All runs

20260831062427-083dfb94

complete Signed by eimu · 3d0163a12475...63d7e589

Started: 2026-08-31 06:24:27 UTC
Generated: 2026-08-31 06:25:10 UTC
Measured samples: 144
Combinations: 1

Machine

Captured from the runner's system manifest.

Host eimu Linux · kernel 6.19.6

PHP 8.5.3

Processor AMD Ryzen 9 5950X 16-Core Processor 32 cores · x86_64

Memory 31.2 GB DIMM_A2 16 GiB 2666 MT/s DDR4 F4-4266C19-16GVK DIMM_B2 16 GiB 2666 MT/s DDR4 F4-4266C19-16GVK

Disk 3.5 TB Device: /dev/nvme0n1 Capacity: 3.5 TB

Runtime envelope

End-to-end time per scenario run across all non-warmup samples on sqlite-memory.

Min sample: 10.6 ms
Mean sample: 86.0 ms
Max sample: 330 ms
Cumulative time: 12.384s
Throughput: 11.628/s

Runtime comparison

Mean time per scenario run in this upload, warmup runs excluded, sorted by runtime.

Init Insert Read Update Delete

DivergenceV3 3.2.0 sqlite-memory 86.0 ms

Where the time goes

Each bar is one framework's mean scenario run on sqlite-memory, split into its phases and sorted by total. Bars share a scale within each engine. Hover a bar for the numbers.

Init Insert Read Update Delete

Mean split across 1 combinations

Insert 22% Read 19% Update 53%

sqlite-memory

DivergenceV33.2.0 86.0 ms

Exact numbers


DivergenceV3 3.2.0	sqlite-memory	144	86.0 ms	19.2 ms	16.7 ms	45.3 ms	3.97 ms

Scenario families

Mean across every framework and engine in this upload.

simple 11.2 ms
string_fixed 73.4 ms
float_fixed 74.3 ms
string_variable 81.8 ms
int_fixed 91.3 ms
canary 134 ms

What the workloads measure

The row shape and operation mix decide where each framework spends its time.

Scenario families

simple baseline row shape: Shows framework overhead with minimal field complexity, so object creation and query setup dominate.
canary mixed field and nullable stress: Wide mixed records amplify casting, hydration, dirty tracking, and adapter metadata overhead.
int_fixed pure scalar writes: A low-noise floor. Slow results here usually mean framework machinery is expensive even before text and cast-heavy payloads show up.
float_fixed numeric conversion paths: Decimal and float normalization can widen update and verification cost, especially for ORMs that compare serialized values.
string_fixed predictable text hydration: Column count and text assignment matter more than payload variability, which makes hydration strategy easier to isolate.
string_variable application-like text workload: Variable strings and note fields make this the closest to real CRUD traffic and usually punish heavier active-record layers the most.

Operation shapes

Warmup vs hot runs: Run 1 is warmup. Hot-path averages exclude it so boot and metadata setup do not dominate steady-state rankings.
Insert and indexed schemas: Indexed tables make writes cost more because every insert and many updates must maintain secondary indexes as well as the row itself.
Read all vs read probes: Read All rewards bulk hydration efficiency; random and tail probes expose per-query setup, lookup overhead, and tiny-result inefficiency.
Update with verification: Each update shape is followed by a verification read, so dirty-checking, write SQL generation, hydration, and comparison all affect the final time.
Delete with confirmation: Deletes are checked afterward. Frameworks that pay extra for follow-up lookups or teardown lose more time here than a bare SQL path.
Total runtime: The total runtime metric is effectively a benchmark of the machine plus adapter behavior across the full workload, not just one CRUD verb.