Neo4j Schema Reference

Scope and Validation

• Validated on 2026-01-04 against bolt://localhost:30687 (Neo4j 5.26.4).
• Counts from: CALL db.stats.retrieve('GRAPH COUNTS').
• Types from: CALL apoc.meta.nodeTypeProperties and CALL apoc.meta.relTypeProperties.
• Counts are for orientation only (data coverage), not for logic.

Read This First (No Assumptions)

1. Returns live on relationships, not nodes: PRIMARY_FILER, INFLUENCES, REFERENCED_IN.
2. All timestamps are Strings (Report.created, News.created, Transcript.conference_datetime, Date.*).
3. JSON is stored as Strings (Report.items, Report.exhibit_contents, Report.extracted_sections, News.channels/tags/authors).
4. Numeric-looking fields are Strings (Company.mkt_cap, Company.shares_out, Company.employees, Fact.value).
5. XBRL booleans are Strings: Fact.is_numeric/is_nil, Dimension.is_explicit/is_typed, Unit.is_simple_unit/is_divide.
6. Returns are percentages (5.06 means 5.06%, not 0.0506).
7. INFLUENCES alias trap: using inf as the alias breaks (inf is treated as infinity). Use r.
8. NaN exists in return fields (PRIMARY_FILER: 4 rows; INFLUENCES: 1 row). Filter with isNaN().
9. hourly_stock is usually Float but appears once as LIST OF FLOAT on PRIMARY_FILER; handle list-or-float.
10. Some Facts lack Context: 12,939 Facts have no IN_CONTEXT relationship; exclude unless needed.
11. Period end_date can be string 'null' for period_type='instant' (2,776 rows).

Schema Discovery (Run First If Unsure)

CALL db.labels() YIELD label RETURN label ORDER BY label
CALL db.relationshipTypes() YIELD relationshipType RETURN relationshipType ORDER BY relationshipType
CALL apoc.meta.schema()

Core Labels (Counts + Key Properties)

Event + Company

Market + Calendar

Relationship Overview (Counts + Purpose)

Returns and Impact

Event Content

HAS_SECTION (Report -> ExtractedSectionContent), HAS_EXHIBIT, HAS_FINANCIAL_STATEMENT, HAS_FILING_TEXT, HAS_TRANSCRIPT, HAS_QA_EXCHANGE, HAS_PREPARED_REMARKS, HAS_FULL_TEXT, HAS_QA_SECTION, NEXT_EXCHANGE, IN_CATEGORY (Report -> AdminReport), HAS_SUB_REPORT.

Company Classification and Actions

BELONGS_TO, RELATED_TO (properties: relationship_type, source_ticker, target_ticker, bidirectional), DECLARED_DIVIDEND, HAS_DIVIDEND, DECLARED_SPLIT, HAS_SPLIT, NEXT (Date -> Date).

Returns: What Lives Where (Validated Counts)

INFLUENCES returns depend on target label:

• To Company: total 190,593; daily_stock 188,806; daily_industry 190,363; daily_sector 190,593; daily_macro 190,593.
• To Industry: total 224,561; daily_industry 224,304; daily_stock 0.
• To Sector: total 224,540; daily_sector 224,540; daily_stock 0.
• To MarketIndex: total 224,540; daily_macro 224,540; daily_stock 0.

Rule: Use the Company-target INFLUENCES edge if you want both stock and benchmark returns on one relationship.

• daily_stock only on Company targets: By design, Sector/Industry/MarketIndex edges don't have daily_stock.
• Data gaps: Some Company edges missing daily_stock (News: 1,746; Transcript: 41).

Data Type Alerts and Anomalies (Validated)

• Report.items is JSON string. Example: ["Item 2.02: Results of Operations and Financial Condition", "Item 9.01: Financial Statements and Exhibits"].
• News.channels is JSON string. Example: ["News", "Guidance"].
• Report.created is ISO string with TZ. Example: 2023-01-04T13:48:33-05:00.
• NaN returns exist. Use WHERE r.daily_stock IS NOT NULL AND NOT isNaN(r.daily_stock).
• hourly_stock appears once as list-of-float on PRIMARY_FILER; handle list-or-float.

Indexes and Search Capabilities

• Range/unique indexes: Report.id, News.id, Transcript.id, Company.id.
• Fulltext: abstract_ft (Abstract.label), concept_ft (Concept.label/qname), company_ft (Company.name/displayLabel), exhibit_content_ft (ExhibitContent.content/exhibit_number), extracted_section_content_ft (ExtractedSectionContent.content/section_name), fact_textblock_ft (Fact.value/qname), filing_text_content_ft (FilingTextContent.content/form_type), financial_statement_content_ft (FinancialStatementContent.value/statement_type), full_transcript_ft (FullTranscriptText.content), news_ft (News.title/body/teaser), prepared_remarks_ft (PreparedRemark.content), qa_exchange_ft (QAExchange.exchanges), question_answer_ft (QuestionAnswer.content), search (Memory.name/type/observations).
• Vector: news_vector_index (News.embedding), qaexchange_vector_idx (QAExchange.embedding).
• No index on Company.ticker. Company count is small; exact match scans are acceptable.

Generic fulltext query:

CALL db.index.fulltext.queryNodes($index_name, $query)
YIELD node, score
RETURN labels(node), node.id, score
ORDER BY score DESC
LIMIT 20

IGNORE these Labels and Relationships (0 count)

Labels: AdminSection, FinancialStatement, Guidance, HyperCube, LineItems, Other, Memory, XBRLMapping, ExtractionPattern, ConceptMapping, GeoMapping, ProductMapping, ConceptPattern, GeoPattern, MetricRule, GeoRule, Extraction, Pattern, NewsExtraction, TranscriptExtraction, TestAbstract, TestFact, Framework, ConceptGroup, Benefit, Tutorial, LearningStep, ExampleApplication, Bloom_Perspective, Bloom_Scene, TestNode, TestContentNode_1754141732. Relationships: HAS_EXTRACTION, MAPS_TO_FACT, COULD_MAP_TO_REAL_FACT, POTENTIAL_XBRL_MATCH, TEST_REL, TEST_EDGE, REFERENCES, Bloom_HAS_SCENE.

Validation Queries (Re-run if DB changes)

CALL db.stats.retrieve('GRAPH COUNTS')

CALL apoc.meta.nodeTypeProperties({includeLabels: ['Report','Company','News','Transcript','Fact','Concept','XBRLNode']})
CALL apoc.meta.relTypeProperties({includeRels: ['PRIMARY_FILER','INFLUENCES','REFERENCED_IN','HAS_PRICE']})

MATCH (n:News)-[r:INFLUENCES]->(:Company)
WHERE r.daily_stock IS NULL AND r.daily_industry IS NOT NULL
RETURN count(r) AS news_anomaly_count

MATCH ()-[r:PRIMARY_FILER]->()
WHERE r.hourly_stock IS NOT NULL
RETURN apoc.meta.cypher.type(r.hourly_stock) as t, count(*) as c

Domain Query Skills

For query patterns beyond schema discovery, see these specialized skills:

Usage: Load neo4j-schema + domain skill for targeted queries. Schema provides structure; domain skills provide patterns.

Known Data Gaps

Version 3.4 | 2026-01-11 | Added self-improvement protocol