ila

Guidance for writing, rewriting, reviewing, or outlining database systems research papers for VLDB, SIGMOD, PVLDB, CIDR, and similar venues. Use when working on paper structure, introductions, contributions, research questions, evaluation sections, figures, tables, related work, limitations, or reviewer-facing clarity for systems papers.

Multi-perspective code review of the most recent commit. Launches 5 parallel review agents covering correctness, reliability, performance, simplicity, and style. Reports findings with severity and confidence scores.

Structured feature planning with interrogation rounds and confirmation gate. Use when starting a non-trivial feature to align on requirements, scope, and edge cases before writing code.

Subsystem-by-subsystem codebase audit. Discovers source directories, reviews each from 5 perspectives, and produces prioritized findings. Use for periodic quality sweeps or before releases.

Behavioral guidelines to reduce common LLM coding mistakes. Use when writing, reviewing, or refactoring code to avoid overcomplication, make surgical changes, surface assumptions, and define verifiable success criteria.

Reference for SIMD, vectorized query execution, and compiled query execution in database systems. Auto-loaded when discussing SIMD intrinsics, vectorization, query compilation, DuckDB execution model, pipeline breakers, morsel-driven parallelism, vector-at-a-time processing, or data-centric code generation.

Reference for data lake storage formats and table formats — Parquet, Delta Lake, Apache Iceberg, Apache Hudi, DuckLake, Lance, F3 (Faster Frozen Format), ORC, and their interaction with DuckDB. Auto-loaded when discussing storage formats, data lakes, table formats, Parquet, Iceberg, Delta Lake, DuckLake, or change data capture.

Generic development rules for DuckDB extension projects. Auto-loaded when writing code, reviewing changes, or discussing development practices. Covers testing, code style, git safety, debugging, and collaboration.

Deep research reference on view matching (answering queries using views, materialized-view query rewriting, subsumption-based rewriting) spanning classical PODS/SIGMOD theory, the Goldstein-Larson matcher, Calcite's implementation, commercial systems (Oracle, SQL Server, DB2, Snowflake, BigQuery, Redshift, Databricks, Firebolt, ClickHouse), adjacent result/semantic caching, and Microsoft/Google computation-reuse work. Auto-loaded when discussing view matching, query rewriting with MVs, subsumption, cache-aware optimization, or integrating a matcher into OpenIVM.

Reference for Z-sets — the integer-weighted multiset algebra at the heart of DBSP, differential dataflow, and OpenIVM's delta tables. Focused on the algebra itself (definition, group structure, operators) and on its lineage (Green/Tannen K-relations, Koch's Z-relations, Naiad/differential dataflow). Mirrors the structure of Feldera's `01-zsets-and-dbsp.md` worked example. For DBSP system details see `explain-related-systems`; for the broader IVM operator theory see `explain-ivm`; for compilation/cost work see `incrementalization-strategies`.

Comprehensive reference for Incremental View Maintenance (IVM) algebraic rules, delta rules for all relational operators, multiplicity approaches, efficiency techniques, and correctness considerations. Auto-loaded when discussing IVM theory, delta computation, view maintenance algorithms, or operator incrementalization.

Reference material for IVM-related systems and papers — DBSP, DBToaster, pg_ivm, Snowflake Dynamic Tables, Materialize. Auto-loaded when discussing related work, comparisons, or IVM literature.

Reference for SQL MERGE statement and its application to IVM upsert patterns. Auto-loaded when discussing upserts, delta application, MERGE, INSERT OR REPLACE, or the CompileAggregateGroups/CompileSimpleAggregates/CompileProjectionsFilters functions.

Deep research reference on incrementalization strategies for IVM — higher-order deltas, factorized IVM, compilation-based approaches, algebraic optimizations, cost models, constraint exploitation, and emerging techniques. Auto-loaded when discussing how to make IVM faster, delta optimization, auxiliary views, higher-order IVM, F-IVM, CROWN, compilation for IVM, IVM cost models, or research directions.

Reference for dbt (data build tool) and LinkedIn Coral — SQL transformation frameworks relevant to IVM. Auto-loaded when discussing dbt models, incremental models, Coral SQL translation, cross-engine SQL compilation, or data transformation pipelines.

Research map for SIDRA-related literature: decentralized analytics, personal data stores, local-first systems, materialized view maintenance, streaming DP, SQL DP, PAC privacy, secure computation, and privacy-preserving database systems.

PAC Privacy and PAC DuckDB extension reference. Auto-loaded when discussing PAC privacy, instance-based privacy, membership inference resistance, mutual-information bounds, PAC SQL rewriting, or integrating PAC/privacy extension ideas into SIDRA.

SIDRA project and paper reference. Auto-loaded when working on SIDRA SQL, decentralized tables/materialized views, staging-to-central flushes, privacy annotations, PDS/client-server flow, or the SIDRA paper in SIDRA.pdf.