Insights
Release notes, research reads, and field notes from building GraphForge and Verdant.
Insights collects Curate Labs writing on knowledge extraction, graph engines, and evaluation: release notes from GraphForge and Verdant, reads from the document-to-graph research literature, and notes from client work.
Latest insights
Agentic Coding Is Creative Destruction, Not a Trap
Agentic coding is not a trap. It is creative destruction: a production shift that will expose bad process, change expertise, and expand who can build software.
Community Reading: GLiNER-Relex for Zero-Shot Joint Extraction
GLiNER-Relex extends the GLiNER family into a unified zero-shot NER and relation extraction model.
WACCY v0.1.0: A Three-Statement Modeling Slice
WACCY v0.1.0 is the first released vertical slice of our small-business financial modeling stack: extraction contracts, validated datasets, model builders, and exporters.
GraphForge v0.4.0: Three-Surface API
GraphForge v0.4.0 completed the three-surface architecture: Cypher execution, graph algorithms, and hybrid search.
GraphForge v0.3.10: Analytics Integration
GraphForge v0.3.10 made the engine easier to integrate with analytical and agentic graph workflows.
GraphForge v0.3.9: Performance
GraphForge v0.3.9 kept full TCK compliance while delivering performance improvements across the stack.
GraphForge v0.3.8: Full openCypher TCK Compliance
GraphForge v0.3.8 reached 100% openCypher TCK compliance with 3,885 of 3,885 scenarios passing.
GraphForge v0.3.7: Function Coverage
GraphForge v0.3.7 expanded function coverage and moved reported TCK compliance to 3,235 of 3,885 scenarios.
GraphForge v0.3.6: TCK Harness and Temporal Fixes
GraphForge v0.3.6 was a correctness release that moved TCK coverage from 1,694 to 2,507 passing scenarios.
GraphForge v0.3.4: Operators and String Functions
GraphForge v0.3.4 filled in operator and string-function gaps, raising reported feature completeness to 92%.
GraphForge v0.3.5: Math, Aggregation, and TCK Steps
GraphForge v0.3.5 added sqrt, rand, pow, aggregation test fixes, and missing TCK step definitions.
GraphForge v0.3.1: Predicate Functions
GraphForge v0.3.1 completed predicate functions and improved NULL-aware list and property testing.
GraphForge v0.3.2: List Operations
GraphForge v0.3.2 completed list functions, adding filter, extract, and reduce with proper scoping and NULL handling.
GraphForge v0.3.3: Pattern and CALL Features
GraphForge v0.3.3 pushed deeper into expressive graph queries with pattern predicates, CALL subqueries, and pattern comprehension.
GraphForge v0.3.0: Major Cypher Feature Release
GraphForge v0.3.0 was a major query-language release that substantially expanded openCypher compatibility.
Community Reading: DOREMI for Long-Tail DocRE
DOREMI focuses on the deployment problem of rare relation performance rather than proposing another relation model.
GraphForge v0.2.1: Dataset Loading
GraphForge v0.2.1 made it easier to start analysis quickly by adding cached loading for real-world SNAP graph datasets.
GraphForge v0.2.0: Core Cypher Complete
GraphForge v0.2.0 moved the query language from early capability toward practical core Cypher coverage.
GraphForge v0.1.4: CI/CD Excellence
GraphForge v0.1.4 made the project faster to iterate on by moving coverage feedback and test analytics closer to the developer.
GraphForge v0.1.2: Infrastructure Improvements
GraphForge v0.1.2 focused on the professional project infrastructure needed for a graph database to mature in public.
Community Reading: CoDe-KG and Sentence Complexity
CoDe-KG treats scientific KG construction as a pipeline of coreference, simplification, decomposition, and relation extraction.
Community Reading: GRAPHTREX for Clinical Temporal RE
GRAPHTREX uses span extraction and heterogeneous graph transformers to model long-range temporal relations in clinical notes.
Community Reading: KGGen for Text-to-KG Construction
KGGen pairs an LLM text-to-KG pipeline with a benchmark focused on graph usefulness.
Community Reading: Graph-DPEP for Few-Shot DocRE
Graph-DPEP makes prompted document relation extraction more structured by decomposing relation types and revisiting missed pairs.