# zoo-bench performance report

raw data: `results/zoo-framework-0.9.0/20261011T092749Z.json`

## Run environment

| item | value |
|---|---|
| CPU model | AMD EPYC 7763 64-Core Processor |
| logical cores | 4 |
| platform | Linux-x86_64 |
| operating system | Linux 6.17.0-1022-azure (#22-Ubuntu SMP Mon Jul 27 17:24:03 UTC 2026) |
| Python | CPython 3.13.16 |
| interpreter | /opt/hostedtoolcache/Python/3.13.16/x64/bin/python |
| GIL | default |
| framework under test (distribution metadata) | 0.9.0 |
| framework under test (module __version__) | 0.9.0 (footnote only, not the version criterion) |
| drive-surface generation of the framework under test | current-generation drive surface (scheduling core + scheduling model; the completion signal comes from the result reactor) |
| harness | zoo-bench 0.0.1 @ 9a42be24ac43f75e2e5d7b16434f91ba3d0717b6 |
| command to reproduce | `zoo-bench run --framework zoo-framework==0.9.0` |

> dist_version (the distribution metadata) is the source of truth; module_version is only a footnote

## Workload

- composition: one unit of work = JSON serialize + deserialize + string concatenation, repeated until the target duration is reached; the tiers are declared by body_tiers_us in matrix.yaml
- **the workload is a stand-in, not a real trace**: its shape resembles business actions such as 'take a frame of data -> serialize -> parse -> transform strings', but this repository holds no real trace, so every conclusion is conditional on that

## Measurement methodology

- **the body duration is instrumented within the same run**: the body measures itself internally and reports back; framework overhead = end-to-end - it. **No cross-run subtraction**  -  two runs differ in state (caches, frequencies, scheduling noise), so subtracting introduces systematic bias.
- **warmups are excluded from statistics**: each unit first runs 3 warmup rounds that are discarded, then only the following 30 measured rounds are counted; the sample count equals 'measured rounds x concurrency'.
- **absolute durations are not comparable across runs**: they only show the order of magnitude; across versions/machines read the **relative quantities within the same run** (overhead share, multiples against the baselines), which are insensitive to overall speed.

## Conclusion summary

- Once the framework under test zoo runs a body of about 2700 microseconds or more, its framework overhead drops below 15% of end-to-end; below that tier, the main price of choosing it is the framework overhead itself.
- At concurrency 1 and a 40 microsecond body, the framework under test is 1.83x faster than process_pool (the fastest measured tier at that concurrency).
- At concurrency 1, once the body exceeds about 40 microseconds, apscheduler is no longer faster than the framework under test.
- At concurrency 4, once the body exceeds about 40 microseconds, apscheduler is no longer faster than the framework under test.
- At concurrency 16, once the body exceeds about 40 microseconds, apscheduler is no longer faster than the framework under test.
- At concurrency 64, once the body exceeds about 40 microseconds, apscheduler is no longer faster than the framework under test.
- At concurrency 1, within the measured tiers asyncio_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 4, within the measured tiers asyncio_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 16, within the measured tiers asyncio_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 64, within the measured tiers asyncio_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 1, within the measured tiers bare_thread is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 4, within the measured tiers bare_thread is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 16, within the measured tiers bare_thread is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 64, within the measured tiers bare_thread is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 1, once the body exceeds about 40 microseconds, process_pool is no longer faster than the framework under test.
- At concurrency 4, within the measured tiers process_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 16, within the measured tiers process_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 64, within the measured tiers process_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 1, within the measured tiers thread_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 4, within the measured tiers thread_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 16, within the measured tiers thread_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.
- At concurrency 64, within the measured tiers thread_pool is **always faster** than the framework under test  -  tiers like these are exactly the openly published unfavourable data of this report.

> a single-point speedup has no selection meaning; the conclusion is phrased as 'which scheme to choose at what body duration' (the same framework overhead is 70% of a 40 microsecond task and a few percent of a 10 millisecond one)

### Overhead threshold crossings

| scheme | concurrency | threshold | first tier at or below the threshold (microseconds) | measured tiers (microseconds) | note |
|---|---|---|---|---|---|
| apscheduler | 1 | 15.00% | 2700 | 40, 300, 2700, 10000 | this is the smallest measured tier that satisfies the condition; the true crossing point lies between it and the previous tier |
| apscheduler | 4 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| apscheduler | 16 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| apscheduler | 64 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| asyncio_pool | 1 | 15.00% | 2700 | 40, 300, 2700, 10000 | this is the smallest measured tier that satisfies the condition; the true crossing point lies between it and the previous tier |
| asyncio_pool | 4 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| asyncio_pool | 16 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| asyncio_pool | 64 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| bare_thread | 1 | 15.00% | 2700 | 40, 300, 2700, 10000 | this is the smallest measured tier that satisfies the condition; the true crossing point lies between it and the previous tier |
| bare_thread | 4 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| bare_thread | 16 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| bare_thread | 64 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| process_pool | 1 | 15.00% | 2700 | 40, 300, 2700, 10000 | this is the smallest measured tier that satisfies the condition; the true crossing point lies between it and the previous tier |
| process_pool | 4 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| process_pool | 16 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| process_pool | 64 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| thread_pool | 1 | 15.00% | 300 | 40, 300, 2700, 10000 | this is the smallest measured tier that satisfies the condition; the true crossing point lies between it and the previous tier |
| thread_pool | 4 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| thread_pool | 16 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| thread_pool | 64 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| zoo | 1 | 15.00% | 2700 | 40, 300, 2700, 10000 | this is the smallest measured tier that satisfies the condition; the true crossing point lies between it and the previous tier |
| zoo | 4 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| zoo | 16 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |
| zoo | 64 | 15.00% | - | 40, 300, 2700, 10000 | this concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values). End-to-end, throughput and the body's self-reported values remain valid, so they are given as usual. |

### Relative turning points

| baseline | concurrency | first tier where it is no longer faster (microseconds) | note |
|---|---|---|---|
| apscheduler | 1 | 40 | this is the smallest measured tier that satisfies the condition |
| apscheduler | 4 | 40 | this is the smallest measured tier that satisfies the condition |
| apscheduler | 16 | 40 | this is the smallest measured tier that satisfies the condition |
| apscheduler | 64 | 40 | this is the smallest measured tier that satisfies the condition |
| asyncio_pool | 1 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| asyncio_pool | 4 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| asyncio_pool | 16 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| asyncio_pool | 64 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| bare_thread | 1 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| bare_thread | 4 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| bare_thread | 16 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| bare_thread | 64 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| process_pool | 1 | 40 | this is the smallest measured tier that satisfies the condition |
| process_pool | 4 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| process_pool | 16 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| process_pool | 64 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| thread_pool | 1 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| thread_pool | 4 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| thread_pool | 16 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |
| thread_pool | 64 | - | **not observed**: within the measured tiers this baseline is always faster than the framework under test |

## Charts

![overhead_ratio](figures/en-overhead_ratio.svg)

> Body tiers at concurrency 4 only; ratios hold within a single run. The overhead numbers of concurrency 16, 64 were withheld (beyond this machine's parallelism; see the caliber section) and are not charted

![throughput](figures/en-throughput.svg)

> Body tier 40 microseconds only; throughput holds within a single run

![relative_multiple](figures/en-relative_multiple.svg)

> One panel per measured concurrency (4 in total); multiples hold within a single run

## Dimension: Latency percentiles and jitter

> end-to-end / concurrency; percentiles and relative spread are given in the same table  -  a metric that only gives the median is not acceptable

| concurrency | body tier (microseconds) | zoo | apscheduler | asyncio_pool | bare_thread | process_pool | thread_pool |
|---|---|---|---|---|---|---|---|
| 1 | 40 | 182.57 microseconds | 288.52 microseconds | 140.37 microseconds | 148.91 microseconds | 333.53 microseconds | 75.77 microseconds |
| 1 | 300 | 449.17 microseconds | 542.76 microseconds | 386.47 microseconds | 384.68 microseconds | 607.37 microseconds | 322.77 microseconds |
| 1 | 2700 | 2.754 milliseconds | 2.999 milliseconds | 2.752 milliseconds | 2.672 milliseconds | 3.009 milliseconds | 2.689 milliseconds |
| 1 | 10000 | 10.066 milliseconds | 9.971 milliseconds | 9.482 milliseconds | 9.648 milliseconds | 9.888 milliseconds | 9.361 milliseconds |
| 4 | 40 | 206.14 microseconds | 256.36 microseconds | 173.04 microseconds | 199.87 microseconds | 168.78 microseconds | 96.03 microseconds |
| 4 | 300 | 485.26 microseconds | 511.06 microseconds | 407.48 microseconds | 419.42 microseconds | 304.29 microseconds | 350.77 microseconds |
| 4 | 2700 | 2.946 milliseconds | 2.983 milliseconds | 2.755 milliseconds | 2.706 milliseconds | 1.385 milliseconds | 2.742 milliseconds |
| 4 | 10000 | 10.106 milliseconds | 10.258 milliseconds | 9.867 milliseconds | 9.696 milliseconds | 4.621 milliseconds | 9.813 milliseconds |
| 16 | 40 | 205.75 microseconds | 224.01 microseconds | 134.08 microseconds | 167.78 microseconds | 149.92 microseconds | 90.21 microseconds |
| 16 | 300 | 470.57 microseconds | 505.14 microseconds | 393.00 microseconds | 424.88 microseconds | 278.51 microseconds | 343.15 microseconds |
| 16 | 2700 | 2.899 milliseconds | 2.980 milliseconds | 2.802 milliseconds | 2.711 milliseconds | 1.370 milliseconds | 2.772 milliseconds |
| 16 | 10000 | 10.084 milliseconds | 9.881 milliseconds | 9.935 milliseconds | 9.609 milliseconds | 4.741 milliseconds | 9.803 milliseconds |
| 64 | 40 | 207.28 microseconds | 220.85 microseconds | 120.26 microseconds | 171.52 microseconds | 194.08 microseconds | 89.21 microseconds |
| 64 | 300 | 491.01 microseconds | 501.75 microseconds | 384.40 microseconds | 424.30 microseconds | 299.76 microseconds | 350.35 microseconds |
| 64 | 2700 | 2.953 milliseconds | 2.965 milliseconds | 2.811 milliseconds | 2.675 milliseconds | 1.364 milliseconds | 2.696 milliseconds |
| 64 | 10000 | 10.538 milliseconds | 10.167 milliseconds | 10.407 milliseconds | 9.827 milliseconds | 4.686 milliseconds | 10.244 milliseconds |

per-unit values (percentiles, spread, raw per-round samples) are in 'Appendix: all values' at the end.

## Dimension: Framework overhead share

> overhead = end-to-end / concurrency - the body duration measured within the same run (no cross-run subtraction). **cells with -**: that concurrency exceeds this machine's parallel capacity; the body's self-reported duration contains scheduling waits from oversubscription and is not comparable with the row's end-to-end per task, so no overhead figure is given (end-to-end and throughput live in their own dimensions)

| concurrency | body tier (microseconds) | zoo | apscheduler | asyncio_pool | bare_thread | process_pool | thread_pool |
|---|---|---|---|---|---|---|---|
| 1 | 40 | 71.66% | 79.81% | 67.08% | 64.13% | 83.82% | 41.38% |
| 1 | 300 | 32.12% | 44.93% | 26.24% | 25.34% | 48.87% | 11.10% |
| 1 | 2700 | 5.65% | 9.02% | 4.58% | 4.02% | 11.37% | 2.35% |
| 1 | 10000 | 2.55% | 3.19% | 1.54% | 1.25% | 3.45% | 0.67% |
| 4 | 40 | 73.26% | 78.08% | 68.70% | 70.52% | 67.43% | 48.25% |
| 4 | 300 | 36.28% | 39.64% | 25.85% | 28.21% | - | 14.92% |
| 4 | 2700 | 7.68% | 8.71% | 4.45% | 4.05% | - | 2.85% |
| 4 | 10000 | - | - | - | - | - | - |
| 16 | 40 | - | - | - | - | - | - |
| 16 | 300 | - | - | - | - | - | - |
| 16 | 2700 | - | - | - | - | - | - |
| 16 | 10000 | - | - | - | - | - | - |
| 64 | 40 | - | - | - | - | - | - |
| 64 | 300 | - | - | - | - | - | - |
| 64 | 2700 | - | - | - | - | - | - |
| 64 | 10000 | - | - | - | - | - | - |

per-unit values (percentiles, spread, raw per-round samples) are in 'Appendix: all values' at the end.

## Dimension: Throughput and concurrency scaling

> throughput = concurrency / median end-to-end; whether it keeps rising with concurrency is the scaling behaviour

| concurrency | body tier (microseconds) | zoo | apscheduler | asyncio_pool | bare_thread | process_pool | thread_pool |
|---|---|---|---|---|---|---|---|
| 1 | 40 | 5477.4 | 3466.0 | 7123.9 | 6715.5 | 2998.2 | 13198.4 |
| 1 | 300 | 2226.3 | 1842.4 | 2587.5 | 2599.6 | 1646.4 | 3098.1 |
| 1 | 2700 | 363.1 | 333.4 | 363.4 | 374.3 | 332.3 | 371.9 |
| 1 | 10000 | 99.3 | 100.3 | 105.5 | 103.6 | 101.1 | 106.8 |
| 4 | 40 | 19404.4 | 15602.9 | 23116.5 | 20013.0 | 23699.2 | 41654.8 |
| 4 | 300 | 8242.9 | 7826.8 | 9816.5 | 9537.0 | 13145.4 | 11403.4 |
| 4 | 2700 | 1357.7 | 1341.0 | 1451.9 | 1478.1 | 2888.8 | 1459.0 |
| 4 | 10000 | 395.8 | 389.9 | 405.4 | 412.6 | 865.5 | 407.6 |
| 16 | 40 | 77763.4 | 71424.4 | 119331.0 | 95364.7 | 106727.0 | 177361.2 |
| 16 | 300 | 34001.3 | 31674.5 | 40712.6 | 37658.1 | 57448.9 | 46626.5 |
| 16 | 2700 | 5518.7 | 5369.8 | 5711.1 | 5901.5 | 11678.3 | 5772.6 |
| 16 | 10000 | 1586.7 | 1619.2 | 1610.5 | 1665.2 | 3375.1 | 1632.1 |
| 64 | 40 | 308754.3 | 289794.7 | 532160.1 | 373124.9 | 329761.1 | 717389.6 |
| 64 | 300 | 130343.7 | 127553.4 | 166491.4 | 150836.5 | 213501.9 | 182674.1 |
| 64 | 2700 | 21671.4 | 21585.8 | 22767.3 | 23921.9 | 46909.7 | 23741.9 |
| 64 | 10000 | 6073.5 | 6294.8 | 6149.5 | 6512.8 | 13657.9 | 6247.3 |

per-unit values (percentiles, spread, raw per-round samples) are in 'Appendix: all values' at the end.

## Dimension: Overhead attribution

> scope: splits the per-task end-to-end into submit side / handoff / body / return and, at the same tier and concurrency, **compares each segment against every baseline**, answering 'which segment the excess falls on'. It is **diagnostic, not the main evidence**: it samples a small set of tiers, and does not compare total overhead against the baselines. **A negative handoff is not an error**: it means the submit call had not returned yet while the body was already running  -  schemes that count thread creation as submit side behave this way (measured: bare_thread handoff -343 microseconds). **The return segment includes waiting for the completion signal** (the last stretch before draining returns); it shares the main measurement's scope

**status: ok**

> this dimension is a **diagnosis**, not the main evidence: the sampled tiers and concurrencies are a fixed small batch, not the whole matrix. The concurrency ceiling is the machine's parallel capacity - beyond it, what is measured is contention;

sampled tiers (microseconds): 40, 10000; concurrency: [1, 4]; this machine's parallel capacity: 4

| scheme | concurrency | tier (microseconds) | submit side | handoff | body | return | status |
|---|---|---|---|---|---|---|---|
| apscheduler | 1 | 40 | 108.78 microseconds | 81.29 microseconds | 58.86 microseconds | 42.86 microseconds | ok |
| apscheduler | 4 | 40 | 110.68 microseconds | 519.28 microseconds | 56.18 microseconds | 218.97 microseconds | ok |
| apscheduler | 1 | 10000 | 166.72 microseconds | 122.62 microseconds | 9.782 milliseconds | 59.54 microseconds | ok |
| apscheduler | 4 | 10000 | 119.74 microseconds | 12.884 milliseconds | 14.542 milliseconds | 9.805 milliseconds | ok |
| asyncio_pool | 1 | 40 | 27.40 microseconds | 47.40 microseconds | 56.28 microseconds | 78.36 microseconds | ok |
| asyncio_pool | 4 | 40 | 19.00 microseconds | 259.15 microseconds | 54.13 microseconds | 309.16 microseconds | ok |
| asyncio_pool | 1 | 10000 | 22.65 microseconds | 36.13 microseconds | 9.197 milliseconds | 84.50 microseconds | ok |
| asyncio_pool | 4 | 10000 | 23.82 microseconds | 12.044 milliseconds | 14.259 milliseconds | 9.649 milliseconds | ok |
| bare_thread | 1 | 40 | 141.09 microseconds | -77.35 microseconds | 52.12 microseconds | 31.62 microseconds | ok |
| bare_thread | 4 | 40 | 150.71 microseconds | -74.46 microseconds | 50.79 microseconds | 302.37 microseconds | ok |
| bare_thread | 1 | 10000 | 5.173 milliseconds | -5094.45 microseconds | 9.675 milliseconds | 38.67 microseconds | ok |
| bare_thread | 4 | 10000 | 5.189 milliseconds | -5084.41 microseconds | 14.786 milliseconds | 9.975 milliseconds | ok |
| process_pool | 1 | 40 | - | - | - | - | not_measurable: the body's start/end timestamps are unavailable - it did not run in this process (cross-process units cannot be split this way) |
| process_pool | 4 | 40 | - | - | - | - | not_measurable: the body's start/end timestamps are unavailable - it did not run in this process (cross-process units cannot be split this way) |
| process_pool | 1 | 10000 | - | - | - | - | not_measurable: the body's start/end timestamps are unavailable - it did not run in this process (cross-process units cannot be split this way) |
| process_pool | 4 | 10000 | - | - | - | - | not_measurable: the body's start/end timestamps are unavailable - it did not run in this process (cross-process units cannot be split this way) |
| thread_pool | 1 | 40 | 8.04 microseconds | 10.46 microseconds | 45.22 microseconds | 15.50 microseconds | ok |
| thread_pool | 4 | 40 | 11.13 microseconds | 180.20 microseconds | 49.39 microseconds | 148.07 microseconds | ok |
| thread_pool | 1 | 10000 | 13.34 microseconds | 17.54 microseconds | 9.434 milliseconds | 28.94 microseconds | ok |
| thread_pool | 4 | 10000 | 13.42 microseconds | 10.048 milliseconds | 15.032 milliseconds | 9.237 milliseconds | ok |
| zoo | 1 | 40 | 56.37 microseconds | 24.72 microseconds | 52.56 microseconds | 62.97 microseconds | ok |
| zoo | 4 | 40 | 64.44 microseconds | 392.36 microseconds | 56.29 microseconds | 334.47 microseconds | ok |
| zoo | 1 | 10000 | 93.87 microseconds | 30.59 microseconds | 9.703 milliseconds | 100.45 microseconds | ok |
| zoo | 4 | 10000 | 92.78 microseconds | 10.328 milliseconds | 14.664 milliseconds | 9.974 milliseconds | ok |

### where the excess over each baseline falls

| baseline | concurrency | tier (microseconds) | excess per task | mostly on | gap in that segment |
|---|---|---|---|---|---|
| apscheduler | 1 | 40 | -100.74 microseconds | return | 20.10 microseconds |
| asyncio_pool | 1 | 40 | -13.24 microseconds | submit side | 28.96 microseconds |
| bare_thread | 1 | 40 | 49.06 microseconds | handoff | 102.07 microseconds |
| thread_pool | 1 | 40 | 117.53 microseconds | submit side | 48.33 microseconds |
| apscheduler | 4 | 40 | -134.18 microseconds | return | 115.50 microseconds |
| asyncio_pool | 4 | 40 | 134.12 microseconds | handoff | 133.22 microseconds |
| bare_thread | 4 | 40 | 337.46 microseconds | handoff | 466.82 microseconds |
| thread_pool | 4 | 40 | 404.51 microseconds | handoff | 212.16 microseconds |
| apscheduler | 1 | 10000 | -196.87 microseconds | return | 40.91 microseconds |
| asyncio_pool | 1 | 10000 | 580.65 microseconds | body | 506.69 microseconds |
| bare_thread | 1 | 10000 | 141.08 microseconds | handoff | 5.125 milliseconds |
| thread_pool | 1 | 10000 | 429.60 microseconds | body | 269.02 microseconds |
| apscheduler | 4 | 10000 | 257.50 microseconds | return | 169.34 microseconds |
| asyncio_pool | 4 | 10000 | 1.275 milliseconds | body | 404.28 microseconds |
| bare_thread | 4 | 10000 | 8.348 milliseconds | handoff | 15.412 milliseconds |
| thread_pool | 4 | 10000 | 671.61 microseconds | return | 737.02 microseconds |

> instrumentation seams used (temporarily wrapped on the harness side, no framework code changed): BaseWaiter._dispatch_worker (handing a worker to the scheduling model); WorkerDispatchCore.resolve_period / resolve_phase / resolve_run_timeout (per-round policy lookups); BaseWaiter.execute_service (one scheduling round)

> **the cost of instrumentation itself** (on the framework-under-test side, paired from two rounds in the same subprocess, so it can be discounted honestly): zoo 40 microseconds / concurrency 1: without instrumentation 195.82 microseconds vs with 197.14 microseconds, a difference of 1.32 microseconds; zoo 40 microseconds / concurrency 4: without instrumentation 788.53 microseconds vs with 782.07 microseconds, a difference of 6.46 microseconds (faster with instrumentation: this round's machine drift exceeded the instrumentation cost); zoo 10000 microseconds / concurrency 1: without instrumentation 9.971 milliseconds vs with 9.936 milliseconds, a difference of 34.78 microseconds (faster with instrumentation: this round's machine drift exceeded the instrumentation cost); zoo 10000 microseconds / concurrency 4: without instrumentation 40.412 milliseconds vs with 40.009 milliseconds, a difference of 403.79 microseconds (faster with instrumentation: this round's machine drift exceeded the instrumentation cost)

> **additivity of the sub-measurements**: the sub-measurements are **independent measurements**: each one's duration contains its own preemption waits, so their sum can exceed the submit side they all live in (about 3% measured). Do not read them as an additive partition.measured overlap this round: zoo 40 microseconds / concurrency 1 overlaps 1.3%; zoo 40 microseconds / concurrency 4 overlaps 5.1%; zoo 10000 microseconds / concurrency 4 overlaps 94.1%

### what the submit side of the framework under test consists of

- concurrency 1, tier 40 microseconds: rest of the scheduling round (due checks, locks, list upkeep) 0.25 microseconds; dispatch (handed to the scheduling model, incl. pool enqueue) 26.68 microseconds; per-round policy lookups (period / phase / timeout each read config) 20.73 microseconds; submit side, other (the adapter's own bookkeeping and completion-signal state) 8.48 microseconds
- concurrency 4, tier 40 microseconds: rest of the scheduling round (due checks, locks, list upkeep) 0.00 microseconds; dispatch (handed to the scheduling model, incl. pool enqueue) 36.68 microseconds; per-round policy lookups (period / phase / timeout each read config) 29.29 microseconds; submit side, other (the adapter's own bookkeeping and completion-signal state) 4.30 microseconds
- concurrency 1, tier 10000 microseconds: rest of the scheduling round (due checks, locks, list upkeep) 0.30 microseconds; dispatch (handed to the scheduling model, incl. pool enqueue) 43.21 microseconds; per-round policy lookups (period / phase / timeout each read config) 35.41 microseconds; submit side, other (the adapter's own bookkeeping and completion-signal state) 15.10 microseconds
- concurrency 4, tier 10000 microseconds: rest of the scheduling round (due checks, locks, list upkeep) 0.00 microseconds; dispatch (handed to the scheduling model, incl. pool enqueue) 51.42 microseconds; per-round policy lookups (period / phase / timeout each read config) 49.10 microseconds; submit side, other (the adapter's own bookkeeping and completion-signal state) 3.68 microseconds

### Attribution conclusions

- At concurrency 1 and a 40 microsecond body, the framework under test is not slower overall than apscheduler (-101 microseconds per task): the two sides have different timing structures, and the per-segment differences have both signs, so no 'which segment' is given.
- At concurrency 1 and a 40 microsecond body, the framework under test is not slower overall than asyncio_pool (-13 microseconds per task): the two sides have different timing structures, and the per-segment differences have both signs, so no 'which segment' is given.
- At concurrency 1 and a 40 microsecond body, the framework under test spends 49 microseconds more per task than bare_thread, most of it on 'handoff' (+102 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'submit side' is -85 microseconds, so read them together.
- At concurrency 1 and a 40 microsecond body, the framework under test spends 118 microseconds more per task than thread_pool, most of it on 'submit side' (+48 microseconds).
- In this group the framework's submit side consists of (microseconds): rest of the scheduling round (due checks, locks, list upkeep) 0, dispatch (handed to the scheduling model, incl. pool enqueue) 27, per-round policy lookups (period / phase / timeout each read config) 21, submit side, other (the adapter's own bookkeeping and completion-signal state) 8.
- At concurrency 4 and a 40 microsecond body, the framework under test is not slower overall than apscheduler (-134 microseconds per task): the two sides have different timing structures, and the per-segment differences have both signs, so no 'which segment' is given.
- At concurrency 4 and a 40 microsecond body, the framework under test spends 134 microseconds more per task than asyncio_pool, most of it on 'handoff' (+133 microseconds).
- At concurrency 4 and a 40 microsecond body, the framework under test spends 337 microseconds more per task than bare_thread, most of it on 'handoff' (+467 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'submit side' is -86 microseconds, so read them together.
- At concurrency 4 and a 40 microsecond body, the framework under test spends 405 microseconds more per task than thread_pool, most of it on 'handoff' (+212 microseconds).
- In this group the framework's submit side consists of (microseconds): rest of the scheduling round (due checks, locks, list upkeep) 0, dispatch (handed to the scheduling model, incl. pool enqueue) 37, per-round policy lookups (period / phase / timeout each read config) 29, submit side, other (the adapter's own bookkeeping and completion-signal state) 4.
- At concurrency 1 and a 10000 microsecond body, the framework under test is not slower overall than apscheduler (-197 microseconds per task): the two sides have different timing structures, and the per-segment differences have both signs, so no 'which segment' is given.
- At concurrency 1 and a 10000 microsecond body, the framework under test spends 581 microseconds more per task than asyncio_pool, most of it on 'body' (+507 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'handoff' is -6 microseconds, so read them together.
- At concurrency 1 and a 10000 microsecond body, the framework under test spends 141 microseconds more per task than bare_thread, most of it on 'handoff' (+5125 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'submit side' is -5079 microseconds, so read them together.
- At concurrency 1 and a 10000 microsecond body, the framework under test spends 430 microseconds more per task than thread_pool, most of it on 'body' (+269 microseconds).
- In this group the framework's submit side consists of (microseconds): rest of the scheduling round (due checks, locks, list upkeep) 0, dispatch (handed to the scheduling model, incl. pool enqueue) 43, per-round policy lookups (period / phase / timeout each read config) 35, submit side, other (the adapter's own bookkeeping and completion-signal state) 15.
- At concurrency 4 and a 10000 microsecond body, the framework under test spends 258 microseconds more per task than apscheduler, most of it on 'return' (+169 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'handoff' is -2557 microseconds, so read them together.
- At concurrency 4 and a 10000 microsecond body, the framework under test spends 1275 microseconds more per task than asyncio_pool, most of it on 'body' (+404 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'handoff' is -1716 microseconds, so read them together.
- At concurrency 4 and a 10000 microsecond body, the framework under test spends 8348 microseconds more per task than bare_thread, most of it on 'handoff' (+15412 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'submit side' is -5096 microseconds, so read them together.
- At concurrency 4 and a 10000 microsecond body, the framework under test spends 672 microseconds more per task than thread_pool, most of it on 'return' (+737 microseconds).
-   note: the per-segment differences in this group do not all share a sign  - 'body' is -368 microseconds, so read them together.
- In this group the framework's submit side consists of (microseconds): rest of the scheduling round (due checks, locks, list upkeep) 0, dispatch (handed to the scheduling model, incl. pool enqueue) 51, per-round policy lookups (period / phase / timeout each read config) 49, submit side, other (the adapter's own bookkeeping and completion-signal state) 4.

## Dimension: The cost of scheduling semantics

> scope: this dimension uses an **internal switch comparison** (same workload, semantics all off vs. enabled one by one) and **does not compare across schemes**  -  a bare writeup has no counterpart for priority / timeout / retry, so it can only answer 'what is this framework's semantics worth', not 'how much more expensive than a bare writeup'

**status: ok** (the conclusion was reached on the 'current' drive-surface generation installed during this measurement)

| semantics item | layer | how probed | measurable | evidence / reason |
|---|---|---|---|---|
| timeout | dispatch layer | behavioral probe (driven repeatedly through scheduling rounds; measures the moment the worker leaves the in-flight table) | yes | timeout reap observed: the worker left the in-flight table before the body ended (0.0122s) (body 0.15s, run_timeout 0.01s); judgment action at dispatch_core.py:425; what is in effect is observation and circuit-breaking (recording, marking unhealthy, removing the in-flight registration, stopping dispatch), **not termination** of the still-running worker - CPython cannot safely interrupt a running thread |
| priority | event layer | structural probe (checks whether the symbols required for execution exist) | no | not re-checked: this item's conclusion was reached on the 'previous' generation; taking it as this generation's fact asserts wiring in the report that the current generation does not have. Re-deriving it needs a probe design of its own (each item needs clean on/off states before it can be priced) |
| retry | event layer | structural probe (checks whether the symbols required for execution exist) | no | not re-checked: this item's conclusion was reached on the 'previous' generation; taking it as this generation's fact asserts wiring in the report that the current generation does not have. Re-deriving it needs a probe design of its own (each item needs clean on/off states before it can be priced) |

> this conclusion was reached on the version under test and its coordinates hold for that version only; after appending a version to matrix.yaml, re-run this probe

## Dimension: Native task line: three execution forms of the same subject

> scope: this line is measured on its own: it enters neither the main matrix's pivot tables nor its crossings, and its numbers must not be read against the baseline adapters listed there

**verdict: not_applicable**

the framework under test has no such capability (symbols not found: adaptive_params, dual_arm_worker, native_params, native_task_worker) - this line does not apply to this version, and that is not a failure

equivalence gate: not run (the probe did not reach it)

> loaded task modbus_rtu.parse_response - Modbus RTU response-frame parsing: the input is a bare RTU ADU response frame (no transport framing), and the tier's frame bytes are its length

> tiers R1, R8, R125 (frame bytes 7, 21, 255); concurrencies 1, 4, 16, 64; each cell runs 100 warm-up plus 100 measured rounds, one child process per concurrency (4 in all)

## tiers where the framework under test is faster

> Tiers where the framework under test is ahead (the two sides differ by more than that tier's band). This comes from the **same within-run comparison** as the openly published unfavourable data  -  one side takes the faster tiers, the other the slower ones, so both cover the same ground and neither can pick its own tiers. **This section being empty is not a failure**: it should be empty when nothing is distinguishable or when the framework loses everywhere.

| baseline | concurrency | body tier (microseconds) | gap | median of the framework under test |
|---|---|---|---|---|
| apscheduler | 1 | 40 | the framework under test is 1.58x faster than apscheduler | 182.57 microseconds |
| process_pool | 1 | 40 | the framework under test is 1.83x faster than process_pool | 182.57 microseconds |
| apscheduler | 1 | 300 | the framework under test is 1.21x faster than apscheduler | 449.17 microseconds |
| process_pool | 1 | 300 | the framework under test is 1.35x faster than process_pool | 449.17 microseconds |
| apscheduler | 1 | 2700 | the framework under test is 1.09x faster than apscheduler | 2.754 milliseconds |
| process_pool | 1 | 2700 | the framework under test is 1.09x faster than process_pool | 2.754 milliseconds |
| apscheduler | 4 | 40 | the framework under test is 1.24x faster than apscheduler | 206.14 microseconds |
| apscheduler | 16 | 300 | the framework under test is 1.07x faster than apscheduler | 470.57 microseconds |
| apscheduler | 16 | 2700 | the framework under test is 1.03x faster than apscheduler | 2.899 milliseconds |

## Openly published unfavourable data

> Tiers where the framework under test is behind (the two sides differ by more than that tier's band). **A report missing this part is not acceptable**  -  a report that only shows where it wins gives the reader every reason to distrust it. Tiers whose difference is smaller than the band are not here but in the 'no winner' section: that is the verdict's outcome, not a place to hide things.

| baseline | concurrency | body tier (microseconds) | gap | median of the framework under test |
|---|---|---|---|---|
| asyncio_pool | 1 | 40 | asyncio_pool is 1.30x faster than zoo | 182.57 microseconds |
| thread_pool | 1 | 40 | thread_pool is 2.41x faster than zoo | 182.57 microseconds |
| asyncio_pool | 1 | 300 | asyncio_pool is 1.16x faster than zoo | 449.17 microseconds |
| bare_thread | 1 | 300 | bare_thread is 1.17x faster than zoo | 449.17 microseconds |
| thread_pool | 1 | 300 | thread_pool is 1.39x faster than zoo | 449.17 microseconds |
| bare_thread | 1 | 2700 | bare_thread is 1.03x faster than zoo | 2.754 milliseconds |
| thread_pool | 1 | 2700 | thread_pool is 1.02x faster than zoo | 2.754 milliseconds |
| asyncio_pool | 1 | 10000 | asyncio_pool is 1.06x faster than zoo | 10.066 milliseconds |
| bare_thread | 1 | 10000 | bare_thread is 1.04x faster than zoo | 10.066 milliseconds |
| thread_pool | 1 | 10000 | thread_pool is 1.08x faster than zoo | 10.066 milliseconds |
| thread_pool | 4 | 40 | thread_pool is 2.15x faster than zoo | 206.14 microseconds |
| asyncio_pool | 4 | 300 | asyncio_pool is 1.19x faster than zoo | 485.26 microseconds |
| bare_thread | 4 | 300 | bare_thread is 1.16x faster than zoo | 485.26 microseconds |
| process_pool | 4 | 300 | process_pool is 1.59x faster than zoo | 485.26 microseconds |
| thread_pool | 4 | 300 | thread_pool is 1.38x faster than zoo | 485.26 microseconds |
| asyncio_pool | 4 | 2700 | asyncio_pool is 1.07x faster than zoo | 2.946 milliseconds |
| bare_thread | 4 | 2700 | bare_thread is 1.09x faster than zoo | 2.946 milliseconds |
| process_pool | 4 | 2700 | process_pool is 2.13x faster than zoo | 2.946 milliseconds |
| thread_pool | 4 | 2700 | thread_pool is 1.07x faster than zoo | 2.946 milliseconds |
| asyncio_pool | 4 | 10000 | asyncio_pool is 1.02x faster than zoo | 10.106 milliseconds |
| process_pool | 4 | 10000 | process_pool is 2.19x faster than zoo | 10.106 milliseconds |
| thread_pool | 4 | 10000 | thread_pool is 1.03x faster than zoo | 10.106 milliseconds |
| asyncio_pool | 16 | 40 | asyncio_pool is 1.53x faster than zoo | 205.75 microseconds |
| bare_thread | 16 | 40 | bare_thread is 1.23x faster than zoo | 205.75 microseconds |
| thread_pool | 16 | 40 | thread_pool is 2.28x faster than zoo | 205.75 microseconds |
| asyncio_pool | 16 | 300 | asyncio_pool is 1.20x faster than zoo | 470.57 microseconds |
| bare_thread | 16 | 300 | bare_thread is 1.11x faster than zoo | 470.57 microseconds |
| process_pool | 16 | 300 | process_pool is 1.69x faster than zoo | 470.57 microseconds |
| thread_pool | 16 | 300 | thread_pool is 1.37x faster than zoo | 470.57 microseconds |
| asyncio_pool | 16 | 2700 | asyncio_pool is 1.03x faster than zoo | 2.899 milliseconds |
| bare_thread | 16 | 2700 | bare_thread is 1.07x faster than zoo | 2.899 milliseconds |
| process_pool | 16 | 2700 | process_pool is 2.12x faster than zoo | 2.899 milliseconds |
| thread_pool | 16 | 2700 | thread_pool is 1.05x faster than zoo | 2.899 milliseconds |
| apscheduler | 16 | 10000 | apscheduler is 1.02x faster than zoo | 10.084 milliseconds |
| asyncio_pool | 16 | 10000 | asyncio_pool is 1.02x faster than zoo | 10.084 milliseconds |
| bare_thread | 16 | 10000 | bare_thread is 1.05x faster than zoo | 10.084 milliseconds |
| process_pool | 16 | 10000 | process_pool is 2.13x faster than zoo | 10.084 milliseconds |
| thread_pool | 16 | 10000 | thread_pool is 1.03x faster than zoo | 10.084 milliseconds |
| asyncio_pool | 64 | 40 | asyncio_pool is 1.72x faster than zoo | 207.28 microseconds |
| bare_thread | 64 | 40 | bare_thread is 1.21x faster than zoo | 207.28 microseconds |
| thread_pool | 64 | 40 | thread_pool is 2.32x faster than zoo | 207.28 microseconds |
| asyncio_pool | 64 | 300 | asyncio_pool is 1.28x faster than zoo | 491.01 microseconds |
| bare_thread | 64 | 300 | bare_thread is 1.16x faster than zoo | 491.01 microseconds |
| process_pool | 64 | 300 | process_pool is 1.64x faster than zoo | 491.01 microseconds |
| thread_pool | 64 | 300 | thread_pool is 1.40x faster than zoo | 491.01 microseconds |
| asyncio_pool | 64 | 2700 | asyncio_pool is 1.05x faster than zoo | 2.953 milliseconds |
| bare_thread | 64 | 2700 | bare_thread is 1.10x faster than zoo | 2.953 milliseconds |
| process_pool | 64 | 2700 | process_pool is 2.16x faster than zoo | 2.953 milliseconds |
| thread_pool | 64 | 2700 | thread_pool is 1.10x faster than zoo | 2.953 milliseconds |
| apscheduler | 64 | 10000 | apscheduler is 1.04x faster than zoo | 10.538 milliseconds |
| bare_thread | 64 | 10000 | bare_thread is 1.07x faster than zoo | 10.538 milliseconds |
| process_pool | 64 | 10000 | process_pool is 2.25x faster than zoo | 10.538 milliseconds |

## tiers with no winner

> Tiers where the two sides **cannot be told apart**: the difference is smaller than that tier's band. The band comes from the two sides' own across-round spread (stdev / median of the per-task medians) **within the same run**, so it varies with the sample and the reader can recompute it from the archived per-round samples. It also covers every measured tier and states a reason for each  -  **this section is not a place to hide things**: reporting noise-level differences as '1.00x faster' is.

> criterion: band = the framework's own across-round relative spread plus the baseline's; a single side's across-round relative spread = stdev / median of that unit's per-round per-task medians

band this run: 1.4% to 30.1% (it varies with each tier's spread).

| baseline | concurrency | body tier (microseconds) | ratio | reason |
|---|---|---|---|---|
| bare_thread | 1 | 40 | 0.82x | the two sides differ by 18.4%, less than the band of 30.1%  -  no winner |
| asyncio_pool | 1 | 2700 | 1.00x | the two sides differ by 0.1%, less than the band of 3.1%  -  no winner |
| apscheduler | 1 | 10000 | 0.99x | the two sides differ by 0.9%, less than the band of 4.6%  -  no winner |
| process_pool | 1 | 10000 | 0.98x | the two sides differ by 1.8%, less than the band of 1.9%  -  no winner |
| asyncio_pool | 4 | 40 | 0.84x | the two sides differ by 16.1%, less than the band of 17.3%  -  no winner |
| bare_thread | 4 | 40 | 0.97x | the two sides differ by 3.0%, less than the band of 14.3%  -  no winner |
| process_pool | 4 | 40 | 0.82x | the two sides differ by 18.1%, less than the band of 22.2%  -  no winner |
| apscheduler | 4 | 300 | 1.05x | the two sides differ by 5.3%, less than the band of 7.0%  -  no winner |
| apscheduler | 4 | 2700 | 1.01x | the two sides differ by 1.3%, less than the band of 2.7%  -  no winner |
| apscheduler | 4 | 10000 | 1.02x | the two sides differ by 1.5%, less than the band of 3.2%  -  no winner |
| bare_thread | 4 | 10000 | 0.96x | the two sides differ by 4.1%, less than the band of 4.7%  -  no winner |
| apscheduler | 16 | 40 | 1.09x | the two sides differ by 8.9%, less than the band of 15.6%  -  no winner |
| process_pool | 16 | 40 | 0.73x | the two sides differ by 27.1%, less than the band of 29.0%  -  no winner |
| apscheduler | 64 | 40 | 1.07x | the two sides differ by 6.5%, less than the band of 9.9%  -  no winner |
| process_pool | 64 | 40 | 0.94x | the two sides differ by 6.4%, less than the band of 16.1%  -  no winner |
| apscheduler | 64 | 300 | 1.02x | the two sides differ by 2.2%, less than the band of 5.0%  -  no winner |
| apscheduler | 64 | 2700 | 1.00x | the two sides differ by 0.4%, less than the band of 1.7%  -  no winner |
| asyncio_pool | 64 | 10000 | 0.99x | the two sides differ by 1.2%, less than the band of 3.9%  -  no winner |
| thread_pool | 64 | 10000 | 0.97x | the two sides differ by 2.8%, less than the band of 3.7%  -  no winner |

## caliber limitations and sources of bias

- **measured layer**: dispatch layer: drives BaseWaiter.execute_service() directly (thread-pool mode) instead of Master's full run loop - the latter's service loop fires once per second, which would lift the latency scale from microseconds to seconds. **One scheduling round per submitted task**, so that whole round's bookkeeping is charged to this single task; Master's usage serves several workers per round, which would spread that cost over tasks
- **scope note** (apscheduler): APScheduler's scheduled-job model, including job registration and trigger-decision cost; the library swallows exceptions raised inside jobs, so this tier re-raises them during drain. **Caliber note**: the library offers no job-completion notification, so the completion signal comes from this tier's Condition, and the cost of one notification per task counts toward its end-to-end
- **scope note** (asyncio_pool): the asyncio event loop dispatches **synchronous** bodies (equivalent to loop.run_in_executor), not native coroutine concurrency; the difference from a thread pool is the event loop's own cost
- **scope note** (bare_thread): one threading.Thread per task, no pooling and no reuse; thread creation cost counts toward end-to-end. Bodies' self-reported timings are collected with a lock-free list.append (atomic under the GIL) - **instrumentation must not charge one tier a cost other tiers do not pay**, or the comparison becomes a comparison of instrumentation
- **scope note** (process_pool): ProcessPoolExecutor: pickling the body and its return value across processes counts toward end-to-end, so this tier is not directly comparable to the multi-threaded tiers
- **scope note** (thread_pool): concurrent.futures.ThreadPoolExecutor, pool size = concurrency; threads are reused, with no task-queue semantics
- **scope note** (zoo): end-to-end stops at the worker completing and being deregistered from the in-flight table; the framework's later result delivery splits by generation: the previous generation's delivery path is broken (dispatch looks reactors up by reactor_name rather than by topic, so every call raises), and the completion signal can only come from the framework's internal callback; the current generation has that fixed, and the completion signal comes from the framework's own result reactor. **Caliber note**: the completion signal comes from this tier's Condition, and the cost of one notification per task counts toward its end-to-end; a bias in that direction makes the framework under test look slower, i.e. it works against it rather than for it
- **groups with no overhead figure**: for the (adapter, concurrency) groups below, concurrency exceeds the parallel capacity of the machine this run was on: the body's self-reported duration contains scheduling waits from oversubscription, which is not the same thing as 'wall clock / concurrency', so subtracting them is meaningless (measured, it produced negative values and meaningless change factors). This report therefore **gives them no framework overhead and no crossing point**; end-to-end, throughput and the body's self-reported values remain valid and not one field has been dropped from the raw archive. The criterion is the logical core count from the environment self-report, plus one evidential backstop: a computed negative overhead is likewise treated as unreadable  -  a framework can only add time, never remove it
- **native task line: no verdict**: the native task line came out not_applicable (the framework under test has no such capability (symbols not found: adaptive_params, dual_arm_worker, native_params, native_task_worker) - this line does not apply to this version, and that is not a failure) and produced no figures. It is reported here rather than folded into the self-check: one line's failure must not hold back the other versions' data or the main matrix
- **self-check items excluded from the verdict**: for these units the body-tier deviation did not take part in the verdict (concurrency above the lowest, or a tier so short that wall clock is dominated by scheduling granularity). The raw values are still in place, but the wall clock at those two places is not enough to conclude whether the calibration is right  -  **factor this in when reading those tiers' numbers**

> absolute durations: valid only within a single run; do not compare across runs

## bring your own baseline

the baselines of this harness are **open**: implement the four methods of `zoo_bench.adapters.BaseAdapter` (`setup` / `submit` / `drain` / `teardown`), register with `zoo_bench.adapters.registry.register`, or list your module path in `extra_modules` of `matrix.yaml`  -  **no file in this repository needs to change**. Your implementation goes through the same equivalence verification (submit N bodies, assert all ran and each exactly once); failing it means the data is not trusted. **We deliberately write this path into the report**: a report written by the framework's own maintainer, measuring that framework, gives every reason to suspect a slowed baseline or a chosen caliber; letting anyone submit their own baseline is what makes this report challengeable and falsifiable.

## self-check

- overall measurement self-check: passed
- process isolation: {'parent_pid': 43405, 'child_pids': [43869, 43872, 43875, 43878, 43881, 43887, 43893, 43899, 43905, 43923, 43941, 43959, 43978, 44034, 44090, 44145, 44206, 44208, 44210, 44212, 44214, 44219, 44224, 44229, 44234, 44251, 44268, 44285, 44302, 44363, 44425, 44490, 44555, 44589, 44623, 44657, 44691, 44824, 44959, 45092, 45225, 45754, 46283, 46812, 47341, 49454, 51567, 53680, 55794, 55798, 55802, 55806, 55810, 55817, 55824, 55831, 55838, 55857, 55876, 55895, 55914, 55981, 56048, 56115, 56182, 56184, 56186, 56189, 56191, 56196, 56201, 56206, 56211, 56228, 56245, 56262, 56279, 56335, 56392, 56453, 56519, 56521, 56523, 56525, 56527, 56532, 56537, 56542, 56547, 56564, 56581, 56598, 56615, 56664, 56713, 56758], 'all_distinct': True, 'none_equals_parent': True}

## Appendix: all values

> the body carries only pivot tables for readability; this appendix gives **every field as measured before any trimming** per unit (percentiles, spread, measured body durations, ...). **Not one value is dropped**: the details live inside the report body itself, not behind links, folds or attachments  -  'verifiable' that requires a separate lookup is not verifiable

### Appendix: Latency percentiles and jitter

| scheme | concurrency | body tier (microseconds) | median | p95 | p99 | relative spread |
|---|---|---|---|---|---|---|
| apscheduler | 1 | 40 | 288.52 microseconds | 319.90 microseconds | 323.17 microseconds | 0.053 |
| apscheduler | 1 | 300 | 542.76 microseconds | 584.20 microseconds | 596.60 microseconds | 0.042 |
| apscheduler | 1 | 2700 | 2.999 milliseconds | 3.199 milliseconds | 3.209 milliseconds | 0.027 |
| apscheduler | 1 | 10000 | 9.971 milliseconds | 10.397 milliseconds | 11.362 milliseconds | 0.034 |
| apscheduler | 4 | 40 | 256.36 microseconds | 278.20 microseconds | 299.54 microseconds | 0.047 |
| apscheduler | 4 | 300 | 511.06 microseconds | 526.87 microseconds | 571.99 microseconds | 0.030 |
| apscheduler | 4 | 2700 | 2.983 milliseconds | 3.051 milliseconds | 3.097 milliseconds | 0.012 |
| apscheduler | 4 | 10000 | 10.258 milliseconds | 10.480 milliseconds | 11.282 milliseconds | 0.025 |
| apscheduler | 16 | 40 | 224.01 microseconds | 265.51 microseconds | 278.59 microseconds | 0.084 |
| apscheduler | 16 | 300 | 505.14 microseconds | 532.88 microseconds | 551.13 microseconds | 0.026 |
| apscheduler | 16 | 2700 | 2.980 milliseconds | 3.025 milliseconds | 3.042 milliseconds | 0.010 |
| apscheduler | 16 | 10000 | 9.881 milliseconds | 9.964 milliseconds | 9.988 milliseconds | 0.008 |
| apscheduler | 64 | 40 | 220.85 microseconds | 246.89 microseconds | 281.47 microseconds | 0.068 |
| apscheduler | 64 | 300 | 501.75 microseconds | 548.43 microseconds | 550.58 microseconds | 0.040 |
| apscheduler | 64 | 2700 | 2.965 milliseconds | 3.002 milliseconds | 3.006 milliseconds | 0.010 |
| apscheduler | 64 | 10000 | 10.167 milliseconds | 10.487 milliseconds | 10.506 milliseconds | 0.014 |
| asyncio_pool | 1 | 40 | 140.37 microseconds | 165.56 microseconds | 181.10 microseconds | 0.083 |
| asyncio_pool | 1 | 300 | 386.47 microseconds | 424.78 microseconds | 434.41 microseconds | 0.046 |
| asyncio_pool | 1 | 2700 | 2.752 milliseconds | 2.835 milliseconds | 2.987 milliseconds | 0.023 |
| asyncio_pool | 1 | 10000 | 9.482 milliseconds | 9.593 milliseconds | 9.606 milliseconds | 0.007 |
| asyncio_pool | 4 | 40 | 173.04 microseconds | 188.13 microseconds | 194.49 microseconds | 0.098 |
| asyncio_pool | 4 | 300 | 407.48 microseconds | 430.31 microseconds | 434.63 microseconds | 0.025 |
| asyncio_pool | 4 | 2700 | 2.755 milliseconds | 2.843 milliseconds | 2.872 milliseconds | 0.020 |
| asyncio_pool | 4 | 10000 | 9.867 milliseconds | 10.018 milliseconds | 10.030 milliseconds | 0.007 |
| asyncio_pool | 16 | 40 | 134.08 microseconds | 145.26 microseconds | 147.06 microseconds | 0.048 |
| asyncio_pool | 16 | 300 | 393.00 microseconds | 410.43 microseconds | 413.08 microseconds | 0.025 |
| asyncio_pool | 16 | 2700 | 2.802 milliseconds | 2.880 milliseconds | 2.901 milliseconds | 0.014 |
| asyncio_pool | 16 | 10000 | 9.935 milliseconds | 10.005 milliseconds | 10.035 milliseconds | 0.006 |
| asyncio_pool | 64 | 40 | 120.26 microseconds | 127.13 microseconds | 155.48 microseconds | 0.075 |
| asyncio_pool | 64 | 300 | 384.40 microseconds | 394.23 microseconds | 419.68 microseconds | 0.023 |
| asyncio_pool | 64 | 2700 | 2.811 milliseconds | 2.843 milliseconds | 2.879 milliseconds | 0.008 |
| asyncio_pool | 64 | 10000 | 10.407 milliseconds | 10.532 milliseconds | 10.737 milliseconds | 0.018 |
| bare_thread | 1 | 40 | 148.91 microseconds | 191.86 microseconds | 265.16 microseconds | 0.191 |
| bare_thread | 1 | 300 | 384.68 microseconds | 418.48 microseconds | 436.19 microseconds | 0.043 |
| bare_thread | 1 | 2700 | 2.672 milliseconds | 2.738 milliseconds | 2.766 milliseconds | 0.011 |
| bare_thread | 1 | 10000 | 9.648 milliseconds | 9.807 milliseconds | 9.991 milliseconds | 0.011 |
| bare_thread | 4 | 40 | 199.87 microseconds | 214.91 microseconds | 218.90 microseconds | 0.069 |
| bare_thread | 4 | 300 | 419.42 microseconds | 470.82 microseconds | 476.46 microseconds | 0.050 |
| bare_thread | 4 | 2700 | 2.706 milliseconds | 2.777 milliseconds | 2.816 milliseconds | 0.013 |
| bare_thread | 4 | 10000 | 9.696 milliseconds | 10.520 milliseconds | 11.286 milliseconds | 0.040 |
| bare_thread | 16 | 40 | 167.78 microseconds | 174.20 microseconds | 174.64 microseconds | 0.021 |
| bare_thread | 16 | 300 | 424.88 microseconds | 436.42 microseconds | 439.49 microseconds | 0.012 |
| bare_thread | 16 | 2700 | 2.711 milliseconds | 2.736 milliseconds | 2.749 milliseconds | 0.005 |
| bare_thread | 16 | 10000 | 9.609 milliseconds | 9.734 milliseconds | 10.228 milliseconds | 0.016 |
| bare_thread | 64 | 40 | 171.52 microseconds | 175.30 microseconds | 176.12 microseconds | 0.015 |
| bare_thread | 64 | 300 | 424.30 microseconds | 431.60 microseconds | 433.02 microseconds | 0.009 |
| bare_thread | 64 | 2700 | 2.675 milliseconds | 2.707 milliseconds | 2.771 milliseconds | 0.009 |
| bare_thread | 64 | 10000 | 9.827 milliseconds | 9.887 milliseconds | 10.032 milliseconds | 0.006 |
| process_pool | 1 | 40 | 333.53 microseconds | 361.85 microseconds | 366.69 microseconds | 0.049 |
| process_pool | 1 | 300 | 607.37 microseconds | 640.03 microseconds | 647.18 microseconds | 0.030 |
| process_pool | 1 | 2700 | 3.009 milliseconds | 3.041 milliseconds | 3.069 milliseconds | 0.007 |
| process_pool | 1 | 10000 | 9.888 milliseconds | 10.016 milliseconds | 10.086 milliseconds | 0.007 |
| process_pool | 4 | 40 | 168.78 microseconds | 221.25 microseconds | 235.10 microseconds | 0.147 |
| process_pool | 4 | 300 | 304.29 microseconds | 365.39 microseconds | 376.12 microseconds | 0.099 |
| process_pool | 4 | 2700 | 1.385 milliseconds | 1.723 milliseconds | 1.978 milliseconds | 0.108 |
| process_pool | 4 | 10000 | 4.621 milliseconds | 4.741 milliseconds | 4.757 milliseconds | 0.011 |
| process_pool | 16 | 40 | 149.92 microseconds | 223.33 microseconds | 232.08 microseconds | 0.218 |
| process_pool | 16 | 300 | 278.51 microseconds | 302.38 microseconds | 334.02 microseconds | 0.087 |
| process_pool | 16 | 2700 | 1.370 milliseconds | 1.542 milliseconds | 1.562 milliseconds | 0.049 |
| process_pool | 16 | 10000 | 4.741 milliseconds | 4.855 milliseconds | 4.901 milliseconds | 0.013 |
| process_pool | 64 | 40 | 194.08 microseconds | 242.04 microseconds | 250.15 microseconds | 0.129 |
| process_pool | 64 | 300 | 299.76 microseconds | 331.67 microseconds | 337.00 microseconds | 0.052 |
| process_pool | 64 | 2700 | 1.364 milliseconds | 1.413 milliseconds | 1.451 milliseconds | 0.019 |
| process_pool | 64 | 10000 | 4.686 milliseconds | 4.721 milliseconds | 5.108 milliseconds | 0.023 |
| thread_pool | 1 | 40 | 75.77 microseconds | 96.09 microseconds | 102.57 microseconds | 0.119 |
| thread_pool | 1 | 300 | 322.77 microseconds | 343.49 microseconds | 347.40 microseconds | 0.034 |
| thread_pool | 1 | 2700 | 2.689 milliseconds | 2.738 milliseconds | 2.804 milliseconds | 0.012 |
| thread_pool | 1 | 10000 | 9.361 milliseconds | 9.494 milliseconds | 9.573 milliseconds | 0.007 |
| thread_pool | 4 | 40 | 96.03 microseconds | 103.48 microseconds | 104.00 microseconds | 0.047 |
| thread_pool | 4 | 300 | 350.77 microseconds | 366.87 microseconds | 371.64 microseconds | 0.021 |
| thread_pool | 4 | 2700 | 2.742 milliseconds | 2.787 milliseconds | 2.810 milliseconds | 0.013 |
| thread_pool | 4 | 10000 | 9.813 milliseconds | 9.951 milliseconds | 9.957 milliseconds | 0.008 |
| thread_pool | 16 | 40 | 90.21 microseconds | 96.78 microseconds | 98.07 microseconds | 0.042 |
| thread_pool | 16 | 300 | 343.15 microseconds | 352.67 microseconds | 355.58 microseconds | 0.015 |
| thread_pool | 16 | 2700 | 2.772 milliseconds | 2.806 milliseconds | 2.810 milliseconds | 0.009 |
| thread_pool | 16 | 10000 | 9.803 milliseconds | 9.917 milliseconds | 10.039 milliseconds | 0.007 |
| thread_pool | 64 | 40 | 89.21 microseconds | 91.13 microseconds | 92.53 microseconds | 0.020 |
| thread_pool | 64 | 300 | 350.35 microseconds | 353.38 microseconds | 355.51 microseconds | 0.006 |
| thread_pool | 64 | 2700 | 2.696 milliseconds | 2.730 milliseconds | 2.736 milliseconds | 0.008 |
| thread_pool | 64 | 10000 | 10.244 milliseconds | 10.415 milliseconds | 10.559 milliseconds | 0.016 |
| zoo | 1 | 40 | 182.57 microseconds | 223.66 microseconds | 236.57 microseconds | 0.111 |
| zoo | 1 | 300 | 449.17 microseconds | 484.91 microseconds | 497.19 microseconds | 0.042 |
| zoo | 1 | 2700 | 2.754 milliseconds | 2.792 milliseconds | 2.799 milliseconds | 0.008 |
| zoo | 1 | 10000 | 10.066 milliseconds | 10.274 milliseconds | 10.404 milliseconds | 0.012 |
| zoo | 4 | 40 | 206.14 microseconds | 237.88 microseconds | 243.81 microseconds | 0.075 |
| zoo | 4 | 300 | 485.26 microseconds | 517.91 microseconds | 527.92 microseconds | 0.040 |
| zoo | 4 | 2700 | 2.946 milliseconds | 3.007 milliseconds | 3.025 milliseconds | 0.016 |
| zoo | 4 | 10000 | 10.106 milliseconds | 10.173 milliseconds | 10.206 milliseconds | 0.007 |
| zoo | 16 | 40 | 205.75 microseconds | 235.19 microseconds | 256.29 microseconds | 0.072 |
| zoo | 16 | 300 | 470.57 microseconds | 488.80 microseconds | 491.20 microseconds | 0.019 |
| zoo | 16 | 2700 | 2.899 milliseconds | 2.938 milliseconds | 2.942 milliseconds | 0.010 |
| zoo | 16 | 10000 | 10.084 milliseconds | 10.217 milliseconds | 10.270 milliseconds | 0.008 |
| zoo | 64 | 40 | 207.28 microseconds | 216.84 microseconds | 219.30 microseconds | 0.031 |
| zoo | 64 | 300 | 491.01 microseconds | 498.97 microseconds | 500.32 microseconds | 0.010 |
| zoo | 64 | 2700 | 2.953 milliseconds | 2.994 milliseconds | 2.997 milliseconds | 0.008 |
| zoo | 64 | 10000 | 10.538 milliseconds | 10.920 milliseconds | 11.105 milliseconds | 0.021 |

### Appendix: Framework overhead share

| scheme | concurrency | body tier (microseconds) | framework overhead | overhead share | body as measured |
|---|---|---|---|---|---|
| apscheduler | 1 | 40 | 230.26 microseconds | 79.81% | 58.25 microseconds |
| apscheduler | 1 | 300 | 243.87 microseconds | 44.93% | 298.89 microseconds |
| apscheduler | 1 | 2700 | 270.38 microseconds | 9.02% | 2.729 milliseconds |
| apscheduler | 1 | 10000 | 318.08 microseconds | 3.19% | 9.653 milliseconds |
| apscheduler | 4 | 40 | 200.17 microseconds | 78.08% | 56.20 microseconds |
| apscheduler | 4 | 300 | 202.61 microseconds | 39.64% | 308.46 microseconds |
| apscheduler | 4 | 2700 | 259.91 microseconds | 8.71% | 2.723 milliseconds |
| apscheduler | 4 | 10000 | - | - | 14.702 milliseconds |
| apscheduler | 16 | 40 | - | - | 53.41 microseconds |
| apscheduler | 16 | 300 | - | - | 319.00 microseconds |
| apscheduler | 16 | 2700 | - | - | 2.736 milliseconds |
| apscheduler | 16 | 10000 | - | - | 15.686 milliseconds |
| apscheduler | 64 | 40 | - | - | 50.44 microseconds |
| apscheduler | 64 | 300 | - | - | 306.35 microseconds |
| apscheduler | 64 | 2700 | - | - | 2.714 milliseconds |
| apscheduler | 64 | 10000 | - | - | 14.852 milliseconds |
| asyncio_pool | 1 | 40 | 94.16 microseconds | 67.08% | 46.21 microseconds |
| asyncio_pool | 1 | 300 | 101.41 microseconds | 26.24% | 285.06 microseconds |
| asyncio_pool | 1 | 2700 | 126.14 microseconds | 4.58% | 2.626 milliseconds |
| asyncio_pool | 1 | 10000 | 146.32 microseconds | 1.54% | 9.336 milliseconds |
| asyncio_pool | 4 | 40 | 118.87 microseconds | 68.70% | 54.17 microseconds |
| asyncio_pool | 4 | 300 | 105.33 microseconds | 25.85% | 302.15 microseconds |
| asyncio_pool | 4 | 2700 | 122.61 microseconds | 4.45% | 2.632 milliseconds |
| asyncio_pool | 4 | 10000 | - | - | 12.035 milliseconds |
| asyncio_pool | 16 | 40 | - | - | 53.76 microseconds |
| asyncio_pool | 16 | 300 | - | - | 303.75 microseconds |
| asyncio_pool | 16 | 2700 | - | - | 2.675 milliseconds |
| asyncio_pool | 16 | 10000 | - | - | 19.300 milliseconds |
| asyncio_pool | 64 | 40 | - | - | 52.05 microseconds |
| asyncio_pool | 64 | 300 | - | - | 303.49 microseconds |
| asyncio_pool | 64 | 2700 | - | - | 2.691 milliseconds |
| asyncio_pool | 64 | 10000 | - | - | 95.273 milliseconds |
| bare_thread | 1 | 40 | 95.49 microseconds | 64.13% | 53.41 microseconds |
| bare_thread | 1 | 300 | 97.47 microseconds | 25.34% | 287.21 microseconds |
| bare_thread | 1 | 2700 | 107.34 microseconds | 4.02% | 2.564 milliseconds |
| bare_thread | 1 | 10000 | 120.48 microseconds | 1.25% | 9.528 milliseconds |
| bare_thread | 4 | 40 | 140.95 microseconds | 70.52% | 58.93 microseconds |
| bare_thread | 4 | 300 | 118.32 microseconds | 28.21% | 301.10 microseconds |
| bare_thread | 4 | 2700 | 109.53 microseconds | 4.05% | 2.597 milliseconds |
| bare_thread | 4 | 10000 | - | - | 14.698 milliseconds |
| bare_thread | 16 | 40 | - | - | 50.78 microseconds |
| bare_thread | 16 | 300 | - | - | 300.13 microseconds |
| bare_thread | 16 | 2700 | - | - | 2.584 milliseconds |
| bare_thread | 16 | 10000 | - | - | 14.647 milliseconds |
| bare_thread | 64 | 40 | - | - | 51.21 microseconds |
| bare_thread | 64 | 300 | - | - | 297.44 microseconds |
| bare_thread | 64 | 2700 | - | - | 2.542 milliseconds |
| bare_thread | 64 | 10000 | - | - | 18.793 milliseconds |
| process_pool | 1 | 40 | 279.58 microseconds | 83.82% | 53.95 microseconds |
| process_pool | 1 | 300 | 296.81 microseconds | 48.87% | 310.56 microseconds |
| process_pool | 1 | 2700 | 342.13 microseconds | 11.37% | 2.667 milliseconds |
| process_pool | 1 | 10000 | 341.21 microseconds | 3.45% | 9.547 milliseconds |
| process_pool | 4 | 40 | 113.81 microseconds | 67.43% | 54.97 microseconds |
| process_pool | 4 | 300 | - | - | 577.68 microseconds |
| process_pool | 4 | 2700 | - | - | 4.801 milliseconds |
| process_pool | 4 | 10000 | - | - | 17.775 milliseconds |
| process_pool | 16 | 40 | - | - | 92.00 microseconds |
| process_pool | 16 | 300 | - | - | 570.91 microseconds |
| process_pool | 16 | 2700 | - | - | 9.708 milliseconds |
| process_pool | 16 | 10000 | - | - | 54.326 milliseconds |
| process_pool | 64 | 40 | - | - | 93.26 microseconds |
| process_pool | 64 | 300 | - | - | 583.35 microseconds |
| process_pool | 64 | 2700 | - | - | 27.749 milliseconds |
| process_pool | 64 | 10000 | - | - | 173.846 milliseconds |
| thread_pool | 1 | 40 | 31.35 microseconds | 41.38% | 44.42 microseconds |
| thread_pool | 1 | 300 | 35.83 microseconds | 11.10% | 286.94 microseconds |
| thread_pool | 1 | 2700 | 63.24 microseconds | 2.35% | 2.625 milliseconds |
| thread_pool | 1 | 10000 | 62.33 microseconds | 0.67% | 9.299 milliseconds |
| thread_pool | 4 | 40 | 46.33 microseconds | 48.25% | 49.70 microseconds |
| thread_pool | 4 | 300 | 52.33 microseconds | 14.92% | 298.44 microseconds |
| thread_pool | 4 | 2700 | 78.15 microseconds | 2.85% | 2.663 milliseconds |
| thread_pool | 4 | 10000 | - | - | 19.045 milliseconds |
| thread_pool | 16 | 40 | - | - | 47.32 microseconds |
| thread_pool | 16 | 300 | - | - | 293.68 microseconds |
| thread_pool | 16 | 2700 | - | - | 2.682 milliseconds |
| thread_pool | 16 | 10000 | - | - | 20.052 milliseconds |
| thread_pool | 64 | 40 | - | - | 47.07 microseconds |
| thread_pool | 64 | 300 | - | - | 294.13 microseconds |
| thread_pool | 64 | 2700 | - | - | 2.598 milliseconds |
| thread_pool | 64 | 10000 | - | - | 40.559 milliseconds |
| zoo | 1 | 40 | 130.82 microseconds | 71.66% | 51.75 microseconds |
| zoo | 1 | 300 | 144.25 microseconds | 32.12% | 304.92 microseconds |
| zoo | 1 | 2700 | 155.52 microseconds | 5.65% | 2.599 milliseconds |
| zoo | 1 | 10000 | 256.30 microseconds | 2.55% | 9.810 milliseconds |
| zoo | 4 | 40 | 151.02 microseconds | 73.26% | 55.12 microseconds |
| zoo | 4 | 300 | 176.04 microseconds | 36.28% | 309.23 microseconds |
| zoo | 4 | 2700 | 226.36 microseconds | 7.68% | 2.720 milliseconds |
| zoo | 4 | 10000 | - | - | 15.005 milliseconds |
| zoo | 16 | 40 | - | - | 55.32 microseconds |
| zoo | 16 | 300 | - | - | 304.71 microseconds |
| zoo | 16 | 2700 | - | - | 2.680 milliseconds |
| zoo | 16 | 10000 | - | - | 19.345 milliseconds |
| zoo | 64 | 40 | - | - | 53.22 microseconds |
| zoo | 64 | 300 | - | - | 314.88 microseconds |
| zoo | 64 | 2700 | - | - | 2.722 milliseconds |
| zoo | 64 | 10000 | - | - | 47.155 milliseconds |

### Appendix: Throughput and concurrency scaling

| scheme | concurrency | body tier (microseconds) | throughput (tasks/s) |
|---|---|---|---|
| apscheduler | 1 | 40 | 3466.0 |
| apscheduler | 1 | 300 | 1842.4 |
| apscheduler | 1 | 2700 | 333.4 |
| apscheduler | 1 | 10000 | 100.3 |
| apscheduler | 4 | 40 | 15602.9 |
| apscheduler | 4 | 300 | 7826.8 |
| apscheduler | 4 | 2700 | 1341.0 |
| apscheduler | 4 | 10000 | 389.9 |
| apscheduler | 16 | 40 | 71424.4 |
| apscheduler | 16 | 300 | 31674.5 |
| apscheduler | 16 | 2700 | 5369.8 |
| apscheduler | 16 | 10000 | 1619.2 |
| apscheduler | 64 | 40 | 289794.7 |
| apscheduler | 64 | 300 | 127553.4 |
| apscheduler | 64 | 2700 | 21585.8 |
| apscheduler | 64 | 10000 | 6294.8 |
| asyncio_pool | 1 | 40 | 7123.9 |
| asyncio_pool | 1 | 300 | 2587.5 |
| asyncio_pool | 1 | 2700 | 363.4 |
| asyncio_pool | 1 | 10000 | 105.5 |
| asyncio_pool | 4 | 40 | 23116.5 |
| asyncio_pool | 4 | 300 | 9816.5 |
| asyncio_pool | 4 | 2700 | 1451.9 |
| asyncio_pool | 4 | 10000 | 405.4 |
| asyncio_pool | 16 | 40 | 119331.0 |
| asyncio_pool | 16 | 300 | 40712.6 |
| asyncio_pool | 16 | 2700 | 5711.1 |
| asyncio_pool | 16 | 10000 | 1610.5 |
| asyncio_pool | 64 | 40 | 532160.1 |
| asyncio_pool | 64 | 300 | 166491.4 |
| asyncio_pool | 64 | 2700 | 22767.3 |
| asyncio_pool | 64 | 10000 | 6149.5 |
| bare_thread | 1 | 40 | 6715.5 |
| bare_thread | 1 | 300 | 2599.6 |
| bare_thread | 1 | 2700 | 374.3 |
| bare_thread | 1 | 10000 | 103.6 |
| bare_thread | 4 | 40 | 20013.0 |
| bare_thread | 4 | 300 | 9537.0 |
| bare_thread | 4 | 2700 | 1478.1 |
| bare_thread | 4 | 10000 | 412.6 |
| bare_thread | 16 | 40 | 95364.7 |
| bare_thread | 16 | 300 | 37658.1 |
| bare_thread | 16 | 2700 | 5901.5 |
| bare_thread | 16 | 10000 | 1665.2 |
| bare_thread | 64 | 40 | 373124.9 |
| bare_thread | 64 | 300 | 150836.5 |
| bare_thread | 64 | 2700 | 23921.9 |
| bare_thread | 64 | 10000 | 6512.8 |
| process_pool | 1 | 40 | 2998.2 |
| process_pool | 1 | 300 | 1646.4 |
| process_pool | 1 | 2700 | 332.3 |
| process_pool | 1 | 10000 | 101.1 |
| process_pool | 4 | 40 | 23699.2 |
| process_pool | 4 | 300 | 13145.4 |
| process_pool | 4 | 2700 | 2888.8 |
| process_pool | 4 | 10000 | 865.5 |
| process_pool | 16 | 40 | 106727.0 |
| process_pool | 16 | 300 | 57448.9 |
| process_pool | 16 | 2700 | 11678.3 |
| process_pool | 16 | 10000 | 3375.1 |
| process_pool | 64 | 40 | 329761.1 |
| process_pool | 64 | 300 | 213501.9 |
| process_pool | 64 | 2700 | 46909.7 |
| process_pool | 64 | 10000 | 13657.9 |
| thread_pool | 1 | 40 | 13198.4 |
| thread_pool | 1 | 300 | 3098.1 |
| thread_pool | 1 | 2700 | 371.9 |
| thread_pool | 1 | 10000 | 106.8 |
| thread_pool | 4 | 40 | 41654.8 |
| thread_pool | 4 | 300 | 11403.4 |
| thread_pool | 4 | 2700 | 1459.0 |
| thread_pool | 4 | 10000 | 407.6 |
| thread_pool | 16 | 40 | 177361.2 |
| thread_pool | 16 | 300 | 46626.5 |
| thread_pool | 16 | 2700 | 5772.6 |
| thread_pool | 16 | 10000 | 1632.1 |
| thread_pool | 64 | 40 | 717389.6 |
| thread_pool | 64 | 300 | 182674.1 |
| thread_pool | 64 | 2700 | 23741.9 |
| thread_pool | 64 | 10000 | 6247.3 |
| zoo | 1 | 40 | 5477.4 |
| zoo | 1 | 300 | 2226.3 |
| zoo | 1 | 2700 | 363.1 |
| zoo | 1 | 10000 | 99.3 |
| zoo | 4 | 40 | 19404.4 |
| zoo | 4 | 300 | 8242.9 |
| zoo | 4 | 2700 | 1357.7 |
| zoo | 4 | 10000 | 395.8 |
| zoo | 16 | 40 | 77763.4 |
| zoo | 16 | 300 | 34001.3 |
| zoo | 16 | 2700 | 5518.7 |
| zoo | 16 | 10000 | 1586.7 |
| zoo | 64 | 40 | 308754.3 |
| zoo | 64 | 300 | 130343.7 |
| zoo | 64 | 2700 | 21671.4 |
| zoo | 64 | 10000 | 6073.5 |