Latest bounded qualification result
The result below comes from the project’s physical MPLA Booster scorecard, recorded at . The run was sealed with a SHA-256 manifest and reported both focused operation timing and total harness wall time. No number on this page is a comparison with another product.
| Measurement | Observed value | Declared gate or context |
|---|---|---|
| Warm attachment service | 26.970416 ms | Below the 50 ms target |
| Complete warm preparation | 467.124167 ms | Separate preparation timing |
| Focused P0 qualification | 2.060053459 s | Independent 60 s cap |
| Total harness wall time | 7.025114208 s | Includes Docker/Desktop setup and cleanup |
| Logical fixture size | 8,589,934,592 bytes | 8 GiB logical chain |
| Cached allocations | 8 | Authenticated fixture allocations |
| Required branches | Depth 1, 5, and 8 | All exposed in the attached fixture |
| Payload copied | 0 bytes | Zero-copy attachment condition |
The sealed manifest SHA-256 was fe1f749d0f59bedf3fc73820d069e784cec6a100f1d27a91988bef1c7576b0f9. The harness reported a PASS for this P0 qualification and returned both supported sandbox inventories to exact empty during cleanup.
What the scorecard measured
This case measures warm attachment of a prepared, authenticated LayerStack fixture and then runs a focused qualification. The fixture represents an exact 8 GiB logical chain with eight cached allocations and required depth branches. The attachment path must expose the expected structure without copying fixture payload bytes.
Preparation time and attachment service time are reported separately. The service measurement isolates the actual cached-fixture attachment; the preparation measurement includes the surrounding work needed to make the prepared fixture available. The focused P0 timer is separate again, and total harness time includes Docker Desktop setup and exact cleanup. Combining these timers into one “startup time” would misrepresent the measurement.
bin/run-mpla-booster-scorecard \
--run-id mpla-final6-p0-20260730t181300z \
--evidence-root <evidence-root> \
--case qualificationTiming did not replace correctness checks
The qualification gate also checked the exact fixture branches and allocation count, logical size, zero copied payload bytes, correctness and security conditions, and post-run cleanup. The manifest bound the resulting evidence so the recorded result could be verified rather than inferred from a single console duration.
The repository report also retains failed phases and defects. For example, later activation work encountered capacity and control-cache handling defects that were recorded as failures rather than omitted. Publishing those failures is important: a passing warm-attachment case does not imply that every unrelated performance or lifecycle gate passed on the same day.
What this benchmark does—and does not—show
The result supports a narrow claim: the tested prepared LayerStack fixture could be attached below its 50 ms service target, while exposing the expected 8 GiB logical structure and copying no payload bytes. That property is relevant to multi-agent infrastructure because private sessions can reuse immutable prepared state without duplicating its logical payload at attachment time.
It does not measure model quality, tokens per second, coding-task success, full sandbox creation on arbitrary hardware, user latency, maximum concurrent agents, or a competitor speedup. Results from a Docker Desktop physical harness will vary with host, storage, image, fixture state, build revision, and workload. Use the test as implementation evidence, not as a universal SLA.
Source, revision, and reproduction
The cited result is preserved in the repository test report at core revision ac5c0686807ab40ee7e4ef3ff0f8488b66292221. Review the sealed source report for the command, pass statement, defect field, and surrounding failed phases.
Reproduction requires the repository’s documented benchmark prerequisites and supported sandbox control surfaces. Do not generalize the result to a different revision without rerunning the scorecard and retaining its new evidence. For the storage model behind the test, read the LayerStack architecture.