May 1, 2026 · Accepted · parameter-encoding-and-binding, measurement-system

Experiment 116: Wide batch insert release coverage

Date: 2026-05-01

Status: Accepted

Direction:parameter-encoding-and-binding, measurement-system

Problem

Experiment 113 showed that batch parameter row width is a real write-path

dimension. The existing release write suite still only tracked the narrow

two-parameter insert shape:

 INSERT INTO t(name, value) VALUES (?, ?) 

That narrow shape is useful, but it is not a sufficient guard for generated

statements or ORM-style sync writes that bind many columns per row. Without a

wide release-suite row, future parameter-encoding regressions could land while

the public write dashboard still looked neutral.

Hypothesis

Adding a single 10,000-row x 20-parameter mixed-type batch insert to

Write Performance will preserve the current narrow batch metrics while making

wide batch behavior visible in release results and on the experiments

dashboard.

Accept if the workload runs across all four peers through the existing

BenchmarkPeer.executeBatch path, uses identical schema and parameter values,

and adds a clear metric key without expanding the release suite into every

focused-benchmark size/width combination.

Approach

benchmark/suites/writes.dart now adds one subsection:

 Wide Batch Insert (10000 rows x 20 params) 

The table shape mirrors experiment 113's mixed parameter matrix:

• 20 bound parameters per row.
• Column type cycle: TEXT, INTEGER, REAL, BLOB.
• 10,000 rows per timed executeBatch.
• Same warmup and iteration policy as the existing write suite.

benchmark/drift/writes_db.dart now registers the matching wide_batch table

so the drift peer participates through the same BenchmarkPeer.executeBatch

path as the other peers. The curated metric registry tracks the resqlite row as

wide batch 10K x20 under the write chart.

Results

Command:

 /Users/dan/Coding/flutter_arm64/bin/dart run benchmark/suites/writes.dart 

Wide batch subsection:

The narrow batch sections still execute normally in the same run:

Validation:

 /Users/dan/Coding/flutter_arm64/bin/dart pub get /Users/dan/Coding/flutter_arm64/bin/dart run build_runner build --delete-conflicting-outputs /Users/dan/Coding/flutter_arm64/bin/dart analyze --fatal-infos /Users/dan/Coding/flutter_arm64/bin/dart run benchmark/suites/writes.dart 

Decision

Accepted — measurement coverage.

This experiment intentionally adds no production optimization. Its value is

that the release suite now tracks the exact workload dimension experiment 113

proved was missing: parameter width. Future batch-parameter work can compare

both the common two-parameter insert and the generated-statement-shaped

20-parameter insert without relying on a focused script alone.

Future Notes

Do not add the full 2 / 8 / 20 parameter-width matrix to release mode unless a

future regression needs that granularity. The release suite should stay compact;

one wide mixed-type row is enough to keep the public dashboard honest while

focused experiments remain free to sweep width and row-count combinations.