Module ceb 

CEB adapter (Cardinality Estimation Benchmark, Negi et al.).

Real subset CSV columns (after scripts/fetch_ceb.sh exports the pickle to CSV):

• query_id, subplan_id
• true_rows (ground truth from PostgreSQL EXPLAIN ANALYZE)
• est_rows (PostgreSQL optimiser estimate)
• optional: query_class, template_id

What we extract:

• Cardinality — log10(true_rows / est_rows) per (query_id, subplan_id). This is the only public dataset in our stack with ground-truth cardinalities, which is why we treat its results as the cardinality-mismatch motif’s primary empirical evidence.

What we cannot extract:

• Latency, plan changes over time, contention, cache I/O — CEB is a batch benchmark, not a temporal trace; we synthesise time as t = subplan_index_within_query + query_index * 1.0 so the trace is well-defined.

Structs

Ceb