research: release T740 precision35 kappa refinement

[Gonzalo-Mier]

Result

This releases the next exact 35-digit endpoint for the unchanged T740 six-skip profile:

• C35 = 86486897503213927562144791190867 / (5*10^34) = 0.00172973795006427855124289582381734
• κ = C35 - 10^-57 - 10^-81

It is strictly above the exact PR #3115 claim by 40867/(5*10^34) + 10^-52 + 10^-76 - 10^-57 - 10^-81 (about 8.1734e-31). This is a small arithmetic precision refinement. It does not change the paid mechanism and does not approach κ > 0.01.

Evidence

• The complete unmodified T740 baseline release and complete strict candidate release both pass.
• Both rational moment engines accept C35 and reject the adjacent grid point, unpaid rank-1 and rank-48 controls.
• The candidate retains all 63 strict outer rows, 17 outer rejection controls, and the authenticated hash-bound native receipt. The strict candidate release ran fresh; only the explicitly hash-verified discovery/outer evidence and unchanged native receipt were reused.
• The registered Python 3.12 package replay passes, including five behavioral controls under optimized Python. check_research_commit.py confirms publication bytes match commit 776ae75d640e6a5672e8ea798677a2cd3ae2d53c.

Reproduction and source pins are in research/t740-six-skip-precision35/README.md. The strict release uses the pinned Python environment and native build.

Frontier correction

PR #3133 published a higher exact κ than this T740 refinement. PR #3143 now refines #3133's endpoint again and is the highest public κ claim. This PR remains an improvement over #3115 only within its unchanged T740 six-skip profile; it does not exceed the current public maximum or approach κ > 0.01.

No selected-frontier update, independent human review, formal proof, or hosted CI pass is claimed. All inherited compiler, physical, precision, caller and analytic hypotheses remain. The contribution retains attribution to the T740 six-skip predecessor and its source lineage.