Performance conclusion: inconclusive. One numerical adapter was measured without a comparator or source-mapped limiter. The eight fixtures use different batch sizes; the timings do not establish a speedup or single-call latency.
Frozen algorithm source 994f3ee4310b4087d289e2eb02adbdb9dfd97827; reviewed harness 2f915110ba42515fdc448a0c226dc1290b80f82d. Release MSVC/C++20, binary64 numerical assurance. One excluded warmup and seven measured batches per fixture, 56 measured batches in all; zero failed checked batches.
Host: Intel64 Family 6 Model 151 Stepping 5, GenuineIntel, Windows-11-10.0.22631-SP0; 12 logical processors. Compiler: MSVC 19.44.35215.0; CMake generator: Visual Studio 17 2022. Each public call is checked; process/runtime caches may be warm after the excluded warmup.
The timed interval includes fresh public calls, fixture construction, literal result checks and result construction. It excludes process startup, JSONL serialization, disk persistence, plotting and HTML rendering. Native thread count and internal phase counters were unavailable; one driver worker was requested. Other project worktrees had active local builds/tests during collection, so host contention may affect these exploratory durations.
On narrow screens, scroll the table horizontally to see every column.
| Fixture | Status | Checked calls per batch | Measured batches | Median batch ms | Batch range ms | Input SHA-256 |
|---|---|---|---|---|---|---|
| l19 | numerical_optimal | 10000 | 7 | 196.742 | 183.395–210.982 | f1c6097654913bd8a326370568a7d475e23bd3553b93776a98cf9acb6c9bb910 |
| integrator | numerical_optimal | 5000 | 7 | 75.580 | 73.877–78.382 | 6f1b01083814fe8587d034a10362a21470f3137a44f846212e27b26f786c4206 |
| all_infeasible | infeasible | 5000 | 7 | 69.757 | 68.735–71.645 | 27b2be0e3d8118643466531993b6df1fbcc328bcd83cda020da9f21f6b509f24 |
| empty_catalogue | empty_catalogue | 5000 | 7 | 30.373 | 29.999–31.244 | ecd876e1d4b5711861b0f6aeb44f830cc9681f301a4b004b65926650e90e9cfa |
| no_scenarios | no_scenarios | 5000 | 7 | 37.121 | 36.512–43.870 | 409b79001bbb00ca49aa1f724937b9e0b628d0e37b34a2f20ffa2f1069ed7e7a |
| unsupported | unsupported | 5000 | 7 | 34.495 | 34.253–34.922 | 6f1b01083814fe8587d034a10362a21470f3137a44f846212e27b26f786c4206 |
| cancelled | stopped | 5000 | 7 | 37.932 | 37.440–39.395 | 410fbcf3487259261dda4cc5d5e4aeb8966827eb584be25ed00fd73edaaf03d3 |
| b08_rounded | numerical_optimal | 1 | 7 | 84.898 | 83.338–86.383 | 47ec040f5631dd1b24012318223efb73fa2bb7aa2db76c03c56cb88f2a85e733 |
L19 checks two one-step actions and endpoint maximizers at 0 and 2; the stationary midpoint has lower objective. The integrator has a unique minimizing plan. Infeasible evidence retains both ordered violating successors. Empty catalogue, no scenarios, Unsupported and cancellation are distinct statuses and publish no optimum.
The displayed rounded B08 four-step linear model enumerates 625 fixed open-loop plans and 256 disturbance corners per plan, 160,000 terminal evaluations. Its numerical minimum is about 67.1506011014, with plan indices 17, 33 and 129 tied within 1e-8. This does not establish the historical 67.12, unrounded or nonlinear model, feedback optimum, or validated arithmetic enclosure.
All checked results, source and harness IDs, fixture/input hashes, process outcomes, and build logs are retained in the offline bundle. Publication and live URL verification follow independent saved-result review.
Versioned publication issue3-finite-catalogue-388b4be3a35ac10c9b42bf8f92ac5ae06b2120049dc6003791bc6487628478de. Report envelope · Publication manifest.