[Clear1nsight]

Change and mathematical scope

Retain the completed, source-bound Fraction zero/unit feasibility finding as an isolated research package. The prototype preserves exact outputs but its recorded 0.237-second trace difference is insufficient headroom for the requested 15% complete-stage target. This is a scoped negative finding, with no performance or κ gain claimed.

The frozen PR #5848 qualifier profiled at 38.314 seconds, including 20.053 seconds in Fraction error propagation. One ordered full actual trace comparison took 20.180 seconds for the original and 19.943 seconds for the derived routine: about 0.62% of the profiled complete qualifier, versus roughly 5.75 seconds needed for 15%. These observations are from an unreserved host and do not establish a reliable speedup, CPU saving or universal obstruction. The experiment stopped before constructing a complete candidate or running a complete B/C wall/CPU benchmark.

The package supplies the exact prototype, written value/type argument, finite replay, full actual-audit script, pinned export and raw evidence. The frozen scientific sources, independent integer backend, precision guards and selected result remain unchanged. Archive integration would preserve this finding; it would not imply scientific approval or admission.

Task and overlap

Explicit donor assignment, refining replay-runtime to PR5848/frozen-66fb6c6/Fraction-zero-unit-small-feasibility. Base: de5ec2e7a6074068452f45b1a191b5c72832df4a, priority revision 8. Artifact head: 71f22c77b393e5cc14925c771894dc1a92c8443c. Its immediate input ancestor e5614e94d93945963268e76fbd37f7dec58fde78 retains the complete 97-file official export of frozen PR5848 66fb6c6b8683b37967e9d7f313b3df99a49d2406, including unchanged PR #5831 and PR #5826 archives.

The canonical released handoff and hub pointer originally retained this on the fork after the negative gate. The donor now requests upstream retention through a focused finding PR; the frozen package and result are unchanged. Historical statements that no upstream PR had been opened describe that earlier checkpoint.

PR #5848 and PR #5839 supply the validated qualifier/Linux baseline. This adds a narrower exact shortcut feasibility test. The later contribution-table finding profiles a broader primitive/table algorithm and is a separate package, now submitted as PR #5852. Targeted upstream searches for this head branch and package path found no existing PR at publication preparation. This is bounded coverage and makes no exhaustive novelty claim.

The branch also retains the exact test-only PR #5761 dependency: Ali Romman's ba8a9e640e1983f3c29993fefa3f46e8ed03313a, cherry-picked as 01014cbe772ea5a59ccbc55266222433944c7368, preserving authorship. It changes no scientific artifact.

Validation

Recorded CPython 3.12.3/Linux evidence covers all 241 finite programs / 10,164 events, complete values, signed witnesses, numeric types and unchanged inputs. All 16 source/code/type/mutation/exception/restoration controls pass normally and under optimization. One full actual audit preserves all 25,984 events / 799 coordinates and every trace/witness value and type. Full output SHA256: 954c7448ff74d78753828e9623ad44c46da9e5ce459420b6e50a5d47294250b9.

Two preliminary checks rejected before arithmetic: a stale archive-member uniqueness count and a nested-code comparison that included source locations. Exact failed sources, hashes, receipts, diagnoses and labelled corrections remain visible. Seven of nine guarded attempts passed; total recorded wall time was 87.116 seconds under the original 180-second per-command, 600-second initial-stage, 1,500-second shared and 1 GiB sampled group RSS limits, with no automatic retries. No failed mathematical differential was repaired.

Publication preparation reran metadata and committed-byte/mode checks only: PASS_PACKAGE_METADATA_AND_PINS and PASS_RESEARCH_COMMIT_BYTES for all 61 package files. It read the existing packaged normal/optimized finite receipts; it did not rerun scientific work. The shallow checkout cannot supply full-history preflight.

Portable finite reproduction, Python 3.12 standard library, about one second locally:

python3 -B scripts/research_package.py --package research/fraction-zero-unit-feasibility

The registered command covers fresh finite feasibility only. The opt-in actual_audit.py is about 43 seconds locally and requires a fresh serial resource budget; it is excluded from the registered replay. Complete candidate normal/optimized/sealed validation, full wall/CPU qualification, hosted CI, independent human/kernel acceptance and physical supplier admission are unclaimed.

Credit

Ali Romman supplied the original Fraction/error arithmetic and integer backend. Preserve PR #5831, PR #5826, PR #5819, Adam Gall, ywz65, PR #2878, Mason Hale/PR #5811 and every retained P104/native/compiler/precision notice. PR #5848/PR #5839 supply the qualifier/Linux baseline; PR #5816, PR #5818, PR #5832 and PR #5761 retain guard/custody/tooling credit. New feasibility work: Chris R / Clear1nsight with substantial OpenAI Codex assistance. The original experiment used no subagents; this publication preparation uses donor-requested agents. Apache-2.0.

Open next questions

The package's primitive-profile question was subsequently addressed by the separately scoped contribution-table finding. Any successor must establish fresh source/overlap/coverage gates and pay the complete qualification cost before a performance claim. This archival publication dispatches no new experiment. κ research, Work 1/2 and feeds remain paused.

[Clear1nsight]

PUBLICATION UPDATE / SCOPED NEGATIVE FINDING / RELEASED — upstream archival retention requested by the donor. No new scientific work was run.

1. Research discussion / PR / artifact: PR #5853 retains the unchanged zero/unit artifact. It supplements the original released finding and hub pointer. Their no-upstream-PR statements describe the earlier fork-only checkpoint.
2. Exact commit: 71f22c77b393e5cc14925c771894dc1a92c8443c, branch Clear1nsight:verify/fraction-error-fastpaths-20261027; input ancestor e5614e94d93945963268e76fbd37f7dec58fde78; frozen PR5848 66fb6c6b8683b37967e9d7f313b3df99a49d2406; canonical base de5ec2e7a6074068452f45b1a191b5c72832df4a, priority 8. Targeted head/package searches found no duplicate before submission.
3. What is established: Recorded 241 programs / 10,164 events and 16 controls pass normally/optimized; one actual trace preserves 25,984 events / 799 coordinates and complete typed values/witnesses/inputs. The 20.180 → 19.943 s ordered trace comparison removes only 0.237 s, about 0.62% of the 38.314 s profiled qualifier, insufficient headroom for the 15% complete-stage gate. Publication checks verify all 61 committed bytes/modes, metadata/pins, 97-file/98-member export, immediate ancestry and prior portable receipt stream hashes. Package unchanged; GitHub confirms this PR open at the exact artifact head.
4. What remains uncertain: No complete candidate, B/C wall/CPU benchmark, reliable speedup, universal obstruction, independent human/kernel acceptance or supplier/κ admission. Exact-head hosted planning is in progress; no hosted pass is inferred. Two pre-arithmetic rejections and their diagnoses/corrections remain; seven of nine prior attempts passed in 87.116 s. Full-history preflight remains unavailable in the shallow checkout. Publication checks reran no science.
5. Smallest useful next task: PR #5852 archives the later primitive/table finding that addresses the original profile question. Any successor requires fresh source/overlap/coverage and complete-cost gates. This archival update dispatches none; κ research, Work 1/2 and feeds remain paused.
6. Reproduction command / expected runtime: python3 -B scripts/research_package.py --package research/fraction-zero-unit-feasibility, Python 3.12 standard library, about one second, fresh finite-only coverage. The separate actual audit is about 43 s and requires a new serial budget.
7. Predecessor credit: Ali Romman supplied Fraction/error arithmetic and the integer backend. Retain PR Speed up complete P104 precision replay with proved exact integer arithmetic #5831/Certify current P104 precision with additive COPY and reset residuals #5826/Certify sharp P104 prefix norms and completed-error guards #5819, Adam Gall, ywz65, PR Certify completed-error recovery for the retimed P104 dual #2878, Mason Hale/PR Raise conditional κ to 0.00312418633064 with a fourth accumulated update #5811, PR Qualify precision once per backend: 36% less time, full coverage #5848/Validate PR5831 on Linux: 4.40x complete precision replay, full coverage #5839, PR Validate complete PR #5757 replay on Linux: 2.60x faster with full coverage #5816/Share complete P104 qualification: 33.7% less time, both literal oracles #5818/Reuse literal gate setup: 25.9% less qualification time, full coverage #5832, exact test-only PR test: preserve legacy routing coverage after priority revision 8 #5761 source ba8a9e640e1983f3c29993fefa3f46e8ed03313a / cherry-pick 01014cbe772ea5a59ccbc55266222433944c7368, and all notices. Full links and credit remain in the original canonical handoff and this PR body. Chris R / Clear1nsight with substantial OpenAI Codex assistance. Original experiment used no subagents; donor-requested agents prepared publication. Apache-2.0.

[Clear1nsight]

CI UPDATE — exact head 71f22c77b393e5cc14925c771894dc1a92c8443c is green in hosted run 40465196696. Planning passed 513 shared tests with 22 unchanged platform skips and selected only research/fraction-zero-unit-feasibility. The fresh finite replay passed all 241 programs / 10,164 events / 16 controls, the registered negative-result check and committed-input integrity. This is finite-only hosted coverage, not a fresh full actual-trace/profile replay or complete performance gain. No frozen package, source, branch head or stopping decision changed. The publication handoff and predecessor credit remain applicable. The companion PR #5852 is also green; its new entry-gate follow-up records a further below-15% complete-cost screen. Kappa research remains paused.