresolved placements, their owners and the next review date. A repair is complete only for the destinations that have been checked.
Questions about evidence lineage
Is a bibliography enough?
A source list names references. A provenance chain also shows which source supports each assertion, which changes occurred, who accepted them and which outputs used the result.
Does a detailed chain increase confidence by itself?
It improves inspectability. It cannot make a weak source authoritative or a wrong identity match valid. Assess the evidence separately and keep its limits visible.
Can a source be replaced without losing history?
Yes. Add the replacement source and a new decision linked to the earlier one. Record why it supersedes or supplements the original, and check if the assertion’s scope or effective period changes.
When is provenance most useful?
Use it when different people or tools extract, normalize, merge, map or infer records before they reach content or structured data. The chain lets a reviewer audit, correct or withdraw an assertion without guessing its origin.
Related decisions and sources
Entity Confidence and Review Dates
Assess what the traced evidence supports and when it needs another review.
Entity Page Ownership Registry
Identify the destinations responsible for maintaining an accepted assertion.
Entity Identity Evidence Register
Keep identity evidence attached to merge, split and matching decisions.
Entity Predicate Library
Choose relationship meanings and direction before mapping source wording.
Entity SEO
Return to the related entity methods.
RDF 1.1 Concepts
Read the source reference on graph statements. This page describes an editorial lineage method, not a complete provenance vocabulary or technical specification.
The examples are teaching scenarios, not records of client work. The illustrations explain record separation and revision; they are not evidence of the assertions described. These editorial methods do not guarantee rankings, indexing, traffic, rich results, citations or AI feature inclusion.