Saved algorithm result · GEO-07

Registered dependent generator sum

Two bounded public GeometryPort fixtures with shared factor IDs. The generator algebra is equivalent over real sets; binary64 evaluation has no validated rounding enclosure.

Numerical outcomes

FixtureLiteral physical referenceObserved generator resultChecks
Shared cancellationx = 0Center 0; retained history.0 factor with coefficient 0Three feasible factor witnesses map to 0; −1 and 1 are outside. Independent Minkowski sum rejects shared IDs.
Constrained shared factorsu − v = 0 and u + w = 0 imply (x,y) = (0,0)Joined factors [u,v,w]; coefficients [0,0,0; 0,1,1]; both equality rows retainedThree feasible witnesses map to the origin; (0,1) and (1,0) are outside. Independent sum rejects shared IDs.

Every checked call used fresh handles, the registered public port, prepared typed DependentSum, workspace execution and literal factor/H checks. Evidence is structural with numerical assurance and no error bound. These examples do not prove all geometry inputs or validated floating point equivalence.

Verified batch observations

One excluded warmup and seven alternating measured batches per fixture. Each batch made 2,000 fully checked calls. All 16 warmup and measured batches passed; all 14 measured batches exceeded 20 ms.

FixtureCalls/batchSuccessesMedian msRange ms
Shared cancellation2,0007/725.95725.648–28.288
Constrained shared factors2,0007/732.20131.590–33.118
Verified batch observations010203040Shared cancellationConstrained shared
Seven measured batches per fixture on one shared millisecond scale; warmups excluded.

Performance interpretation is exploratory and inconclusive: no comparator, speedup, source-mapped limiter profile, application-scale workload or isolated native-operation latency is claimed.

The steady_clock interval includes fresh handles and registration, admission, workspace execution, literal checks, independent rejection and per-call JSON result assignment. It excludes startup, JSONL serialization, Python validation, persistence and rendering. Internal phase counters and observed native thread count are unavailable.

Source and host evidence

Algorithm source 5411e37ad456f0d5b3bf775130f149622ef5ebfa; reviewed harness code eaa32fdfdcf42b3f3f3221dd3031d654fe05e576; sealed collection HEAD eef1eda3d326418a392b3cc1e8f8f4b204ca7a73; saved report geo07-dependent-4dcc5a8fe36f17ed. Raw SHA-256 3cc538611451add4eaef7adcbb5f225bbe668eadb8c32bd9e98de298f85fbc0c; records SHA-256 fb4a0bab5953e679ece8e814aeb2a0fcbe68c4816fad3cc097c9e1b48774399c; native manifest SHA-256 08e5bda248ee6eee7dd54732a7427b07f3c27638ef1001e588262c7ff5edc72a.

MSVC 19.44.35215.0, x64 C++20 Release /O2 /Ob2 /DNDEBUG; Windows-11-10.0.22631-SP0, 12 logical processors. Immediately before measurement, the collector observed CPU 0% and zero cmake, MSBuild or cl processes. One sequential driver worker was requested.

Two earlier attempts were rejected and preserved in evidence/rejected-001 and evidence/rejected-002. The first observed three idle MSBuild nodes and stopped before trials. The second executed all literal checks, but its 1,000-call batches fell below the 20 ms admission floor. Neither contributes measured results.

Download and audit

The original process captures, failed attempts, inputs, build evidence, records and rendered tables are available below and in the offline ZIP.