Outer retune of #2260: #1673 capped phase-cell fusion outer on the selective-borrowing supplier, conditional κ = 0.00168341258854…

.gitattributes

@@ -151,3 +151,6 @@ research/p111-minrank-proof-reconciliation/vendor/pr620/geo.py -text -whitespace
 research/community-batch1099-audit/** -text -whitespace
 research/community-frontier1094-audit/** -text -whitespace
 research/abelian-cayley-metric-scope/** -text
+
+# Preserve exact bytes of the PR2260 outer-retune package (vendored manifests hash CRLF originals).
+research/selective-borrowing-outer-retune/** -text -whitespace

research/selective-borrowing-outer-retune/ENDPOINT.json

@@ -0,0 +1,24 @@
+{
+ "schema": "pr2260-complex-endpoint-v1",
+ "endpoint_digits": 200,
+ "complex_lower": "84170629427130373116515987019996848025185486122559439565749892539974576607402329484274052554882976074372640281764827503402818127399337032498416431355363381039395040713096498569825471629631417624971/50000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000",
+ "native_moment_upper_minus_one": "-466613893423209518177921260564632055104983223656392446505474389023721262048553556326995186369512808582930453531274552714040321049531255201448249184773777251000263731719719497243345657319686788193503437796712642056306223183203814278564287827795421499845366525899041634408553199930980324673394005653995514456715570300885490565/55635907749862844386400314505595567251599107970324256757699365672051224279213746123593285972818168186054976565058958882713758619825090273161853126012111375244844087293568574798982611534498532273220965009515410130996422392922644302042543251587903817239657511863583535042948644880667099640152669697790704265082737250094246104040101380096",
+ "conservative_moment_upper_minus_one": "-1604311886227731499673438010683704032301860041213536077101918791653484672422351020721085313388109096199549518898181706420838569638089533/55635907749862844386400314505595567251599107970324256757699365672051224279213746123593285972818168186054976565058958882713758619825090273161853126012111375244844087293568574798982611534498532273220965009515410130996422392922644302042543251587903817239657511863583535042948644880667099640152669697790704265082737250094246104040101380096",
+ "conservative_moment_lower_minus_one_at_next_grid_point": "8782818318892888361895183238291065070329224689158219775274982180462987877949718149496206647447748541722776447615606448679750013968391/27817953874931422193200157252797783625799553985162128378849682836025612139606873061796642986409084093027488282529479441356879309912545136580926563006055687622422043646784287399491305767249266136610482504757705065498211196461322151021271625793951908619828755931791767521474322440333549820076334848895352132541368625047123052020050690048",
+ "profile": {
+  "1": 176252,
+  "2": 30096,
+  "3": 30908,
+  "4": 10152,
+  "5": 4493,
+  "6": 18,
+  "7": 494,
+  "8": 576,
+  "32": 11,
+  "39": 7,
+  "48": 2493
+ },
+ "m": 73,
+ "W": 7168,
+ "surcharge_weight": "5329/640000000000000"
+}

research/selective-borrowing-outer-retune/NOTICE-PR2220-OUTER

@@ -0,0 +1,8 @@
+New research prepared by Rohan Arun with substantial OpenAI Codex assistance. Apache-2.0. Parent words and build recipe method derive from Jeremy Kothe PR1274 and Rohan Arun PR1218, with lineage credited in README.md. All inherited mathematical qualifications are retained.
+Additional pivot555 and567 composition by Rohan Arun with substantial OpenAI Codex assistance.
+
+T74 exact-fill owner construction by Rohan Arun with substantial OpenAI Codex assistance.
+
+PR1377 grid refinement credited to Jeremy Kothe. Outer witness refinement by Rohan Arun with OpenAI Codex assistance; construction unchanged.
+
+Pivot-566 kernel, words, edits, rebuild.py and rebind.py: PR1625 (mdumitrean). Endpoint/outer-witness stacking, fine_cache.py and oracle stage by jeremykothe2 with oh-my-pi assistance.

research/selective-borrowing-outer-retune/README.md

@@ -0,0 +1,61 @@
+# Outer retune of PR2260: the PR1673 capped phase-cell fusion outer on the selective-borrowing supplier
+
+**Conditional final kappa = 0.001683412588542607462330319740399936960503709722451188791314997850799491532148046589685481051097659521487452805635296550068056362547986740649968328627107267620787900814261929971396509432592628352499419999999999999999999998999999999999999999999999**
+
+That is C − 10^-246 − 10^-222, where C is PR2260's complex supplier endpoint certified here to 10^-200. It is about +0.1733% over PR2260's 3361/2000000 = 0.0016805 and about +0.2177% over PR2220's 0.00167975495612554…
+
+**What is new here: nothing in the construction.** This package is an *outer retune*. PR2260 (Dugongue) built the finite complex supplier: the 44-alias selective borrowing on PR2226's architecture, 436 helper roles, and a new physical owner packing of the T8 controller. PR2260 assembled it with the retained balanced-prefix outer, whose final kappa is about b/(1+b). PR2220 instead applied PR1673's capped phase-cell fusion outer to its supplier, with strict slacks retuned to the fixed-grid maximum (PR2180 / PR2161). This package applies that same outer, and the same PR2220 ledger code, unchanged in formulas, to PR2260's supplier. The final kappa is then the supplier endpoint minus a fixed-grid backoff.
+
+## What is checked (`python3 -B verify.py`, assertions on; numpy + SymPy for the vendored outer evidence, as PR2220)
+1. **Manifest.** `MANIFEST.json` pins every file.
+2. **PR2260 supplier, replayed from scratch.** The package is vendored verbatim at its head `fde6b107` (`vendor/pr2260-selective-borrowing-kappa`, its own manifest pinned). The replay runs PR2260's own steps: native local words, owners, the independent family-factor audit, the finite invoice, and its exact arithmetic including its own 47-row outer at 3361/2000000.
+   - *One stated deviation.* PR2260's independent receipt pins the SHA-256 of a gzip byte stream, and those bytes depend on the zlib build: CPython 3.9–3.14 here produce different bytes from PR2260's record. Every other receipt key is compared verbatim with PR2260's EXPECTED.json. The *uncompressed* family-factor payload hash in `SUPPLIER-BINDING.json` is a reproducibility pin taken from this package's own replays, identical under CPython 3.11, 3.12 and 3.14. PR2260 did not publish it, and PR2260's recorded compressed hash could not be reproduced on any zlib build available here.
+3. **Binding.** PR2260's recounted global profile `{1:176252, 2:30096, 3:30908, 4:10152, 5:4493, 6:18, 7:494, 8:576, 32:11, 39:7, 48:2493}` must equal `SPEC.json`. The other bound inputs are m = 73, W = 7168, rank mass 520704 and deficit 2560, plus the retained bit endpoint 0.001733427039589844234749.
+4. **Endpoint** (`certify_endpoint.py`, exact rationals with directed rounding). The moment is Σ_r r H_r/(mW) (m/r)^b.
+   - Both moments are below 1 at C: the native moment, and the moment with PR2260's retained conservative bad-residue surcharge.
+   - The conservative moment exceeds 1 at C + 10^-200.
+   - C = 0.0016834125885426074623303197… PR2260 itself certified only 8417/5000000.
+5. **Outer** (`outer/`). These are PR2220's `check_ledger.py` and `independent_check.py`, regenerated byte-identically, together with PR2220's supplier-independent schedule, structure, gauge-free and fine-padding checks and PR1673's 19 exact evidence replays.
+   - 54 strict rows: PR2220's 63 rows, minus its 11 rows that compare with historical T74 kappas, plus `improvement_over_pr2220` and `improvement_over_pr2260`.
+   - 17 rejecting controls fire.
+   - The fixed-grid maximum kappa = C − 1/F − B is asserted.
+   - *Stale T74 numbers in `outer/*.md`.* Those files are byte-identical to PR2220 (except the pointer note prepended to `outer/GAUGEFREE-INTERIORS.md` and this package's own `outer/ACCRETIVE-SCALE.md`, see below) and quote its T74 instance (DIG = 60000, r0 = 1e-60080, a 2^-199767 grid, 62/63 strict rows, q = 0.0016797549…). The values used here are DIG = 200, R0 = 1e-280 (ZETA = 1e-282, STOP = 1e-286), F = 10^246, B = 1e-222, enclosure grid 2^-1112 and 54 rows (`outer/check_ledger.py`, which derives them from `SPEC.json`; the ledger run prints the 54 rows).
+6. **Bundled proofs and the accretive scale qualification.**
+   - `vendor/pr1786/FINE-PRIME-PADDING.md` (SHA-256 `47d1b1b4df28aadda3cd6bdeb1fbc47290595e9208cd09c1a1f338520d33fdce`) and its retained notice `vendor/pr1786/NOTICE.md` (`dcc1adb2461aa93ca62b6a912dbd8b215599e6af8053dfbacb7423b544dd659a`) are vendored verbatim from PR2220 `fc09a70baf7d6bcff935df226f6fc9e6c6c0131d:research/capped-fusion-t74-k554/vendor/pr1786/`; the proof is byte-identical to PR1786 `c52345caa318fc6b1b6e2659c57d746317a72823:research/fine-prime-padding/FINE-PRIME-PADDING.md`. It is the proof cited by `outer/FINE-PADDING-COMPOSITION.md`; the notice's `outer/check_fine_padding.py` is byte-identical to PR2220's.
+   - GAUGEFREE-INTERIORS Theorem 4.1 is false as a uniform bound in Lambda, mu (maintainer review of PR1758, https://github.com/CrocSwap/integer-mult-bounds/pull/1758#issuecomment-6203527654). This package relies on the corrected form in `outer/ACCRETIVE-SCALE.md`: log(2 + Lambda + 1/mu) in general, and the original ratio form for unit-diagonal kernels (Lambda >= 1, mu <= 1), which every applied kernel G'_s is, including truncated/rounded kernels and descendants. `outer/check_accretive_scale.py` (exact, run by `outer/run_outer.py`) checks the counterexample, the inequalities and the unit diagonal of the applied kernels. No ledger number or kappa changes.
+
+## Why the outer is supplier-generic
+PR1673's theorem (`vendor/pr1673/capped-phase-fusion/phase_fusion_note.tex`, Thm "Conditional capped-fusion transfer" and §"Updated T74 supplier") is "parameterized by the effective supplier rate". PR1673 was first stated on the T8 selected supplier (b_* = 0.001675462676808…), which is the controller class of PR2260's 72-call T8 supplier; T74 was a later substitution.
+
+**Checked facts, not inferences:**
+- *Whole-call row hosting.* PR2260's arithmetic recount (`vendor/pr2260-selective-borrowing-kappa/arithmetic/verify.py`) requires exactly 72 × 436 owner rows, one per (call, helper). Each row has one (owner_replica, child, port) bin and contiguous axes `range(offset, offset+width)`. Roles are therefore whole; "multi-bin repacking" describes the packing solver, not split residuals.
+- *No hidden supplier rate in the outer evidence.* `struct_check.py`, `sched_test.py`, `sched_review.py`, `check_gaugefree.py` and `check_fine_padding.py` contain no supplier rate or kappa literal; their checks are structural.
+- *The gain comes from the outer, not a looser claim.* PR2260's own `above_balance` control tests the balanced-prefix outer's margin eps·q ≈ b/(1+b) ≈ 0.0016805 + ~8e-8; 3361/2000000 is the floor of that margin on PR2260's 1e-7 grid (about 8e-8 below the margin), so the gain comes from the different outer, not from PR2260 having claimed less.
+
+- **Original setting.** It was first stated for the T8 selected supplier (b_* = 0.001675462676808…) and then substituted with T74. The substitution needs only:
+  - a fixed native dimension and maximum child rank (here m = 73 and r_max = 48, with 73² > 2·48²);
+  - restoration of arbitrary dirty helper and spectator coordinates, together with the same normalized global transform at each completed-call boundary. PR2260 checks both native orientations under arbitrary dirty inputs;
+  - the retained interfaces (common-frame / mutable-source realization, whole-call row hosting, semantic precision guard with C_1 = 1, fixed-tape compiler). PR2260 states it uses the same retained interfaces, with C_1 = 1.
+- **What a larger stock changes.** A larger finite stock W, PR2260's fresh C_0 < 2^210000 and its row coefficient 16467 < 20161 change fixed constants, not powers of p.
+- **Ordering.** The ordering q < a_eff < b < a_* also holds here.
+
+**Items of PR1673's own T74 substitution review (`vendor/pr1673/.../INDEPENDENT-SUBSTITUTION-REVIEW.md`), mapped to PR2260.** Items marked *checked* were read against the sources by an independent read-only audit; items marked *inherited* are taken from PR2260's own package (`PROOF.md`, `PROOF-LOCAL.md`) and not re-derived here.
+- *Product-row stock Q_rows = W_complex^D · W_bit^D.* Checked: W = 7168 changes only a constant of the stock, not a power of p.
+- *Semantic-guard / prefix-allowance constant.* T74 used 2048 bits. For PR2260 the guard is its magnitude guard 244λ < 512λ with C_0 < 2^210000. Inherited from PR2260; not re-derived here.
+- *Completed global call, no ambient-p address permutations.* PR2260's dirty borrowing is internal to the T8 parent call and is restored at its boundary, so the supplier is used only as a completed global call on complete fields. Inherited from PR2260's PROOF.md / PROOF-LOCAL.md; not re-derived here.
+- *436 helpers instead of 480.* Checked: an interface-neutral constant; the owner-row recount is 72 × 436 (see above).
+- *Equal-chunk interchange of the bit supplier.* Inherited: the retained bit construction is unchanged from PR2260; not re-derived here.
+
+## Scope and caveats
+- This is a conditional claim, exactly as conditional as PR2260 and PR2220.
+  - It inherits PR2260's supplier premises: the all-size compiler, physical fixed-tape realization, retained bit construction, exact recovery and analytic transfer.
+  - It also inherits every premise of PR1673's capped-fusion outer and of the PR1786 / PR1758 / PR2180 refinements bundled in PR2220's outer.
+- No new finite construction, no Lean proof and no independent human review are claimed. The retune changes only which already-published outer is applied to an already-published supplier.
+- If PR2260's supplier is withdrawn or corrected, this retune falls with it.
+- PR2260 is itself a recent PR that has not yet been reviewed, so this result is conditional on it as well.
+
+## Credits
+- **Supplier:** PR2260 (Dugongue), selective 44-alias dirty borrowing and the new physical owner packing. PR2226 (GitHub 1337isnot1337; Cinna Davis), the anticipatory borrowing architecture. Their predecessors are credited in `vendor/pr2260-selective-borrowing-kappa/NOTICE.md`.
+- **Outer:** PR1673 (Vlado Vojdanovski, Vlado-V), capped phase-cell fusion, vendored verbatim from PR2220. PR2220 (jeremykothe2; prepared in the Rohan Arun T74 series), the retune method, ledger and outer evidence package. PR2180 / PR2161 (dataisfire), the fixed-grid maximum. PR1786 (chafreaky / Chafik Boukhalfa), fine prime padding. PR1758, the gauge-free interiors. PR1699, the classical matrix boundary. Their cited sources, including Le Gall and ADVWXXZ for rectangular matrix exponents and Trudgian for effective primes, are listed in `outer/*.md`.
+
+Prepared by DreamingOfClouds with Claude Opus 5.5 (Anthropic), via Claude Code; Apache-2.0.

research/selective-borrowing-outer-retune/SPEC.json

@@ -0,0 +1,39 @@
+{
+ "schema": "outer-retune-spec/1",
+ "name": "pr2260-selective-borrowing-supplier+pr1673-capped-fusion-outer",
+ "supplier": {
+  "pr": 2260,
+  "head_commit": "fde6b107e26a14da5bac5b66913c914708ea6901",
+  "package": "vendor/pr2260-selective-borrowing-kappa",
+  "supplier_kappa": "3361/2000000",
+  "certified_complex_saving_in_pr2260": "8417/5000000",
+  "bit_lower": "1733427039589844234749/1000000000000000000000000"
+ },
+ "geometry": {
+  "dimension": 73,
+  "calls": 72,
+  "helpers_per_call": 436
+ },
+ "expected": {
+  "global_profile": {
+   "1": 176252,
+   "2": 30096,
+   "3": 30908,
+   "4": 10152,
+   "5": 4493,
+   "6": 18,
+   "7": 494,
+   "8": 576,
+   "32": 11,
+   "39": 7,
+   "48": 2493
+  },
+  "live": 7168,
+  "rank_mass": 520704,
+  "deficit": 2560,
+  "kappa": "3361/2000000"
+ },
+ "outer_refinement": {
+  "endpoint_digits": 200
+ }
+}

research/selective-borrowing-outer-retune/SUPPLIER-BINDING.json

@@ -0,0 +1,6 @@
+{
+ "pr2260_head_commit": "fde6b107e26a14da5bac5b66913c914708ea6901",
+ "pr2260_manifest_sha256": "a7e8b8b8602ccb42ae5534de8ad5feb0f7c4ac46ba2b8f1ce4cb5aa1a3a33718",
+ "independent_actual_factors_uncompressed_sha256": "33d9b4a0744997b975e3f426c737062d783bed0c1065df4dac42b8dc173838b4",
+ "note": "PR2260 pins actual_factors_sha256 of gzip.compress(payload, mtime=0); those bytes depend on the zlib build (CPython 3.9-3.14 on this machine give a06dd8ea.../49c72a31..., PR2260 records 81289c86...). This package pins the uncompressed payload instead; all other receipt keys are compared verbatim."
+}