Semantic compilation
Schema, mapping, migration, plan compilation, and cached reuse
Benchmarks
A reproducible benchmark for the systems thesis
Eight evaluation groups define baselines, controls, source capture, statistics, and the conditions that would falsify CatDB's engineering claims.
Evaluation groups
Execution performance
Native SQL, compiled SQL, federation, materialization, refresh, and provenance
Integration complexity
Mapping count, transformation code, repair surface, and propagation cost
Migration correctness and human study
Seeded semantic faults, recovery, plan correctness, and blinded scoring
Incremental materialization
Update latency, recomputation, freshness, recovery, and equivalence
Distributed behavior
Replay, convergence, conflicts, recovery, staleness, and causal ordering
Versioned workspaces
Storage, checkout, diff, merge, historical queries, and isolation
Federation and communication-aware routing
Planning, latency, transfer, calls, ablation, and prediction error
Results require traceable source data
Every future chart must link to its harness revision, dataset, environment, raw runs, statistical analysis, and failed trials. The protocol makes that evidence contract explicit before measurement.
Read the benchmark planSee the related limitations