Conditional κ = 0.001672780159952542: reverse one entrance promotion to fill T16 gaps

research/demote635-t16/CLAIM.json

@@ -0,0 +1,15 @@
+{
+  "track": "complex supplier, continuation authorized after seed17 bit work",
+  "seed": 17,
+  "base": "2921874911956718fcaaaf9396fe8267b3cfa9f0",
+  "source": "589d1cedf5ef5f231182f4ec86408616311a8af0",
+  "object": "Reverse E5 to E4 promotion on helpers 635 637 638 639 at 109 calls, fill unit owner gaps through within-call permutations",
+  "method": "Restore 12 literal known pre-promotion gate frames; unchanged scalar construction; independently re-audit both words and mixed global admission",
+  "deliverable": "Conditional frontier improvement with exact reproducible certificate if verification succeeds",
+  "expected_finish": "Within 45 minutes of this local claim",
+  "at": "2026-10-13T18:07:52.559099+00:00",
+  "overlap": "PR664 introduced promotions; PR685 composed them; this reverses one group on new k8 T16 geometry and is distinct from PR1090 smaller ends. No pinned coordination issue.",
+  "result": "PASS_DEMOTE635_T16_COMPLETE_ADMISSION",
+  "kappa": "0.001672780159952542",
+  "completed_at_utc": "2026-10-13T18:22:43.591531+00:00"
+}

research/demote635-t16/NOTICE.md

@@ -0,0 +1,25 @@
+New material is licensed under Apache-2.0, as required by CONTRIBUTING.md.
+Prepared for seslly / shipstuff with substantial OpenAI Codex assistance in
+research, implementation, verification and writing.
+
+This contribution depends on Jeremy Kothe / jeremykothe2's PR1087 owner map
+and seven-slot k8 package; the package discloses Claude assistance. Credit
+PR664's entrance-promotion construction and PR685's composed promotion and
+owner work, including chafreaky / Chafik Boukhalfa. The literal k8 word retains
+ColdWitness / icekylinx PR1005, PR1018 including rohanarun PR987, NewBArch /
+Jeremy Kothe PR869, and the PR799/741/685/645/635/589 lineage and all source
+notices preserved in the pinned source packages.
+
+The independent owner checker is adapted from Hao Yan / FanPUYen's PR858
+indep16.py, through the existing mixed-provider adaptation for seslly. It
+continues to reconstruct literal event windows, physical charts and the
+invoice without importing the owner constructor or controller auditor.
+The native word and phase checks use the pinned PR635 and retained main
+sources without changing their mathematical predicates. All consumed source
+packages retain their original licenses, contributor notices and AI disclosures.
+
+Alexander Schwartzman's PR1030, and its PR1090 composition, are credited as
+concurrent work on narrower ends; they are not dependencies of this opposite
+entrance change. Rohan Arun's PR1094 is a concurrent owner-only improvement
+used for frontier comparison, not as a source. No exclusive priority is claimed
+for general entrance tradeoffs, packing, frame lowering, or independent replay.

research/demote635-t16/README.md

@@ -0,0 +1,115 @@
+# Conditional κ = 0.001672780159952542 from four lower entrance frames
+
+The exact conditional certificate is **836390079976271 / 500000000000000000**.
+It improves pinned PR1087's `0.001672654541898561` by
+`125618053981 / 10^18`. The conservative certificate is
+`0.001672780157102838`; the native complex lower endpoint is
+`1675583042021589509879 / 10^24`. The retained PR589 bit supplier is unchanged
+and freshly repriced. This is a finite conditional construction, not an
+unconditional multiplication theorem or a claim of κ > 0.01.
+
+The change reverses one of the entrance promotions introduced by PR664 and
+retained in the k8 construction. Helpers **635, 637, 638, 639** enter at
+`span(298,170,102,30)` instead of `span(298,170,102,30,1)`; twelve phase-B
+frames, at indices 372 through 383, decrease accordingly. Every scalar
+instruction, source access, center, helper end and cleanup instruction is
+unchanged. The new helper widths are five instead of four.
+
+A wider helper is useful here because PR1087 leaves many one-rank gaps next
+to width-four roles. At each of 109 eligible complete calls, a deterministic
+permutation of equal-width owner positions puts the four affected helpers in
+four such gaps, and their increased widths fill those gaps. The permutation
+stays within a call, preserving its occupied families and event windows.
+Affected bins are repacked contiguously and checked in full. The other 35
+calls use the original word. There is no split helper or additional call.
+
+## Complete cost
+
+Writing H_r for a paid rank-r child, the local change per altered call is
+`-8 H1 + 8 H2 + 4 H3 - 4 H4`, with rank mass +4. Filling four unit gaps
+removes another `4 H1`. The global delta is therefore exactly
+`109 * (-12 H1 + 8 H2 + 4 H3 - 4 H4)`, with zero change in total rank mass.
+This is a change to the paid recursive profile, not merely an ADD/XOR count.
+The exact moment engines price the literal aggregate rather than extrapolating
+from a numerical local score.
+
+| Quantity | Certificate |
+|---|---:|
+| Complete calls / whole helper roles | 144 / 69,120 |
+| Original / changed provider calls | 35 / 109 |
+| Paid rank mass / deficit | 1,041,408 / 5,120 |
+| Idle children | 1,844 H1 + 378 H3 + 6 H7 + 4,866 H48 |
+| Exact prefix / charged prefix bits | 397 / 512 |
+| Physical row coefficient / retained budget | 16,469 / 20,161 |
+
+The full global profile is
+`348300 H1 + 61064 H2 + 63022 H3 + 20012 H4 + 11808 H5 + 6 H7 + 1152 H8 + 4866 H48`.
+The unchanged controller checks source access, both orientations, wrappers,
+restoration, charts, every recursive child, the finite invoice and all 47
+strict outer inequalities. No purported saving is credited outside that invoice.
+
+## Reproduce
+
+From a full repository checkout, with Python 3.12+ and Git:
+
+```sh
+python3 -B research/demote635-t16/reproduce.py --out /tmp/demote635-fresh
+```
+
+The output directory must be new. Expected final line:
+
+```text
+PASS_DEMOTE635_T16_COMPLETE_ADMISSION 0.001672780159952542
+```
+
+The first run fetches missing immutable source commits. `--main /path/to/repo`
+can supply the main dependency tree when this package is checked out sparsely.
+No optimizer or third-party Python package is needed. `SOURCES.json` pins
+base **2921874911956718fcaaaf9396fe8267b3cfa9f0**, PR1087 at
+**589d1cedf5ef5f231182f4ec86408616311a8af0**, and the PR635 native verifier at
+**a6895a69f8ad98ab45777b03e9e6f48888c39f6f**. The parent manifest and every
+consumed main-file pin are checked. The literal baseline word is bound to
+PR1087's construction payload hash.
+
+The command reconstructs the changed word and owner permutation, freshly
+replays both 704-column scalar words in both orientations, both sets of
+2,896 exact phase edges, wrappers and independent frame/cone audits, and then
+replays the parent and mixed global admissions. It rejects 19 additional
+frame/scalar/profile mutations, 23 controller mutations, 114 admission
+mutations, 12 bit-supplier mutations and six independent owner mutations.
+Each native word also carries its 25 native controls. Optimized Python is
+rejected before any output is created. Regenerated containers are compared
+through the inherited semantic identities; committed source containers retain
+exact hashes.
+
+The portable command starts from the pinned literal word. It does not rerun
+the historical k8 search/generator chain or the unchanged PR589 bit circuit.
+Those are inherited evidence, not claimed as fresh work. The affected native
+and global checks are fresh. This package changes no shared generator,
+selected-frontier artifact or `upstream/` file; repository-wide historical
+suites and hosted CI are separate from its scoped verification.
+
+## Scope, prior work and overlap
+
+PR664 established entrance promotion; PR685 and the subsequent k8 lineage
+composed it with other changes. This contribution reverses one group against
+the new T16 owner geometry. The older observation that promoting every group
+can lose to packing costs is retained. The new evidence is the explicit
+opposite transformation, within-call placement and admitted complete profile
+on PR1087. It does not claim a new general optimization theorem.
+
+PR1090 independently constructs narrower k8 ends and explicitly reports a
+smaller pending κ. Our saved earlier helper543 narrow-end experiment was
+also superseded. Neither is a dependency of this word. PR1094's verified
+`0.001672660197076918` and PR1107's reported fully replayed
+`0.001672765980186726` were checked during preparation and are below this
+result. `VALIDATION.json` records the final metadata refresh. Neither
+concurrent owner map is consumed. Historical and bounded unsuccessful investigations are
+summarized in `SEARCH.json`; none is an impossibility proof.
+
+All-size recursive compilation, common labels and mutable-source realization,
+whole-call physical hosting, precision/tape accounting, exact recovery,
+analytic transfer and PR589 bit assumptions H1–H5 remain inherited obligations.
+The finite checks neither discharge these contracts nor constitute a full
+Lean proof. Contributor credit and substantial AI assistance are disclosed in
+`NOTICE.md`.

research/demote635-t16/SEARCH.json

@@ -0,0 +1,46 @@
+{
+  "assignment_seed": 17,
+  "randomness": "OS-generated in the original sprint assignment; retained for this continuation",
+  "source_word_sha256": "c2b294300760e1fd6eeab8c22c4d13e4633dca40e52cac8da0d68163a76e3ca8",
+  "selection": "After negative fixed-cell searches and a superseded narrower-end contribution, test the opposite entrance/packing tradeoff. Helpers 635,637,638,639 are the first group in the source promotion frame recipe.",
+  "completed_coverage": [
+    {
+      "object": "21-gate leaver cells on new k8 word",
+      "domain": "All 720 simultaneous S6 source/target relabelings in each of eight cells; fixed external frame boundaries",
+      "attempted": 5760,
+      "feasible": 5760,
+      "result": "No paid-profile improvement in this class",
+      "method": "Finite subspace-factor elimination with numerical nomination costs; no formal optimum theorem claimed",
+      "limits": "180 seconds per cell; 200000 assignments per separator; neither limit reached"
+    },
+    {
+      "object": "Alternative pair-only 21-gate cell",
+      "domain": "All 720 permutations in cells 01+ and 23+ with fixed boundaries",
+      "result": "No feasible frame schedule in these boundary-restricted classes"
+    },
+    {
+      "object": "Forward frame closure for cell 23+",
+      "domain": "720 original and 720 alternative pair-only permutations",
+      "result": "Original: 720 feasible, no improvement. Alternative: none feasible with fixed ends."
+    },
+    {
+      "object": "Entrance promotions on new word",
+      "domain": "64 single rank-one promotions and same-cap groups/consecutive pairs, 96 distinct candidate sets",
+      "result": "96 forward-frame candidates; no profitable unmodified-owner placement on PR1087"
+    },
+    {
+      "object": "One four-helper promotion plus local owner reassignment",
+      "domain": "24 seeded neighborhoods of 31 owner bins on PR1087, 15 second MILP limit each",
+      "result": "No complete-cost improvement found; time limits are not infeasibility proofs"
+    },
+    {
+      "object": "Opposite entrance change selected here",
+      "domain": "Reverse one known four-helper promotion; inspect all 144 calls for four width-four roles in distinct one-rank-gap owner positions",
+      "attempted": 144,
+      "eligible": 109,
+      "result": "Literal recipe plus deterministic within-call permutations; exact native and complete assembly replay",
+      "stopping_rule": "Stop exploration and package the first fully admitted frontier improvement"
+    }
+  ],
+  "scope": "A finite search record, not an impossibility or global optimality claim. Earlier smaller-end evidence is retained on a separate local branch and is not submitted in this contribution."
+}

research/demote635-t16/SOURCES.json

@@ -0,0 +1,12 @@
+{
+  "base_commit": "2921874911956718fcaaaf9396fe8267b3cfa9f0",
+  "repository": "https://github.com/CrocSwap/integer-mult-bounds.git",
+  "parent_pr": 1087,
+  "parent_commit": "589d1cedf5ef5f231182f4ec86408616311a8af0",
+  "parent_package": "research/gauge44-slots-t16",
+  "parent_manifest_sha256": "fe55cb2812dc863a4bcd9e6083474f7aa986d2095ef66ae210ba85b251fe86c9",
+  "native_commit": "a6895a69f8ad98ab45777b03e9e6f48888c39f6f",
+  "native_package": "research/complex-571-cold14-st2",
+  "comparison_kappa": "0.001672654541898561",
+  "baseline_payload_sha256": "c2b294300760e1fd6eeab8c22c4d13e4633dca40e52cac8da0d68163a76e3ca8"
+}