Conditional κ = 0.001469603966017302: rank-sector102 on112 ports, fully rebuilt

.gitattributes

@@ -18,3 +18,7 @@ research/coordinated-frames-and-entrance-banks/** -text -whitespace
 # Frozen PR210 package and round-nine evidence retain their audited bytes.
 research/five-stage-source527-banks/** -text -whitespace
 research/community-round9-audit/** -text -whitespace
+
+# Preserve the exact frozen rank-sector/112-port supplier bytes.
+research/ranksector-kernel102-port112/** -text
+research/ranksector-kernel102-port112-validation/** -text

.github/workflows/ranksector-kernel102-port112.yml

@@ -0,0 +1,16 @@
+name: Rank-sector kernel102 on 112 ports
+on:
+  pull_request:
+  push:
+jobs:
+  verify:
+    runs-on: ubuntu-latest
+    timeout-minutes: 45
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.14'
+      - run: sudo apt-get update && sudo apt-get install -y libboost-dev
+      - run: python -m pip install numpy sympy
+      - run: python -B research/ranksector-kernel102-port112/verify.py "$RUNNER_TEMP/partial-own-ranksector112" --boost-include /usr/include

research/ranksector-kernel102-port112-validation/COMPARISON.json

@@ -0,0 +1,34 @@
+{
+  "status": "PASS_TWO_FRESH_COLD_REPLAYS_WITH_IDENTICAL_STABLE_RESULT",
+  "source_manifest_sha256": "cd2b2e47b2dbd994d4fd506d36d4c1e561d9615faaddada72aa9d1734b88d47c",
+  "platforms": {
+    "macos": {
+      "stages": {
+        "root": 18,
+        "dependency": 21,
+        "local": 13
+      },
+      "python": "3.14.8",
+      "root_result_sha256": "4f2fd17cad01f9580b67c08575954e8bf3286f8de94940b4e99767040bddd4aa",
+      "kernel_omission_controls": 102,
+      "scalar_columns_each_orientation": 800,
+      "global_prefix_bits": 210,
+      "charged_prefix_bits": 512
+    },
+    "linux": {
+      "stages": {
+        "root": 18,
+        "dependency": 21,
+        "local": 13
+      },
+      "python": "3.14.8",
+      "root_result_sha256": "4f2fd17cad01f9580b67c08575954e8bf3286f8de94940b4e99767040bddd4aa",
+      "kernel_omission_controls": 102,
+      "scalar_columns_each_orientation": 800,
+      "global_prefix_bits": 210,
+      "charged_prefix_bits": 512
+    }
+  },
+  "comparison_head": "2080903b9e6bd01e3f64cc4c7c62ea1b3af0aa6a",
+  "scope": "Both complete suppliers rebuilt, then rank-sector102word, physical gates, literal disjoint112projection, exact phases, actual whole-owner controller, independent raw reconstruction, finite invoice and2engines/47outer. No cached supplier outputs accepted. All-size contracts remain conditional."
+}

research/ranksector-kernel102-port112-validation/README.md

@@ -0,0 +1,44 @@
+# Fresh rank-sector102/112-port validation
+
+Both complete cold replays passed on macOS arm64 and Linux aarch64 with Python
+3.14.8. Each passed 18 root stages, 21 supplier stages and 13 changed local
+stages. Complete RESULT files are byte-identical across platforms.
+COMPARISON.json binds them to the frozen source manifest. Every artifact listed
+in each root receipt was checked against its hash after collection.
+
+Both the bit and complex suppliers are freshly reconstructed. The new 120-port
+word receives independent 800-column checks in both orientations and all 102
+omitted-correction controls, followed by fresh scalar, phases, wrappers,
+common-label and operation-by-operation guarded physical checks.
+
+The restricted 112-port word receives 784-column checks in both orientations,
+new phases and wrappers. PROGRAM-PROJECTION.json proves that every retained
+expanded instruction is identical to the guarded 120-port program under an
+injective register rename. All removed registers form a disjoint component.
+Thus this is an explicit transfer of the checked operation bounds, rather than
+a claim that the 120-specific physical driver was rerun directly on 112 ports.
+All three projection controls fail.
+
+Global restriction rejects 13 mutants; the raw auditor independently reconstructs
+literal routes, owner windows, physical charts and every paid child. All 40,320
+helper roles have whole owners. The independent scalar/Horner invoice measures
+20/25-bit local prefixes and a 210-bit global bound under the 512-bit charge.
+Two rational engines, six bootstrap levels, four outer controls and 47 strict
+outer inequalities pass. The four preoutput package controls also reject
+optimized Python, missing input, changed source and an unexpected file.
+
+The retained pilot failures concern fixture expectations or launching:
+- A bank-owner mutant assigned its existing value 7168 and therefore did nothing.
+  The replacement changes the actual value by one; all other controls remain.
+- An initial launcher used Python without NumPy and failed before allocation.
+  The discovery allocator then ran with the established environment.
+- The raw audit expected the old residual inventory. It now checks the actual
+  fixed inventory 378 width 9, 164 width 8 and 18 width 7. Scalar and physical tests
+  are retained; no acceptance condition is bypassed.
+
+The final conditional κ is 0.001469603966017302. PR527 at
+2080903b9e6bd01e3f64cc4c7c62ea1b3af0aa6a announces 0.001467289440740487.
+The new witness is about 0.158% higher than that published value. The frozen package README
+also retains its earlier comparison with PR523. No claim to a global optimum,
+fresh Lean theorem or unconditional all-size bound follows. GitHub CI is
+separate from these completed local and remote cold replays.