Apr 23, 2026 · Rejected

Experiment 096: Direct batch parameter encoding

Date: 2026-04-23

Status: Rejected

Problem

executeBatch flattens List<List<Object?>> into a temporary Dart

List<Object?> before encoding native parameter structs. For large batches,

that extra list is avoidable.

Hypothesis

Writing native parameter structs directly from the nested parameter sets should

reduce large batch write overhead by avoiding the intermediate flattened list.

Approach

Added nested-param allocation/free helpers and routed both top-level

executeBatchWrite and nested transaction batch writes through them. The C

batch API was unchanged. The experiment was run in isolation and then reverted.

Results

Artifacts:

• benchmark/profile/results/exp096-direct-batch-param-encoding-dispatch.log
• benchmark/results/2026-04-23T19-28-04-exp096-direct-batch-param-encoding.md
• benchmark/results/2026-04-23T19-28-04-exp096-direct-batch-param-encoding.json

Focused dispatch did not move:

The full release comparison reported 0 wins, 1 regression, 152 neutral.

Large batch medians trended in the right direction in the structured artifact,

but the harness did not classify the improvement as a reliable win and the code

added a substantial second parameter-encoding path.

Decision

Rejected.

Avoiding the flattened list is a plausible cleanup only if batch payload

construction becomes a proven bottleneck. With the current suite, the added

allocator complexity is not justified.