[tugrapaydiner]

Continue from the user's actual latest code

Base: 098a8477f6cc8bb696423e78c4d6b6a26a95a0e1 on faster-queries.
The user's automatic selection, prepared-root proofs, direct SAT serialization, reusable proof scratch, Windows support and seven-arm benchmark remain intact. Main and previous research PRs are not merged or replaced.

Tested implementation: 8d28a63707a1e3d56d5d16688924cd3985ee80d7.
Native SHA-256: 67a8b5c8c55dd4fdc5070f2308cabf37a463a7a1294f12a8dd496d792852fb25.
Subsequent commits only remove one-shot publication machinery and add/correct reporting. The retained workflow is read-only.

Code

Remove redundant implication-arc and observer-conflict global sorting after canonical input admission. A documented ordering invariant shows stable CSR scattering produces exactly the old forward/reverse arrays. Diagnostic native builds compare all arrays and arc counts with the sorted reference. Raw mode gets the same applicable improvement. No ABI, format, cap, policy threshold or default changes.

Completed checks

• 455 local compiled-query tests passed in recorded groups/fresh processes; 152 focused tests are a subset, not extra tests.
• 152 focused tests on each of x86 Python 3.11, x86 Python 3.13 and native ARM64 Python 3.13.
• 156 Windows x64/MSVC Python 3.13 tests passed.
• 112 exact contracts under UBSan and 112 under ASan+UBSan; leak detection disabled.
• 1,200 five-mode comparisons preserve full output bytes AND returned metadata against the user checkpoint.
• All seven jobs in run 38781025834 passed, including publication of the tested code and three full comparisons. All three downloaded audits replay identically under Python -S without loading a native solver.

Measured against the immediate parent auto engine

Same 48 exposed graphs, five lengths, seven arms, nine paired repetitions and complete checked ready-packet endpoint. Each environment completes 15,120 timed calls and 48 workers.

Environment	Single-request new/parent time	1,024-request new/parent time	New auto / new raw at 1,024
Linux x86 Py3.11	0.552124	0.931766	0.878810
Linux x86 Py3.13	0.621886	0.960653	0.897270
Linux ARM64 Py3.13	0.663518	0.945146	0.882249

All three retain the unchanged 10% development target. The user's preceding run already passed this target; that accomplishment is not attributed to this PR.

Limitations retained

Exposed synthetic development, not new confirmation or outside-team replication. Ready-packet timing excludes packing, compilation, process startup and offline replay. Explicit route still loses on many short-call cells. The isolated auto/raw slow cells, initial incomplete combined local test run, and all older failures remain documented. No full repository/model-suite rerun, universal no-regression claim, leak-checking result, Windows timing panel or algorithmic priority claim.

Full report: experiments/compiled_query/LINEAR_REVIEW_20261015.md.
Construction proof: experiments/compiled_query/LINEAR_CONSTRUCTION.md.
Completed workflow: https://github.com/tugrapaydiner/SPECTRA/actions/runs/38781025834

Draft for review; no merge or default promotion requested.