Propose conditional κ ≈ 0.001729737530919310 via T148 carrier skips (full replay pending)

.github/workflows/t148-carrier-full-release.yml

@@ -0,0 +1,59 @@
+name: T148 carrier complete verification
+on:
+  push:
+    branches: [ci-t148-carrier-20261016]
+    paths: [.github/workflows/t148-carrier-full-release.yml]
+  workflow_dispatch:
+  workflow_call:
+permissions:
+  contents: read
+jobs:
+  complete-release:
+    if: github.repository == 'dataisfire/integer-mult-bounds'
+    runs-on: ubuntu-24.04
+    timeout-minutes: 150
+    env:
+      PYTHONDONTWRITEBYTECODE: '1'
+      PYTHONINTMAXSTRDIGITS: '0'
+      OPENBLAS_NUM_THREADS: '1'
+      OMP_NUM_THREADS: '1'
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+          sparse-checkout: |
+            .github
+            scripts
+            docs
+            certificates
+            research/rank7-joint-endframes
+            research/rank7-t740-release
+            research/rank7-t148-restore-release
+            research/two-sided-helper-extraction
+            research/t148-carrier-skips
+      - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065
+        with:
+          python-version: '3.12'
+      - name: Install exact scientific dependencies
+        run: python -m pip install numpy==1.26.4 scipy==1.15.3 sympy==1.14.0 mpmath==1.3.0 gmpy2==2.3.2
+      - name: Fetch exact public source commits
+        run: |
+          python - <<'PY'
+          import json, re, subprocess
+          rows = json.load(open('research/t148-carrier-skips/GIT-SOURCES.json'))
+          heads = sorted({r['head'] for r in rows})
+          assert all(re.fullmatch('[0-9a-f]{40}', h) for h in heads)
+          subprocess.run(['git','fetch','--no-tags','https://github.com/CrocSwap/integer-mult-bounds.git',*heads],check=True)
+          PY
+      - name: Run both complete releases without weakening any scientific check
+        run: python -B research/t148-carrier-skips/ci_full.py --work "$RUNNER_TEMP/t148-complete-verification"
+      - name: Registered finite replay after both completed releases
+        run: python -B scripts/research_package.py --package research/t148-carrier-skips
+      - uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02
+        with:
+          name: verified-t148-carrier-package
+          path: research/t148-carrier-skips/
+          retention-days: 7
+          if-no-files-found: error
+          include-hidden-files: true

research/rank7-t148-restore-release/.gitattributes

@@ -0,0 +1 @@
+* -text

research/rank7-t148-restore-release/GENERAL-CEILING.json

@@ -0,0 +1,21 @@
+{
+  "model": "nine-stage coverage, one unpaid coordinate/family, centre64, idle rank<=51",
+  "local_profile": {
+    "1": 2443,
+    "2": 707,
+    "3": 232,
+    "4": 108,
+    "5": 18,
+    "8": 8
+  },
+  "primitive_saving_cap": [
+    "439088585719/250000000000000",
+    "1756354342877/1000000000000000"
+  ],
+  "realized_fraction_lower": "8645761685714791828251/8781771714385000000000",
+  "scope": "Primitive-moment cap only. The older retained a/(1+a) transfer is not substituted for the source release modern outer contract.",
+  "source_commit": "d8e0f83f00191df62174b641e9fff055c357f641",
+  "source_path": "research/complex-local-word-ceiling/ceiling.py",
+  "source_sha256": "86a1432870b3b21d8454106ec9162fddc0471472d9214c590a1420f2426cedfc",
+  "call": "ceiling(local, 73, 896, 9, 64)"
+}

research/rank7-t148-restore-release/README.md

@@ -0,0 +1,195 @@
+# T148 packing composed with the R1 restore edit
+
+This package takes PR3022's released admission handoff. It composes its literal
+T148 allocation with Jeremy Kothe's PR3015 R1 word edit, using PR3021's complete
+release recipe. The selected result and all frozen predecessors stay unchanged.
+
+A fresh complete finite source/native/admission/outer release passed at
+
+```text
+kappa = 0.001729152337142958365650 - 10^-46 - 10^-70
+PR3021 = 0.001729151203983018442998 - 10^-46 - 10^-70
+gain = 283289984980663 / 250000000000000000000000
+     = 1.133159939922652e-9
+```
+
+The finite complete-moment root lies strictly between
+`0.0017291523371429583656502` and `0.0017291523371429583656503`.
+This is an author-checked conditional candidate, not an adopted supplier or
+selected-frontier update. `evidence/RESULT.json` binds the exact released point.
+
+## What changes
+
+PR3021 combines R1 with PR3018's T74 journal. Here the unchanged edited providers
+run on PR3022's doubled order, fill and dispatch. These are the only three
+scientific arguments changed in the source's complete-admission command.
+Every preceding source derivation and native-provider command is retained.
+The source admission retains its pinned PR2969 comparator. This wrapper also
+checks strict improvement over the newer pinned PR3021 endpoint; the two gain
+fields therefore have different baselines. Some inherited directory and status
+names still say `packing74` or `T74`; actual geometry is checked as 148 replicas,
+1,332 calls and 133,117 physical families.
+
+The complete profile is
+
+```text
+{1:3288708, 2:941724, 3:318948, 4:143856,
+ 5:27416, 8:10656, 32:1, 48:56380}.
+```
+
+Relative to exactly twice PR3021, the difference is
+`4 H1 + 4 H3 - 4 H5 - 2 H14 + H32`.
+The new empty rank-32 bin is paid. Every one of 543,456 helper roles and all
+91 private cleanup buffers remain charged. The scalar words, helper widths,
+source derivation and native operations are exactly PR3015's, not a new word
+search. The inherited geometry and moment kernels are reused and credited;
+this package does not call them independent new implementations.
+
+## Composition and cost argument
+
+1. PR3018's journal applies literally to PR3015's unchanged original fill.
+   The complete T74 invoice with the edited providers equals PR3021's profile.
+2. Map each old call time `t` to `2t+c` and replica `r` to `r+74c`, for
+   `c=0,1`. Each replica projects to the same nine source stages. Exact doubling
+   alone leaves the normalized moment unchanged. Apply PR3022's 86-row journal
+   only after that lift. Full geometry checks every resulting role, source
+   window, family exclusion, mask and compact prefix.
+3. R1 changes the local per-call profile by `-2 H1 + H2`, without changing the
+   ordered helper widths. It therefore composes with the owner journal. The
+   geometry checker recomputes the sum of actual local calls
+   and actual idle remainders; it does not infer that sum from rank mass.
+4. With `H_r(b)=(r/73)^(1-b)`, compare the complete moment divided by `896T`.
+   Directed integer intervals certify both endpoint signs. Every child rank
+   is below 73, so this moment is strictly increasing and the bracket isolates
+   its unique root. The new root exceeds PR3021's by between
+   `1.1331599399226515e-9` and `1.1331599399226517e-9`. This interval is for the
+   composed edited words, not PR3022's unedited-word interval.
+5. Replica-dependent costs are not held fixed. Live families increase from
+   66,304 to 132,608; complete physical families increase from 66,813 to 133,117.
+   The serial scratch namespace remains 408 reserved helper tableaux,
+   8 centres, 91 private buffers and 2 coefficient families. Both 1,332 local
+   sweeps and 1,332 bridges are charged. Completed-event depth per replica
+   remains 18 under the inherited block-energy contract.
+6. Compute the group exactly as
+   `|Sp(72,2)| = 2^1296 product(j=1..36, 2^(2j)-1)`.
+   The physical-stock bit length increases from 2644 to 2645, so the charged
+   stock-to-L2 exponent increases from 1322 to 1323. The conservative prefix
+   reserve becomes `128+37+18+1323+1+4 = 1511`, up from 1510, still below the
+   retained 2048-bit payload. Thus this composition needs no precision-cap
+   edit. The complete release checks these bounds against the emitted physical
+   schedule and actual native prefixes. The observed native
+   prefix is 93 bits, giving an actual global prefix of 1476 bits. The stored
+   complete invoice and explicit one-bit-undercharge controls bind these values.
+
+The last step is conditional on the inherited universal block-energy lemma.
+Finite premise checking is not a proof of that universal implication.
+
+PR3015's source-specific fixed-inventory fractional packing cap is bracketed by
+`0.0017291527560043309345350` and `0.0017291527560043309345351`.
+The candidate realizes at least `0.9999997577649683` of that cap and remains about
+`4.188613726e-10` below it. This is the inherited 2,551-pattern capacity-8/16/32/48
+relaxation, with the cap certificate's local profile and inventory checked
+against both edited words. It is more restrictive than a rank-mass-only packing
+relaxation. The new package does not independently recertify the cap's dual
+proof; exact attainment and a more general owner optimum are not claimed.
+
+The broader `complex-local-word-ceiling/ceiling.py` calculation on this paid local
+profile gives primitive-moment cap `[0.001756354342876, 0.001756354342877]` under
+its nine-stage coverage and rank-51 halving relaxation. The candidate realizes
+at least `0.9845122336250876` of that broader cap. `GENERAL-CEILING.json` records
+that call with `local={1:2443,2:707,3:232,4:108,5:18,8:8}`, `m=73`,
+`families=896`, `stages=9`, `centre=64`. The bounded verifier separately checks
+both signs after clearing the fractional rank-51 count by scaling all counts
+by 51. The source's modern outer release, not the older calculator's optional
+`a/(1+a)` planning transfer, determines the final conditional kappa.
+
+## Reproduction
+
+The fresh chain's measured stage time was 1494.85 seconds on the recorded
+Linux/Python 3.12.15/NumPy 1.26.4 environment. It covers a fresh retained
+bootstrap, both current native-provider chains, actual owners, parent lift,
+rebind, discovery, independent outer checks and the complete modern release.
+The final wrapper's stronger payload/cap/invoice postchecks were rerun against
+the completed artifacts. Its normalized scientific recipe is identical to the
+recipe that ran. `evidence/VALIDATION.json` records that distinction.
+
+The bounded registered replay recomputes geometry and exact moment arithmetic,
+checks the stored complete-release relations, runs 13 behavioral tests and
+rejects ten receipt/actual-invoice corruptions. It does not rerun the full
+supplier chain. Optimized Python 3.12 and ordinary Python 3.13 produce the same
+finite receipt; warning-as-error execution and Ruff correctness checks pass.
+
+```sh
+python3.12 -B scripts/research_package.py \
+  --package research/rank7-t148-restore-release
+```
+
+For the full fresh chain, use Python 3.12 with NumPy 1.26.4. The recipe exports
+historical exact Git objects, including PR3015, PR2894 and PR2833; the complete
+predecessor pins are retained in the exported PR3015 recipe. If an object is
+absent, fetch its exact public commit before replay. No fetch is hidden inside
+the verifier.
+
+```sh
+git fetch origin e244296debb45fe6f4da84d29e4d64fbab55fda0
+git fetch origin 18d03c447886895ef9abf888398178e0428533ca
+git fetch origin e9307b48d7d1b690939935ecef0f8e524e6d92f4
+python3.12 -B research/rank7-t148-restore-release/reproduce.py \
+  --repo "$PWD" --work /tmp/fresh-t148-r1-release
+```
+
+Use a fresh external workspace. POSIX process-group cleanup and enabled
+assertions are required by the inherited chain. `--dry-run` checks exact source
+exports and the changed recipe only. Container hashes bind this particular run;
+canonical decompressed fill and JSON semantics define cross-environment equality.
+
+`VENDOR.json` pins exact source Git blobs and hashes. `SOURCES.json` pins live
+inputs and historical provenance. The package changes no shared validator.
+
+## Limits and engineering notes
+
+This is a finite conditional construction, not a complete multiplication
+proof, practical speedup, independent human review or selected-frontier update.
+All-size/compiler, physical backing and movement, precision/fixed-tape,
+analytic transfer and universal block-energy obligations remain inherited.
+Replica-dependent storage doubles; this is not a storage-saving claim.
+There is no integral-optimality claim. PR3024's T740 cap attainment and its
+active R1/admission follow-up have a different, larger-stock scope. They are
+not preempted or validated here.
+
+The new code has bounded input counts, finite bisection and source-authenticated
+imports. Retained kernels execute compiled authenticated source bytes rather
+than possibly stale default-import bytecode. Python requires managed allocation.
+Geometry limits live collections to 1,332 calls, 543,456 roles and 298,368 bins. The three retained import targets
+are fixed by hashes. Existing source exporters and native validators retain
+their original assertion and dynamic-code mechanisms rather than undergoing an
+unrelated rewrite. Full stages have a 1,800-second timeout with process-group
+cleanup; the registered finite commands have 120-second limits. These are
+bounded research-runtime deviations from fixed allocation and direct dispatch.
+Future changes to the source interface must revalidate those bounds and pins.
+
+## Open next questions
+
+Independently rerun and review this exact complete T148+R1 release before
+supplier adoption. Inspect the emitted stock-norm and prefix increase alongside
+the actual owner schedule, then review the inherited universal block-energy
+premise separately. A fresh replay took about 25 minutes on this development
+host; another host may differ. Exact source access and the documented environment
+are prerequisites. Structured details are in `TASK.json`.
+
+## Sources and credit
+
+- Jeremy Kothe, PR3015 `e244296debb45fe6f4da84d29e4d64fbab55fda0`, supplies R1.
+- Nick Petalas, PR3021 `217edac2ebfa6250536bd6df153de14901a69395`, supplies the
+  T74 composition, exporter, process cleanup and source-chain postchecks.
+- Nick Petalas, PR3022 `553dc9d8fc8c1489aa2b0378331a0de4b5e9b44e`, supplies the
+  doubled journal and retained geometry. PR3018 supplies the earlier journal
+  and directed moment kernel.
+- Marius-Paul Dumitrean's PR2979 supplies the joint rank-seven construction.
+  Mason Hale, Jeremy Diviney, Yiyang Lu, Tekalign, Ross Richardson, Dugongue,
+  icekylinx, themilkman216 and the retained word/native/geometry/owner lineage
+  keep their predecessor credit and source notices.
+
+New here is the T148+R1 composition, rederived replica-dependent bill and its
+source-bound release. Substantial automated assistance was used. No independent
+human or formal-kernel review is claimed. Apache-2.0.

research/rank7-t148-restore-release/SOURCES.json

@@ -0,0 +1,54 @@
+{
+  "schema_version": 1,
+  "base_commit": "d8e0f83f00191df62174b641e9fff055c357f641",
+  "files": [
+    {
+      "path": "certificates/selected-result.json",
+      "usage": "provenance",
+      "git_blob": "b1fea0c94436b236cc2b7d41a7be032811b94c81",
+      "sha256": "bb2eaa58647b353f88d0aec2eafb9d30df1be53589231b23e12ceef3e4bad129"
+    },
+    {
+      "path": "docs/research/agent-task-catalog.json",
+      "usage": "provenance",
+      "git_blob": "67f085acd86acd7d4e03922d9cf4b26bca66fed5",
+      "sha256": "47b42433f5a85809a3aadd29c2c339da2c6ed728e4f7762df6b88bccb7874cf0"
+    },
+    {
+      "path": "research/rank7-joint-endframes/allocation/DISPATCH.json",
+      "usage": "input",
+      "git_blob": "9544fee4e6ee1f202cee22bfaeaf229ee3b2515e",
+      "sha256": "65e435bd68e97243c003677271b88485a29d4d0fae2dff1354100404a145e74c"
+    },
+    {
+      "path": "research/rank7-joint-endframes/allocation/FILL.json.gz",
+      "usage": "input",
+      "git_blob": "8d4fd644d60ba2c98106dc770803b647afdf0963",
+      "sha256": "c2136be9434cda511a1a8a44d98f9f72ea5376830a8426674253b4ee351c4c39"
+    },
+    {
+      "path": "research/rank7-joint-endframes/allocation/ORDER.json",
+      "usage": "input",
+      "git_blob": "7409c4df1092241729739596c7648bda504e6aad",
+      "sha256": "519d1d5dbb997a219fcfcad34c0b03e7838bef24199784d7e964fa10090f66e1"
+    },
+    {
+      "path": "research/rank7-joint-endframes/allocation/vendor/PLAN.json.gz",
+      "usage": "input",
+      "git_blob": "7a1db55a4d53ee99280be4f3ab7d29f7b2033adf",
+      "sha256": "d566cfb87b71639d4138e1f89b543e2dd6a7772070226a2a14c799a6835bf21e"
+    },
+    {
+      "path": "research/rank7-joint-endframes/allocation/vendor/TEMPLATE.json.gz",
+      "usage": "input",
+      "git_blob": "114808c32b1799da53524bdc25cea67efb60009e",
+      "sha256": "21297b20f1f2dc76eb9dd8124416068c3cb096bab9df17fa4da19bc68dfe082e"
+    },
+    {
+      "path": "research/complex-local-word-ceiling/ceiling.py",
+      "usage": "provenance",
+      "git_blob": "6c0f54527d879396f689523bfde19928a9b6157c",
+      "sha256": "86a1432870b3b21d8454106ec9162fddc0471472d9214c590a1420f2426cedfc"
+    }
+  ]
+}