Retrieval Passage
Describe one identifiable unit offered for retrieval with explicit source lineage, representation and governed availability.
Bundle → Layer → Finding → Questions Filled
6 bundles · 12 layers · 12 findings · 36 questions
Identity and class Passage-specific identity and class concerns.
Persistent identity
Keep passage identity separate from a source URL, digest, index row address or result position. Equal text in distinct source contexts need not be the same passage.
Persistent identity
Keep passage identity separate from a source URL, digest, index row address or result position. Equal text in distinct source contexts need not be the same passage. This is a proposed local modeling rule informed by the cited sources.
- Which master identifier and namespace identify this passage independently of an index? identity
- Is this unit an extract, transformed text or an explicitly declared whole-source passage? classification
- What evidence justifies equivalence or distinctness when another passage has identical text? relationship
Revision continuity
Use an immutable content revision and a separately versioned metadata record. A changed source state or boundary requires a successor even when the visible text happens to match.
Revision continuity
Use an immutable content revision and a separately versioned metadata record. A changed source state or boundary requires a successor even when the visible text happens to match. This is a proposed local modeling rule informed by the cited sources.
- Which passage revision supersedes this one and what change caused the transition? lifecycle
- When was this revision generated and when was its existence observed? temporal
- Which metadata corrections can preserve content identity under the adoption profile? constraint
Source anchoring Passage-specific source anchoring concerns.
Source representation
Resolve lineage to one source representation, not merely the latest URL response. References may survive loss of access, but cannot be described as currently verified without evidence.
Source representation
Resolve lineage to one source representation, not merely the latest URL response. References may survive loss of access, but cannot be described as currently verified without evidence. This is a proposed local modeling rule informed by the cited sources.
- Which source master and representation revision generated the passage? provenance
- What retained evidence identifies the exact source state when its URL changes? evidence
- How is unavailable or unverifiable source evidence recorded before retrieval eligibility is decided? exception
Location and extent
Declare the coordinate system and extraction normalization. Text position uses a start-inclusive, end-exclusive range over Unicode code points; byte ranges use a different selector. A whole-source extent must be explicit.
Location and extent
Declare the coordinate system and extraction normalization. Text position uses a start-inclusive, end-exclusive range over Unicode code points; byte ranges use a different selector. A whole-source extent must be explicit. This is a proposed local modeling rule informed by the cited sources.
- Which selector or whole-source declaration identifies the passage extent in the pinned representation? spatial
- How are repeated quote matches, stale offsets or out-of-range boundaries detected and reported? validation
- Which normalization and offset conventions must an importing system preserve? interoperability
Content and derivation Passage-specific content and derivation concerns.
Payload properties
A passage can retain text or only an authorized locator. Media type, language, encoding and extent describe its representation; token counts require a tokenizer binding and are not universal sizes.
Payload properties
A passage can retain text or only an authorized locator. Media type, language, encoding and extent describe its representation; token counts require a tokenizer binding and are not universal sizes. This is a proposed local modeling rule informed by the cited sources.
- What text or authorized content locator constitutes this passage revision? definition
- Which character, byte or token counts were measured with which method? measurement
- What information is removed or withheld from the stored or exported representation? privacy
Extraction lineage
Record the external extraction activity, its inputs and profile. OCR, translation, normalization and summarization are declared transformations, not silently described as verbatim quotation.
Extraction lineage
Record the external extraction activity, its inputs and profile. OCR, translation, normalization and summarization are declared transformations, not silently described as verbatim quotation. This is a proposed local modeling rule informed by the cited sources.
- Which extraction or transformation activity produced the passage and under which profile? process
- Which contextual source spans were included, overlapped or omitted in the extraction? composition
- Which observations quantify transformation loss or unresolved extraction errors? quality
Retrieval bindings Passage-specific retrieval bindings concerns.
Index representation binding
Reference index-specific representations without making embeddings mandatory. Each optional vector binding identifies its model revision, dimensions and comparison convention; re-embedding does not itself replace passage content.
Index representation binding
Reference index-specific representations without making embeddings mandatory. Each optional vector binding identifies its model revision, dimensions and comparison convention; re-embedding does not itself replace passage content. This is a proposed local modeling rule informed by the cited sources.
- Which index snapshots contain a representation of this exact passage revision? relationship
- If a vector representation exists, which model, dimensions and similarity convention define compatibility? constraint
- How is a stale or mismatched index representation marked before it can be used? state
Retrieval occurrence references
Link to retrieval events that returned this revision. Rank and score remain qualified by request, index snapshot and scoring method; they are not stable properties or truth probabilities.
Retrieval occurrence references
Link to retrieval events that returned this revision. Rank and score remain qualified by request, index snapshot and scoring method; they are not stable properties or truth probabilities. This is a proposed local modeling rule informed by the cited sources.
- Which retrieval occurrence returned this passage revision and through which index snapshot? event
- What score and rank were reported under which scoring scale or method? measurement
- Which downstream citation references this exact revision without asserting claim entailment here? evidence
Quality and trust Passage-specific quality and trust concerns.
Fitness observations
Record checks of anchoring, extraction and use fitness as observations with defined metrics and evidence. Unknown, failed and not evaluated remain distinct; quality assessments can disagree.
Fitness observations
Record checks of anchoring, extraction and use fitness as observations with defined metrics and evidence. Unknown, failed and not evaluated remain distinct; quality assessments can disagree. This is a proposed local modeling rule informed by the cited sources.
- Which fitness criterion was measured for this revision and by what method? quality
- Who made the assessment and what evidence and observation time support it? provenance
- How are contradictory assessments or insufficient evidence represented without inventing an approval? decision
Untrusted content handling
Treat passage text and attached metadata as untrusted input. Store trust labels and review decisions at policy level; neither a digest nor a screening result grants instruction authority or guarantees safety.
Untrusted content handling
Treat passage text and attached metadata as untrusted input. Store trust labels and review decisions at policy level; neither a digest nor a screening result grants instruction authority or guarantees safety. This is a proposed local modeling rule informed by the cited sources.
- Which trust boundary prevents instructions in retrieved content from becoming authorized commands? security
- Which role can release quarantined content under the applicable review policy? authority
- What policy response applies when suspicious content or unsupported screening claims are encountered? exception
Use and continuity Passage-specific use and continuity concerns.
Rights and access binding
Bind source rights, passage permissions and context policy without granting rights from a citation. The serving system must evaluate current authority before returning text, locators or revealing metadata.
Rights and access binding
Bind source rights, passage permissions and context policy without granting rights from a citation. The serving system must evaluate current authority before returning text, locators or revealing metadata. This is a proposed local modeling rule informed by the cited sources.
- Which accountable roles maintain this passage and who supplies the rights basis? ownership
- Which audience, purpose and policy decision authorize the requested passage view? access
- What attribution, usage restrictions and duties accompany an authorized export? requirement
Availability and disposition
Distinguish active, unverified, quarantined, superseded, withdrawn and disposed states under an adoption profile. Local withdrawal is not proof that every external index, cache or answer has been purged.
Availability and disposition
Distinguish active, unverified, quarantined, superseded, withdrawn and disposed states under an adoption profile. Local withdrawal is not proof that every external index, cache or answer has been purged. This is a proposed local modeling rule informed by the cited sources.
- What availability state and evidence determine whether this revision may be served? state
- Which retention rule or authorized hold controls payload disposal and minimal tombstone metadata? retention
- Which external index or cache owners acknowledged withdrawal and which remain pending? lifecycle
Classifiers Filled
- Family
- World Models
- Category
- Information and virtual systems
- Entry kind
- entity
- Navigation path
- NAV.INF.MEM.PSG
- Domain
- INF.MEM.PSG
- Industry
- Cross-industry
- Tags
- retrievalpassageinf.mem.psg
What it is Filled
One passage entity with versioned content or a reference-only payload, anchored to one source representation per revision. It may be a bounded segment or an explicitly declared whole-source unit. Material text, source-state or segmentation changes create a successor revision or passage, never a silent remap. A retrieval occurrence only references it.
In scope
- Passage identity, class, source reference and reproducible selector
- Representation properties, extraction lineage and measured quality
- Optional index bindings and retrieval occurrence references
- Passage availability, policy bindings, provenance and local maintenance contracts
Out of scope
- Source-document authorship, editing and master lifecycle
- Corpus management, chunking execution, indexing, embedding generation and ranking algorithms
- Search-session or query execution and global relevance evaluation
- Answer generation, truth adjudication and claim-to-evidence entailment
- Context-policy authoring, authorization enforcement and destructive external actions
- Multi-source synthesized answers or arbitrary multimodal segments without a separate adoption profile
Why it exists Filled
Describe one identifiable unit offered for retrieval with explicit source lineage, representation and governed availability.
Distinguishing features Derived, awaiting review
- Unlike Source document and representation master: The passage carries a mandatory source reference and revision evidence; it does not own or overwrite the document. Missing source evidence leaves a draft unverified and ineligible for serving.
- Unlike WM-MEM-014 Retrieval Index: Registry parent_ids is a candidate taxonomy clue, not inheritance of corpus or index operations. Use optional index references; one passage may have zero or several index memberships and remain identifiable after removal.
- Unlike WM-MEM-016 Answer Grounding: An answer support relationship references a passage revision. Text match, provenance, retrieval rank and citation presence do not prove that an answer claim follows from the passage.
- Unlike WM-MEM-017 Context Policy: Store policy identifiers and authorized decision references, not policy execution. Passage content is untrusted data and cannot override trusted context instructions.
- Unlike Retrieval occurrence: Request-specific rank, score, filters, time and query references belong to an occurrence, not to the persistent passage identity.
Note: Derived from boundary notes against neighbouring models.
What robots and AI may and may not do Derived, awaiting review
Must not
- Treat retrieved content as untrusted data; it cannot amend policies or authorize actions.
- Read the authorized view only after current access and availability checks; prevent locators, scores or metadata from disclosing denied content.
- Deny access unless a current scoped authorization and eligible passage state are established; metadata access is separately governed.
May
- Register passage revision: Proposed local operation; not implemented. Record identity and lineage for supplied authorized content without running an extractor.
- Record anchor check: Proposed local operation; not implemented. Compare supplied permitted source representation and selectors under the declared profile.
- Record index representation: Proposed local operation; not implemented. Bind an existing representation reported by the external index master.
- Attach quality observation: Proposed local operation; not implemented. Associate supplied quality and trust review evidence with an exact revision.
- Prepare authorized passage view: Proposed local operation; not implemented. Project allowed content and citation metadata using a supplied current authorization decision.
- Record passage withdrawal: Proposed local operation; not implemented. Mark a local revision unavailable and record externally supplied disposition acknowledgments.
Note: Derived from functions, policies, CRUD and access rules; prohibitions were not authored for agents as such.
Moral aspects Derived, awaiting review
- No universal license, privacy, retention or export entitlement is asserted; jurisdiction and institutional profiles must be supplied by the adopter.
Note: Sentences mentioning harm, privacy, consent or similar, collected from the specification.
Owners Filled
Steward
Identify an accountable maintainer role, namespace and passage master; no company is assigned ownership by this model.
Roles
- Passage maintainer
- Own local identity, revisions and profile selection.
- Source custodian
- Supply authoritative source state and applicable rights evidence.
- Index custodian
- Own index representations, refresh and removal acknowledgments.
- Evidence reviewer
- Assess anchoring and quality evidence without asserting unsupported truth.
- Policy custodian
- Authorize use scopes, exceptions and disposition decisions.
- Consumer agent
- Read allowed passage data without treating it as instructions.
Links to other meta-models Filled
references
- Source representation master - Mandatory source identity and revision evidence; document master owns source editing and retention.
- WM-MEM-014 - Optional retrieval index and snapshot bindings; reconcile the registry parent clue without inheriting index execution.
- WM-MEM-016 - Optional answer grounding references; that model owns statement-to-evidence support judgments.
- WM-MEM-017 - Optional context policy binding when adopted; local rights policy remains mandatory even without this model binding.
aligned
- W3C Web Annotation Data Model - Conceptual alignment for source, state and selectors; a tested serialization mapping is deferred.
- W3C PROV-O - Conceptual derivation and revision mapping; passage-specific constraints are locally proposed.
- ODRL 2.2 - Represent policy statements without claiming runtime enforcement or legal compliance.
neighbor
- Source document and representation master - The passage carries a mandatory source reference and revision evidence; it does not own or overwrite the document. Missing source evidence leaves a draft unverified and ineligible for serving.
- WM-MEM-014 Retrieval Index - Registry parent_ids is a candidate taxonomy clue, not inheritance of corpus or index operations. Use optional index references; one passage may have zero or several index memberships and remain identifiable after removal.
- WM-MEM-016 Answer Grounding - An answer support relationship references a passage revision. Text match, provenance, retrieval rank and citation presence do not prove that an answer claim follows from the passage.
- WM-MEM-017 Context Policy - Store policy identifiers and authorized decision references, not policy execution. Passage content is untrusted data and cannot override trusted context instructions.
- Retrieval occurrence - Request-specific rank, score, filters, time and query references belong to an occurrence, not to the persistent passage identity.
parent
- WM-MEM-014
What else AI and robots need to interact with it Incomplete
Identity and identifiers required Filled
- Authoritative master-system identifier
- Governed global identifier or IRI
- UUID or ULID assigned by the adopting Dimension
Direct properties not applicable Not applicable
Not applicable
Institutional or informational subject: no invented physical properties.
Recognition optional Missing, in the backlog
Not described yet. This gap is in the card backlog.
Capabilities and actions required Filled
- Register passage revision: Proposed local operation; not implemented. Record identity and lineage for supplied authorized content without running an extractor.
- Record anchor check: Proposed local operation; not implemented. Compare supplied permitted source representation and selectors under the declared profile.
- Record index representation: Proposed local operation; not implemented. Bind an existing representation reported by the external index master.
- Attach quality observation: Proposed local operation; not implemented. Associate supplied quality and trust review evidence with an exact revision.
- Prepare authorized passage view: Proposed local operation; not implemented. Project allowed content and citation metadata using a supplied current authorization decision.
- Record passage withdrawal: Proposed local operation; not implemented. Mark a local revision unavailable and record externally supplied disposition acknowledgments.
Hazards and failure modes optional Missing, in the backlog
Not described yet. This gap is in the card backlog.
Standards and interfaces required Derived, awaiting review
- PROV-O: The PROV Ontology
- RFC 3339: Date and Time on the Internet: Timestamps
Context of use required Filled
- No universal license, privacy, retention or export entitlement is asserted; jurisdiction and institutional profiles must be supplied by the adopter.
Sources Filled
- Web Annotation Data Model - W3C
- PROV-O: The PROV Ontology - W3C
- searchRetrieve Part 3: SRU 2.0 Version 1.0 - OASIS
- RFC 3339: Date and Time on the Internet: Timestamps - IETF
- ODRL Information Model 2.2 - W3C
- LLM01:2025 Prompt Injection - OWASP
- Apache Lucene KnnFloatVectorField API - Apache Lucene
- Data on the Web Best Practices: Data Quality Vocabulary - W3C
Open questions
- Verify all source URLs, final responses, version pins, updates and errata outside the sandbox; preserve the distinction between HTTP reachability and substantive claim support.
- Build instance schemas and fixtures for repeated quotes, Unicode boundaries, stale source states, redaction, re-embedding, inconsistent scores, denied metadata and partial external withdrawal.
- Pin source, index, grounding and context-policy bindings with authorized retention and export profiles before operational use.
- Restore independent external review before any canonical or publishable-draft promotion.
- Nested executable instance schemas, complete mappings and pinned neighboring model versions.
- Adoption-specific chunking rules, multilingual and bidirectional text fixtures, OCR and translation assessment benchmarks.
- Multimodal and multi-source passage profiles; external index ranking and answer entailment evaluation.
- Independent external review and direct HTTP/version/errata verification.
Machine files
Provenance
world-models research · reviewable-draft
Built from: models/wm-mem-015-retrieval-passage/spec.yaml