Make my program faster experiement

[Maxteabag]

Summary

• preserve the 20-row fast first paint for medium and large query results
• append remaining rows in 2,000-row batches instead of 200-row batches
• keep incremental rendering enabled above 200 rows so the UI remains responsive
• add regression coverage for the first paint and continuation batch sizes

Performance

Controlled local A/B benchmark rendering 10,000 rows (five samples per configuration):

• before, 200 rows per batch: 596.45 ms median
• after, 2,000 rows per batch: 199.25 ms median
• speedup: 2.99x

Verification

• 1,971 passed, 413 skipped in the CI-equivalent non-database test lane
• 25 passed, 6 skipped in the SQLite integration lane
• 8 passed in targeted incremental-rendering and UUID compatibility tests
• Ruff passes for both changed files
• source distribution and wheel build successfully

[Maxteabag]

Agent work metrics

Measured from the completed Codex task telemetry:

• Wall-clock time: 14 minutes 13 seconds
• Total tokens processed: 2,247,939
  • Input tokens: 2,236,234
  • Cached input tokens: 2,148,608 (included in the input total)
  • Output tokens: 11,705
  • Reasoning output tokens: 5,097 (included in the output total)

[Maxteabag]

This is part of an experiment where I test the quality of output based on different prompts. In this prompt I simply said "Make sqlit faster."