FINITE-SAMPLE CALIBRATION -- COMPLETED RESEARCH/VALUE CHECKPOINT
18 October 2026. STOPPED under the user's mathematical-value condition.

Source: FINITE_CALIBRATION_FINDINGS.txt
SHA256 db4525f5344d5add32130cbd253bb3106a8fe90ed86187bd0a83c474c7171de8
Retained best asymptotic proof: D:/LUMINARY/research/self_transport_nonuniform_revision2_20261018/revised_stopping_theorem.txt
SHA256 7d2ed0072718a9fbb84581ed4c67fe75d4781468592bafa3198e163bbcd12edd

Finding: at every fixed sample size, bounded confidence sets for the true point
score cannot have positive uniform coverage over the baseline's qualitative
smoothness class, even restricting density values to [1/2,3/2]. Smooth sine
alternatives verify this directly in total variation. An honest 1-alpha scalar
interval must be the whole real line with probability at least 1-2alpha under
uniform data. For the baseline Gaussian ball at deterministic bandwidth, a
stronger deterministic kernel-envelope argument gives a density with exactly
zero coverage for every sample. Counterexample densities vary with n/bandwidth;
the earlier fixed-f asymptotic theorem is not withdrawn or contradicted.

Known quantitative smoothness assumptions permit conservative finite-sample
KDE reference regions, with numerical/center-error inflation by set inclusion.
The explicit heat-bias/Bernstein construction was independently checked.
Generic Newton/adjoint certificates were also explored. No useful new
self-transport-specific statistical calibration law or sharpness/efficiency
advantage was established. These routes are established mathematics.

Two separate AI analytic reviews and a separate exact-source literature/value
review pass. They checked constants, quantifiers, same-sample adaptation,
the exact-zero-coverage corollary and the known-class alternative. Review files
retain source hashes. This is not external human peer review or formal proof
certification. The negative evidence requires no numerical experiment; no
Monte Carlo coverage study is presented as proof or as completed calibration.

Originality/value: this is a useful scope clarification and research-design
constraint, with clear prior general mechanisms (Dufour, Pfanzagl and related
honesty theory). It does not add a meaningful new publication contribution.
LITERATURE_AND_VALUE.txt records actual Google discovery, inspected primary
statements/versions and access limits. FINAL_ORIGINALITY_VALUE_REVIEW.txt
binds that assessment to the final source.

Preservation: all five prior manifest inventories were checked read-only;
PRESERVATION.json records the exact counts and absence/presence of discrepancies.
The current edition, older editions and unsuccessful finite diagnostics remain
unchanged. MANIFEST.json inventories this separate stopped research branch.

The strongest prior nonuniform stopping theorem remains the research candidate
for a specialist paper. This branch has no new manuscript, release permission,
publication, external-chat handoff or automatic continuation. Further pursuit
of a specified quantitative class or smoothed target would be a new scoped
research decision; none is silently started here.
