What did each step change in the engine?

Which step moved what

Step each change added, in points of the baseline value: −13 pts means the metric dropped by 13 % of what it was on the baseline. Everything is a cost, so down is better. An empty cell: the step did not measure that metric.

A step inside the run-to-run range of the two columns it compares, or under 3 points, is noise on this machine and is shown as ≈. Nothing here blocks a change; the page only shows its effect.

Engine cost after each step

Left to right: the baseline, then each step in the order it was recorded; every column includes the ones before it. Above each bar: the value, how many times lower than the baseline, and the step from the previous column.

Robustness checks

Each check runs a failure against the real engine on a fresh state file and looks at what is left.

What needs a look

Every check that fails on the last step, and every step that raised a cost beyond noise.

Table

Median of the repeats; the range is in the tooltip of each cell. Green: the value meets the design target of the last row.

Volumes

bench_volumes.py stores molecule payloads two ways and times the same reads on both: SQLite blobs (RT) against append-only volumes (serial: HT no fsync, H fsync per shard; parallel: VP the parent writes, VC one segment per chunk, PJ a real ms_flow job whose rows carry Payloads, PJZ/PJD the same with zstd, without and with a dictionary per series; PS the real PayloadStore with one fsync per commit, PSN the same without it). Warm reads run right after the import, with the data in the page cache; cold ones run after fsync and posix_fadvise(DONTNEED) on every file, so they come from the disk. A blank cell was not measured (the 2026-10-15 run has no cold reads). Plan: 08-plan-volumenes.md.

How this was measured

Workload. Empty tasks, so what is timed is the orchestration alone: with real tasks the same cost is added on top of the task time. Every run starts from a new state file and a temporary HOME.

Reading it. The numbers are from one machine and move a few percent between runs; compare columns of this page with each other, not with another machine.

Record a step: python -m benchmarks.refactor_optimization.bench --label <step>, add its line to STEPS in report.py, then python -m benchmarks.refactor_optimization.report.