OPT-07 ordered hull membership

Native record opt07-bbb4aee749d643ff; measured source commit 52444e305684c0b59c049cce9f9de7ea09eb4a4a. Algorithm csl.hull.ordered-membership version 1.0.0; backend csl.lp.highs adapter 1.0.0+highs.1.11.0.

Three fixed binary64 clouds exercise signed-score ordering, a zero-width duplicate, and a late exterior point. Every one of 2,400 calls passed the independent exact-dyadic certificate and literal support oracle. This is evidence for these represented inputs, not a general or validated-real hull theorem.

Measured batches

Each marker is the median of seven measured 100-call batches; the line spans minimum to maximum. One warmup batch per fixture is excluded. Durations are for registered run_with calls and output moves; fresh workspace construction, independent validation, JSON and HTML are outside the interval. Lazy native initialization inside the calls is included. No per-call latency is inferred by division.

case_id	points	nonwarmup_batches	nonwarmup_calls	median_batch_ns	min_batch_ns	max_batch_ns
signed-score	6	7	700	52116200	51192500	53811700
zero-width-duplicate	4	7	700	35886100	35349400	41002000
late-outside	4	7	700	36184800	35798400	37316000

All batch outcomes

case_id	round	warmup	calls	verified_calls	elapsed_ns	status
signed-score	0	True	100	100	54493400	complete
signed-score	1	False	100	100	53691000	complete
signed-score	2	False	100	100	52385400	complete
signed-score	3	False	100	100	51580600	complete
signed-score	4	False	100	100	52116200	complete
signed-score	5	False	100	100	51192500	complete
signed-score	6	False	100	100	52023400	complete
signed-score	7	False	100	100	53811700	complete
zero-width-duplicate	0	True	100	100	35420500	complete
zero-width-duplicate	1	False	100	100	41002000	complete
zero-width-duplicate	2	False	100	100	35886100	complete
zero-width-duplicate	3	False	100	100	35802800	complete
zero-width-duplicate	4	False	100	100	37373700	complete
zero-width-duplicate	5	False	100	100	36054200	complete
zero-width-duplicate	6	False	100	100	35782800	complete
zero-width-duplicate	7	False	100	100	35349400	complete
late-outside	0	True	100	100	36101400	complete
late-outside	1	False	100	100	37072200	complete
late-outside	2	False	100	100	35798400	complete
late-outside	3	False	100	100	36088900	complete
late-outside	4	False	100	100	36184800	complete
late-outside	5	False	100	100	36447300	complete
late-outside	6	False	100	100	37316000	complete
late-outside	7	False	100	100	36124100	complete

Performance interpretation: descriptive single-adapter batches; no comparator, speedup, source-mapped limiter, or scalability claim. Internal LP phase durations, pivots and observed native thread count are unavailable. Configured native threads: 1. OS: Windows-11-10.0.22631-SP0; machine: AMD64; logical processors: 12. Release compiler project: stdcpp20 with MaxSpeed and NDEBUG. Exact compile command is in the saved sanitized Release build log. Concurrent-work note: Pre-run observation (2026-10-16 23:22 Europe/Moscow): Windows host with 12 logical processors; Win32_Processor LoadPercentage 7. Three resident MSBuild processes used 0.00 CPU seconds over a two-second check after the clean Release build. No other deliberate local compilation or benchmark is running. GitHub Actions checks are remote; helper agents are idle. One excluded warmup and seven measured 100-call batches per fixture are planned..

Saved evidence

Generated from saved bytes without running the algorithm. The raw local build-log SHA-256 is preserved in collection provenance; its public copy is sanitized and separately hashed.

Publication opt07-6e77b298e159fa0b80d5d69ccd8c14233eb8830513a5d060f98ee76d8e414c96; native report opt07-bbb4aee749d643ff. Versioned envelope · Publication manifest.