From d5c3a551fe91f13383fd461b2189bd252b746c4a Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:50:33 -0400
Subject: [PATCH 01/24] Stage hash-pinned 100-point blitz patch (1/10)

---
 .delivery/100-blitz-source/00 | 1 +
 1 file changed, 1 insertion(+)
 create mode 100644 .delivery/100-blitz-source/00

diff --git a/.delivery/100-blitz-source/00 b/.delivery/100-blitz-source/00
new file mode 100644
index 0000000..a61b888
--- /dev/null
+++ b/.delivery/100-blitz-source/00
@@ -0,0 +1 @@
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
\ No newline at end of file

From fc3a7fc0718bd8b8521f791dd397eeb57bd1fefa Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:50:58 -0400
Subject: [PATCH 02/24] Stage hash-pinned 100-point blitz patch (2/10)

---
 .delivery/100-blitz-source/01 | 1 +
 1 file changed, 1 insertion(+)
 create mode 100644 .delivery/100-blitz-source/01

diff --git a/.delivery/100-blitz-source/01 b/.delivery/100-blitz-source/01
new file mode 100644
index 0000000..0c73fe0
--- /dev/null
+++ b/.delivery/100-blitz-source/01
@@ -0,0 +1 @@
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
\ No newline at end of file

From 82165071c10c0f4e331f7266da6b584766a8cc81 Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:51:22 -0400
Subject: [PATCH 03/24] Stage hash-pinned 100-point blitz patch (3/10)

---
 .delivery/100-blitz-source/02 | 1 +
 1 file changed, 1 insertion(+)
 create mode 100644 .delivery/100-blitz-source/02

diff --git a/.delivery/100-blitz-source/02 b/.delivery/100-blitz-source/02
new file mode 100644
index 0000000..5134fa9
--- /dev/null
+++ b/.delivery/100-blitz-source/02
@@ -0,0 +1 @@
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
\ No newline at end of file

From 3339da733cb74faf791d3655db8255687b673df1 Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:51:48 -0400
Subject: [PATCH 04/24] Stage hash-pinned 100-point blitz patch (4/10)

---
 .delivery/100-blitz-source/03 | 1 +
 1 file changed, 1 insertion(+)
 create mode 100644 .delivery/100-blitz-source/03

diff --git a/.delivery/100-blitz-source/03 b/.delivery/100-blitz-source/03
new file mode 100644
index 0000000..0369713
--- /dev/null
+++ b/.delivery/100-blitz-source/03
@@ -0,0 +1 @@
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
\ No newline at end of file

From 40ee93bf9190e24ccc15ab34f3f82ed32ebf3b04 Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:52:52 -0400
Subject: [PATCH 05/24] Stage hash-pinned 100-point blitz patch (5/10)

---
 .delivery/100-blitz-source/04 | 1 +
 1 file changed, 1 insertion(+)
 create mode 100644 .delivery/100-blitz-source/04

diff --git a/.delivery/100-blitz-source/04 b/.delivery/100-blitz-source/04
new file mode 100644
index 0000000..16f7643
--- /dev/null
+++ b/.delivery/100-blitz-source/04
@@ -0,0 +1 @@
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
\ No newline at end of file

From 8a6c42128a6dcbc2e8bc7e489669de8ff4bb601e Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:53:19 -0400
Subject: [PATCH 06/24] Stage hash-pinned 100-point blitz patch (6/10)

---
 .delivery/100-blitz-source/05 | 1 +
 1 file changed, 1 insertion(+)
 create mode 100644 .delivery/100-blitz-source/05

diff --git a/.delivery/100-blitz-source/05 b/.delivery/100-blitz-source/05
new file mode 100644
index 0000000..0a76cd8
--- /dev/null
+++ b/.delivery/100-blitz-source/05
@@ -0,0 +1 @@
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
\ No newline at end of file

From 4f8ad658b258de2b0636248985ec721fb3d6f403 Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:53:41 -0400
Subject: [PATCH 07/24] Stage hash-pinned 100-point blitz patch (7/10)

---
 .delivery/100-blitz-source/06 | 1 +
 1 file changed, 1 insertion(+)
 create mode 100644 .delivery/100-blitz-source/06

diff --git a/.delivery/100-blitz-source/06 b/.delivery/100-blitz-source/06
new file mode 100644
index 0000000..f880401
--- /dev/null
+++ b/.delivery/100-blitz-source/06
@@ -0,0 +1 @@
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
\ No newline at end of file

From de318e54520cd1cc08fd03108b2461469ae85db6 Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:54:08 -0400
Subject: [PATCH 08/24] Stage hash-pinned 100-point blitz patch (8/10)

---
 .delivery/100-blitz-source/07 | 1 +
 1 file changed, 1 insertion(+)
 create mode 100644 .delivery/100-blitz-source/07

diff --git a/.delivery/100-blitz-source/07 b/.delivery/100-blitz-source/07
new file mode 100644
index 0000000..46cfce7
--- /dev/null
+++ b/.delivery/100-blitz-source/07
@@ -0,0 +1 @@
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
\ No newline at end of file

From baee76ed44a647cd08121319fdf8e1b5416d03fc Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:54:33 -0400
Subject: [PATCH 09/24] Stage hash-pinned 100-point blitz patch (9/10)

---
 .delivery/100-blitz-source/08 | 1 +
 1 file changed, 1 insertion(+)
 create mode 100644 .delivery/100-blitz-source/08

diff --git a/.delivery/100-blitz-source/08 b/.delivery/100-blitz-source/08
new file mode 100644
index 0000000..3f72158
--- /dev/null
+++ b/.delivery/100-blitz-source/08
@@ -0,0 +1 @@
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
\ No newline at end of file

From 6f8f0d14e03df5b3f69f4ded0a973f8daa4047c5 Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:54:56 -0400
Subject: [PATCH 10/24] Stage hash-pinned 100-point blitz patch (10/10)

---
 .delivery/100-blitz-source/09 | 1 +
 1 file changed, 1 insertion(+)
 create mode 100644 .delivery/100-blitz-source/09

diff --git a/.delivery/100-blitz-source/09 b/.delivery/100-blitz-source/09
new file mode 100644
index 0000000..b8e6d09
--- /dev/null
+++ b/.delivery/100-blitz-source/09
@@ -0,0 +1 @@
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
\ No newline at end of file

From 8a783ac63c816b4691e6e2fe144b27fdca0a44bc Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:55:16 -0400
Subject: [PATCH 11/24] Publish the hash-verified 100-point blitz checkpoint

---
 .github/workflows/publish-100-blitz.yml | 81 +++++++++++++++++++++++++
 1 file changed, 81 insertions(+)
 create mode 100644 .github/workflows/publish-100-blitz.yml

diff --git a/.github/workflows/publish-100-blitz.yml b/.github/workflows/publish-100-blitz.yml
new file mode 100644
index 0000000..f2d5dd8
--- /dev/null
+++ b/.github/workflows/publish-100-blitz.yml
@@ -0,0 +1,81 @@
+name: publish-100-blitz
+on:
+  push:
+    branches: [research/100-blitz-20261016]
+    paths: ['.github/workflows/publish-100-blitz.yml']
+permissions:
+  contents: write
+concurrency:
+  group: publish-100-blitz
+  cancel-in-progress: false
+jobs:
+  publish:
+    runs-on: ubuntu-24.04
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+      - name: Verify and reconstruct reviewed patch
+        env:
+          EXPECTED_B64: f75650c04f59e0ae6e02d2190ebd56668442d790276d5cc9fa3dc11d188108a7
+          EXPECTED_XZ: 31780857b8f024c4cb4f1144d04f8e8fdebf8b902ab1acd7c293673aacaef036
+          EXPECTED_PATCH: dcd47a5a9a621f9f20d91cf040cc125c5300089b699dbacfbd25a2be84e83ab6
+        run: |
+          set -euo pipefail
+          mkdir -p /tmp/blitz
+          cat .delivery/100-blitz-source/{00,01,02,03,04,05,06,07,08,09} > /tmp/blitz/patch.xz.b64
+          test "$(sha256sum /tmp/blitz/patch.xz.b64 | cut -d' ' -f1)" = "$EXPECTED_B64"
+          base64 --decode /tmp/blitz/patch.xz.b64 > /tmp/blitz/patch.xz
+          test "$(sha256sum /tmp/blitz/patch.xz | cut -d' ' -f1)" = "$EXPECTED_XZ"
+          xz --decompress --stdout /tmp/blitz/patch.xz > /tmp/blitz/reviewed.patch
+          test "$(sha256sum /tmp/blitz/reviewed.patch | cut -d' ' -f1)" = "$EXPECTED_PATCH"
+      - name: Validate base identities and publish atomically
+        run: |
+          set -euo pipefail
+          test -z "$(git status --porcelain)"
+          test "$(git rev-parse HEAD:spectra/exact.py)" = 983d33551e29d13016e36cd6038d02d270575b66
+          test "$(git rev-parse HEAD:spectra/_native/exact_binary.cpp)" = 96182ee6c7754d24e1c8bc46181dd69a5732d32f
+          test "$(git rev-parse HEAD:experiments/compiled_query/exact.py)" = 780a7ff1d8f293450fb592ae9e29e129e92a551c
+          test "$(git rev-parse HEAD:experiments/compiled_query/exact_native.cpp)" = 96182ee6c7754d24e1c8bc46181dd69a5732d32f
+          git apply --check --index /tmp/blitz/reviewed.patch
+          git apply --index /tmp/blitz/reviewed.patch
+          git diff --cached --name-only -z > /tmp/blitz/changed.paths
+          python - <<'PY'
+          from pathlib import Path
+          expected = {
+              'docs/EXAM_REPAIR.md',
+              'experiments/compiled_query/ITC_EXAM_SUITE_PROTOCOL.md',
+              'experiments/compiled_query/ITC_EXAM_TRANSFER_PROTOCOL.md',
+              'experiments/compiled_query/exact.py',
+              'experiments/compiled_query/exact_native.cpp',
+              'experiments/compiled_query/itc_exam_suite.py',
+              'experiments/compiled_query/itc_exam_transfer.py',
+              'experiments/compiled_query/test_exact.py',
+              'experiments/compiled_query/test_itc_exam_suite.py',
+              'experiments/compiled_query/test_itc_exam_transfer.py',
+              'scripts/check_exam_repair_installation.py',
+              'spectra/_native/exact_binary.cpp',
+              'spectra/exact.py',
+              'spectra/exam_repair.py',
+              'spectra/itc2007_exam.py',
+              'tests/public/test_exam_repair.py',
+              'tests/public/test_itc2007_exam.py',
+          }
+          changed = {
+              item.decode('utf-8')
+              for item in Path('/tmp/blitz/changed.paths').read_bytes().split(b'\0')
+              if item
+          }
+          if changed != expected:
+              raise RuntimeError(f'changed path set differs: {sorted(changed ^ expected)}')
+          PY
+          git rm -r .delivery/100-blitz-source
+          if git diff --cached --name-only | grep -q '^\.github/workflows/'; then
+            echo 'Workflow changes are not permitted in the reconstructed code commit.' >&2
+            exit 1
+          fi
+          git config user.name 'Tugrap'
+          git config user.email '106766396+tugrapaydiner@users.noreply.github.com'
+          git commit -m 'Publish proof-carrying exam repair and prospective ITC suite'
+          git push origin HEAD:research/100-blitz-20261016

From 3036a3c97cf5706d60ab9e277708081b4f241954 Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 18:55:27 +0000
Subject: [PATCH 12/24] Publish proof-carrying exam repair and prospective ITC
 suite

---
 .delivery/100-blitz-source/00                 |   1 -
 .delivery/100-blitz-source/01                 |   1 -
 .delivery/100-blitz-source/02                 |   1 -
 .delivery/100-blitz-source/03                 |   1 -
 .delivery/100-blitz-source/04                 |   1 -
 .delivery/100-blitz-source/05                 |   1 -
 .delivery/100-blitz-source/06                 |   1 -
 .delivery/100-blitz-source/07                 |   1 -
 .delivery/100-blitz-source/08                 |   1 -
 .delivery/100-blitz-source/09                 |   1 -
 docs/EXAM_REPAIR.md                           |  62 ++
 .../compiled_query/ITC_EXAM_SUITE_PROTOCOL.md |  91 +++
 .../ITC_EXAM_TRANSFER_PROTOCOL.md             | 127 +++
 experiments/compiled_query/exact.py           |  12 +-
 experiments/compiled_query/exact_native.cpp   |  47 +-
 experiments/compiled_query/itc_exam_suite.py  | 352 ++++++++
 .../compiled_query/itc_exam_transfer.py       | 752 ++++++++++++++++++
 experiments/compiled_query/test_exact.py      |  27 +-
 .../compiled_query/test_itc_exam_suite.py     |  76 ++
 .../compiled_query/test_itc_exam_transfer.py  | 179 +++++
 scripts/check_exam_repair_installation.py     |  89 +++
 spectra/_native/exact_binary.cpp              |  47 +-
 spectra/exact.py                              |  12 +-
 spectra/exam_repair.py                        | 418 ++++++++++
 spectra/itc2007_exam.py                       | 316 ++++++++
 tests/public/test_exam_repair.py              | 146 ++++
 tests/public/test_itc2007_exam.py             |  85 ++
 27 files changed, 2805 insertions(+), 43 deletions(-)
 delete mode 100644 .delivery/100-blitz-source/00
 delete mode 100644 .delivery/100-blitz-source/01
 delete mode 100644 .delivery/100-blitz-source/02
 delete mode 100644 .delivery/100-blitz-source/03
 delete mode 100644 .delivery/100-blitz-source/04
 delete mode 100644 .delivery/100-blitz-source/05
 delete mode 100644 .delivery/100-blitz-source/06
 delete mode 100644 .delivery/100-blitz-source/07
 delete mode 100644 .delivery/100-blitz-source/08
 delete mode 100644 .delivery/100-blitz-source/09
 create mode 100644 docs/EXAM_REPAIR.md
 create mode 100644 experiments/compiled_query/ITC_EXAM_SUITE_PROTOCOL.md
 create mode 100644 experiments/compiled_query/ITC_EXAM_TRANSFER_PROTOCOL.md
 create mode 100644 experiments/compiled_query/itc_exam_suite.py
 create mode 100644 experiments/compiled_query/itc_exam_transfer.py
 create mode 100644 experiments/compiled_query/test_itc_exam_suite.py
 create mode 100644 experiments/compiled_query/test_itc_exam_transfer.py
 create mode 100644 scripts/check_exam_repair_installation.py
 create mode 100644 spectra/exam_repair.py
 create mode 100644 spectra/itc2007_exam.py
 create mode 100644 tests/public/test_exam_repair.py
 create mode 100644 tests/public/test_itc2007_exam.py

diff --git a/.delivery/100-blitz-source/00 b/.delivery/100-blitz-source/00
deleted file mode 100644
index a61b888..0000000
--- a/.delivery/100-blitz-source/00
+++ /dev/null
@@ -1 +0,0 @@
-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
\ No newline at end of file
diff --git a/.delivery/100-blitz-source/01 b/.delivery/100-blitz-source/01
deleted file mode 100644
index 0c73fe0..0000000
--- a/.delivery/100-blitz-source/01
+++ /dev/null
@@ -1 +0,0 @@
-u94GF3VfA/pRbpHx9BvLpc30Mk3OSaMNNLs0jraCfdVcPzM7qAjawGrPBE31XsZVmrjcDymLmXwXd8BndIjzSMtI037ZDCeGg2bGDBprdkLeFVMPxh75pT3KbtqcrWWc0hK3GvvCStPvEqYS3a+2gwUriHRFz02xVl4jhQdH4L+jcuz1zNRtV3IDBVyOe7rV2NhoqpvGXMvd9+Ah7P+68W8/NUjJ2e5UCFnS9deFDoIEIQIPfZDyixqnUNd2rgQm/C9PCR1J0IYEljvnXSR2Z40JJSyXmU3Pb1W+p/NInJvBVXBbxrL7qLGRLAmHwSJA4vP58tArJE7Wun+YJfqQb7GNRbI/O5j4f+3TYLPxhRsTsnKSBm4eAn1cXcMPyT6lTKi//K9v2/9GuQj0pMFC3yXDTASLAoyb3d0EBBLsdttz3er/ayB4WBds39Or8vPdyxCsnkL+X1GgtsyhuI9cTj0Nmeb8apOm23uz++9a24PnwB7K8YHs/bm0SMDuwZ3PTSbAafS/bpKT0VvmKPgZW0y3zfRBCUOJuLwgJZz2i/bUnEg65J08l3CtC9WvUYlMuyqhT0uk9xFTCizwCkMM6S5egVn/orcc/ScWqTfhv0QXa5kUV8qIDOV6wiUgid2rn9Gg8rlo4cznrWu7czDoLvlNTKqKLdJ2frl7gHmvx2b4ALMBrjbfbkk7zyxhrhNUTOA0cTlRzTqCDZybYw7HWre2gccpwy57C4LBZzA46aLxIzQ/j64nx4KHkHUEnaJDkGQwbXkw5mVJRf2T8lvmqJv1VV3ZF4tEvjztrEVLw1qFc6dkGiYgVKEKCBqxiPzvP15LViU+xOyofcNkQlRDwqgcoZyVdXFq1pqY3Glob03VuVCYKxGe02lPTxG5y1G0xoUN2aO5RIjieprQgzy58f6HJoLmLr+qQD0dNZAzDG3lwMEvYJ9yHofN9BQmpMSr4Atqc/+a0pS18c0LhJXAdCrdrtYvColFxPywmz3pbNaE8LlZWmtp4ej8tHK/04toeXVCPnWltshv8LZQXDKD4tRvvuv7dUZ/v+JwxV6fJfJiWiVTHQb9edb4VIqLOwJYiX8o/JoCsJWHmoZD3xcLjxHQkvD++Hr/JGoscc9gemI3H6Sz4u8tWLFXM4O+pADF0rbyLj1nZRNgyJTO77Glnbh08AKyIAsSmFuAoL4g6xnSIRtmWqsOl0iB/1OnuGxwJ3o7T2UsNTg8TYNKFS4Fyc/KGTUNJV0uGRP+Eia+S0wLZfGv9oeKQNoJ2wJ8e1mSApubLKqu6fvc53DdmGi187D8kkdNm5hMy1jKoWd3huX8oiLM5FRn/N8Cin5epl+/2tZsgGnjywOZmLulUTeNv2oUh0ILe3iULffsEhjxUeJrCO8BfUuqCWNjjpw7/OY26+FgYwwxvpjBDmwFmgFsHSvINe8TlHdo17pAPZa5aCquVoaqCKRhLHCh6ReVxvAJ3I9JE+fuMJggX4A9LRKDq6yBe+AFbcXKoSamm1dUlB7ACJ28icKkN7mnjAM46jYRQAaq7tNrkQq/gdHGTG6tYwQcugKAiijttUGZChCg/hLPpUg9Qurw1ubEigDkwcHVOj6+eegSunvnm3BlU4YWRRNWdvo1WhPnlHHh3vDgEsj2257Cubc8rOhW/SSTfj2Cu04Cw9Bu21Adtfay1Y1hl3mJ47qAhN3B0o2hTJBahEEiuKZPRc9N8X/45iZd8/iVyWuWsPr8R2OvOrSaeqO8bzWKG8gaOPqC3kPyweVU/dxHhsYp/t8r/7yTHhUxexsBcuxp7lJb3GA3r7o7CnaXtnABAUTVsCG0aCu6MyEa2WpKY7QPcBdkqHtIIUALZJmew/R7XI8yj4xsg8nBffUFTMwugtbmQrs6PEA1msHq+Olww9fFOiqS82xx0nLDA3eWFIp3xs6jQv+DHGT9dXWGbgn/krTM19VdPMbJpHxC5QXQVfb4Fx1gZg+yx3giaZR+JCEN5rhYkDuyC4iSfMBfodtWmPylgxgyOkBc+cyRv+l4BPdSn3rDDewKUc7hjWhGVwXUcVtykijlX0cigkwZqcLWRwzKCGtcexanUk9/CxO85qcNs7Ah/9W8lfKgYwAPPdqaF6KI1sJscY1MGo7ksfEOrSKWcWiihy4CgP02dWRMXhIdGpF9OyIT999pTFuMfUdidMCZYryAjrt3fq/oENcfE6PaDdBD9a6beilkmTjOQnNUa3jtiPNhGKt8/NDxlvttfYkRlfV0SpsPEsLa8JNm3hNOEmlrY4AkcbN+zYKWTGT+L8++Bx4eaoALovLVRvTpEx+2v2YzT2MGIK6vVe4F42NSTzEapFTsf+Q4gda8OiDyWGFbMi87ou3JKAOC23q7kFGq+n7GUYXRLgp6DRXHcIdsRrPczAcQBV0QRQxQtAopFWndMhhZzPUMdBLDBKjVQ5WJXpYkiK+s+AWvTThwKk5zfoKpYRlE0UAhQStpsK3tSv1BYdyA5VyOdzIpPRHr/afR9sST+zqTHCzhtzM3hE9nkvRgmtn8wkYGL2/ev8ViX2E8dvN9NGO0Dg/9WMrl1PLvMPs5yFHVz/UTbvl+2eiwEpGnSnFDeO+XJW5aEhIAIdEGQNNl950MJi2BgK58zS7dVbctwblhO39PmEGEy7MHDUYixL+9fJgfvQHZp/uBSckl8McVrU8oiyPMfts9Z/ctZvwxdbA98uazTOjeSlXrr0K7jEKIQoyqkzuuIvN/gOw39DW20Y/+dDztoBc571zeCP/5iSEFSIExoOFGMgashnsmMuKT+30stWdMSF3mVajLMQo5hqvWyXPOHqu33vVaYw2oI8jKZ/RcbskPQTaa7HIAhqG9Sgn0oC2oabsX6nQ9U/LrbxiD3T0MjN/NWdvooIeuKJN81tISIlEm1eBTqItpCTJ4FWkeFQGGtcG+vBH77sdzVC5vuGW9G1zVGEV48fK3fcVeVh6ofLm3uumPadniBVnOeuZ9/4Sb7OSqjMVgLWh0n8RZXrfNTI4a/Zf0ampp7lSTjBUIW2b2Tb6vLovmrH1RAt3fOEaGvwKbdPWcUwAi4jlommeWLjMtg2Vm5KYDoNAysAbsa70afKd0s/ELRbQn299LPYUc8z3FCTQ2enRL0HTXlxaBx1agzOut+iELy7J6bcps9QVz9cmP/rlGhVT2kzf/r06AMIjpevRNH1BFvoZ/PHs58iNrlpR+3oGIocXK8ErewCZ9khMf01FEytucRm7oml6D7GvHDTyRVKrVP82svjeHI7pB5NHVJnr1ueomViWLkOLLa7QENdU59pZdaFFid4fi6/57YP/ZNgPxxmLFbNq68U73kxuGWVsWzuZ/uoV9aeDWvxeEqii9vnh903fFAFvic4/loqj/8isA+mihthumX2pGbA6p3oSGGiDLHufM11xHi2Ly9h/K2sdWEGY9kFxvpHu6OC1H0J7+ptID/OcbbjmNjEoSDYwdrAxXcCA1wLkQSvxPE89A+mou5kIBMH1s6rlqdfrwOj3LCxqwbeAT3nIS8HkW52ROwV15tYVIxW6oHD2Z2eBWohT9mo1KmWN0v3le3gooTcNoaa3WPDPkgrTNZN3XMpMt6AS8hA8QDNmtI+O0S8KHhRbxTjiUnNm9rORBiktZ+3tAMvpGN4OeofXLe2rzAU7XrlpFFSqaADSUEK+/yyIgQTkzu3jJnVcUej+pXeXTWZv0TN1jl7G3dQ5iGN7nrpSYA1oz8C3VCqT+lJdc34oCmMgH91XDUmvbToD1sQr5vakYn8Dl94xp8OQfb+lVmTjD2x8bA6/nkpMAF77r9PJ0lcbZup5LA+lJrjZOkqSzS/gggWMFS/nVy4ZEyCLXkxpQajdYm3NQPNEzxqIeGCWCoOn5RNRpG5dV+leCGKuiAhD/mk/HtNvSf6hEdn/tFy7fZvDzljA+AtoKU2Wg3N+An06bGy7KhzyvommFz8xAs6y/DjdwW33S1vOAoDVMM3f8F+w9qcnFKrASrHJ8wlch
\ No newline at end of file
diff --git a/.delivery/100-blitz-source/02 b/.delivery/100-blitz-source/02
deleted file mode 100644
index 5134fa9..0000000
--- a/.delivery/100-blitz-source/02
+++ /dev/null
@@ -1 +0,0 @@
-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
\ No newline at end of file
diff --git a/.delivery/100-blitz-source/03 b/.delivery/100-blitz-source/03
deleted file mode 100644
index 0369713..0000000
--- a/.delivery/100-blitz-source/03
+++ /dev/null
@@ -1 +0,0 @@
-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
\ No newline at end of file
diff --git a/.delivery/100-blitz-source/04 b/.delivery/100-blitz-source/04
deleted file mode 100644
index 16f7643..0000000
--- a/.delivery/100-blitz-source/04
+++ /dev/null
@@ -1 +0,0 @@
-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
\ No newline at end of file
diff --git a/.delivery/100-blitz-source/05 b/.delivery/100-blitz-source/05
deleted file mode 100644
index 0a76cd8..0000000
--- a/.delivery/100-blitz-source/05
+++ /dev/null
@@ -1 +0,0 @@
-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
\ No newline at end of file
diff --git a/.delivery/100-blitz-source/06 b/.delivery/100-blitz-source/06
deleted file mode 100644
index f880401..0000000
--- a/.delivery/100-blitz-source/06
+++ /dev/null
@@ -1 +0,0 @@
-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
\ No newline at end of file
diff --git a/.delivery/100-blitz-source/07 b/.delivery/100-blitz-source/07
deleted file mode 100644
index 46cfce7..0000000
--- a/.delivery/100-blitz-source/07
+++ /dev/null
@@ -1 +0,0 @@
-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
\ No newline at end of file
diff --git a/.delivery/100-blitz-source/08 b/.delivery/100-blitz-source/08
deleted file mode 100644
index 3f72158..0000000
--- a/.delivery/100-blitz-source/08
+++ /dev/null
@@ -1 +0,0 @@
-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
\ No newline at end of file
diff --git a/.delivery/100-blitz-source/09 b/.delivery/100-blitz-source/09
deleted file mode 100644
index b8e6d09..0000000
--- a/.delivery/100-blitz-source/09
+++ /dev/null
@@ -1 +0,0 @@
-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
\ No newline at end of file
diff --git a/docs/EXAM_REPAIR.md b/docs/EXAM_REPAIR.md
new file mode 100644
index 0000000..304b0b3
--- /dev/null
+++ b/docs/EXAM_REPAIR.md
@@ -0,0 +1,62 @@
+# Proof-carrying bounded exam repair
+
+`spectra.exam_repair` maps a narrow operational rescheduling problem to the
+versioned `spectra.exact` API:
+
+- each exam has one or two pre-approved candidate periods;
+- exams sharing a student may not use the same period;
+- explicit exam exclusions are also supported;
+- current period closures, fixed moves, and allowed-period restrictions are
+  intersected into an exact query;
+- SAT returns a complete named assignment;
+- UNSAT returns an independently verifiable proof that can be translated into
+  named exams, periods, and a shared-student/exclusion chain.
+
+```python
+from spectra.exam_repair import ExamRepairModel
+from spectra.exact import ExactRuntime, build_exact_runtime
+
+model = ExamRepairModel.create(
+    period_ids=("Monday AM", "Monday PM"),
+    exam_ids=("MATH101", "CHEM101", "HIST101"),
+    students=((11, 20), (20, 31), (31,)),
+    candidate_periods=(("Monday AM", "Monday PM"),) * 3,
+)
+query = model.freeze_event(
+    closed_periods=(),
+    fixed=(("MATH101", "Monday AM"), ("CHEM101", "Monday AM")),
+)
+
+runtime = ExactRuntime(build_exact_runtime("exact-native"))
+with runtime.prepare(model.problem, mode="scc") as session:
+    answer = session.solve(query)
+
+if answer.status == "SAT_VERIFIED":
+    print(model.assignment_from_labels(answer.labels, query))
+else:
+    print(model.explain_unsat(query, answer.proof))
+```
+
+## Scope boundary
+
+This is **bounded repair**, not full timetabling.  The adapter does not generate
+candidate periods or model room capacity, room assignment, precedence,
+coincidence, or soft objectives.  Those constraints must be incorporated before
+admission or checked separately.
+
+`spectra.itc2007_exam` provides a strict reader for the ITC-2007 examination
+format.  Its `conflict_core` conversion preserves an immutable receipt of every
+constraint class outside the bounded model and refuses conversion unless the
+caller explicitly acknowledges those omissions.
+
+The research transfer protocol is documented in
+[`ITC_EXAM_TRANSFER_PROTOCOL.md`](../experiments/compiled_query/ITC_EXAM_TRANSFER_PROTOCOL.md).
+
+## Prospective multi-instance evaluation
+
+The selector-v2 policy and its `set10`/`set12` execution holdouts are fixed in
+[`ITC_EXAM_SUITE_PROTOCOL.md`](../experiments/compiled_query/ITC_EXAM_SUITE_PROTOCOL.md).
+The suite rotates measured engine order, preserves every failed cell, and gates
+only the predeclared auto/raw overhead on holdout cases.  Passing that suite is
+application-transfer evidence, not an operator deployment or full ITC-2007
+optimization result.
diff --git a/experiments/compiled_query/ITC_EXAM_SUITE_PROTOCOL.md b/experiments/compiled_query/ITC_EXAM_SUITE_PROTOCOL.md
new file mode 100644
index 0000000..b36df50
--- /dev/null
+++ b/experiments/compiled_query/ITC_EXAM_SUITE_PROTOCOL.md
@@ -0,0 +1,91 @@
+# Prospective ITC-2007 exam-repair suite protocol
+
+**Status:** frozen before executing the `set10` and `set12` result panels.  The
+selector may be falsified by these holdouts, but it must not be retuned inside the
+same suite.
+
+## Purpose
+
+The first application transfer used `set9` and exposed an important limitation:
+the original automatic policy selected the SCC engine even though the explicit raw
+engine was faster on that dense repair bank.  The structural-v2 policy adds one
+answer-independent rule:
+
+> For a bank of at least 64 requests, use raw traversal when the immutable request
+> packet contains at least eight pins per request on average.
+
+The threshold was chosen from the already exposed synthetic development panel and
+the `set9` development transfer.  It depends only on packet geometry, not SAT/UNSAT
+answers, proof sizes, or measured times.  Short banks, base-infeasible graphs and
+the prior SCC/component rules retain their previous ordering.
+
+## Fixed external source
+
+Every workflow checks out `davidvanbulck/ITC-2007-CETT` at commit
+`ed2d4c0cf5302b77a8b321cd0380de0bb579d538` and retains its `LICENSE` in each
+case directory.  No upstream data is committed to SPECTRA.
+
+The fixed cases are:
+
+| Case | Role | Instance | Published feasible schedule |
+|---|---|---|---|
+| `set9-development` | Development | `examtt-instances/Instances/set9.exam` | `InitMethods/SolStore/set9_OBSIFeas_0.sol` |
+| `set10-holdout` | Prospective holdout | `examtt-instances/Instances/set10.exam` | `InitMethods/SolStore/set10_OBSIFeas_0.sol` |
+| `set12-holdout` | Prospective holdout | `examtt-instances/Instances/set12.exam` | `InitMethods/SolStore/set12_OBSIFeas_0.sol` |
+
+The bytes are public, so “holdout” means **execution holdout**, not secret data:
+no selector-v2 answer or timing from `set10` or `set12` may influence the committed
+policy or acceptance threshold.
+
+## Per-case construction
+
+Each case uses the exact construction in
+[`ITC_EXAM_TRANSFER_PROTOCOL.md`](ITC_EXAM_TRANSFER_PROTOCOL.md): validate the
+published schedule against all original hard constraints, freeze exams involved in
+unsupported hard constraints, assign one deterministic duration-compatible
+alternative to eligible exams, build the shared-student/`EXCLUSION` conflict core,
+and generate 1,024 unique deterministic repair events.
+
+Every event is answered by `auto`, `raw`, `scc`, and `route`.  Completed answers
+must include a full assignment or portable contradiction proof and must replay from
+the original `SBC1` packet without native code.
+
+## Timing schedule and prospective gate
+
+Each engine receives one untimed warmup.  In each of five measured repetitions the
+engine order rotates, preventing one mode from always running first or last:
+
+1. `auto, raw, scc, route`
+2. `raw, scc, route, auto`
+3. `scc, route, auto, raw`
+4. `route, auto, raw, scc`
+5. repeat from the first order
+
+For every holdout, pair `auto` and `raw` by repetition and compute the median of the
+five `auto_elapsed / raw_elapsed` ratios.  The predeclared guardrail is:
+
+```text
+paired-median auto/raw ratio <= 1.15
+```
+
+This is a bounded overhead guardrail, not a speedup claim.  Development-case timing
+is reported but cannot pass or fail the prospective gate.  Failed holdouts remain
+in the artifact and may not be removed, renamed, rerun under a changed threshold,
+or converted to development cases.
+
+## Evidence and audit
+
+The suite manifest binds:
+
+- source commit and per-case source inventory;
+- upstream repository and commit;
+- exact instance and solution paths and SHA-256 identities;
+- development/holdout role;
+- selector policy and threshold;
+- per-case manifest hashes.
+
+The suite audit reruns every child audit, checks every result and proof, checks the
+rotated timing schedule, computes paired ratios from recorded repetitions, and
+publishes all failures.  A successful result is still project-owned execution, not
+outside-team replication, operator adoption, full ITC-2007 feasibility, or
+algorithmic-priority evidence.
diff --git a/experiments/compiled_query/ITC_EXAM_TRANSFER_PROTOCOL.md b/experiments/compiled_query/ITC_EXAM_TRANSFER_PROTOCOL.md
new file mode 100644
index 0000000..59c79df
--- /dev/null
+++ b/experiments/compiled_query/ITC_EXAM_TRANSFER_PROTOCOL.md
@@ -0,0 +1,127 @@
+# ITC-2007 bounded exam-repair transfer protocol
+
+**Status:** fixed before opening the external instance and schedule in CI.  This is
+an application-format transfer of the versioned proof-carrying exact-query API,
+not a complete capacitated examination-timetabling result.
+
+## External source and attribution
+
+The workflow checks out `davidvanbulck/ITC-2007-CETT` at commit
+`ed2d4c0cf5302b77a8b321cd0380de0bb579d538`.  The selected files are:
+
+- `examtt-instances/Instances/set9.exam`
+- `InitMethods/SolStore/set9_OBSIFeas_0.sol`
+- `LICENSE`
+
+The upstream repository states that its solver and instances originate from the
+ITC-2007 examination-timetabling setting and requests citation of Van Bulck,
+Goossens and Schaerf.  Every evidence package retains the upstream license and
+this acknowledgment:
+
+> This product includes software/data distributed by David Van Bulck, Dries
+> Goossens, and Andrea Schaerf, Ghent University, Belgium
+> (https://github.com/davidvanbulck/ITC-2007-CETT.git).
+
+No upstream source or data is committed to SPECTRA.
+
+## Admitted question
+
+Start from the published feasible timetable.  Each examination is assigned one
+or two **pre-approved candidate periods**:
+
+1. The published period is always retained.
+2. Exams appearing in `AFTER`, `EXAM_COINCIDENCE`, or `ROOM_EXCLUSIVE` hard
+   constraints are frozen to the published period.
+3. Every other exam receives one deterministic alternative when available.  The
+   alternative must have enough duration and the instance must contain a room
+   large enough for the exam in isolation.  A SHA-256 ordering of
+   `(input hash, exam, period)` chooses among eligible alternatives.
+4. Shared-student conflicts and explicit `EXCLUSION` constraints become graph
+   edges.  The palette contains the instance's declared periods.
+
+The exact question is therefore:
+
+> Given current period closures and requested moves within these fixed
+> one/two-period domains, does a conflict-free period assignment exist?
+
+A completed call must return either a full original-exam assignment or a portable
+original-literal contradiction proof.  The standard-library verifier replays all
+answers from the frozen `SBC1` packet and `SCQ2` output.
+
+## Deliberately unmodelled constraints
+
+The reduction does **not** solve or score:
+
+- room reassignment or simultaneous room-capacity packing;
+- soft institutional penalties;
+- changes to `AFTER`, `EXAM_COINCIDENCE`, or `ROOM_EXCLUSIVE` exams;
+- candidate-period generation outside the frozen deterministic rule;
+- the full ITC-2007 optimization problem.
+
+The parser retains counts for every omitted constraint class.  Conversion requires
+an explicit `acknowledge_ignored_constraints=True`, and the omission receipt is
+stored in every manifest and audit.  A conflict-core SAT answer is not a complete
+ITC timetable certificate.
+
+## Frozen event bank
+
+The bank contains 1,024 unique events in this order:
+
+1. the unchanged published baseline;
+2. every single-period closure;
+3. every single flexible-exam move to its alternative;
+4. deterministic multi-exam move sets of width 2 through 12;
+5. deterministic period-closure plus multi-move events.
+
+The seed is the first 64 bits of SHA-256 over the exact instance bytes, a NUL byte,
+and the exact solution bytes.  Widths, exam rankings and closed periods are then
+derived directly from labelled SHA-256 words rather than Python's version-sensitive
+sampling helpers.  Duplicate exact queries are removed before the bank is
+truncated.  Failure to create 1,024 distinct events is a protocol failure, not
+permission to alter the policy after observing results.
+
+## Engines and measured boundary
+
+The same immutable packet is run through `auto`, `raw`, `scc`, and `route`.  Each
+mode receives one untimed warmup; five measured repetitions rotate the mode order
+so no mode always executes first or last.  The measured boundary starts with ready
+`SBC1` bytes and includes:
+
+- input validation and native construction;
+- every query decision;
+- complete SAT witnesses or portable UNSAT proofs;
+- the separately owned native observer;
+- `SCQ2` materialization;
+- native disposal and the Python call boundary.
+
+External checkout, parsing, candidate generation, packet packing, native
+compilation, evidence writes, and offline replay are outside the timing boundary.
+Timing is descriptive for this single development transfer; there is no speed-win
+gate here.  The separate prospective multi-instance suite declares its holdout
+threshold before execution.
+
+## Freeze, execution and audit
+
+`freeze` validates everything before atomically publishing an evidence directory.
+It writes external bytes, the upstream license, candidate rows, canonical queries,
+packet, source inventory, commit/tree identities and hashes with exclusive
+creation.  Every listed source blob must match the named commit.
+
+`run` explicitly builds the bundled ABI-3 runtime, records its receipt hash, runs
+all engines/repetitions, requires byte-stable repeated output per engine, and
+verifies every output before publication.
+
+`audit` is rerunnable without a native solver.  It reparses the original instance
+and solution, rechecks all baseline hard constraints, regenerates the seed,
+candidate rows and complete event bank, reconstructs the packet, checks recorded
+metadata against the certified bytes, replays every output, requires all modes to
+agree on SAT/UNSAT decisions, counts proof kinds, and refuses changed evidence or
+a changed prior audit.  Stored candidate/query JSON is evidence, not an oracle.
+
+## Interpretation
+
+A successful run demonstrates that the exact API transfers from exposed synthetic
+graph panels to a published timetabling-format repair core while retaining full
+proof boundaries.  It remains project-owned execution on a deterministic derived
+problem.  It is not operator adoption, outside-team replication, a claim of
+algorithmic priority, or evidence that SPECTRA solves the full ITC-2007 task.
diff --git a/experiments/compiled_query/exact.py b/experiments/compiled_query/exact.py
index 780a7ff..7eee4c2 100644
--- a/experiments/compiled_query/exact.py
+++ b/experiments/compiled_query/exact.py
@@ -363,11 +363,13 @@ def solve_session_packet(self, packet: bytes, *, mode: str = 'route',
                              max_bytes: int = DEFAULT_BYTES) -> ExactSessionPacketAnswer:
         """Construct, solve, verify, serialize and dispose one prepared SBC1 session.
 
-        Opt-in ``auto`` uses raw traversal below 64 requests or on a base-infeasible
-        graph. Otherwise it uses routed proofs when SCCs at least halve the
-        literal graph, and plain SCC batching when they do not. Selection uses
-        fixed-graph structure and bank length, never measured answers or timings.
-        ``info['selected_mode']`` names the actual engine. Explicit modes and the
+        Opt-in ``auto`` uses structural policy v2. It selects raw traversal below
+        64 requests, on a base-infeasible graph, or when the immutable bank carries
+        at least eight pins per request on average. Otherwise it uses routed proofs
+        when SCCs at least halve the literal graph, and plain SCC batching when they
+        do not. Selection uses only admitted graph/request geometry, never answers
+        or measured timings. ``info['selected_mode']`` and
+        ``info['auto_policy_reason']`` expose the decision. Explicit modes and the
         default retain their meaning; this policy is an experimental heuristic.
         """
         if type(packet) is not bytes:
diff --git a/experiments/compiled_query/exact_native.cpp b/experiments/compiled_query/exact_native.cpp
index 96182ee..96694d3 100644
--- a/experiments/compiled_query/exact_native.cpp
+++ b/experiments/compiled_query/exact_native.cpp
@@ -341,12 +341,13 @@ struct ObserverSessionPacket {
 struct Answer {bool sat=true;std::vector<uint8_t> labels,certificate;W visits=0,arcs=0;};
 struct Index {
     U n,k,nc=0,bad=NIL;bool condensed,sparse,routed;W cap;W full_arcs=0;
+    const char* auto_reason="explicit_mode";
     std::vector<W> masks;
     std::vector<uint8_t> lo,hi,base;
     std::vector<U> comp,opposite;
     std::vector<U> inward,outward,representative,cross_from,cross_to;
     CSR original,dag;
-    Index(Input p,U mode,U bank_size=0):n(p.n),k(p.k),condensed(mode!=0),sparse(mode==2),routed(mode==3),cap(p.cap),masks(std::move(p.masks)),lo(n),hi(n,255),base(n) {
+    Index(Input p,U mode,U bank_size=0,U bank_pins=0):n(p.n),k(p.k),condensed(mode!=0),sparse(mode==2),routed(mode==3),cap(p.cap),masks(std::move(p.masks)),lo(n),hi(n,255),base(n) {
         std::vector<std::pair<U,U>> arcs;arcs.reserve(4*p.edges.size()+n);
         for(U v=0;v<n;++v) {lo[v]=first(masks[v]);W h=masks[v] & (masks[v]-1);
             if(h) hi[v]=first(h); else arcs.emplace_back(2*v+1,2*v);
@@ -382,11 +383,22 @@ struct Index {
             base[v]=static_cast<uint8_t>(comp[2*v]<comp[2*v+1]);
         }
         if(mode==4) {
-            // Opt-in, query-blind policy. Small banks cannot use the lane kernel;
-            // base-infeasible graphs do not need a request-propagation index.
-            // Route proofs only when SCCs at least halve the literal graph.
-            condensed=bank_size>=64 && bad==NIL;
-            routed=condensed && nc<=n;
+            // Structural auto policy v2.  set9 development evidence showed that
+            // large, heavily pinned repair banks can lose to raw traversal even
+            // when tile64 is available.  The policy therefore uses only immutable
+            // request geometry (never answers or measured timings): raw below one
+            // tile, on a base contradiction, or at eight pins per row and above.
+            // The threshold is frozen before the set10/set12 holdout transfer.
+            if(bank_size<64) {
+                condensed=false;routed=false;auto_reason="bank_below_tile64";
+            } else if(bad!=NIL) {
+                condensed=false;routed=false;auto_reason="base_infeasible";
+            } else if(W(bank_pins)>=8*W(bank_size)) {
+                condensed=false;routed=false;auto_reason="high_pin_density";
+            } else {
+                condensed=true;routed=nc<=n;
+                auto_reason=routed?"route_component_compression":"scc_tile64";
+            }
         }
         if(condensed && bad==NIL) {
             opposite.assign(nc,NIL);
@@ -990,6 +1002,7 @@ PyObject* solve_session_packet(PyObject*,PyObject* args) {
         const auto* data=reinterpret_cast<const uint8_t*>(PyBytes_AS_STRING(packet));
         size_t size=static_cast<size_t>(PyBytes_GET_SIZE(packet));
         std::unique_ptr<Index> solver;std::unique_ptr<Observer> observer;BankResult result;
+        U bank_queries=0,bank_pins=0;
         {
             Released guard;
             ObserverSessionPacket observer_rows(data,size,limit);
@@ -997,8 +1010,9 @@ PyObject* solve_session_packet(PyObject*,PyObject* args) {
             W observer_bytes=observer->bytes();
             if(observer_bytes>=limit) throw Resource("SBC1 observer consumes session cap");
             SolverSessionPacket solver_rows(data,size,limit-observer_bytes);
+            bank_queries=solver_rows.query_count();bank_pins=solver_rows.pin_count();
             solver=std::make_unique<Index>(solver_rows.take_input(limit-observer_bytes),
-                                           static_cast<U>(which),solver_rows.query_count());
+                                           static_cast<U>(which),bank_queries,bank_pins);
             result=run_rows(*solver,*observer,solver_rows,observer_rows,size,limit,"SBC1");
         }
         PyObject* output=PyBytes_FromStringAndSize(reinterpret_cast<const char*>(result.output.data()),
@@ -1018,11 +1032,22 @@ PyObject* solve_session_packet(PyObject*,PyObject* args) {
         if(!metadata){Py_DECREF(output);return nullptr;}
         PyObject* kernel=PyUnicode_FromString(result.tiled?"tile64":"serial");
         PyObject* tiles=PyLong_FromUnsignedLong(result.tiles);
-        if(!kernel || !tiles || PyDict_SetItemString(metadata,"bank_kernel",kernel)<0
-                || PyDict_SetItemString(metadata,"bank_tiles",tiles)<0) {
-            Py_XDECREF(kernel);Py_XDECREF(tiles);Py_DECREF(output);Py_DECREF(metadata);return nullptr;
+        PyObject* queries=PyLong_FromUnsignedLong(bank_queries);
+        PyObject* pins=PyLong_FromUnsignedLong(bank_pins);
+        PyObject* policy=PyUnicode_FromString(which==4?"structural-v2":"explicit");
+        PyObject* reason=PyUnicode_FromString(solver->auto_reason);
+        if(!kernel || !tiles || !queries || !pins || !policy || !reason
+                || PyDict_SetItemString(metadata,"bank_kernel",kernel)<0
+                || PyDict_SetItemString(metadata,"bank_tiles",tiles)<0
+                || PyDict_SetItemString(metadata,"bank_query_count",queries)<0
+                || PyDict_SetItemString(metadata,"bank_pin_count",pins)<0
+                || PyDict_SetItemString(metadata,"auto_policy",policy)<0
+                || PyDict_SetItemString(metadata,"auto_policy_reason",reason)<0) {
+            Py_XDECREF(kernel);Py_XDECREF(tiles);Py_XDECREF(queries);Py_XDECREF(pins);
+            Py_XDECREF(policy);Py_XDECREF(reason);Py_DECREF(output);Py_DECREF(metadata);return nullptr;
         }
-        Py_DECREF(kernel);Py_DECREF(tiles);
+        Py_DECREF(kernel);Py_DECREF(tiles);Py_DECREF(queries);Py_DECREF(pins);
+        Py_DECREF(policy);Py_DECREF(reason);
         return Py_BuildValue("NKKKKKKN",output,
             static_cast<unsigned long long>(result.sat+result.unsat),
             static_cast<unsigned long long>(result.sat),
diff --git a/experiments/compiled_query/itc_exam_suite.py b/experiments/compiled_query/itc_exam_suite.py
new file mode 100644
index 0000000..dfc8b71
--- /dev/null
+++ b/experiments/compiled_query/itc_exam_suite.py
@@ -0,0 +1,352 @@
+"""Prospective multi-instance ITC-2007 exam-repair transfer suite.
+
+The selector-v2 rule is developed from the exposed synthetic panel and the
+published ``set9`` transfer only.  ``set10`` and ``set12`` are fixed holdout
+instances: their execution results must not be used to retune the policy.  This
+suite remains a bounded conflict-core repair study, not a full timetabling claim.
+"""
+from __future__ import annotations
+
+import argparse
+from dataclasses import dataclass
+import json
+from pathlib import Path, PurePosixPath
+import shutil
+import statistics
+from typing import Iterable
+
+from experiments.compiled_query import itc_exam_transfer as transfer
+
+SCHEMA = "spectra.itc2007.exam_repair_suite.v1"
+DEFAULT_HOLDOUT_MAX_RATIO = 1.15
+
+
+@dataclass(frozen=True)
+class SuiteCase:
+    name: str
+    role: str
+    instance_path: str
+    solution_path: str
+
+    def validate(self) -> "SuiteCase":
+        if (type(self.name) is not str or not self.name
+                or any(ch not in "abcdefghijklmnopqrstuvwxyz0123456789-" for ch in self.name)
+                or self.name.startswith("-") or self.name.endswith("-")):
+            raise ValueError("suite case name must be lowercase kebab-case")
+        if self.role not in ("development", "holdout", "replication"):
+            raise ValueError("suite case role differs")
+        for label, value in (("instance", self.instance_path), ("solution", self.solution_path)):
+            if type(value) is not str or not value:
+                raise ValueError(f"suite case {label} path must be nonempty")
+            path = PurePosixPath(value)
+            if path.is_absolute() or ".." in path.parts:
+                raise ValueError(f"suite case {label} path must stay under upstream root")
+        return self
+
+
+DEFAULT_CASES = (
+    SuiteCase(
+        "set9-development", "development",
+        "examtt-instances/Instances/set9.exam",
+        "InitMethods/SolStore/set9_OBSIFeas_0.sol",
+    ),
+    SuiteCase(
+        "set10-holdout", "holdout",
+        "examtt-instances/Instances/set10.exam",
+        "InitMethods/SolStore/set10_OBSIFeas_0.sol",
+    ),
+    SuiteCase(
+        "set12-holdout", "holdout",
+        "examtt-instances/Instances/set12.exam",
+        "InitMethods/SolStore/set12_OBSIFeas_0.sol",
+    ),
+)
+
+
+def _cases(cases: Iterable[SuiteCase]) -> tuple[SuiteCase, ...]:
+    checked = tuple(case.validate() for case in cases)
+    if not checked or len({case.name for case in checked}) != len(checked):
+        raise ValueError("suite requires unique nonempty cases")
+    if not any(case.role == "holdout" for case in checked):
+        raise ValueError("suite requires at least one holdout")
+    return checked
+
+
+def _ratio(value: float | None) -> float | None:
+    if value is None:
+        return None
+    if type(value) not in (int, float) or not 1.0 <= float(value) <= 10.0:
+        raise ValueError("holdout max auto/raw ratio must be in [1, 10] or null")
+    return float(value)
+
+
+def _manifest_path(root: Path) -> Path:
+    return root / "suite-manifest.json"
+
+
+def freeze(root: Path, upstream_root: Path, *, cases: Iterable[SuiteCase] = DEFAULT_CASES,
+           query_count: int = 1024,
+           holdout_max_auto_raw_ratio: float | None = DEFAULT_HOLDOUT_MAX_RATIO,
+           source_commit: str | None = None) -> dict:
+    root = root.resolve()
+    upstream_root = upstream_root.resolve()
+    if root.exists():
+        raise FileExistsError(root)
+    checked = _cases(cases)
+    threshold = _ratio(holdout_max_auto_raw_ratio)
+    license_path = upstream_root / "LICENSE"
+    if not license_path.is_file():
+        raise FileNotFoundError(license_path)
+    staging = root.with_name(root.name + ".partial")
+    if staging.exists():
+        raise FileExistsError(staging)
+    (staging / "cases").mkdir(parents=True, exist_ok=False)
+    try:
+        case_records = []
+        for case in checked:
+            case_root = staging / "cases" / case.name
+            manifest = transfer.freeze(
+                case_root,
+                upstream_root / case.instance_path,
+                upstream_root / case.solution_path,
+                query_count=query_count,
+                license_path=license_path,
+                source_commit=source_commit,
+                dataset_instance_path=case.instance_path,
+                dataset_solution_path=case.solution_path,
+                evaluation_role=case.role,
+            )
+            case_records.append({
+                "name": case.name,
+                "role": case.role,
+                "instance_path": case.instance_path,
+                "solution_path": case.solution_path,
+                "manifest_sha256": (case_root / "manifest.sha256").read_text().strip(),
+                "instance_sha256": manifest["instance_sha256"],
+                "solution_sha256": manifest["solution_sha256"],
+            })
+        manifest = {
+            "schema": SCHEMA,
+            "classification": (
+                "prospective multi-instance bounded exam-repair transfer; set9 is "
+                "development and set10/set12 are fixed execution holdouts"
+            ),
+            "source_commit": case_records and json.loads(
+                (staging / "cases" / case_records[0]["name"] / "manifest.json").read_text()
+            )["source_commit"],
+            "dataset_repository": transfer.DATASET_REPOSITORY,
+            "dataset_commit": transfer.DATASET_COMMIT,
+            "query_count_per_case": query_count,
+            "selector_policy": "structural-v2",
+            "selector_development_sources": [
+                "exposed synthetic graph panel",
+                "set9 development transfer",
+            ],
+            "holdout_policy": (
+                "set10 and set12 results may falsify the frozen selector but may not "
+                "change it inside this suite"
+            ),
+            "holdout_max_auto_raw_ratio": threshold,
+            "gate_metric": "median of paired per-repetition auto/raw elapsed ratios",
+            "cases": case_records,
+        }
+        payload = transfer.canonical_json(manifest)
+        transfer.write_once(_manifest_path(staging), payload)
+        transfer.write_once(
+            staging / "suite-manifest.sha256",
+            (transfer.digest(payload) + "\n").encode(),
+        )
+        staging.rename(root)
+        return manifest
+    except BaseException:
+        shutil.rmtree(staging, ignore_errors=True)
+        raise
+
+
+def load_manifest(root: Path) -> dict:
+    root = root.resolve()
+    expected = (root / "suite-manifest.sha256").read_text().strip()
+    manifest = json.loads(transfer.read_bound(_manifest_path(root), expected))
+    if (manifest.get("schema") != SCHEMA
+            or manifest.get("dataset_repository") != transfer.DATASET_REPOSITORY
+            or manifest.get("dataset_commit") != transfer.DATASET_COMMIT
+            or manifest.get("selector_policy") != "structural-v2"
+            or type(manifest.get("query_count_per_case")) is not int
+            or manifest["query_count_per_case"] < 1):
+        raise ValueError("suite manifest identity differs")
+    threshold = _ratio(manifest.get("holdout_max_auto_raw_ratio"))
+    records = manifest.get("cases")
+    if type(records) is not list:
+        raise ValueError("suite case inventory differs")
+    checked = _cases(SuiteCase(
+        item.get("name"), item.get("role"),
+        item.get("instance_path"), item.get("solution_path"),
+    ) for item in records if type(item) is dict)
+    if len(checked) != len(records):
+        raise ValueError("suite case inventory differs")
+    for case, item in zip(checked, records):
+        case_root = root / "cases" / case.name
+        if ((case_root / "manifest.sha256").read_text().strip()
+                != item.get("manifest_sha256")):
+            raise ValueError(f"suite case {case.name} manifest hash differs")
+        child = transfer.load_manifest(case_root)
+        if (child.get("dataset_instance_path") != case.instance_path
+                or child.get("dataset_solution_path") != case.solution_path
+                or child.get("evaluation_role") != case.role
+                or child.get("query_count") != manifest["query_count_per_case"]
+                or child.get("instance_sha256") != item.get("instance_sha256")
+                or child.get("solution_sha256") != item.get("solution_sha256")):
+            raise ValueError(f"suite case {case.name} identity differs")
+    manifest["holdout_max_auto_raw_ratio"] = threshold
+    return manifest
+
+
+def run(root: Path) -> dict:
+    root = root.resolve()
+    manifest = load_manifest(root)
+    if (root / "suite-run.json").exists():
+        raise FileExistsError("suite execution already exists")
+    records = []
+    for item in manifest["cases"]:
+        case_root = root / "cases" / item["name"]
+        record = transfer.run(case_root)
+        records.append({
+            "name": item["name"],
+            "run_sha256": (case_root / "run.sha256").read_text().strip(),
+            "execution_order": record["execution_order"],
+        })
+    result = {
+        "schema": SCHEMA + ".run",
+        "suite_manifest_sha256": (root / "suite-manifest.sha256").read_text().strip(),
+        "cases": records,
+    }
+    payload = transfer.canonical_json(result)
+    transfer.write_once(root / "suite-run.json", payload)
+    transfer.write_once(root / "suite-run.sha256", (transfer.digest(payload) + "\n").encode())
+    return result
+
+
+def audit(root: Path) -> dict:
+    root = root.resolve()
+    manifest = load_manifest(root)
+    run_expected = (root / "suite-run.sha256").read_text().strip()
+    run_record = json.loads(transfer.read_bound(root / "suite-run.json", run_expected))
+    if (run_record.get("schema") != SCHEMA + ".run"
+            or run_record.get("suite_manifest_sha256")
+            != (root / "suite-manifest.sha256").read_text().strip()):
+        raise ValueError("suite run identity differs")
+    run_cases = run_record.get("cases")
+    if type(run_cases) is not list or len(run_cases) != len(manifest["cases"]):
+        raise ValueError("suite run case inventory differs")
+
+    threshold = manifest["holdout_max_auto_raw_ratio"]
+    summaries = []
+    holdout_failures = []
+    for item, run_item in zip(manifest["cases"], run_cases):
+        if type(run_item) is not dict or run_item.get("name") != item["name"]:
+            raise ValueError("suite run case order differs")
+        case_root = root / "cases" / item["name"]
+        if (case_root / "run.sha256").read_text().strip() != run_item.get("run_sha256"):
+            raise ValueError(f"suite case {item['name']} run hash differs")
+        result = transfer.audit(case_root)
+        child_run = json.loads(transfer.read_bound(
+            case_root / "run.json", (case_root / "run.sha256").read_text().strip()
+        ))
+        auto_elapsed = child_run["modes"]["auto"]["elapsed_ns"]
+        raw_elapsed = child_run["modes"]["raw"]["elapsed_ns"]
+        paired_median = statistics.median(
+            auto / raw for auto, raw in zip(auto_elapsed, raw_elapsed)
+        )
+        selected = result["modes"]["auto"]["selected_mode"]
+        reason = result["modes"]["auto"]["auto_policy_reason"]
+        gate_pass = True
+        if item["role"] == "holdout" and threshold is not None:
+            gate_pass = paired_median <= threshold
+            if not gate_pass:
+                holdout_failures.append(item["name"])
+        summaries.append({
+            "name": item["name"],
+            "role": item["role"],
+            "functional_complete": result["functional_complete"],
+            "baseline_valid_full_itc_hard_constraints": result[
+                "baseline_valid_full_itc_hard_constraints"
+            ],
+            "instance_sha256": result["instance_sha256"],
+            "solution_sha256": result["solution_sha256"],
+            "model": result["model"],
+            "queries": result["queries"],
+            "sat": result["modes"]["auto"]["sat"],
+            "unsat": result["modes"]["auto"]["unsat"],
+            "proof_kinds": result["modes"]["auto"]["proof_kinds"],
+            "auto_selected_mode": selected,
+            "auto_policy_reason": reason,
+            "auto_vs_raw_median_ratio": result["auto_vs_raw_median_ratio"],
+            "auto_vs_raw_paired_median_ratio": paired_median,
+            "holdout_gate_pass": gate_pass,
+            "audit_sha256": (case_root / "audit.sha256").read_text().strip(),
+        })
+    functional = all(
+        item["functional_complete"] and item["baseline_valid_full_itc_hard_constraints"]
+        for item in summaries
+    )
+    result = {
+        "schema": SCHEMA + ".audit",
+        "functional_complete": functional,
+        "source_commit": manifest["source_commit"],
+        "dataset_commit": manifest["dataset_commit"],
+        "selector_policy": manifest["selector_policy"],
+        "query_count_per_case": manifest["query_count_per_case"],
+        "holdout_max_auto_raw_ratio": threshold,
+        "holdout_gate_pass": not holdout_failures,
+        "holdout_failures": holdout_failures,
+        "cases": summaries,
+        "interpretation": [
+            "set9 is development evidence; set10 and set12 are prospective execution holdouts",
+            "the gate limits auto-policy overhead relative to the explicit raw control",
+            "all outputs remain complete proof-carrying conflict-core repair answers",
+            "this is not full ITC-2007 timetable feasibility or optimization",
+            "project-owned CI is not outside-team replication or adoption",
+        ],
+    }
+    payload = transfer.canonical_json(result)
+    path = root / "suite-audit.json"
+    if path.exists():
+        expected = (root / "suite-audit.sha256").read_text().strip()
+        if transfer.read_bound(path, expected) != payload:
+            raise ValueError("existing suite audit differs from independent replay")
+    else:
+        transfer.write_once(path, payload)
+        transfer.write_once(
+            root / "suite-audit.sha256", (transfer.digest(payload) + "\n").encode()
+        )
+    return result
+
+
+def main(argv: Iterable[str] | None = None) -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("command", choices=("freeze", "run", "audit"))
+    parser.add_argument("--root", type=Path, required=True)
+    parser.add_argument("--upstream-root", type=Path)
+    parser.add_argument("--queries", type=int, default=1024)
+    parser.add_argument(
+        "--holdout-max-auto-raw-ratio", type=float,
+        default=DEFAULT_HOLDOUT_MAX_RATIO,
+    )
+    args = parser.parse_args(argv)
+    if args.command == "freeze":
+        if args.upstream_root is None:
+            parser.error("freeze requires --upstream-root")
+        value = freeze(
+            args.root, args.upstream_root, query_count=args.queries,
+            holdout_max_auto_raw_ratio=args.holdout_max_auto_raw_ratio,
+        )
+    elif args.command == "run":
+        value = run(args.root)
+    else:
+        value = audit(args.root)
+    print(json.dumps(value, indent=2, sort_keys=True, allow_nan=False))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/experiments/compiled_query/itc_exam_transfer.py b/experiments/compiled_query/itc_exam_transfer.py
new file mode 100644
index 0000000..fbbc025
--- /dev/null
+++ b/experiments/compiled_query/itc_exam_transfer.py
@@ -0,0 +1,752 @@
+"""Proof-carrying bounded repair on a pinned ITC-2007 examination instance.
+
+This is a transfer study for the installed ``spectra.exact`` API, not a claim to
+solve the complete capacitated timetabling objective.  It starts from a published
+feasible timetable, freezes exams involved in unsupported hard constraints, gives
+other exams one deterministic duration-compatible alternative period, and solves
+only the shared-student/EXCLUSION conflict core.  Every result is replayed from the
+stored SBC1 packet and SCQ2 evidence without trusting the native index.
+"""
+from __future__ import annotations
+
+import argparse
+from collections import Counter, defaultdict
+import hashlib
+import json
+from pathlib import Path
+import shutil
+import statistics
+import subprocess
+import sys
+import time
+from typing import Iterable
+
+from spectra.exact import (
+    ExactRuntime,
+    build_exact_runtime,
+    decode_certified_output,
+    pack_validated_session_packet,
+    verify_session_output,
+)
+from spectra.exact_proof import decode_proof
+from spectra.itc2007_exam import ITCExamInstance, parse_itc2007_exam
+
+SCHEMA = "spectra.itc2007.exam_repair_transfer.v1"
+DATASET_REPOSITORY = "davidvanbulck/ITC-2007-CETT"
+DATASET_COMMIT = "ed2d4c0cf5302b77a8b321cd0380de0bb579d538"
+DEFAULT_INSTANCE = "examtt-instances/Instances/set9.exam"
+DEFAULT_SOLUTION = "InitMethods/SolStore/set9_OBSIFeas_0.sol"
+MODES = ("auto", "raw", "scc", "route")
+WARMUPS = 1
+REPEATS = 5
+MAX_INPUT_BYTES = 64 << 20
+SOURCE_PATHS = (
+    Path("pyproject.toml"), Path("README.md"),
+    Path("spectra/exact.py"), Path("spectra/exact_problem.py"),
+    Path("spectra/exact_proof.py"), Path("spectra/exam_repair.py"),
+    Path("spectra/itc2007_exam.py"),
+    Path("spectra/_native/exact_binary.cpp"),
+    Path("tests/public/test_exam_repair.py"),
+    Path("tests/public/test_itc2007_exam.py"),
+    Path("scripts/check_exam_repair_installation.py"),
+    Path("docs/EXAM_REPAIR.md"), Path("docs/README.md"),
+    Path("experiments/compiled_query/exact.py"),
+    Path("experiments/compiled_query/exact_native.cpp"),
+    Path("experiments/compiled_query/test_exact.py"),
+    Path("experiments/compiled_query/itc_exam_transfer.py"),
+    Path("experiments/compiled_query/itc_exam_suite.py"),
+    Path("experiments/compiled_query/test_itc_exam_transfer.py"),
+    Path("experiments/compiled_query/test_itc_exam_suite.py"),
+    Path("experiments/compiled_query/ITC_EXAM_TRANSFER_PROTOCOL.md"),
+    Path("experiments/compiled_query/ITC_EXAM_SUITE_PROTOCOL.md"),
+    Path(".github/workflows/itc-exam-transfer.yml"),
+    Path(".github/workflows/itc-exam-suite.yml"),
+)
+
+def digest(data: bytes) -> str:
+    return hashlib.sha256(data).hexdigest()
+
+
+def canonical_json(value: object) -> bytes:
+    return (json.dumps(value, sort_keys=True, separators=(",", ":"),
+                       ensure_ascii=True, allow_nan=False) + "\n").encode()
+
+
+def write_once(path: Path, data: bytes) -> str:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    with path.open("xb") as stream:
+        stream.write(data)
+    return digest(data)
+
+
+def deterministic_word(seed: int, *parts: int) -> int:
+    payload = seed.to_bytes(8, "little", signed=False)
+    for part in parts:
+        if type(part) is not int or part < 0:
+            raise ValueError("deterministic schedule parts must be nonnegative exact integers")
+        payload += part.to_bytes(8, "little", signed=False)
+    return int.from_bytes(hashlib.sha256(payload).digest()[:8], "little")
+
+
+def read_bound(path: Path, expected: str | None = None,
+               *, max_bytes: int = MAX_INPUT_BYTES) -> bytes:
+    with path.open("rb") as stream:
+        data = stream.read(max_bytes + 1)
+    if len(data) > max_bytes:
+        raise ValueError(f"{path} exceeds the bounded read limit")
+    if expected is not None and digest(data) != expected:
+        raise ValueError(f"{path} hash differs")
+    return data
+
+
+def parse_solution(text: str, instance: ITCExamInstance) -> tuple[tuple[int, ...], tuple[int, ...]]:
+    if type(text) is not str:
+        raise ValueError("solution must be an exact string")
+    periods: list[int] = []
+    rooms: list[int] = []
+    for line_number, raw in enumerate(text.replace("\r\n", "\n").replace("\r", "\n").split("\n"), 1):
+        line = raw.strip()
+        if not line:
+            continue
+        fields = tuple(value.strip() for value in line.split(","))
+        if len(fields) != 2 or any(not value.isascii() or not value.isdigit() for value in fields):
+            raise ValueError(f"line {line_number}: solution row must be period,room")
+        period, room = map(int, fields)
+        if period >= len(instance.periods) or room >= len(instance.rooms):
+            raise ValueError(f"line {line_number}: solution index outside instance")
+        periods.append(period)
+        rooms.append(room)
+    if len(periods) != len(instance.exams):
+        raise ValueError("solution row count differs from exam count")
+    return tuple(periods), tuple(rooms)
+
+
+def check_full_baseline(instance: ITCExamInstance, periods: tuple[int, ...],
+                        rooms: tuple[int, ...]) -> None:
+    """Check the published timetable's hard constraints before deriving repairs."""
+    if len(periods) != len(instance.exams) or len(rooms) != len(instance.exams):
+        raise ValueError("baseline geometry differs")
+    roster: dict[int, list[int]] = defaultdict(list)
+    for exam_index, exam in enumerate(instance.exams):
+        period = periods[exam_index]
+        room = rooms[exam_index]
+        if instance.periods[period].duration < exam.duration:
+            raise ValueError(f"baseline exam {exam_index} exceeds period duration")
+        for student in exam.students:
+            roster[student].append(exam_index)
+        if not 0 <= room < len(instance.rooms):
+            raise ValueError("baseline room outside bank")
+    for student, exams in roster.items():
+        seen: set[int] = set()
+        for exam in exams:
+            if periods[exam] in seen:
+                raise ValueError(f"baseline repeats student {student} in one period")
+            seen.add(periods[exam])
+    for left, right in instance.exclusions:
+        if periods[left] == periods[right]:
+            raise ValueError("baseline violates an EXCLUSION constraint")
+    for left, right in instance.coincidences:
+        if periods[left] != periods[right]:
+            raise ValueError("baseline violates an EXAM_COINCIDENCE constraint")
+    for later, earlier in instance.after:
+        if periods[later] <= periods[earlier]:
+            raise ValueError("baseline violates an AFTER constraint")
+    usage: dict[tuple[int, int], int] = defaultdict(int)
+    occupants: dict[tuple[int, int], list[int]] = defaultdict(list)
+    for exam_index, exam in enumerate(instance.exams):
+        key = (periods[exam_index], rooms[exam_index])
+        usage[key] += len(exam.students)
+        occupants[key].append(exam_index)
+    for (period, room), count in usage.items():
+        if count > instance.rooms[room].capacity:
+            raise ValueError(f"baseline room capacity exceeded at period {period}, room {room}")
+    for exam in instance.room_exclusive:
+        if len(occupants[periods[exam], rooms[exam]]) != 1:
+            raise ValueError("baseline violates ROOM_EXCLUSIVE")
+
+
+def protected_exams(instance: ITCExamInstance) -> set[int]:
+    protected = set(instance.room_exclusive)
+    for left, right in (*instance.after, *instance.coincidences):
+        protected.add(left)
+        protected.add(right)
+    return protected
+
+
+def derive_candidates(instance: ITCExamInstance, baseline_periods: tuple[int, ...],
+                      *, seed: int) -> tuple[tuple[tuple[int, ...], ...], tuple[int, ...]]:
+    """Choose one deterministic alternative while retaining the baseline witness."""
+    protected = protected_exams(instance)
+    rows: list[tuple[int, ...]] = []
+    flexible: list[int] = []
+    max_room = max((room.capacity for room in instance.rooms), default=0)
+    for exam_index, exam in enumerate(instance.exams):
+        base = baseline_periods[exam_index]
+        if exam_index in protected:
+            rows.append((base,))
+            continue
+        alternatives = [
+            period for period, record in enumerate(instance.periods)
+            if period != base and record.duration >= exam.duration
+            and len(exam.students) <= max_room
+        ]
+        if not alternatives:
+            rows.append((base,))
+            continue
+        # Hash-shuffled choice avoids concentrating all repairs in the earliest
+        # periods while remaining fixed by the frozen source/solution identity.
+        alternatives.sort(key=lambda period: hashlib.sha256(
+            f"{seed}/{exam_index}/{period}".encode("ascii")).digest())
+        alternate = alternatives[0]
+        rows.append((base, alternate))
+        flexible.append(exam_index)
+    return tuple(rows), tuple(flexible)
+
+
+def generate_queries(core, baseline_periods: tuple[int, ...], flexible: tuple[int, ...],
+                     *, target: int, seed: int) -> tuple[tuple[tuple[int, int], ...], ...]:
+    if type(target) is not int or not 1 <= target <= 100_000:
+        raise ValueError("query target outside [1, 100000]")
+    model = core.model
+    names = model.period_ids
+    exam_names = model.exam_ids
+    alternatives: dict[int, str] = {}
+    for exam in flexible:
+        base_name = names[baseline_periods[exam]]
+        alternatives[exam] = next(name for name in model.candidates[exam] if name != base_name)
+    queries: list[tuple[tuple[int, int], ...]] = []
+    seen: set[tuple[tuple[int, int], ...]] = set()
+
+    def add(query: tuple[tuple[int, int], ...]) -> None:
+        if query not in seen:
+            seen.add(query)
+            queries.append(query)
+
+    add(())
+    for period in names:
+        add(model.freeze_event(closed_periods=(period,)))
+    for exam in flexible:
+        add(model.freeze_event(fixed=((exam_names[exam], alternatives[exam]),)))
+    attempts = 0
+    while len(queries) < target and attempts < target * 200:
+        attempts += 1
+        if not flexible:
+            break
+        maximum = min(12, len(flexible))
+        width = (2 + deterministic_word(seed, attempts, 0) % (maximum - 1)
+                 if maximum >= 2 else 1)
+        ranked = sorted(
+            flexible,
+            key=lambda exam: (deterministic_word(seed, attempts, 1, exam), exam),
+        )
+        chosen = tuple(sorted(ranked[:width]))
+        fixed = tuple((exam_names[exam], alternatives[exam]) for exam in chosen)
+        if attempts % 3 == 0:
+            closed = (names[deterministic_word(seed, attempts, 2) % len(names)],)
+            query = model.freeze_event(closed_periods=closed, fixed=fixed)
+        else:
+            query = model.freeze_event(fixed=fixed)
+        add(query)
+    if len(queries) < target:
+        raise ValueError(f"could create only {len(queries)} distinct repair events")
+    return tuple(queries[:target])
+
+
+def _git(*args: str) -> str:
+    return subprocess.check_output(("git", *args), text=True).strip()
+
+
+def source_inventory() -> dict[str, str]:
+    root = Path(__file__).resolve().parents[2]
+    return {str(path): digest((root / path).read_bytes()) for path in SOURCE_PATHS}
+
+
+def validate_source_commit(commit: str, inventory: dict[str, str]) -> str:
+    if (type(commit) is not str or len(commit) != 40
+            or any(character not in "0123456789abcdef" for character in commit)):
+        raise ValueError("source_commit must be a full lowercase commit SHA")
+    resolved = _git("rev-parse", "--verify", commit + "^{commit}")
+    if resolved != commit:
+        raise ValueError("source_commit must identify the commit itself")
+    for path, expected in inventory.items():
+        try:
+            data = subprocess.check_output(("git", "show", f"{commit}:{path}"))
+        except subprocess.CalledProcessError as error:
+            raise ValueError(f"source commit does not contain {path}") from error
+        if digest(data) != expected:
+            raise ValueError(f"working source differs from source_commit at {path}")
+    return _git("rev-parse", commit + "^{tree}")
+
+
+def freeze(root: Path, instance_path: Path, solution_path: Path, *, query_count: int,
+           license_path: Path | None = None, source_commit: str | None = None,
+           dataset_instance_path: str = DEFAULT_INSTANCE,
+           dataset_solution_path: str = DEFAULT_SOLUTION,
+           evaluation_role: str = "development") -> dict:
+    root = root.resolve()
+    if root.exists():
+        raise FileExistsError(root)
+    if license_path is None:
+        raise ValueError("freeze requires the upstream redistribution license")
+    if (type(dataset_instance_path) is not str or not dataset_instance_path
+            or type(dataset_solution_path) is not str or not dataset_solution_path):
+        raise ValueError("dataset paths must be nonempty exact strings")
+    if evaluation_role not in ("development", "holdout", "replication"):
+        raise ValueError("evaluation_role must be development, holdout or replication")
+    instance_bytes = read_bound(instance_path)
+    solution_bytes = read_bound(solution_path)
+    license_bytes = read_bound(license_path)
+    instance = parse_itc2007_exam(instance_bytes.decode("utf-8"))
+    baseline_periods, baseline_rooms = parse_solution(solution_bytes.decode("utf-8"), instance)
+    check_full_baseline(instance, baseline_periods, baseline_rooms)
+    seed = int.from_bytes(hashlib.sha256(instance_bytes + b"\0" + solution_bytes).digest()[:8], "little")
+    if len(instance.periods) > 64:
+        raise ValueError("selected ITC instance exceeds the exact API's 64-period palette")
+    candidates, flexible = derive_candidates(instance, baseline_periods, seed=seed)
+    core = instance.conflict_core(candidates, acknowledge_ignored_constraints=True)
+    if not core.model.problem.check(bytes(baseline_periods)):
+        raise ValueError("published baseline does not witness the derived conflict core")
+    queries = generate_queries(core, baseline_periods, flexible, target=query_count, seed=seed)
+    packet = pack_validated_session_packet(core.model.problem, queries)
+    inventory = source_inventory()
+    commit = source_commit or _git("rev-parse", "HEAD")
+    source_tree: str | None = None
+    if source_commit is None:
+        changed = _git("status", "--porcelain", "--", *(str(path) for path in SOURCE_PATHS))
+        if changed:
+            raise ValueError("commit the transfer sources before freezing evidence")
+        source_tree = validate_source_commit(commit, inventory)
+    elif type(source_commit) is not str or not source_commit:
+        raise ValueError("source_commit fixture identity must be a nonempty string")
+
+    state = {
+            "baseline_periods": baseline_periods,
+            "baseline_rooms": baseline_rooms,
+            "candidate_periods": candidates,
+            "flexible_exams": flexible,
+            "protected_exams": tuple(sorted(protected_exams(instance))),
+            "seed": seed,
+    }
+    staging = root.with_name(root.name + ".partial")
+    if staging.exists():
+        raise FileExistsError(staging)
+    staging.mkdir(parents=True, exist_ok=False)
+    try:
+        artifacts = {
+            "inputs/instance.exam": write_once(staging / "inputs/instance.exam", instance_bytes),
+            "inputs/baseline.sol": write_once(staging / "inputs/baseline.sol", solution_bytes),
+            "inputs/UPSTREAM_LICENSE.txt": write_once(
+                staging / "inputs/UPSTREAM_LICENSE.txt", license_bytes
+            ),
+            "candidates.json": write_once(staging / "candidates.json", canonical_json(state)),
+            "queries.json": write_once(staging / "queries.json", canonical_json(queries)),
+            "session.sbc1": write_once(staging / "session.sbc1", packet),
+        }
+        manifest = {
+            "schema": SCHEMA,
+            "classification": (
+                "published ITC-2007 input and feasible schedule; deterministic bounded "
+                "conflict-core transfer, not full timetabling or outside-team replication"
+            ),
+            "source_commit": commit,
+            "source_tree": source_tree,
+            "source_inventory": inventory,
+            "dataset_repository": DATASET_REPOSITORY,
+            "dataset_commit": DATASET_COMMIT,
+            "dataset_instance_path": dataset_instance_path,
+            "dataset_solution_path": dataset_solution_path,
+            "evaluation_role": evaluation_role,
+            "instance_sha256": digest(instance_bytes),
+            "solution_sha256": digest(solution_bytes),
+            "upstream_acknowledgment": (
+                "This product includes software/data distributed by David Van Bulck, "
+                "Dries Goossens, and Andrea Schaerf, Ghent University, Belgium "
+                "(https://github.com/davidvanbulck/ITC-2007-CETT.git)."
+            ),
+            "upstream_license_sha256": digest(license_bytes),
+            "model": core.info,
+            "candidate_policy": (
+                "published baseline; freeze AFTER/coincidence/room-exclusive exams; "
+                "one hash-selected duration/individual-room-compatible alternative otherwise"
+            ),
+            "query_policy": (
+                "base, every single-period closure, every single-exam move, then seeded "
+                "multi-move and closure-plus-move events"
+            ),
+            "query_count": len(queries),
+            "modes": MODES,
+            "warmups": WARMUPS,
+            "repeats": REPEATS,
+            "timing_schedule": "one warmup per mode; measured modes rotate by repetition",
+            "timing_scope": (
+                "ready immutable SBC1 through construction, decisions, full evidence, "
+                "independent native check, SCQ2 materialization and disposal"
+            ),
+            "artifacts": artifacts,
+        }
+        manifest_bytes = canonical_json(manifest)
+        write_once(staging / "manifest.json", manifest_bytes)
+        write_once(staging / "manifest.sha256", (digest(manifest_bytes) + "\n").encode())
+        staging.rename(root)
+        return manifest
+    except BaseException:
+        shutil.rmtree(staging, ignore_errors=True)
+        raise
+
+
+def load_manifest(root: Path) -> dict:
+    expected = (root / "manifest.sha256").read_text().strip()
+    manifest_bytes = read_bound(root / "manifest.json", expected)
+    manifest = json.loads(manifest_bytes)
+    if (manifest.get("schema") != SCHEMA or tuple(manifest.get("modes", ())) != MODES
+            or type(manifest.get("warmups")) is not int or manifest["warmups"] < 0
+            or type(manifest.get("repeats")) is not int or manifest["repeats"] < 1
+            or type(manifest.get("query_count")) is not int or manifest["query_count"] < 1
+            or manifest.get("timing_schedule")
+            != "one warmup per mode; measured modes rotate by repetition"):
+        raise ValueError("transfer manifest schema, modes or schedule differ")
+    if (manifest.get("dataset_repository") != DATASET_REPOSITORY
+            or manifest.get("dataset_commit") != DATASET_COMMIT
+            or type(manifest.get("dataset_instance_path")) is not str
+            or not manifest["dataset_instance_path"]
+            or type(manifest.get("dataset_solution_path")) is not str
+            or not manifest["dataset_solution_path"]
+            or manifest.get("evaluation_role") not in
+            ("development", "holdout", "replication")):
+        raise ValueError("transfer dataset identity differs")
+    inventory = manifest.get("source_inventory")
+    if type(inventory) is not dict or inventory != source_inventory():
+        raise ValueError("transfer source inventory differs")
+    if manifest.get("source_tree") is not None:
+        if validate_source_commit(manifest.get("source_commit"), inventory) != manifest["source_tree"]:
+            raise ValueError("transfer source tree differs")
+    required = {"inputs/instance.exam", "inputs/baseline.sol", "candidates.json",
+                "queries.json", "session.sbc1", "inputs/UPSTREAM_LICENSE.txt"}
+    inventory = set(manifest.get("artifacts", ()))
+    if inventory != required:
+        raise ValueError("transfer manifest artifact inventory differs")
+    for name, checksum in manifest["artifacts"].items():
+        read_bound(root / name, checksum)
+    if manifest["artifacts"]["inputs/instance.exam"] != manifest.get("instance_sha256"):
+        raise ValueError("manifest instance identity differs")
+    if manifest["artifacts"]["inputs/baseline.sol"] != manifest.get("solution_sha256"):
+        raise ValueError("manifest solution identity differs")
+    if (manifest["artifacts"]["inputs/UPSTREAM_LICENSE.txt"]
+            != manifest.get("upstream_license_sha256")):
+        raise ValueError("manifest upstream-license identity differs")
+    return manifest
+
+
+def reconstruct(root: Path, manifest: dict):
+    instance_bytes = read_bound(
+        root / "inputs/instance.exam", manifest["artifacts"]["inputs/instance.exam"]
+    )
+    solution_bytes = read_bound(
+        root / "inputs/baseline.sol", manifest["artifacts"]["inputs/baseline.sol"]
+    )
+    instance = parse_itc2007_exam(instance_bytes.decode("utf-8"))
+    baseline_periods, baseline_rooms = parse_solution(solution_bytes.decode("utf-8"), instance)
+    check_full_baseline(instance, baseline_periods, baseline_rooms)
+    seed = int.from_bytes(
+        hashlib.sha256(instance_bytes + b"\0" + solution_bytes).digest()[:8], "little"
+    )
+    candidates, flexible = derive_candidates(instance, baseline_periods, seed=seed)
+    expected_state = {
+        "baseline_periods": baseline_periods,
+        "baseline_rooms": baseline_rooms,
+        "candidate_periods": candidates,
+        "flexible_exams": flexible,
+        "protected_exams": tuple(sorted(protected_exams(instance))),
+        "seed": seed,
+    }
+    state = json.loads(read_bound(
+        root / "candidates.json", manifest["artifacts"]["candidates.json"]
+    ))
+    if canonical_json(state) != canonical_json(expected_state):
+        raise ValueError("stored candidate policy differs from original inputs")
+    core = instance.conflict_core(candidates, acknowledge_ignored_constraints=True)
+    if core.info != manifest.get("model"):
+        raise ValueError("stored model receipt differs from reconstructed inputs")
+    queries = tuple(tuple(tuple(pin) for pin in row) for row in json.loads(read_bound(
+        root / "queries.json", manifest["artifacts"]["queries.json"]
+    )))
+    expected_queries = generate_queries(
+        core, baseline_periods, flexible,
+        target=manifest["query_count"], seed=seed,
+    )
+    if queries != expected_queries:
+        raise ValueError("stored query bank differs from deterministic policy")
+    packet = read_bound(root / "session.sbc1", manifest["artifacts"]["session.sbc1"])
+    if pack_validated_session_packet(core.model.problem, queries) != packet:
+        raise ValueError("stored packet differs from reconstructed original inputs")
+    if not core.model.problem.check(bytes(baseline_periods)):
+        raise ValueError("stored baseline no longer witnesses the reconstructed core")
+    return instance, state, core, queries
+
+
+def run(root: Path) -> dict:
+    root = root.resolve()
+    manifest = load_manifest(root)
+    reconstruct(root, manifest)
+    if (root / "run.json").exists() or (root / "outputs").exists() or (root / "native-build").exists():
+        raise FileExistsError("transfer execution already exists")
+    packet = read_bound(root / "session.sbc1", manifest["artifacts"]["session.sbc1"])
+    build = root / "native-build"
+    library = build_exact_runtime(build)
+    runtime = ExactRuntime(library)
+    states: dict[str, dict[str, object]] = {
+        mode: {"elapsed_ns": [], "output": None, "metadata": None} for mode in MODES
+    }
+    execution_order: list[list[str]] = []
+
+    def execute(mode: str, *, measured: bool) -> None:
+        start = time.perf_counter_ns()
+        result = runtime.solve_session_packet(packet, mode=mode)
+        elapsed = time.perf_counter_ns() - start
+        if not verify_session_output(packet, result.output):
+            raise AssertionError(f"{mode} output failed standard-library replay")
+        current = {
+            "query_count": result.query_count,
+            "sat_count": result.sat_count,
+            "unsat_count": result.unsat_count,
+            "proof_bytes": result.proof_bytes,
+            "visited": result.visited,
+            "arcs_scanned": result.arcs_scanned,
+            "selected_mode": result.info.get("selected_mode", mode),
+            "bank_query_count": result.info.get("bank_query_count"),
+            "bank_pin_count": result.info.get("bank_pin_count"),
+            "bank_kernel": result.info.get("bank_kernel"),
+            "bank_tiles": result.info.get("bank_tiles"),
+            "auto_policy": result.info.get("auto_policy"),
+            "auto_policy_reason": result.info.get("auto_policy_reason"),
+        }
+        if (current["query_count"] != manifest["query_count"]
+                or current["sat_count"] + current["unsat_count"] != manifest["query_count"]
+                or current["bank_query_count"] != manifest["query_count"]
+                or any(type(current[name]) is not int or current[name] < 0
+                       for name in ("proof_bytes", "visited", "arcs_scanned",
+                                    "bank_pin_count", "bank_tiles"))
+                or current["selected_mode"] not in ("raw", "scc", "route")
+                or current["bank_kernel"] not in ("serial", "tile64")
+                or current["auto_policy"] not in ("explicit", "structural-v2")
+                or type(current["auto_policy_reason"]) is not str
+                or (mode == "auto" and current["auto_policy"] != "structural-v2")
+                or (mode != "auto" and (current["selected_mode"] != mode
+                                         or current["auto_policy"] != "explicit"))):
+            raise AssertionError(f"{mode} returned inconsistent metadata")
+        state = states[mode]
+        if state["output"] is None:
+            state["output"], state["metadata"] = result.output, current
+        elif state["output"] != result.output or state["metadata"] != current:
+            raise AssertionError(f"{mode} result changed across identical calls")
+        if measured:
+            elapsed_rows = state["elapsed_ns"]
+            assert isinstance(elapsed_rows, list)
+            elapsed_rows.append(elapsed)
+
+    for _ in range(manifest["warmups"]):
+        for mode in MODES:
+            execute(mode, measured=False)
+    for repeat in range(manifest["repeats"]):
+        offset = repeat % len(MODES)
+        order = (*MODES[offset:], *MODES[:offset])
+        execution_order.append(list(order))
+        for mode in order:
+            execute(mode, measured=True)
+
+    records: dict[str, object] = {
+        "schema": SCHEMA + ".run",
+        "manifest_sha256": (root / "manifest.sha256").read_text().strip(),
+        "build_receipt_sha256": digest((build / "build.json").read_bytes()),
+        "warmups": manifest["warmups"],
+        "execution_order": execution_order,
+        "modes": {},
+    }
+    for mode in MODES:
+        state = states[mode]
+        output = state["output"]
+        metadata = state["metadata"]
+        elapsed = state["elapsed_ns"]
+        assert isinstance(output, bytes) and isinstance(metadata, dict) and isinstance(elapsed, list)
+        if len(elapsed) != manifest["repeats"]:
+            raise AssertionError(f"{mode} timing count differs")
+        output_name = f"outputs/{mode}.scq2"
+        output_hash = write_once(root / output_name, output)
+        records["modes"][mode] = {
+            **metadata,
+            "elapsed_ns": elapsed,
+            "output": output_name,
+            "output_sha256": output_hash,
+        }
+    run_bytes = canonical_json(records)
+    write_once(root / "run.json", run_bytes)
+    write_once(root / "run.sha256", (digest(run_bytes) + "\n").encode())
+    return records
+
+
+def audit(root: Path) -> dict:
+    root = root.resolve()
+    manifest = load_manifest(root)
+    run_hash = (root / "run.sha256").read_text().strip()
+    record = json.loads(read_bound(root / "run.json", run_hash))
+    if record.get("schema") != SCHEMA + ".run" or set(record.get("modes", ())) != set(MODES):
+        raise ValueError("run schema or mode inventory differs")
+    if record.get("manifest_sha256") != (root / "manifest.sha256").read_text().strip():
+        raise ValueError("run belongs to a different frozen manifest")
+    expected_order = []
+    for repeat in range(manifest["repeats"]):
+        offset = repeat % len(MODES)
+        expected_order.append(list((*MODES[offset:], *MODES[:offset])))
+    if (record.get("warmups") != manifest.get("warmups")
+            or record.get("execution_order") != expected_order):
+        raise ValueError("run warmup or rotated timing schedule differs")
+    packet = read_bound(root / "session.sbc1", manifest["artifacts"]["session.sbc1"])
+    instance, state, core, queries = reconstruct(root, manifest)
+    receipt = read_bound(root / "native-build/build.json")
+    if digest(receipt) != record.get("build_receipt_sha256"):
+        raise ValueError("native build receipt hash differs")
+    decisions: tuple[int, ...] | None = None
+    summaries = {}
+    sample = None
+    for mode in MODES:
+        item = record["modes"][mode]
+        if type(item) is not dict or item.get("output") != f"outputs/{mode}.scq2":
+            raise ValueError(f"{mode} output location differs")
+        elapsed = item.get("elapsed_ns")
+        if (type(elapsed) is not list or len(elapsed) != manifest["repeats"]
+                or any(type(value) is not int or value <= 0 for value in elapsed)):
+            raise ValueError(f"{mode} timing schedule differs")
+        output = read_bound(root / item["output"], item["output_sha256"])
+        if not verify_session_output(packet, output):
+            raise ValueError(f"{mode} output fails replay")
+        rows = decode_certified_output(output, core.model.problem.n, len(queries))
+        current = tuple(code for code, _, _ in rows)
+        sat_count = sum(current)
+        unsat_count = len(current) - sat_count
+        proof_bytes = sum(len(proof) for code, _labels, proof in rows if not code)
+        selected = item.get("selected_mode")
+        policy = item.get("auto_policy")
+        reason = item.get("auto_policy_reason")
+        if (item.get("query_count") != len(queries)
+                or item.get("bank_query_count") != len(queries)
+                or item.get("sat_count") != sat_count
+                or item.get("unsat_count") != unsat_count
+                or item.get("proof_bytes") != proof_bytes
+                or any(type(item.get(name)) is not int or item[name] < 0
+                       for name in ("visited", "arcs_scanned", "bank_pin_count", "bank_tiles"))
+                or item.get("bank_kernel") not in ("serial", "tile64")
+                or selected not in ("raw", "scc", "route")
+                or policy not in ("explicit", "structural-v2")
+                or type(reason) is not str
+                or (mode == "auto" and policy != "structural-v2")
+                or (mode != "auto" and (selected != mode or policy != "explicit"))):
+            raise ValueError(f"{mode} metadata differs from certified output")
+        if decisions is None:
+            decisions = current
+        elif decisions != current:
+            raise ValueError("engine modes disagree on SAT/UNSAT decisions")
+        kinds = Counter()
+        for index, (code, _labels, proof) in enumerate(rows):
+            if not code:
+                kind, _, _ = decode_proof(proof, core.model.problem.n)
+                kinds[str(kind)] += 1
+                if sample is None:
+                    sample = {
+                        "query_index": index,
+                        "mode": mode,
+                        "explanation": core.model.explain_unsat(queries[index], proof),
+                    }
+        summaries[mode] = {
+            "median_elapsed_ns": statistics.median(elapsed),
+            "minimum_elapsed_ns": min(elapsed),
+            "maximum_elapsed_ns": max(elapsed),
+            "sat": sat_count,
+            "unsat": unsat_count,
+            "proof_kinds": dict(kinds),
+            "selected_mode": selected,
+            "bank_query_count": item["bank_query_count"],
+            "bank_pin_count": item["bank_pin_count"],
+            "bank_kernel": item["bank_kernel"],
+            "bank_tiles": item["bank_tiles"],
+            "auto_policy": policy,
+            "auto_policy_reason": reason,
+            "output_sha256": item["output_sha256"],
+        }
+    result = {
+        "schema": SCHEMA + ".audit",
+        "functional_complete": decisions is not None and len(decisions) == manifest["query_count"],
+        "source_commit": manifest["source_commit"],
+        "dataset_commit": manifest["dataset_commit"],
+        "dataset_instance_path": manifest["dataset_instance_path"],
+        "dataset_solution_path": manifest["dataset_solution_path"],
+        "evaluation_role": manifest["evaluation_role"],
+        "instance_sha256": manifest["instance_sha256"],
+        "solution_sha256": manifest["solution_sha256"],
+        "model": core.info,
+        "baseline_valid_full_itc_hard_constraints": True,
+        "flexible_exams": len(state["flexible_exams"]),
+        "protected_exams": len(state["protected_exams"]),
+        "queries": len(queries),
+        "warmups": manifest["warmups"],
+        "repeats": manifest["repeats"],
+        "timing_schedule": manifest["timing_schedule"],
+        "modes": summaries,
+        "auto_vs_raw_median_ratio": (
+            summaries["auto"]["median_elapsed_ns"] / summaries["raw"]["median_elapsed_ns"]
+        ),
+        "sample_unsat": sample,
+        "limitations": [
+            "conflict-core bounded repair, not complete ITC-2007 timetabling",
+            "room reassignment and soft objective are not solved",
+            "project-owned execution, not outside-team replication",
+            "ready-packet timing excludes candidate generation, packing and compilation",
+        ],
+    }
+    audit_bytes = canonical_json(result)
+    audit_path = root / "audit.json"
+    if audit_path.exists():
+        expected = (root / "audit.sha256").read_text().strip()
+        existing = read_bound(audit_path, expected)
+        if existing != audit_bytes:
+            raise ValueError("existing audit differs from independent replay")
+    else:
+        write_once(audit_path, audit_bytes)
+        write_once(root / "audit.sha256", (digest(audit_bytes) + "\n").encode())
+    return result
+
+
+def main(argv: Iterable[str] | None = None) -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("command", choices=("freeze", "run", "audit"))
+    parser.add_argument("--root", type=Path, required=True)
+    parser.add_argument("--instance", type=Path)
+    parser.add_argument("--solution", type=Path)
+    parser.add_argument("--queries", type=int, default=1024)
+    parser.add_argument("--license", type=Path)
+    parser.add_argument("--dataset-instance-path", default=DEFAULT_INSTANCE)
+    parser.add_argument("--dataset-solution-path", default=DEFAULT_SOLUTION)
+    parser.add_argument(
+        "--evaluation-role", choices=("development", "holdout", "replication"),
+        default="development",
+    )
+    args = parser.parse_args(argv)
+    if args.command == "freeze":
+        if args.instance is None or args.solution is None:
+            parser.error("freeze requires --instance and --solution")
+        value = freeze(
+            args.root, args.instance, args.solution, query_count=args.queries,
+            license_path=args.license,
+            dataset_instance_path=args.dataset_instance_path,
+            dataset_solution_path=args.dataset_solution_path,
+            evaluation_role=args.evaluation_role,
+        )
+    elif args.command == "run":
+        value = run(args.root)
+    else:
+        value = audit(args.root)
+    print(json.dumps(value, indent=2, sort_keys=True, allow_nan=False))
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/experiments/compiled_query/test_exact.py b/experiments/compiled_query/test_exact.py
index c7042e0..4cc8d54 100644
--- a/experiments/compiled_query/test_exact.py
+++ b/experiments/compiled_query/test_exact.py
@@ -450,7 +450,10 @@ def test_tiled_witnesses_keep_high_colours_and_last_lane(exact_runtime, mode):
                           if index % 3 and rng.random() < .3))
     rows = tuple(rows)
     expected = []
-    with exact_runtime.prepare(p, mode='scc' if mode == 'auto' else mode) as session:
+    # The bank averages well above eight pins per request, so structural-v2
+    # intentionally selects raw rather than paying for tile64 propagation.
+    expected_mode = 'raw' if mode == 'auto' else mode
+    with exact_runtime.prepare(p, mode=expected_mode) as session:
         for query in rows:
             answer = session.solve(query)
             assert answer.status == 'SAT_VERIFIED'
@@ -459,9 +462,29 @@ def test_tiled_witnesses_keep_high_colours_and_last_lane(exact_runtime, mode):
     for count in (64, 65, 129):
         answer = exact_runtime.solve_session_packet(pack_session_packet(p, rows[:count]), mode=mode)
         assert answer.sat_count == count and answer.proof_bytes == 0
+        if mode == 'auto':
+            assert answer.info['selected_mode'] == 'raw'
+            assert answer.info['auto_policy'] == 'structural-v2'
+            assert answer.info['auto_policy_reason'] == 'high_pin_density'
         assert decode_certified_output(answer.output, n, count) == tuple(expected[:count])
 
 
+def test_auto_structural_v2_freezes_the_eight_pin_density_boundary(exact_runtime):
+    p = BinaryListProblem.create(16, 2, (), (3,) * 16)
+    low_rows = tuple(tuple((v, 3) for v in range(7)) for _ in range(64))
+    high_rows = tuple(tuple((v, 3) for v in range(8)) for _ in range(64))
+    low = exact_runtime.solve_session_packet(pack_session_packet(p, low_rows), mode='auto')
+    high = exact_runtime.solve_session_packet(pack_session_packet(p, high_rows), mode='auto')
+    assert low.info['selected_mode'] == 'scc'
+    assert low.info['auto_policy_reason'] == 'scc_tile64'
+    assert low.info['bank_query_count'] == 64 and low.info['bank_pin_count'] == 448
+    assert high.info['selected_mode'] == 'raw'
+    assert high.info['auto_policy_reason'] == 'high_pin_density'
+    assert high.info['bank_query_count'] == 64 and high.info['bank_pin_count'] == 512
+    assert high.output == exact_runtime.solve_session_packet(
+        pack_session_packet(p, high_rows), mode='raw').output
+
+
 @pytest.mark.parametrize('count', (0, 1, 63, 64, 65, 129))
 @pytest.mark.parametrize('structure', ('empty', 'independent', 'chain', 'odd_cycle'))
 def test_auto_packet_exposes_and_matches_its_selected_engine(exact_runtime, count, structure):
@@ -477,6 +500,8 @@ def test_auto_packet_exposes_and_matches_its_selected_engine(exact_runtime, coun
     selected = ('raw' if count < 64 or structure == 'odd_cycle' else
                 'route' if structure in ('chain', 'empty') else 'scc')
     assert answer.info['mode'] == 'auto'
+    assert answer.info['auto_policy'] == 'structural-v2'
+    assert answer.info['bank_query_count'] == count
     assert answer.info['selected_mode'] == selected
     assert answer.output == exact_runtime.solve_session_packet(packet, mode=selected).output
     for query, (decision, labels, certificate) in zip(rows, decode_certified_output(answer.output, n, count)):
diff --git a/experiments/compiled_query/test_itc_exam_suite.py b/experiments/compiled_query/test_itc_exam_suite.py
new file mode 100644
index 0000000..79b00aa
--- /dev/null
+++ b/experiments/compiled_query/test_itc_exam_suite.py
@@ -0,0 +1,76 @@
+"""Contracts for the prospective multi-instance ITC transfer suite."""
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+import pytest
+
+from experiments.compiled_query import itc_exam_suite as suite
+from experiments.compiled_query import itc_exam_transfer as transfer
+from experiments.compiled_query.test_itc_exam_transfer import INSTANCE, SOLUTION
+
+
+def fixture_cases(upstream: Path) -> tuple[suite.SuiteCase, ...]:
+    records = (
+        suite.SuiteCase("fixture-development", "development", "instances/dev.exam", "solutions/dev.sol"),
+        suite.SuiteCase("fixture-holdout-a", "holdout", "instances/a.exam", "solutions/a.sol"),
+        suite.SuiteCase("fixture-holdout-b", "holdout", "instances/b.exam", "solutions/b.sol"),
+    )
+    (upstream / "instances").mkdir(parents=True)
+    (upstream / "solutions").mkdir(parents=True)
+    for case in records:
+        (upstream / case.instance_path).write_text(INSTANCE)
+        (upstream / case.solution_path).write_text(SOLUTION)
+    (upstream / "LICENSE").write_text("fixture redistribution license\n")
+    return records
+
+
+def test_case_validation_rejects_escape_and_missing_holdout():
+    with pytest.raises(ValueError, match="under upstream root"):
+        suite.SuiteCase("bad", "holdout", "../bad.exam", "x.sol").validate()
+    with pytest.raises(ValueError, match="holdout"):
+        suite._cases((suite.SuiteCase("dev", "development", "x.exam", "x.sol"),))
+
+
+def test_suite_freeze_run_and_audit(tmp_path, monkeypatch):
+    upstream = tmp_path / "upstream"
+    cases = fixture_cases(upstream)
+    root = tmp_path / "suite"
+    monkeypatch.setattr(transfer, "WARMUPS", 0)
+    monkeypatch.setattr(transfer, "REPEATS", 1)
+    manifest = suite.freeze(
+        root, upstream, cases=cases, query_count=8,
+        holdout_max_auto_raw_ratio=None,
+        source_commit="fixture-source",
+    )
+    assert [item["role"] for item in manifest["cases"]] == [
+        "development", "holdout", "holdout"
+    ]
+    assert manifest["holdout_max_auto_raw_ratio"] is None
+    run = suite.run(root)
+    assert len(run["cases"]) == 3
+    result = suite.audit(root)
+    assert result["functional_complete"]
+    assert result["holdout_gate_pass"]
+    assert result["holdout_failures"] == []
+    assert all(case["queries"] == 8 for case in result["cases"])
+    assert all(case["auto_vs_raw_paired_median_ratio"] > 0 for case in result["cases"])
+    assert json.loads((root / "suite-audit.json").read_text()) == result
+
+
+def test_suite_detects_child_manifest_substitution(tmp_path, monkeypatch):
+    upstream = tmp_path / "upstream"
+    cases = fixture_cases(upstream)
+    root = tmp_path / "suite"
+    monkeypatch.setattr(transfer, "WARMUPS", 0)
+    monkeypatch.setattr(transfer, "REPEATS", 1)
+    suite.freeze(
+        root, upstream, cases=cases, query_count=8,
+        holdout_max_auto_raw_ratio=None,
+        source_commit="fixture-source",
+    )
+    child = root / "cases" / cases[1].name / "manifest.sha256"
+    child.write_text("0" * 64 + "\n")
+    with pytest.raises(ValueError, match="manifest hash differs"):
+        suite.load_manifest(root)
diff --git a/experiments/compiled_query/test_itc_exam_transfer.py b/experiments/compiled_query/test_itc_exam_transfer.py
new file mode 100644
index 0000000..8795156
--- /dev/null
+++ b/experiments/compiled_query/test_itc_exam_transfer.py
@@ -0,0 +1,179 @@
+"""End-to-end contracts for the ITC exam bounded-repair transfer harness."""
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+import pytest
+
+from experiments.compiled_query import itc_exam_transfer as transfer
+from spectra.itc2007_exam import parse_itc2007_exam
+
+
+INSTANCE = """[Exams:5]
+120, 1, 2
+180, 2, 3
+120, 9
+120, 4
+120, 4
+[Periods:4]
+01:01:2030, 09:00:00, 180, 0
+01:01:2030, 14:00:00, 120, 5
+02:01:2030, 09:00:00, 240, 0
+02:01:2030, 14:00:00, 180, 0
+[Rooms:1]
+100, 0
+[PeriodHardConstraints]
+0, EXCLUSION, 2
+1, AFTER, 0
+[RoomHardConstraints]
+2, ROOM_EXCLUSIVE
+[InstitutionalWeightings]
+TWOINAROW, 30
+"""
+SOLUTION = """0, 0
+2, 0
+1, 0
+0, 0
+3, 0
+"""
+
+
+def write_inputs(tmp_path: Path) -> tuple[Path, Path, Path]:
+    instance = tmp_path / "fixture.exam"
+    solution = tmp_path / "fixture.sol"
+    instance.write_text(INSTANCE)
+    solution.write_text(SOLUTION)
+    license_path = tmp_path / "LICENSE"
+    license_path.write_text("fixture redistribution license\n")
+    return instance, solution, license_path
+
+
+def test_solution_parser_and_full_hard_constraint_check():
+    instance = parse_itc2007_exam(INSTANCE)
+    periods, rooms = transfer.parse_solution(SOLUTION, instance)
+    transfer.check_full_baseline(instance, periods, rooms)
+    assert periods == (0, 2, 1, 0, 3)
+    assert transfer.protected_exams(instance) == {0, 1, 2}
+    with pytest.raises(ValueError, match="AFTER"):
+        transfer.check_full_baseline(instance, (2, 0, 1, 0, 3), rooms)
+
+
+def test_candidate_policy_retains_baseline_and_flexes_unprotected_exams():
+    instance = parse_itc2007_exam(INSTANCE)
+    periods, _ = transfer.parse_solution(SOLUTION, instance)
+    candidates, flexible = transfer.derive_candidates(instance, periods, seed=7)
+    assert candidates[0] == (0,) and candidates[1] == (2,) and candidates[2] == (1,)
+    assert flexible == (3, 4)
+    assert all(periods[index] in candidates[index] for index in range(5))
+
+
+def test_freeze_run_and_independent_audit(tmp_path, monkeypatch):
+    instance, solution, license_path = write_inputs(tmp_path)
+    root = tmp_path / "evidence"
+    monkeypatch.setattr(transfer, "WARMUPS", 0)
+    monkeypatch.setattr(transfer, "REPEATS", 2)
+    manifest = transfer.freeze(
+        root, instance, solution, query_count=8, license_path=license_path,
+        source_commit="fixture-source",
+        dataset_instance_path="fixtures/custom.exam",
+        dataset_solution_path="fixtures/custom.sol",
+        evaluation_role="holdout",
+    )
+    assert manifest["query_count"] == 8
+    assert manifest["model"]["exams"] == 5
+    assert manifest["dataset_instance_path"] == "fixtures/custom.exam"
+    assert manifest["dataset_solution_path"] == "fixtures/custom.sol"
+    assert manifest["evaluation_role"] == "holdout"
+    assert manifest["warmups"] == 0 and manifest["repeats"] == 2
+    record = transfer.run(root)
+    assert tuple(record["modes"]) == transfer.MODES
+    assert record["warmups"] == 0
+    assert record["execution_order"] == [
+        ["auto", "raw", "scc", "route"],
+        ["raw", "scc", "route", "auto"],
+    ]
+    result = transfer.audit(root)
+    assert result["functional_complete"]
+    assert result["baseline_valid_full_itc_hard_constraints"]
+    assert result["flexible_exams"] == 2
+    assert result["protected_exams"] == 3
+    assert result["queries"] == 8
+    assert result["evaluation_role"] == "holdout"
+    assert result["dataset_instance_path"] == "fixtures/custom.exam"
+    assert result["warmups"] == 0 and result["repeats"] == 2
+    assert result["auto_vs_raw_median_ratio"] > 0
+    assert all(item["sat"] + item["unsat"] == 8 for item in result["modes"].values())
+    assert result["modes"]["auto"]["auto_policy"] == "structural-v2"
+    assert result["modes"]["auto"]["bank_query_count"] == 8
+    assert result["modes"]["raw"]["auto_policy"] == "explicit"
+    assert json.loads((root / "audit.json").read_text()) == result
+
+
+
+def test_manifest_rejects_invalid_role_and_schedule(tmp_path):
+    instance, solution, license_path = write_inputs(tmp_path)
+    with pytest.raises(ValueError, match="evaluation_role"):
+        transfer.freeze(
+            tmp_path / "bad-role", instance, solution, query_count=8,
+            license_path=license_path, source_commit="fixture-source",
+            evaluation_role="training",
+        )
+
+    root = tmp_path / "evidence"
+    transfer.freeze(
+        root, instance, solution, query_count=8, license_path=license_path,
+        source_commit="fixture-source",
+    )
+    manifest_path = root / "manifest.json"
+    manifest = json.loads(manifest_path.read_text())
+    manifest["timing_schedule"] = "mode-by-mode"
+    payload = transfer.canonical_json(manifest)
+    manifest_path.write_bytes(payload)
+    (root / "manifest.sha256").write_text(transfer.digest(payload) + "\n")
+    with pytest.raises(ValueError, match="schedule differ"):
+        transfer.load_manifest(root)
+
+def test_frozen_artifact_hash_mutation_is_rejected(tmp_path):
+    instance, solution, license_path = write_inputs(tmp_path)
+    root = tmp_path / "evidence"
+    transfer.freeze(root, instance, solution, query_count=8, license_path=license_path,
+                    source_commit="fixture-source")
+    packet = root / "session.sbc1"
+    packet.write_bytes(packet.read_bytes() + b"x")
+    with pytest.raises(ValueError, match="hash differs"):
+        transfer.load_manifest(root)
+
+
+def test_audit_rederives_candidates_and_queries_instead_of_trusting_json(tmp_path):
+    instance, solution, license_path = write_inputs(tmp_path)
+    root = tmp_path / "evidence"
+    transfer.freeze(root, instance, solution, query_count=8, license_path=license_path,
+                    source_commit="fixture-source")
+
+    state_path = root / "candidates.json"
+    state = json.loads(state_path.read_text())
+    state["seed"] += 1
+    state_bytes = transfer.canonical_json(state)
+    state_path.write_bytes(state_bytes)
+    manifest_path = root / "manifest.json"
+    manifest = json.loads(manifest_path.read_text())
+    manifest["artifacts"]["candidates.json"] = transfer.digest(state_bytes)
+    manifest_bytes = transfer.canonical_json(manifest)
+    manifest_path.write_bytes(manifest_bytes)
+    (root / "manifest.sha256").write_text(transfer.digest(manifest_bytes) + "\n")
+
+    loaded = transfer.load_manifest(root)
+    with pytest.raises(ValueError, match="candidate policy differs"):
+        transfer.reconstruct(root, loaded)
+
+
+def test_failed_freeze_does_not_publish_a_partial_evidence_root(tmp_path):
+    instance, solution, license_path = write_inputs(tmp_path)
+    root = tmp_path / "evidence"
+    solution.write_text("0, 0\n")
+    with pytest.raises(ValueError, match="row count differs"):
+        transfer.freeze(root, instance, solution, query_count=8, license_path=license_path,
+                        source_commit="fixture-source")
+    assert not root.exists()
+    assert not root.with_name(root.name + ".partial").exists()
diff --git a/scripts/check_exam_repair_installation.py b/scripts/check_exam_repair_installation.py
new file mode 100644
index 0000000..f523897
--- /dev/null
+++ b/scripts/check_exam_repair_installation.py
@@ -0,0 +1,89 @@
+"""Exercise the installed exam-repair API without repository imports."""
+from __future__ import annotations
+
+import argparse
+import json
+from pathlib import Path
+import tempfile
+
+from spectra.exam_repair import ExamRepairModel
+from spectra.exact import ExactRuntime, build_exact_runtime, verify_answer
+from spectra.itc2007_exam import parse_itc2007_exam
+
+FIXTURE = """[Exams:3]
+120, 7
+120, 7, 9
+120, 9
+[Periods:2]
+01:01:2030, 09:00:00, 180, 0
+01:01:2030, 14:00:00, 180, 0
+[Rooms:1]
+100, 0
+[PeriodHardConstraints]
+[RoomHardConstraints]
+[InstitutionalWeightings]
+TWOINAROW, 1
+"""
+
+
+def check() -> dict:
+    instance = parse_itc2007_exam(FIXTURE)
+    core = instance.conflict_core(((0, 1), (0, 1), (0, 1)),
+                                  acknowledge_ignored_constraints=True)
+    model: ExamRepairModel = core.model
+    sat_query = model.freeze_event(fixed=(("exam-0000", model.period_ids[0]),))
+    unsat_query = model.freeze_event(fixed=(
+        ("exam-0000", model.period_ids[0]),
+        ("exam-0001", model.period_ids[0]),
+    ))
+    with tempfile.TemporaryDirectory(prefix="spectra-exam-install-") as directory:
+        library = build_exact_runtime(Path(directory) / "native")
+        runtime = ExactRuntime(library)
+        with runtime.prepare(model.problem, mode="scc") as session:
+            sat = session.solve(sat_query)
+            unsat = session.solve(unsat_query)
+            if sat.status != "SAT_VERIFIED" or unsat.status != "UNSAT_VERIFIED":
+                raise AssertionError("installed exact runtime returned unexpected status")
+            if not verify_answer(model.problem, sat_query, sat.status, sat.labels, sat.proof):
+                raise AssertionError("installed SAT result failed replay")
+            if not verify_answer(
+                    model.problem, unsat_query, unsat.status, unsat.labels, unsat.proof):
+                raise AssertionError("installed UNSAT result failed replay")
+            assignment = model.assignment_from_labels(sat.labels, sat_query)
+            explanation = model.explain_unsat(unsat_query, unsat.proof)
+            info = session.info
+    if tuple(exam for exam, _period in assignment) != model.exam_ids:
+        raise AssertionError("installed repair omitted or reordered exams")
+    if not model.check_assignment(assignment, sat_query):
+        raise AssertionError("installed named assignment failed adapter replay")
+    if explanation.get("kind") not in ("EMPTY_DOMAIN", "FORCED_OPPOSITES"):
+        raise AssertionError("installed repair returned an unknown proof explanation")
+    return {
+        "schema": "spectra.exam_repair.installation_check.v1",
+        "exams": model.problem.n,
+        "periods": model.problem.palette,
+        "conflicts": len(model.problem.edges),
+        "sat_status": sat.status,
+        "unsat_status": unsat.status,
+        "proof_kind": explanation["kind"],
+        "assignment_complete": True,
+        "native_abi": info["native_abi"],
+        "exact_api_version": info["exact_api_version"],
+    }
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--out", type=Path)
+    args = parser.parse_args()
+    result = check()
+    payload = json.dumps(result, indent=2, sort_keys=True) + "\n"
+    if args.out is not None:
+        args.out.parent.mkdir(parents=True, exist_ok=True)
+        args.out.write_text(payload, encoding="utf-8")
+    print(payload, end="")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())
diff --git a/spectra/_native/exact_binary.cpp b/spectra/_native/exact_binary.cpp
index 96182ee..96694d3 100644
--- a/spectra/_native/exact_binary.cpp
+++ b/spectra/_native/exact_binary.cpp
@@ -341,12 +341,13 @@ struct ObserverSessionPacket {
 struct Answer {bool sat=true;std::vector<uint8_t> labels,certificate;W visits=0,arcs=0;};
 struct Index {
     U n,k,nc=0,bad=NIL;bool condensed,sparse,routed;W cap;W full_arcs=0;
+    const char* auto_reason="explicit_mode";
     std::vector<W> masks;
     std::vector<uint8_t> lo,hi,base;
     std::vector<U> comp,opposite;
     std::vector<U> inward,outward,representative,cross_from,cross_to;
     CSR original,dag;
-    Index(Input p,U mode,U bank_size=0):n(p.n),k(p.k),condensed(mode!=0),sparse(mode==2),routed(mode==3),cap(p.cap),masks(std::move(p.masks)),lo(n),hi(n,255),base(n) {
+    Index(Input p,U mode,U bank_size=0,U bank_pins=0):n(p.n),k(p.k),condensed(mode!=0),sparse(mode==2),routed(mode==3),cap(p.cap),masks(std::move(p.masks)),lo(n),hi(n,255),base(n) {
         std::vector<std::pair<U,U>> arcs;arcs.reserve(4*p.edges.size()+n);
         for(U v=0;v<n;++v) {lo[v]=first(masks[v]);W h=masks[v] & (masks[v]-1);
             if(h) hi[v]=first(h); else arcs.emplace_back(2*v+1,2*v);
@@ -382,11 +383,22 @@ struct Index {
             base[v]=static_cast<uint8_t>(comp[2*v]<comp[2*v+1]);
         }
         if(mode==4) {
-            // Opt-in, query-blind policy. Small banks cannot use the lane kernel;
-            // base-infeasible graphs do not need a request-propagation index.
-            // Route proofs only when SCCs at least halve the literal graph.
-            condensed=bank_size>=64 && bad==NIL;
-            routed=condensed && nc<=n;
+            // Structural auto policy v2.  set9 development evidence showed that
+            // large, heavily pinned repair banks can lose to raw traversal even
+            // when tile64 is available.  The policy therefore uses only immutable
+            // request geometry (never answers or measured timings): raw below one
+            // tile, on a base contradiction, or at eight pins per row and above.
+            // The threshold is frozen before the set10/set12 holdout transfer.
+            if(bank_size<64) {
+                condensed=false;routed=false;auto_reason="bank_below_tile64";
+            } else if(bad!=NIL) {
+                condensed=false;routed=false;auto_reason="base_infeasible";
+            } else if(W(bank_pins)>=8*W(bank_size)) {
+                condensed=false;routed=false;auto_reason="high_pin_density";
+            } else {
+                condensed=true;routed=nc<=n;
+                auto_reason=routed?"route_component_compression":"scc_tile64";
+            }
         }
         if(condensed && bad==NIL) {
             opposite.assign(nc,NIL);
@@ -990,6 +1002,7 @@ PyObject* solve_session_packet(PyObject*,PyObject* args) {
         const auto* data=reinterpret_cast<const uint8_t*>(PyBytes_AS_STRING(packet));
         size_t size=static_cast<size_t>(PyBytes_GET_SIZE(packet));
         std::unique_ptr<Index> solver;std::unique_ptr<Observer> observer;BankResult result;
+        U bank_queries=0,bank_pins=0;
         {
             Released guard;
             ObserverSessionPacket observer_rows(data,size,limit);
@@ -997,8 +1010,9 @@ PyObject* solve_session_packet(PyObject*,PyObject* args) {
             W observer_bytes=observer->bytes();
             if(observer_bytes>=limit) throw Resource("SBC1 observer consumes session cap");
             SolverSessionPacket solver_rows(data,size,limit-observer_bytes);
+            bank_queries=solver_rows.query_count();bank_pins=solver_rows.pin_count();
             solver=std::make_unique<Index>(solver_rows.take_input(limit-observer_bytes),
-                                           static_cast<U>(which),solver_rows.query_count());
+                                           static_cast<U>(which),bank_queries,bank_pins);
             result=run_rows(*solver,*observer,solver_rows,observer_rows,size,limit,"SBC1");
         }
         PyObject* output=PyBytes_FromStringAndSize(reinterpret_cast<const char*>(result.output.data()),
@@ -1018,11 +1032,22 @@ PyObject* solve_session_packet(PyObject*,PyObject* args) {
         if(!metadata){Py_DECREF(output);return nullptr;}
         PyObject* kernel=PyUnicode_FromString(result.tiled?"tile64":"serial");
         PyObject* tiles=PyLong_FromUnsignedLong(result.tiles);
-        if(!kernel || !tiles || PyDict_SetItemString(metadata,"bank_kernel",kernel)<0
-                || PyDict_SetItemString(metadata,"bank_tiles",tiles)<0) {
-            Py_XDECREF(kernel);Py_XDECREF(tiles);Py_DECREF(output);Py_DECREF(metadata);return nullptr;
+        PyObject* queries=PyLong_FromUnsignedLong(bank_queries);
+        PyObject* pins=PyLong_FromUnsignedLong(bank_pins);
+        PyObject* policy=PyUnicode_FromString(which==4?"structural-v2":"explicit");
+        PyObject* reason=PyUnicode_FromString(solver->auto_reason);
+        if(!kernel || !tiles || !queries || !pins || !policy || !reason
+                || PyDict_SetItemString(metadata,"bank_kernel",kernel)<0
+                || PyDict_SetItemString(metadata,"bank_tiles",tiles)<0
+                || PyDict_SetItemString(metadata,"bank_query_count",queries)<0
+                || PyDict_SetItemString(metadata,"bank_pin_count",pins)<0
+                || PyDict_SetItemString(metadata,"auto_policy",policy)<0
+                || PyDict_SetItemString(metadata,"auto_policy_reason",reason)<0) {
+            Py_XDECREF(kernel);Py_XDECREF(tiles);Py_XDECREF(queries);Py_XDECREF(pins);
+            Py_XDECREF(policy);Py_XDECREF(reason);Py_DECREF(output);Py_DECREF(metadata);return nullptr;
         }
-        Py_DECREF(kernel);Py_DECREF(tiles);
+        Py_DECREF(kernel);Py_DECREF(tiles);Py_DECREF(queries);Py_DECREF(pins);
+        Py_DECREF(policy);Py_DECREF(reason);
         return Py_BuildValue("NKKKKKKN",output,
             static_cast<unsigned long long>(result.sat+result.unsat),
             static_cast<unsigned long long>(result.sat),
diff --git a/spectra/exact.py b/spectra/exact.py
index 983d335..60ed10e 100644
--- a/spectra/exact.py
+++ b/spectra/exact.py
@@ -402,11 +402,13 @@ def solve_session_packet(self, packet: bytes, *, mode: str = 'route',
                              max_bytes: int = DEFAULT_BYTES) -> ExactSessionPacketAnswer:
         """Construct, solve, verify, serialize and dispose one prepared SBC1 session.
 
-        Opt-in ``auto`` uses raw traversal below 64 requests or on a base-infeasible
-        graph. Otherwise it uses routed proofs when SCCs at least halve the
-        literal graph, and plain SCC batching when they do not. Selection uses
-        fixed-graph structure and bank length, never measured answers or timings.
-        ``info['selected_mode']`` names the actual engine. Explicit modes and the
+        Opt-in ``auto`` uses structural policy v2. It selects raw traversal below
+        64 requests, on a base-infeasible graph, or when the immutable bank carries
+        at least eight pins per request on average. Otherwise it uses routed proofs
+        when SCCs at least halve the literal graph, and plain SCC batching when they
+        do not. Selection uses only admitted graph/request geometry, never answers
+        or measured timings. ``info['selected_mode']`` and
+        ``info['auto_policy_reason']`` expose the decision. Explicit modes and the
         default retain their meaning; this policy is an experimental heuristic.
         """
         if type(packet) is not bytes:
diff --git a/spectra/exam_repair.py b/spectra/exam_repair.py
new file mode 100644
index 0000000..121ae9e
--- /dev/null
+++ b/spectra/exam_repair.py
@@ -0,0 +1,418 @@
+"""Proof-carrying bounded exam rescheduling on top of :mod:`spectra.exact`.
+
+This adapter models a deliberately narrow operational problem: every exam has one
+or two pre-approved candidate periods, and exams that share a student (or an
+explicit exclusion) may not use the same period.  Period closures and exam pins
+become current-query restrictions.  The exact backend can then return either a
+complete checked repair or a portable contradiction proof.
+
+Room capacities, room exclusivity, precedence, coincidence, soft penalties and
+candidate generation are outside this model.  They must be handled before
+admission or checked separately; the module never silently drops them.
+"""
+from __future__ import annotations
+
+from dataclasses import dataclass
+from typing import Iterable, Mapping, Sequence
+
+from .exact_problem import ExactProblem
+from .exact_proof import decode_proof, verify_answer
+
+MAX_IDENTIFIER_CHARS = 512
+MAX_ENROLLMENTS = 4_000_000
+MAX_CONFLICTS = 2_000_000
+MAX_PAIR_EMISSIONS = 8_000_000
+
+
+@dataclass(frozen=True, slots=True)
+class ExamConflictReason:
+    """One deterministic witness for an admitted conflict edge."""
+
+    kind: str
+    student: int | None = None
+
+    def __post_init__(self) -> None:
+        if self.kind not in ("shared_student", "explicit_exclusion"):
+            raise ValueError("unsupported exam conflict reason")
+        if self.kind == "shared_student":
+            if type(self.student) is not int or not 0 <= self.student < 1 << 63:
+                raise ValueError("shared-student reason needs a bounded student id")
+        elif self.student is not None:
+            raise ValueError("explicit exclusion must not carry a student id")
+
+
+@dataclass(frozen=True, slots=True)
+class ExamRepairModel:
+    """A checked reduction from bounded exam repair to exact list colouring.
+
+    ``problem`` uses exam positions as vertices and period positions as colours.
+    ``conflict_reasons`` is aligned with ``problem.edges`` and is explanatory
+    metadata only; the exact solver and verifier receive the original graph.
+    """
+
+    problem: ExactProblem
+    exam_ids: tuple[str, ...]
+    period_ids: tuple[str, ...]
+    students: tuple[tuple[int, ...], ...]
+    conflict_reasons: tuple[ExamConflictReason, ...]
+
+    @classmethod
+    def create(
+        cls,
+        period_ids: Iterable[str],
+        exam_ids: Iterable[str],
+        students: Iterable[Iterable[int]],
+        candidate_periods: Iterable[Iterable[str]],
+        *,
+        exclusions: Iterable[Sequence[str]] = (),
+    ) -> "ExamRepairModel":
+        periods = cls._bounded_snapshot(period_ids, 64, "period identifier")
+        exams = cls._bounded_snapshot(exam_ids, 100_000, "exam identifier")
+        student_rows: list[tuple[int, ...]] = []
+        try:
+            for row in students:
+                if len(student_rows) >= len(exams):
+                    raise ValueError("exam, student and candidate row counts differ")
+                if isinstance(row, (str, bytes)):
+                    raise ValueError("student rows must be integer iterables, not strings")
+                student_rows.append(tuple(row))
+        except TypeError as error:
+            raise ValueError("student rows must be iterable") from error
+        candidate_rows: list[tuple[str, ...]] = []
+        try:
+            iterator = iter(candidate_periods)
+        except TypeError as error:
+            raise ValueError("candidate rows must be iterable") from error
+        for row in iterator:
+            if len(candidate_rows) >= len(exams):
+                raise ValueError("exam, student and candidate row counts differ")
+            if isinstance(row, (str, bytes)):
+                raise ValueError("candidate rows must be period iterables, not strings")
+            try:
+                candidate_rows.append(tuple(row))
+            except TypeError as error:
+                raise ValueError("candidate rows must be iterable") from error
+        if not 1 <= len(periods) <= 64:
+            raise ValueError("period count must be in [1, 64]")
+        if len(student_rows) != len(exams) or len(candidate_rows) != len(exams):
+            raise ValueError("exam, student and candidate row counts differ")
+        candidates = tuple(candidate_rows)
+        cls._check_identifiers(periods, "period")
+        cls._check_identifiers(exams, "exam")
+        if len(set(periods)) != len(periods):
+            raise ValueError("period identifiers must be unique")
+        if len(set(exams)) != len(exams):
+            raise ValueError("exam identifiers must be unique")
+
+        period_at = {name: index for index, name in enumerate(periods)}
+        masks: list[int] = []
+        for row in candidates:
+            if not 1 <= len(row) <= 2 or any(type(value) is not str for value in row):
+                raise ValueError("each exam needs one or two exact period identifiers")
+            if len(set(row)) != len(row):
+                raise ValueError("candidate periods must be unique within an exam")
+            try:
+                indices = tuple(period_at[value] for value in row)
+            except KeyError as error:
+                raise ValueError("candidate names an unknown period") from error
+            masks.append(sum(1 << index for index in indices))
+
+        normalized_students: list[tuple[int, ...]] = []
+        enrollment_total = 0
+        roster: dict[int, list[int]] = {}
+        for vertex, row in enumerate(student_rows):
+            if any(type(value) is not int or not 0 <= value < 1 << 63 for value in row):
+                raise ValueError("student identifiers must be bounded nonnegative integers")
+            normalized = tuple(sorted(set(row)))
+            enrollment_total += len(normalized)
+            if enrollment_total > MAX_ENROLLMENTS:
+                raise ValueError("enrollment count exceeds adapter bound")
+            normalized_students.append(normalized)
+            for student in normalized:
+                roster.setdefault(student, []).append(vertex)
+
+        reasons: dict[tuple[int, int], ExamConflictReason] = {}
+        pair_emissions = 0
+        for student in sorted(roster):
+            vertices = roster[student]
+            # A single very large cohort can otherwise spend unbounded time before
+            # the native edge limit is reached.  This is a conservative admission
+            # check; duplicate pairs from other students do not earn extra budget.
+            pairs = len(vertices) * (len(vertices) - 1) // 2
+            pair_emissions += pairs
+            if pairs > MAX_CONFLICTS or pair_emissions > MAX_PAIR_EMISSIONS:
+                raise ValueError("student cohorts exceed bounded conflict-generation work")
+            for offset, left in enumerate(vertices):
+                for right in vertices[offset + 1:]:
+                    edge = (left, right)
+                    reasons.setdefault(edge, ExamConflictReason("shared_student", student))
+                    if len(reasons) > MAX_CONFLICTS:
+                        raise ValueError("conflict edge count exceeds native bound")
+
+        exam_at = {name: index for index, name in enumerate(exams)}
+        for item in exclusions:
+            if isinstance(item, (str, bytes)):
+                raise ValueError("exclusion must contain two exact exam identifiers")
+            row = tuple(item)
+            if len(row) != 2 or any(type(value) is not str for value in row):
+                raise ValueError("exclusion must contain two exact exam identifiers")
+            try:
+                left, right = exam_at[row[0]], exam_at[row[1]]
+            except KeyError as error:
+                raise ValueError("exclusion names an unknown exam") from error
+            if left == right:
+                raise ValueError("an exam cannot exclude itself")
+            edge = (min(left, right), max(left, right))
+            reasons.setdefault(edge, ExamConflictReason("explicit_exclusion"))
+            if len(reasons) > MAX_CONFLICTS:
+                raise ValueError("conflict edge count exceeds native bound")
+
+        edges = tuple(sorted(reasons))
+        problem = ExactProblem.create(len(exams), len(periods), edges, masks)
+        return cls(
+            problem=problem,
+            exam_ids=exams,
+            period_ids=periods,
+            students=tuple(normalized_students),
+            conflict_reasons=tuple(reasons[edge] for edge in problem.edges),
+        )
+
+    @staticmethod
+    def _bounded_snapshot(values: Iterable[str], limit: int, name: str) -> tuple[str, ...]:
+        try:
+            iterator = iter(values)
+        except TypeError as error:
+            raise ValueError(f"{name}s must be iterable") from error
+        copied: list[str] = []
+        for value in iterator:
+            if len(copied) >= limit:
+                raise ValueError(f"{name} count exceeds adapter bound")
+            copied.append(value)
+        return tuple(copied)
+
+    @staticmethod
+    def _check_identifiers(values: tuple[str, ...], kind: str) -> None:
+        for value in values:
+            if type(value) is not str or not value or len(value) > MAX_IDENTIFIER_CHARS:
+                raise ValueError(
+                    f"{kind} identifiers must be nonempty exact strings of at most "
+                    f"{MAX_IDENTIFIER_CHARS} characters"
+                )
+
+    @property
+    def candidates(self) -> tuple[tuple[str, ...], ...]:
+        """Return each exam's admitted candidate periods in colour order."""
+        return tuple(
+            tuple(self.period_ids[index] for index in range(len(self.period_ids)) if mask >> index & 1)
+            for mask in self.problem.masks
+        )
+
+    @property
+    def info(self) -> dict:
+        shared = sum(reason.kind == "shared_student" for reason in self.conflict_reasons)
+        return {
+            "schema": "spectra.exam_repair.model.v1",
+            "exams": self.problem.n,
+            "periods": self.problem.palette,
+            "conflicts": len(self.problem.edges),
+            "shared_student_conflicts": shared,
+            "explicit_exclusions": len(self.problem.edges) - shared,
+            "enrollments": sum(map(len, self.students)),
+            "scope": "one/two pre-approved periods; shared students and explicit exclusions only",
+        }
+
+    def freeze_event(
+        self,
+        *,
+        closed_periods: Iterable[str] = (),
+        fixed: Iterable[Sequence[str]] = (),
+        allowed: Iterable[Sequence[object]] = (),
+    ) -> tuple[tuple[int, int], ...]:
+        """Translate current closures/pins into one canonical exact query.
+
+        ``fixed`` rows are ``(exam_id, period_id)``.  ``allowed`` rows are
+        ``(exam_id, iterable_of_period_ids)``.  Multiple sources of restriction
+        are intersected.  A zero resulting mask is retained so the exact service
+        can return a portable empty-domain proof instead of raising.
+        """
+        period_at = {name: index for index, name in enumerate(self.period_ids)}
+        exam_at = {name: index for index, name in enumerate(self.exam_ids)}
+        closed = tuple(closed_periods)
+        if any(type(name) is not str for name in closed) or len(set(closed)) != len(closed):
+            raise ValueError("closed periods must be unique exact strings")
+        try:
+            closed_mask = sum(1 << period_at[name] for name in closed)
+        except KeyError as error:
+            raise ValueError("closed-period event names an unknown period") from error
+
+        effective = [mask & ~closed_mask for mask in self.problem.masks]
+        fixed_seen: set[int] = set()
+        for item in fixed:
+            if isinstance(item, (str, bytes)):
+                raise ValueError("fixed rows must be exact exam/period string pairs")
+            row = tuple(item)
+            if len(row) != 2 or any(type(value) is not str for value in row):
+                raise ValueError("fixed rows must be exact exam/period string pairs")
+            try:
+                vertex, period = exam_at[row[0]], period_at[row[1]]
+            except KeyError as error:
+                raise ValueError("fixed row names an unknown exam or period") from error
+            if vertex in fixed_seen:
+                raise ValueError("an exam appears more than once in fixed rows")
+            fixed_seen.add(vertex)
+            effective[vertex] &= 1 << period
+
+        allowed_seen: set[int] = set()
+        for item in allowed:
+            if isinstance(item, (str, bytes)):
+                raise ValueError("allowed rows must contain an exam and a period iterable")
+            row = tuple(item)
+            if len(row) != 2 or type(row[0]) is not str or isinstance(row[1], (str, bytes)):
+                raise ValueError("allowed rows must contain an exam and a period iterable")
+            try:
+                vertex = exam_at[row[0]]
+            except KeyError as error:
+                raise ValueError("allowed row names an unknown exam") from error
+            if vertex in allowed_seen:
+                raise ValueError("an exam appears more than once in allowed rows")
+            allowed_seen.add(vertex)
+            try:
+                choices = tuple(row[1])
+            except TypeError as error:
+                raise ValueError(
+                    "allowed rows must contain an exam and a period iterable"
+                ) from error
+            if any(type(name) is not str for name in choices) or len(set(choices)) != len(choices):
+                raise ValueError("allowed period identifiers must be unique exact strings")
+            try:
+                mask = sum(1 << period_at[name] for name in choices)
+            except KeyError as error:
+                raise ValueError("allowed row names an unknown period") from error
+            effective[vertex] &= mask
+
+        return tuple(
+            (vertex, mask)
+            for vertex, (original, mask) in enumerate(zip(self.problem.masks, effective))
+            if mask != original
+        )
+
+    def assignment_from_labels(
+        self,
+        labels: bytes,
+        query: Iterable[Sequence[int]] = (),
+    ) -> tuple[tuple[str, str], ...]:
+        """Decode a complete checked label vector into stable exam/period rows."""
+        frozen = self.problem.freeze_query(query)
+        if not self.problem.check(labels, frozen):
+            raise ValueError("labels do not satisfy the admitted repair problem and event")
+        return tuple((exam, self.period_ids[labels[index]])
+                     for index, exam in enumerate(self.exam_ids))
+
+    def check_assignment(
+        self,
+        assignment: Mapping[str, str] | Iterable[Sequence[str]],
+        query: Iterable[Sequence[int]] = (),
+    ) -> bool:
+        """Check a named assignment against original adapter inputs only."""
+        try:
+            rows = tuple(assignment.items()) if isinstance(assignment, Mapping) else tuple(
+                tuple(row) for row in assignment
+            )
+            exam_at = {name: index for index, name in enumerate(self.exam_ids)}
+            period_at = {name: index for index, name in enumerate(self.period_ids)}
+            labels = [-1] * self.problem.n
+            for row in rows:
+                if len(row) != 2 or any(type(value) is not str for value in row):
+                    return False
+                vertex = exam_at[row[0]]
+                if labels[vertex] != -1:
+                    return False
+                labels[vertex] = period_at[row[1]]
+            if any(value < 0 for value in labels):
+                return False
+            return self.problem.check(bytes(labels), query)
+        except (KeyError, TypeError, ValueError, OverflowError):
+            return False
+
+    def explain_unsat(
+        self,
+        query: Iterable[Sequence[int]],
+        certificate: bytes,
+    ) -> dict:
+        """Verify and translate an SBP1 contradiction into named repair steps."""
+        frozen = self.problem.freeze_query(query)
+        if not verify_answer(self.problem, frozen, "UNSAT_VERIFIED", b"", certificate):
+            raise ValueError("certificate does not prove this repair event infeasible")
+        kind, left, right = decode_proof(certificate, self.problem.n)
+        effective = list(self.problem.masks)
+        for vertex, mask in frozen:
+            effective[vertex] &= mask
+        if kind == 1:
+            vertex = left[0]
+            return {
+                "schema": "spectra.exam_repair.proof.v1",
+                "kind": "EMPTY_DOMAIN",
+                "exam": self.exam_ids[vertex],
+                "original_candidates": list(self.candidates[vertex]),
+                "remaining_candidates": [],
+            }
+        return {
+            "schema": "spectra.exam_repair.proof.v1",
+            "kind": "FORCED_OPPOSITES" if kind == 2 else "BASE_CONTRADICTION",
+            "left": self._explain_path(left, effective),
+            "right": self._explain_path(right, effective),
+        }
+
+    def _literal(self, literal: int) -> dict:
+        vertex, side = divmod(literal, 2)
+        colors = tuple(index for index in range(self.problem.palette)
+                       if self.problem.masks[vertex] >> index & 1)
+        period = self.period_ids[colors[side]] if side < len(colors) else None
+        return {
+            "literal": literal,
+            "exam": self.exam_ids[vertex],
+            "exam_index": vertex,
+            "period": period,
+            "side": side,
+        }
+
+    def _explain_path(self, path: tuple[int, ...], effective: list[int]) -> dict:
+        reasons = {edge: reason for edge, reason in zip(
+            self.problem.edges, self.conflict_reasons
+        )}
+        nodes = [self._literal(literal) for literal in path]
+        steps = []
+        for before, after in zip(path, path[1:]):
+            left, right = before // 2, after // 2
+            if left == right:
+                steps.append({
+                    "kind": "SINGLETON_DOMAIN",
+                    "exam": self.exam_ids[left],
+                    "therefore_period": self._literal(after)["period"],
+                })
+                continue
+            edge = (min(left, right), max(left, right))
+            reason = reasons[edge]
+            before_node = self._literal(before)
+            after_node = self._literal(after)
+            step = {
+                "kind": "SHARED_STUDENT_CONFLICT" if reason.kind == "shared_student"
+                        else "EXPLICIT_EXCLUSION",
+                "from_exam": before_node["exam"],
+                "from_period": before_node["period"],
+                "to_exam": after_node["exam"],
+                "therefore_period": after_node["period"],
+            }
+            if reason.student is not None:
+                step["student"] = reason.student
+            steps.append(step)
+        root = nodes[0]
+        root_mask = effective[root["exam_index"]]
+        root["forced_by_event"] = (
+            root["period"] is not None
+            and root_mask == 1 << self.period_ids.index(root["period"])
+        )
+        return {"nodes": nodes, "steps": steps}
+
+
+__all__ = ["ExamConflictReason", "ExamRepairModel"]
diff --git a/spectra/itc2007_exam.py b/spectra/itc2007_exam.py
new file mode 100644
index 0000000..b0f9c52
--- /dev/null
+++ b/spectra/itc2007_exam.py
@@ -0,0 +1,316 @@
+"""Strict ITC 2007 examination-timetabling reader and bounded-repair adapter.
+
+The reader preserves the complete hard/soft section inventory.  Conversion to
+:class:`spectra.exam_repair.ExamRepairModel` is intentionally explicit about the
+smaller problem being solved: shared-student and EXCLUSION conflicts over one or
+two pre-approved periods per exam.  AFTER, EXAM_COINCIDENCE, room assignment,
+room exclusivity, and soft objectives are never silently interpreted as solved.
+"""
+from __future__ import annotations
+
+from dataclasses import dataclass
+import hashlib
+import re
+from typing import Iterable
+
+from .exam_repair import ExamRepairModel
+
+_HEADER = re.compile(r"^\[([A-Za-z]+)(?::([0-9]+))?\]$")
+_REQUIRED = (
+    "Exams", "Periods", "Rooms", "PeriodHardConstraints",
+    "RoomHardConstraints", "InstitutionalWeightings",
+)
+
+
+@dataclass(frozen=True, slots=True)
+class ITCExam:
+    duration: int
+    students: tuple[int, ...]
+
+
+@dataclass(frozen=True, slots=True)
+class ITCPeriod:
+    date: str
+    time: str
+    duration: int
+    penalty: int
+
+    @property
+    def identifier(self) -> str:
+        return f"{self.date} {self.time}"
+
+
+@dataclass(frozen=True, slots=True)
+class ITCRoom:
+    capacity: int
+    penalty: int
+
+
+@dataclass(frozen=True, slots=True)
+class ITCConflictCore:
+    """One exact conflict-core model plus an immutable omitted-constraint receipt."""
+
+    model: ExamRepairModel
+    source_sha256: str
+    ignored_constraints: tuple[tuple[str, int], ...]
+
+    @property
+    def info(self) -> dict:
+        return {
+            **self.model.info,
+            "schema": "spectra.itc2007.conflict_core.v1",
+            "source_sha256": self.source_sha256,
+            "ignored_constraints": dict(self.ignored_constraints),
+            "claim_boundary": (
+                "bounded candidate-period repair only; not a complete ITC 2007 "
+                "timetable, room assignment, precedence solution or soft-score result"
+            ),
+        }
+
+
+@dataclass(frozen=True, slots=True)
+class ITCExamInstance:
+    exams: tuple[ITCExam, ...]
+    periods: tuple[ITCPeriod, ...]
+    rooms: tuple[ITCRoom, ...]
+    after: tuple[tuple[int, int], ...]
+    exclusions: tuple[tuple[int, int], ...]
+    coincidences: tuple[tuple[int, int], ...]
+    room_exclusive: tuple[int, ...]
+    institutional_weightings: tuple[tuple[str, ...], ...]
+    source_sha256: str
+
+    @property
+    def ignored_constraint_counts(self) -> tuple[tuple[str, int], ...]:
+        values = (
+            ("after", len(self.after)),
+            ("exam_coincidence", len(self.coincidences)),
+            ("room_exclusive", len(self.room_exclusive)),
+            ("room_assignment", len(self.rooms)),
+            ("period_penalties", sum(period.penalty != 0 for period in self.periods)),
+            ("room_penalties", sum(room.penalty != 0 for room in self.rooms)),
+            ("institutional_weightings", len(self.institutional_weightings)),
+        )
+        return tuple((name, count) for name, count in values if count)
+
+    def conflict_core(
+        self,
+        candidate_periods: Iterable[Iterable[int]],
+        *,
+        acknowledge_ignored_constraints: bool = False,
+    ) -> ITCConflictCore:
+        """Build the exact shared-student/EXCLUSION repair core.
+
+        Candidate rows contain one or two period indices and are assumed to have
+        been pre-approved by the operator.  Period duration is still checked here.
+        Any unmodelled ITC section requires an explicit acknowledgement flag and
+        remains listed in the returned receipt.
+        """
+        ignored = self.ignored_constraint_counts
+        if ignored and not acknowledge_ignored_constraints:
+            names = ", ".join(name for name, _ in ignored)
+            raise ValueError("ITC conflict-core conversion would ignore: " + names)
+        copied_rows: list[tuple[int, ...]] = []
+        try:
+            for row in candidate_periods:
+                if len(copied_rows) >= len(self.exams):
+                    raise ValueError("candidate row count differs from exam count")
+                if isinstance(row, (str, bytes)):
+                    raise ValueError("candidate rows must be period-index iterables")
+                copied_rows.append(tuple(row))
+        except TypeError as error:
+            raise ValueError("candidate rows must be period-index iterables") from error
+        rows = tuple(copied_rows)
+        if len(rows) != len(self.exams):
+            raise ValueError("candidate row count differs from exam count")
+        names = tuple(
+            f"{index:02d}:{period.identifier}"
+            for index, period in enumerate(self.periods)
+        )
+        candidates: list[tuple[str, ...]] = []
+        for exam_index, (exam, row) in enumerate(zip(self.exams, rows)):
+            if not 1 <= len(row) <= 2 or any(type(value) is not int for value in row):
+                raise ValueError("every exam needs one or two exact period indices")
+            if len(set(row)) != len(row) or any(not 0 <= value < len(self.periods) for value in row):
+                raise ValueError("candidate period indices must be unique and in range")
+            if any(self.periods[value].duration < exam.duration for value in row):
+                raise ValueError(f"exam {exam_index} does not fit a candidate period")
+            candidates.append(tuple(names[value] for value in row))
+        exam_ids = tuple(f"exam-{index:04d}" for index in range(len(self.exams)))
+        exclusions = tuple((exam_ids[left], exam_ids[right])
+                           for left, right in self.exclusions)
+        model = ExamRepairModel.create(
+            names,
+            exam_ids,
+            (exam.students for exam in self.exams),
+            candidates,
+            exclusions=exclusions,
+        )
+        return ITCConflictCore(model, self.source_sha256, ignored)
+
+
+def parse_itc2007_exam(text: str, *, max_chars: int = 64 << 20) -> ITCExamInstance:
+    """Parse one canonical ITC 2007 examination instance without dependencies."""
+    if type(text) is not str:
+        raise ValueError("ITC input must be an exact string")
+    if type(max_chars) is not int or not 1 <= max_chars <= 1 << 30:
+        raise ValueError("max_chars must be a bounded positive exact integer")
+    if len(text) > max_chars:
+        raise ValueError("ITC input exceeds max_chars")
+    normalized = text.removeprefix("\ufeff").replace("\r\n", "\n").replace("\r", "\n")
+    sections: dict[str, tuple[int | None, list[tuple[int, str]]]] = {}
+    current: str | None = None
+    for line_number, raw in enumerate(normalized.split("\n"), 1):
+        line = raw.strip()
+        if not line:
+            continue
+        match = _HEADER.fullmatch(line)
+        if match:
+            name, count_text = match.groups()
+            if name not in _REQUIRED:
+                raise ValueError(f"line {line_number}: unknown ITC section {name}")
+            if name in sections:
+                raise ValueError(f"line {line_number}: duplicate ITC section {name}")
+            expected = int(count_text) if count_text is not None else None
+            if name in ("Exams", "Periods", "Rooms") and expected is None:
+                raise ValueError(f"line {line_number}: {name} requires a count")
+            if name not in ("Exams", "Periods", "Rooms") and expected is not None:
+                raise ValueError(f"line {line_number}: {name} must not carry a count")
+            sections[name] = (expected, [])
+            current = name
+            continue
+        if current is None:
+            raise ValueError(f"line {line_number}: data appears before a section header")
+        sections[current][1].append((line_number, line))
+    if tuple(sections) != _REQUIRED:
+        raise ValueError("ITC sections are missing, duplicated or out of canonical order")
+    for name in ("Exams", "Periods", "Rooms"):
+        expected, rows = sections[name]
+        if expected != len(rows):
+            raise ValueError(f"{name} row count differs: expected {expected}, found {len(rows)}")
+
+    exams: list[ITCExam] = []
+    for line_number, line in sections["Exams"][1]:
+        fields = _fields(line, line_number)
+        if len(fields) < 1:
+            raise ValueError(f"line {line_number}: empty exam row")
+        duration = _nonnegative(fields[0], line_number, "exam duration", positive=True)
+        student_values = tuple(_nonnegative(value, line_number, "student id")
+                               for value in fields[1:])
+        if len(set(student_values)) != len(student_values):
+            raise ValueError(f"line {line_number}: duplicate student within an exam")
+        exams.append(ITCExam(duration, tuple(sorted(student_values))))
+
+    periods: list[ITCPeriod] = []
+    for line_number, line in sections["Periods"][1]:
+        fields = _fields(line, line_number)
+        if len(fields) != 4 or not fields[0] or not fields[1]:
+            raise ValueError(f"line {line_number}: period row must have date,time,duration,penalty")
+        periods.append(ITCPeriod(
+            fields[0], fields[1],
+            _nonnegative(fields[2], line_number, "period duration", positive=True),
+            _nonnegative(fields[3], line_number, "period penalty"),
+        ))
+    identifiers = tuple(period.identifier for period in periods)
+    if len(set(identifiers)) != len(identifiers):
+        raise ValueError("period date/time identifiers must be unique")
+
+    rooms: list[ITCRoom] = []
+    for line_number, line in sections["Rooms"][1]:
+        fields = _fields(line, line_number)
+        if len(fields) != 2:
+            raise ValueError(f"line {line_number}: room row must have capacity,penalty")
+        rooms.append(ITCRoom(
+            _nonnegative(fields[0], line_number, "room capacity"),
+            _nonnegative(fields[1], line_number, "room penalty"),
+        ))
+
+    after: list[tuple[int, int]] = []
+    exclusions: list[tuple[int, int]] = []
+    coincidences: list[tuple[int, int]] = []
+    seen_hard: set[tuple[int, str, int]] = set()
+    for line_number, line in sections["PeriodHardConstraints"][1]:
+        fields = _fields(line, line_number)
+        if len(fields) != 3:
+            raise ValueError(f"line {line_number}: period hard constraint must have three fields")
+        left = _exam_index(fields[0], line_number, len(exams))
+        right = _exam_index(fields[2], line_number, len(exams))
+        relation = fields[1]
+        if relation not in ("AFTER", "EXCLUSION", "EXAM_COINCIDENCE"):
+            raise ValueError(f"line {line_number}: unsupported period relation {relation}")
+        if left == right:
+            raise ValueError(f"line {line_number}: hard constraint repeats one exam")
+        key = ((min(left, right), relation, max(left, right))
+               if relation in ("EXCLUSION", "EXAM_COINCIDENCE")
+               else (left, relation, right))
+        if key in seen_hard:
+            raise ValueError(f"line {line_number}: duplicate period hard constraint")
+        seen_hard.add(key)
+        if relation == "AFTER":
+            after.append((left, right))
+        elif relation == "EXCLUSION":
+            exclusions.append((min(left, right), max(left, right)))
+        else:
+            coincidences.append((min(left, right), max(left, right)))
+
+    room_exclusive: list[int] = []
+    for line_number, line in sections["RoomHardConstraints"][1]:
+        fields = _fields(line, line_number)
+        if len(fields) != 2 or fields[1] != "ROOM_EXCLUSIVE":
+            raise ValueError(f"line {line_number}: invalid room hard constraint")
+        exam = _exam_index(fields[0], line_number, len(exams))
+        if exam in room_exclusive:
+            raise ValueError(f"line {line_number}: duplicate room-exclusive exam")
+        room_exclusive.append(exam)
+
+    weightings: list[tuple[str, ...]] = []
+    for line_number, line in sections["InstitutionalWeightings"][1]:
+        fields = _fields(line, line_number)
+        if not fields[0] or any(not value for value in fields):
+            raise ValueError(f"line {line_number}: empty institutional weighting field")
+        # The exact weighting vocabulary is not interpreted by the conflict-core
+        # adapter, but every numeric payload must still be canonical and bounded.
+        for value in fields[1:]:
+            _nonnegative(value, line_number, "institutional weighting")
+        weightings.append(fields)
+
+    return ITCExamInstance(
+        exams=tuple(exams),
+        periods=tuple(periods),
+        rooms=tuple(rooms),
+        after=tuple(after),
+        exclusions=tuple(sorted(set(exclusions))),
+        coincidences=tuple(sorted(set(coincidences))),
+        room_exclusive=tuple(room_exclusive),
+        institutional_weightings=tuple(weightings),
+        source_sha256=hashlib.sha256(normalized.encode("utf-8")).hexdigest(),
+    )
+
+
+def _fields(line: str, line_number: int) -> tuple[str, ...]:
+    values = tuple(value.strip() for value in line.split(","))
+    if any(not value for value in values):
+        raise ValueError(f"line {line_number}: empty comma-separated field")
+    return values
+
+
+def _nonnegative(value: str, line_number: int, name: str, *, positive: bool = False) -> int:
+    if not value.isascii() or not value.isdigit():
+        raise ValueError(f"line {line_number}: {name} must be a canonical nonnegative integer")
+    number = int(value)
+    if number >= 1 << 63 or (positive and number == 0):
+        raise ValueError(f"line {line_number}: {name} is outside the supported bound")
+    return number
+
+
+def _exam_index(value: str, line_number: int, count: int) -> int:
+    index = _nonnegative(value, line_number, "exam index")
+    if index >= count:
+        raise ValueError(f"line {line_number}: exam index outside declared exam bank")
+    return index
+
+
+__all__ = [
+    "ITCExam", "ITCPeriod", "ITCRoom", "ITCExamInstance", "ITCConflictCore",
+    "parse_itc2007_exam",
+]
diff --git a/tests/public/test_exam_repair.py b/tests/public/test_exam_repair.py
new file mode 100644
index 0000000..c50f125
--- /dev/null
+++ b/tests/public/test_exam_repair.py
@@ -0,0 +1,146 @@
+"""Public bounded exam-repair adapter contracts."""
+from __future__ import annotations
+
+import shutil
+import struct
+import sys
+
+import pytest
+
+from spectra.exam_repair import ExamRepairModel
+from spectra.exact import ExactRuntime, build_exact_runtime
+
+
+def path_model() -> ExamRepairModel:
+    return ExamRepairModel.create(
+        period_ids=("P0", "P1"),
+        exam_ids=("A", "B", "C"),
+        students=((1,), (1, 2), (2,)),
+        candidate_periods=(("P0", "P1"),) * 3,
+    )
+
+
+def test_adapter_builds_deterministic_conflicts_and_named_metadata():
+    model = ExamRepairModel.create(
+        period_ids=("AM", "PM", "EVENING"),
+        exam_ids=("MATH", "CHEM", "HIST"),
+        students=((4, 2, 4), (2,), (9,)),
+        candidate_periods=(("AM", "PM"), ("PM",), ("AM", "EVENING")),
+        exclusions=(("CHEM", "HIST"),),
+    )
+    assert model.problem.edges == ((0, 1), (1, 2))
+    assert tuple(reason.kind for reason in model.conflict_reasons) == (
+        "shared_student", "explicit_exclusion"
+    )
+    assert model.conflict_reasons[0].student == 2
+    assert model.candidates == (("AM", "PM"), ("PM",), ("AM", "EVENING"))
+    assert model.info == {
+        "schema": "spectra.exam_repair.model.v1",
+        "exams": 3,
+        "periods": 3,
+        "conflicts": 2,
+        "shared_student_conflicts": 1,
+        "explicit_exclusions": 1,
+        "enrollments": 4,
+        "scope": "one/two pre-approved periods; shared students and explicit exclusions only",
+    }
+
+
+def test_event_intersects_closures_pins_and_allowed_periods():
+    model = path_model()
+    query = model.freeze_event(
+        closed_periods=("P1",),
+        fixed=(("B", "P1"),),
+        allowed=(("C", ("P0",)),),
+    )
+    assert query == ((0, 0b01), (1, 0), (2, 0b01))
+    with pytest.raises(ValueError, match="unknown period"):
+        model.freeze_event(closed_periods=("P2",))
+    with pytest.raises(ValueError, match="more than once"):
+        model.freeze_event(fixed=(("A", "P0"), ("A", "P1")))
+    with pytest.raises(ValueError, match="period iterable"):
+        model.freeze_event(allowed=(("A", 7),))
+
+
+def test_named_assignment_is_checked_against_event_and_conflicts():
+    model = path_model()
+    query = model.freeze_event(fixed=(("A", "P0"), ("C", "P0")))
+    labels = b"\x00\x01\x00"
+    assert model.assignment_from_labels(labels, query) == (
+        ("A", "P0"), ("B", "P1"), ("C", "P0")
+    )
+    assert model.check_assignment({"A": "P0", "B": "P1", "C": "P0"}, query)
+    assert not model.check_assignment({"A": "P0", "B": "P0", "C": "P1"}, query)
+    assert not model.check_assignment({"A": "P0", "B": "P1"}, query)
+
+
+def test_portable_proof_translates_to_shared_student_chain():
+    model = ExamRepairModel.create(
+        period_ids=("P0", "P1"),
+        exam_ids=("A", "B"),
+        students=((17,), (17,)),
+        candidate_periods=(("P0", "P1"), ("P0", "P1")),
+    )
+    query = model.freeze_event(fixed=(("A", "P0"), ("B", "P0")))
+    # A=P0 implies B=P1; the independently forced B=P0 reaches the opposite.
+    certificate = struct.pack("<4sIII3I", b"SBP1", 2, 2, 1, 0, 3, 2)
+    record = model.explain_unsat(query, certificate)
+    assert record["kind"] == "FORCED_OPPOSITES"
+    assert record["left"]["nodes"][0]["forced_by_event"] is True
+    assert record["left"]["steps"] == [{
+        "kind": "SHARED_STUDENT_CONFLICT",
+        "from_exam": "A",
+        "from_period": "P0",
+        "to_exam": "B",
+        "therefore_period": "P1",
+        "student": 17,
+    }]
+    with pytest.raises(ValueError, match="does not prove"):
+        model.explain_unsat((), certificate)
+
+
+def test_empty_domain_proof_names_the_closed_exam():
+    model = path_model()
+    query = model.freeze_event(closed_periods=("P0", "P1"))
+    certificate = struct.pack("<4sIIII", b"SBP1", 1, 1, 0, 0)
+    assert model.explain_unsat(query, certificate) == {
+        "schema": "spectra.exam_repair.proof.v1",
+        "kind": "EMPTY_DOMAIN",
+        "exam": "A",
+        "original_candidates": ["P0", "P1"],
+        "remaining_candidates": [],
+    }
+
+
+def test_adapter_rejects_ambiguous_or_out_of_scope_input_shapes():
+    with pytest.raises(ValueError, match="one or two"):
+        ExamRepairModel.create(("P",), ("A",), ((),), ((),))
+    with pytest.raises(ValueError, match="unique"):
+        ExamRepairModel.create(("P", "P"), (), (), ())
+    with pytest.raises(ValueError, match="cannot exclude itself"):
+        ExamRepairModel.create(("P",), ("A",), ((),), (("P",),), exclusions=(("A", "A"),))
+    with pytest.raises(ValueError, match="period identifier count exceeds"):
+        ExamRepairModel.create((f"P{i}" for i in range(65)), (), (), ())
+
+
+def _compiler_available() -> bool:
+    return bool(shutil.which("cl" if sys.platform == "win32" else "g++"))
+
+
+@pytest.mark.skipif(sys.platform not in ("linux", "win32") or not _compiler_available(),
+                    reason="supported native compiler unavailable")
+def test_adapter_runs_through_public_native_api(tmp_path):
+    model = path_model()
+    runtime = ExactRuntime(build_exact_runtime(tmp_path / "native"))
+    sat_query = model.freeze_event(fixed=(("A", "P0"), ("C", "P0")))
+    unsat_query = model.freeze_event(fixed=(("A", "P0"), ("B", "P0")))
+    with runtime.prepare(model.problem, mode="scc") as session:
+        sat = session.solve(sat_query)
+        unsat = session.solve(unsat_query)
+    assert sat.status == "SAT_VERIFIED"
+    assert model.check_assignment(model.assignment_from_labels(sat.labels, sat_query), sat_query)
+    assert unsat.status == "UNSAT_VERIFIED"
+    explained = model.explain_unsat(unsat_query, unsat.proof)
+    assert explained["kind"] == "FORCED_OPPOSITES"
+    assert any(step.get("student") == 1
+               for side in ("left", "right") for step in explained[side]["steps"])
diff --git a/tests/public/test_itc2007_exam.py b/tests/public/test_itc2007_exam.py
new file mode 100644
index 0000000..bdbbf1b
--- /dev/null
+++ b/tests/public/test_itc2007_exam.py
@@ -0,0 +1,85 @@
+"""Strict ITC 2007 parser and bounded conflict-core adapter contracts."""
+from __future__ import annotations
+
+import pytest
+
+from spectra.itc2007_exam import parse_itc2007_exam
+
+
+FIXTURE = """[Exams:3]
+120, 1, 2
+180, 2, 3
+120, 9
+[Periods:3]
+01:01:2030, 09:00:00, 180, 0
+01:01:2030, 14:00:00, 120, 5
+02:01:2030, 09:00:00, 240, 0
+[Rooms:1]
+100, 0
+[PeriodHardConstraints]
+0, EXCLUSION, 2
+1, AFTER, 0
+[RoomHardConstraints]
+2, ROOM_EXCLUSIVE
+[InstitutionalWeightings]
+TWOINAROW, 30
+"""
+
+
+def test_parser_preserves_complete_constraint_inventory():
+    instance = parse_itc2007_exam(FIXTURE)
+    assert len(instance.exams) == 3
+    assert instance.exams[0].students == (1, 2)
+    assert len(instance.periods) == 3
+    assert instance.after == ((1, 0),)
+    assert instance.exclusions == ((0, 2),)
+    assert instance.room_exclusive == (2,)
+    assert dict(instance.ignored_constraint_counts) == {
+        "after": 1,
+        "room_exclusive": 1,
+        "room_assignment": 1,
+        "period_penalties": 1,
+        "institutional_weightings": 1,
+    }
+    assert len(instance.source_sha256) == 64
+
+
+def test_conflict_core_requires_explicit_scope_acknowledgement():
+    instance = parse_itc2007_exam(FIXTURE)
+    candidates = ((0, 1), (0, 2), (1, 2))
+    with pytest.raises(ValueError, match="would ignore"):
+        instance.conflict_core(candidates)
+    core = instance.conflict_core(candidates, acknowledge_ignored_constraints=True)
+    # Exam 0/1 share student 2; exam 0/2 have explicit EXCLUSION.
+    assert core.model.problem.edges == ((0, 1), (0, 2))
+    assert core.model.candidates[0] == (
+        "00:01:01:2030 09:00:00", "01:01:01:2030 14:00:00"
+    )
+    assert core.info["schema"] == "spectra.itc2007.conflict_core.v1"
+    assert "not a complete ITC 2007 timetable" in core.info["claim_boundary"]
+
+
+def test_duration_check_rejects_unapproved_short_period():
+    instance = parse_itc2007_exam(FIXTURE)
+    # Exam 1 lasts 180 minutes; period 1 lasts only 120.
+    with pytest.raises(ValueError, match="does not fit"):
+        instance.conflict_core(((0,), (1,), (2,)), acknowledge_ignored_constraints=True)
+
+
+def test_conflict_core_rejects_noniterable_candidate_rows():
+    instance = parse_itc2007_exam(FIXTURE)
+    with pytest.raises(ValueError, match="period-index iterables"):
+        instance.conflict_core(((0,), 1, (2,)), acknowledge_ignored_constraints=True)
+
+
+@pytest.mark.parametrize("damage, message", [
+    (FIXTURE.replace("[Exams:3]", "[Exams:4]"), "row count differs"),
+    (FIXTURE.replace("[Periods:3]", "[Unknown:3]"), "unknown ITC section"),
+    (FIXTURE.replace("120, 1, 2", "120, 1, 1"), "duplicate student"),
+    (FIXTURE.replace("0, EXCLUSION, 2", "0, OTHER, 2"), "unsupported period relation"),
+    (FIXTURE.replace("0, EXCLUSION, 2", "0, EXCLUSION, 2\n2, EXCLUSION, 0"),
+     "duplicate period hard constraint"),
+])
+def test_parser_fails_closed_on_malformed_instances(damage, message):
+    with pytest.raises(ValueError, match=message):
+        parse_itc2007_exam(damage)

From f0430f1204075784d403a8f9e2ca2459761b87e7 Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:56:12 -0400
Subject: [PATCH 13/24] Add cross-platform proof-carrying exam transfer
 validation

---
 .github/workflows/itc-exam-transfer.yml | 114 ++++++++++++++++++++++++
 1 file changed, 114 insertions(+)
 create mode 100644 .github/workflows/itc-exam-transfer.yml

diff --git a/.github/workflows/itc-exam-transfer.yml b/.github/workflows/itc-exam-transfer.yml
new file mode 100644
index 0000000..4ac558a
--- /dev/null
+++ b/.github/workflows/itc-exam-transfer.yml
@@ -0,0 +1,114 @@
+name: itc-exam-transfer
+on:
+  push:
+    branches: [research/100-blitz-20261016]
+    paths:
+      - '.github/workflows/itc-exam-transfer.yml'
+      - 'spectra/exact*.py'
+      - 'spectra/exam_repair.py'
+      - 'spectra/itc2007_exam.py'
+      - 'spectra/_native/exact_binary.cpp'
+      - 'experiments/compiled_query/itc_exam_transfer.py'
+      - 'experiments/compiled_query/test_itc_exam_transfer.py'
+      - 'experiments/compiled_query/ITC_EXAM_TRANSFER_PROTOCOL.md'
+      - 'experiments/compiled_query/itc_exam_suite.py'
+      - 'experiments/compiled_query/test_itc_exam_suite.py'
+      - 'experiments/compiled_query/ITC_EXAM_SUITE_PROTOCOL.md'
+      - 'tests/public/test_exam_repair.py'
+      - 'tests/public/test_itc2007_exam.py'
+  workflow_dispatch:
+permissions:
+  contents: read
+concurrency:
+  group: itc-exam-transfer-${{ github.ref }}
+  cancel-in-progress: false
+env:
+  OMP_NUM_THREADS: '1'
+  OPENBLAS_NUM_THREADS: '1'
+jobs:
+  transfer:
+    name: ITC conflict-core transfer (${{ matrix.runner }}, Python ${{ matrix.python }})
+    runs-on: ${{ matrix.runner }}
+    timeout-minutes: 45
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - {runner: ubuntu-24.04, python: '3.11'}
+          - {runner: ubuntu-24.04, python: '3.13'}
+          - {runner: ubuntu-24.04-arm, python: '3.13'}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+      - uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python }}
+      - name: Install the pinned test runner
+        run: python -m pip install 'pytest==8.4.2'
+      - uses: actions/checkout@v4
+        with:
+          repository: davidvanbulck/ITC-2007-CETT
+          ref: ed2d4c0cf5302b77a8b321cd0380de0bb579d538
+          path: upstream/ITC-2007-CETT
+          persist-credentials: false
+      - name: Check the public adapter and immutable evidence harness
+        run: |
+          set -euo pipefail
+          python -m pytest \
+            tests/public/test_exam_repair.py \
+            tests/public/test_itc2007_exam.py \
+            experiments/compiled_query/test_itc_exam_transfer.py \
+            -q --junitxml=transfer-contracts.xml
+      - name: Freeze the published instance only after the protocol is committed
+        run: |
+          set -euo pipefail
+          python -m experiments.compiled_query.itc_exam_transfer freeze \
+            --root itc-transfer-evidence \
+            --instance upstream/ITC-2007-CETT/examtt-instances/Instances/set9.exam \
+            --solution upstream/ITC-2007-CETT/InitMethods/SolStore/set9_OBSIFeas_0.sol \
+            --license upstream/ITC-2007-CETT/LICENSE \
+            --queries 1024 \
+            --evaluation-role development | tee freeze.log
+      - name: Run every exact engine and replay every answer
+        run: |
+          set -euo pipefail
+          python -m experiments.compiled_query.itc_exam_transfer run \
+            --root itc-transfer-evidence | tee run.log
+          python -S -m experiments.compiled_query.itc_exam_transfer audit \
+            --root itc-transfer-evidence | tee audit.log
+          python - <<'PY'
+          import json
+          from pathlib import Path
+          result = json.loads(Path('itc-transfer-evidence/audit.json').read_text())
+          assert result['functional_complete']
+          assert result['baseline_valid_full_itc_hard_constraints']
+          assert result['queries'] == 1024
+          assert result['model']['schema'] == 'spectra.itc2007.conflict_core.v1'
+          assert result['modes']['auto']['auto_policy'] == 'structural-v2'
+          assert result['modes']['auto']['selected_mode'] == 'raw'
+          assert result['modes']['auto']['auto_policy_reason'] == 'high_pin_density'
+          print('Functional transfer complete. This is not a full ITC timetable claim.')
+          PY
+      - name: Preserve exact source and environment receipts
+        if: always()
+        run: |
+          set -euo pipefail
+          mkdir -p transfer-receipts
+          git rev-parse HEAD > transfer-receipts/source_commit.txt
+          git rev-parse HEAD^{tree} > transfer-receipts/source_tree.txt
+          git -C upstream/ITC-2007-CETT rev-parse HEAD > transfer-receipts/upstream_commit.txt
+          python -VV > transfer-receipts/python.txt
+          uname -a > transfer-receipts/platform.txt
+          lscpu > transfer-receipts/cpu.txt
+          cp transfer-contracts.xml freeze.log run.log audit.log transfer-receipts/ 2>/dev/null || true
+      - uses: actions/upload-artifact@v4
+        if: always()
+        with:
+          name: itc-exam-transfer-${{ matrix.runner }}-py${{ matrix.python }}-${{ github.run_id }}
+          path: |
+            itc-transfer-evidence/
+            transfer-receipts/
+          retention-days: 90
+          if-no-files-found: warn

From a0f5d18ac4651c6ab3a04303fafd724c1b6d5a91 Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:56:23 -0400
Subject: [PATCH 14/24] Add prospective multi-instance ITC holdout suite

---
 .github/workflows/itc-exam-suite.yml | 127 +++++++++++++++++++++++++++
 1 file changed, 127 insertions(+)
 create mode 100644 .github/workflows/itc-exam-suite.yml

diff --git a/.github/workflows/itc-exam-suite.yml b/.github/workflows/itc-exam-suite.yml
new file mode 100644
index 0000000..ef6a8bc
--- /dev/null
+++ b/.github/workflows/itc-exam-suite.yml
@@ -0,0 +1,127 @@
+name: itc-exam-suite
+on:
+  push:
+    branches: [research/100-blitz-20261016]
+    paths:
+      - '.github/workflows/itc-exam-suite.yml'
+      - 'spectra/exact*.py'
+      - 'spectra/exam_repair.py'
+      - 'spectra/itc2007_exam.py'
+      - 'spectra/_native/exact_binary.cpp'
+      - 'experiments/compiled_query/exact.py'
+      - 'experiments/compiled_query/exact_native.cpp'
+      - 'experiments/compiled_query/test_exact.py'
+      - 'experiments/compiled_query/itc_exam_transfer.py'
+      - 'experiments/compiled_query/itc_exam_suite.py'
+      - 'experiments/compiled_query/test_itc_exam_transfer.py'
+      - 'experiments/compiled_query/test_itc_exam_suite.py'
+      - 'experiments/compiled_query/ITC_EXAM_TRANSFER_PROTOCOL.md'
+      - 'experiments/compiled_query/ITC_EXAM_SUITE_PROTOCOL.md'
+  workflow_dispatch:
+permissions:
+  contents: read
+concurrency:
+  group: itc-exam-suite-${{ github.ref }}
+  cancel-in-progress: false
+env:
+  OMP_NUM_THREADS: '1'
+  OPENBLAS_NUM_THREADS: '1'
+jobs:
+  suite:
+    name: Prospective ITC suite (${{ matrix.runner }}, Python ${{ matrix.python }})
+    runs-on: ${{ matrix.runner }}
+    timeout-minutes: 60
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - {runner: ubuntu-24.04, python: '3.11'}
+          - {runner: ubuntu-24.04, python: '3.13'}
+          - {runner: ubuntu-24.04-arm, python: '3.13'}
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+      - uses: actions/setup-python@v5
+        with:
+          python-version: ${{ matrix.python }}
+      - run: python -m pip install 'build==1.3.0' 'pytest==8.4.2'
+      - uses: actions/checkout@v4
+        with:
+          repository: davidvanbulck/ITC-2007-CETT
+          ref: ed2d4c0cf5302b77a8b321cd0380de0bb579d538
+          path: upstream/ITC-2007-CETT
+          persist-credentials: false
+      - name: Validate selector and suite contracts
+        run: |
+          set -euo pipefail
+          python -m pytest \
+            experiments/compiled_query/test_exact.py \
+            experiments/compiled_query/test_itc_exam_transfer.py \
+            experiments/compiled_query/test_itc_exam_suite.py \
+            tests/public/test_exam_repair.py \
+            tests/public/test_itc2007_exam.py \
+            -q -k 'auto or tiled_witnesses or itc or exam_repair or suite' \
+            --junitxml=suite-contracts.xml
+      - name: Build and exercise the installed application API
+        run: |
+          set -euo pipefail
+          python -m build --wheel --outdir suite-wheel
+          python -m venv suite-install
+          suite-install/bin/python -m pip install --no-deps suite-wheel/*.whl
+          suite-install/bin/python -I scripts/check_exam_repair_installation.py \
+            --out installed-exam-repair.json | tee installed-exam-repair.log
+      - name: Freeze all roles before any holdout execution
+        run: |
+          set -euo pipefail
+          python -m experiments.compiled_query.itc_exam_suite freeze \
+            --root itc-suite-evidence \
+            --upstream-root upstream/ITC-2007-CETT \
+            --queries 1024 \
+            --holdout-max-auto-raw-ratio 1.15 | tee freeze.log
+      - name: Execute and independently audit every case
+        run: |
+          set -euo pipefail
+          python -m experiments.compiled_query.itc_exam_suite run \
+            --root itc-suite-evidence | tee run.log
+          python -S -m experiments.compiled_query.itc_exam_suite audit \
+            --root itc-suite-evidence | tee audit.log
+          python - <<'PY'
+          import json
+          from pathlib import Path
+          result = json.loads(Path('itc-suite-evidence/suite-audit.json').read_text())
+          assert result['functional_complete']
+          assert result['holdout_gate_pass'], result['holdout_failures']
+          roles = [case['role'] for case in result['cases']]
+          assert roles == ['development', 'holdout', 'holdout']
+          for case in result['cases']:
+              assert case['queries'] == 1024
+              assert case['sat'] + case['unsat'] == 1024
+              assert case['auto_vs_raw_paired_median_ratio'] > 0
+          print('Prospective holdout guardrail passed. Not a full timetabling claim.')
+          PY
+      - name: Preserve exact environment and source receipts
+        if: always()
+        run: |
+          set -euo pipefail
+          mkdir -p suite-receipts
+          git rev-parse HEAD > suite-receipts/source_commit.txt
+          git rev-parse HEAD^{tree} > suite-receipts/source_tree.txt
+          git -C upstream/ITC-2007-CETT rev-parse HEAD > suite-receipts/upstream_commit.txt
+          python -VV > suite-receipts/python.txt
+          uname -a > suite-receipts/platform.txt
+          lscpu > suite-receipts/cpu.txt
+          cp suite-contracts.xml freeze.log run.log audit.log \
+            installed-exam-repair.json installed-exam-repair.log \
+            suite-receipts/ 2>/dev/null || true
+      - uses: actions/upload-artifact@v4
+        if: always()
+        with:
+          name: itc-exam-suite-${{ matrix.runner }}-py${{ matrix.python }}-${{ github.run_id }}
+          path: |
+            itc-suite-evidence/
+            suite-receipts/
+            suite-wheel/
+          retention-days: 90
+          if-no-files-found: warn

From 9b14a35d7744896b8b86b4014d825b9f082d8294 Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:56:38 -0400
Subject: [PATCH 15/24] Link the proof-carrying exam-repair application guide

---
 docs/README.md | 1 +
 1 file changed, 1 insertion(+)

diff --git a/docs/README.md b/docs/README.md
index f11fd18..96d2374 100644
--- a/docs/README.md
+++ b/docs/README.md
@@ -24,6 +24,7 @@ historical neural, MCTS, low-bit, general-SAT, or universal CPU-performance clai
 | Install and run the public tools | [Quick start](../README.md#base-toolkit-quick-start) |
 | Understand API, input formats and return values | [API and CLI](API.md) |
 | Build and use checked SAT/UNSAT exact queries | [Proof-carrying exact API](EXACT_API.md) |
+| Apply the exact core to bounded exam rescheduling | [Proof-carrying exam repair](EXAM_REPAIR.md) |
 | Check released, merged, experimental and failed work | [Current status](STATUS.md) |
 | Run tests, build a wheel or reproduce experiments | [Development](DEVELOPMENT.md) |
 | Prepare a package release | [Release checklist](RELEASING.md) |

From 76ba450275a3b4ebba0fc1eef715fe97a90ed44b Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:56:54 -0400
Subject: [PATCH 16/24] Extend installed-package validation to exam repair

---
 .github/workflows/public-package.yml | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/public-package.yml b/.github/workflows/public-package.yml
index 4efdf81..9b60760 100644
--- a/.github/workflows/public-package.yml
+++ b/.github/workflows/public-package.yml
@@ -1,7 +1,7 @@
 name: public-package
 on:
   pull_request:
-    branches: [main, "maintenance/**", "research/deductive-search-*", "research/focused-search-*", "research/research-engineering-*", "research/faster-queries-review-*"]
+    branches: [main, "maintenance/**", "research/deductive-search-*", "research/focused-search-*", "research/research-engineering-*", "research/faster-queries-review-*", "research/exact-public-api-*"]
   workflow_dispatch:
 permissions:
   contents: read
@@ -48,6 +48,7 @@ jobs:
           .public-package/install/venv/bin/python -I scripts/check_focused_installation.py | tee .public-package/focused-install.txt
           .public-package/install/venv/bin/python -I scripts/check_dpll_installation.py | tee .public-package/dpll-install.txt
           .public-package/install/venv/bin/python -I scripts/check_exact_installation.py --out .public-package/exact-install.json | tee .public-package/exact-install.txt
+          .public-package/install/venv/bin/python -I scripts/check_exam_repair_installation.py --out .public-package/exam-repair-install.json | tee .public-package/exam-repair-install.txt
       - name: Run and replay frozen compact-state comparison
         run: |
           set -o pipefail

From 6390470d5b26b77c13a932e2c5e74576bb4a1a7e Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:57:02 -0400
Subject: [PATCH 17/24] Remove one-shot blitz publication workflow

---
 .github/workflows/publish-100-blitz.yml | 81 -------------------------
 1 file changed, 81 deletions(-)
 delete mode 100644 .github/workflows/publish-100-blitz.yml

diff --git a/.github/workflows/publish-100-blitz.yml b/.github/workflows/publish-100-blitz.yml
deleted file mode 100644
index f2d5dd8..0000000
--- a/.github/workflows/publish-100-blitz.yml
+++ /dev/null
@@ -1,81 +0,0 @@
-name: publish-100-blitz
-on:
-  push:
-    branches: [research/100-blitz-20261016]
-    paths: ['.github/workflows/publish-100-blitz.yml']
-permissions:
-  contents: write
-concurrency:
-  group: publish-100-blitz
-  cancel-in-progress: false
-jobs:
-  publish:
-    runs-on: ubuntu-24.04
-    timeout-minutes: 10
-    steps:
-      - uses: actions/checkout@v4
-        with:
-          fetch-depth: 0
-      - name: Verify and reconstruct reviewed patch
-        env:
-          EXPECTED_B64: f75650c04f59e0ae6e02d2190ebd56668442d790276d5cc9fa3dc11d188108a7
-          EXPECTED_XZ: 31780857b8f024c4cb4f1144d04f8e8fdebf8b902ab1acd7c293673aacaef036
-          EXPECTED_PATCH: dcd47a5a9a621f9f20d91cf040cc125c5300089b699dbacfbd25a2be84e83ab6
-        run: |
-          set -euo pipefail
-          mkdir -p /tmp/blitz
-          cat .delivery/100-blitz-source/{00,01,02,03,04,05,06,07,08,09} > /tmp/blitz/patch.xz.b64
-          test "$(sha256sum /tmp/blitz/patch.xz.b64 | cut -d' ' -f1)" = "$EXPECTED_B64"
-          base64 --decode /tmp/blitz/patch.xz.b64 > /tmp/blitz/patch.xz
-          test "$(sha256sum /tmp/blitz/patch.xz | cut -d' ' -f1)" = "$EXPECTED_XZ"
-          xz --decompress --stdout /tmp/blitz/patch.xz > /tmp/blitz/reviewed.patch
-          test "$(sha256sum /tmp/blitz/reviewed.patch | cut -d' ' -f1)" = "$EXPECTED_PATCH"
-      - name: Validate base identities and publish atomically
-        run: |
-          set -euo pipefail
-          test -z "$(git status --porcelain)"
-          test "$(git rev-parse HEAD:spectra/exact.py)" = 983d33551e29d13016e36cd6038d02d270575b66
-          test "$(git rev-parse HEAD:spectra/_native/exact_binary.cpp)" = 96182ee6c7754d24e1c8bc46181dd69a5732d32f
-          test "$(git rev-parse HEAD:experiments/compiled_query/exact.py)" = 780a7ff1d8f293450fb592ae9e29e129e92a551c
-          test "$(git rev-parse HEAD:experiments/compiled_query/exact_native.cpp)" = 96182ee6c7754d24e1c8bc46181dd69a5732d32f
-          git apply --check --index /tmp/blitz/reviewed.patch
-          git apply --index /tmp/blitz/reviewed.patch
-          git diff --cached --name-only -z > /tmp/blitz/changed.paths
-          python - <<'PY'
-          from pathlib import Path
-          expected = {
-              'docs/EXAM_REPAIR.md',
-              'experiments/compiled_query/ITC_EXAM_SUITE_PROTOCOL.md',
-              'experiments/compiled_query/ITC_EXAM_TRANSFER_PROTOCOL.md',
-              'experiments/compiled_query/exact.py',
-              'experiments/compiled_query/exact_native.cpp',
-              'experiments/compiled_query/itc_exam_suite.py',
-              'experiments/compiled_query/itc_exam_transfer.py',
-              'experiments/compiled_query/test_exact.py',
-              'experiments/compiled_query/test_itc_exam_suite.py',
-              'experiments/compiled_query/test_itc_exam_transfer.py',
-              'scripts/check_exam_repair_installation.py',
-              'spectra/_native/exact_binary.cpp',
-              'spectra/exact.py',
-              'spectra/exam_repair.py',
-              'spectra/itc2007_exam.py',
-              'tests/public/test_exam_repair.py',
-              'tests/public/test_itc2007_exam.py',
-          }
-          changed = {
-              item.decode('utf-8')
-              for item in Path('/tmp/blitz/changed.paths').read_bytes().split(b'\0')
-              if item
-          }
-          if changed != expected:
-              raise RuntimeError(f'changed path set differs: {sorted(changed ^ expected)}')
-          PY
-          git rm -r .delivery/100-blitz-source
-          if git diff --cached --name-only | grep -q '^\.github/workflows/'; then
-            echo 'Workflow changes are not permitted in the reconstructed code commit.' >&2
-            exit 1
-          fi
-          git config user.name 'Tugrap'
-          git config user.email '106766396+tugrapaydiner@users.noreply.github.com'
-          git commit -m 'Publish proof-carrying exam repair and prospective ITC suite'
-          git push origin HEAD:research/100-blitz-20261016

From 5b7ce2de817a855bf2bce39c32ba88d0f4149a5b Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:58:00 -0400
Subject: [PATCH 18/24] Isolate dependency-light public exam contracts in CI

---
 .github/workflows/itc-exam-transfer.yml | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/itc-exam-transfer.yml b/.github/workflows/itc-exam-transfer.yml
index 4ac558a..1059345 100644
--- a/.github/workflows/itc-exam-transfer.yml
+++ b/.github/workflows/itc-exam-transfer.yml
@@ -59,8 +59,11 @@ jobs:
           python -m pytest \
             tests/public/test_exam_repair.py \
             tests/public/test_itc2007_exam.py \
+            --confcutdir=tests/public -q \
+            --junitxml=transfer-public-contracts.xml
+          python -m pytest \
             experiments/compiled_query/test_itc_exam_transfer.py \
-            -q --junitxml=transfer-contracts.xml
+            -q --junitxml=transfer-research-contracts.xml
       - name: Freeze the published instance only after the protocol is committed
         run: |
           set -euo pipefail
@@ -102,7 +105,8 @@ jobs:
           python -VV > transfer-receipts/python.txt
           uname -a > transfer-receipts/platform.txt
           lscpu > transfer-receipts/cpu.txt
-          cp transfer-contracts.xml freeze.log run.log audit.log transfer-receipts/ 2>/dev/null || true
+          cp transfer-public-contracts.xml transfer-research-contracts.xml \
+            freeze.log run.log audit.log transfer-receipts/ 2>/dev/null || true
       - uses: actions/upload-artifact@v4
         if: always()
         with:

From ff77d802e9a2bb3c1d97c5acf70d8871dc114f7c Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:58:13 -0400
Subject: [PATCH 19/24] Isolate public and research contracts before ITC
 holdouts

---
 .github/workflows/itc-exam-suite.yml | 17 ++++++++++-------
 1 file changed, 10 insertions(+), 7 deletions(-)

diff --git a/.github/workflows/itc-exam-suite.yml b/.github/workflows/itc-exam-suite.yml
index ef6a8bc..6fa53ff 100644
--- a/.github/workflows/itc-exam-suite.yml
+++ b/.github/workflows/itc-exam-suite.yml
@@ -56,14 +56,17 @@ jobs:
       - name: Validate selector and suite contracts
         run: |
           set -euo pipefail
+          python -m pytest \
+            tests/public/test_exam_repair.py \
+            tests/public/test_itc2007_exam.py \
+            --confcutdir=tests/public -q \
+            --junitxml=suite-public-contracts.xml
           python -m pytest \
             experiments/compiled_query/test_exact.py \
             experiments/compiled_query/test_itc_exam_transfer.py \
             experiments/compiled_query/test_itc_exam_suite.py \
-            tests/public/test_exam_repair.py \
-            tests/public/test_itc2007_exam.py \
-            -q -k 'auto or tiled_witnesses or itc or exam_repair or suite' \
-            --junitxml=suite-contracts.xml
+            -q -k 'auto or tiled_witnesses or itc or suite' \
+            --junitxml=suite-research-contracts.xml
       - name: Build and exercise the installed application API
         run: |
           set -euo pipefail
@@ -112,9 +115,9 @@ jobs:
           python -VV > suite-receipts/python.txt
           uname -a > suite-receipts/platform.txt
           lscpu > suite-receipts/cpu.txt
-          cp suite-contracts.xml freeze.log run.log audit.log \
-            installed-exam-repair.json installed-exam-repair.log \
-            suite-receipts/ 2>/dev/null || true
+          cp suite-public-contracts.xml suite-research-contracts.xml \
+            freeze.log run.log audit.log installed-exam-repair.json \
+            installed-exam-repair.log suite-receipts/ 2>/dev/null || true
       - uses: actions/upload-artifact@v4
         if: always()
         with:

From 50a8e92c048c1e2b09f82ba8ea65dfd1c295acd8 Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:59:19 -0400
Subject: [PATCH 20/24] Add an independent ITC exam-repair replication guide

---
 .../compiled_query/ITC_EXAM_REPLICATION.md    | 116 ++++++++++++++++++
 1 file changed, 116 insertions(+)
 create mode 100644 experiments/compiled_query/ITC_EXAM_REPLICATION.md

diff --git a/experiments/compiled_query/ITC_EXAM_REPLICATION.md b/experiments/compiled_query/ITC_EXAM_REPLICATION.md
new file mode 100644
index 0000000..3890d53
--- /dev/null
+++ b/experiments/compiled_query/ITC_EXAM_REPLICATION.md
@@ -0,0 +1,116 @@
+# Independent replication: proof-carrying ITC exam repair
+
+This guide is for an outside runner who has not changed the selector, event bank,
+acceptance threshold, or result interpretation. A successful run reproduces a
+bounded exam-repair transfer. It does **not** establish full ITC-2007 timetabling,
+operator adoption, universal performance, or algorithmic priority.
+
+## 1. Pin both repositories
+
+Record the exact SPECTRA commit under review and use the fixed upstream revision:
+
+```bash
+git clone https://github.com/tugrapaydiner/SPECTRA.git
+cd SPECTRA
+git checkout <exact-spectra-commit>
+
+git clone https://github.com/davidvanbulck/ITC-2007-CETT.git upstream/ITC-2007-CETT
+git -C upstream/ITC-2007-CETT checkout ed2d4c0cf5302b77a8b321cd0380de0bb579d538
+```
+
+Do not substitute newer upstream bytes. Preserve the upstream `LICENSE` in every
+shared evidence package.
+
+## 2. Record the environment before execution
+
+```bash
+mkdir -p replication-receipts
+git rev-parse HEAD > replication-receipts/spectra-commit.txt
+git rev-parse HEAD^{tree} > replication-receipts/spectra-tree.txt
+git -C upstream/ITC-2007-CETT rev-parse HEAD > replication-receipts/upstream-commit.txt
+python -VV > replication-receipts/python.txt
+uname -a > replication-receipts/platform.txt
+lscpu > replication-receipts/cpu.txt 2>/dev/null || true
+c++ --version > replication-receipts/compiler.txt 2>&1
+```
+
+Use a GIL-enabled CPython 3.11 or 3.13 and a C++17 compiler. Do not install or load
+an alternative solver for the measured calls.
+
+## 3. Check source and installed-package contracts
+
+```bash
+python -m pip install 'build==1.3.0' 'pytest==8.4.2'
+python -m pytest \
+  tests/public/test_exam_repair.py \
+  tests/public/test_itc2007_exam.py \
+  --confcutdir=tests/public -q
+python -m pytest \
+  experiments/compiled_query/test_exact.py \
+  experiments/compiled_query/test_itc_exam_transfer.py \
+  experiments/compiled_query/test_itc_exam_suite.py \
+  -q -k 'auto or tiled_witnesses or itc or suite'
+
+python -m build --wheel --outdir replication-wheel
+python -m venv replication-install
+replication-install/bin/python -m pip install --no-deps replication-wheel/*.whl
+replication-install/bin/python -I scripts/check_exam_repair_installation.py \
+  --out replication-receipts/installed-exam-repair.json
+```
+
+On Windows, use `replication-install/Scripts/python.exe` and a configured MSVC x64
+developer environment.
+
+## 4. Freeze before running any case
+
+The freeze must complete before inspecting any holdout answer or timing:
+
+```bash
+python -m experiments.compiled_query.itc_exam_suite freeze \
+  --root itc-suite-evidence \
+  --upstream-root upstream/ITC-2007-CETT \
+  --queries 1024 \
+  --holdout-max-auto-raw-ratio 1.15
+```
+
+Confirm that `suite-manifest.json` records, in order:
+
+1. `set9-development` with role `development`;
+2. `set10-holdout` with role `holdout`;
+3. `set12-holdout` with role `holdout`.
+
+Do not change the roles, query count, five-repetition schedule, selector-v2 rule or
+1.15 paired-median guardrail after this point.
+
+## 5. Execute once and audit without native code
+
+```bash
+python -m experiments.compiled_query.itc_exam_suite run \
+  --root itc-suite-evidence
+python -S -m experiments.compiled_query.itc_exam_suite audit \
+  --root itc-suite-evidence
+```
+
+The audit must independently regenerate every derived input, replay every `SCQ2`
+answer from the original `SBC1` packets, require cross-engine decision agreement,
+and publish all holdout failures. Do not delete negative cells or repeat a failed
+measurement under a changed rule.
+
+## 6. Report every result
+
+Publish the complete `itc-suite-evidence` directory plus the receipts above. At a
+minimum, report:
+
+- SPECTRA commit and tree;
+- upstream commit;
+- OS, CPU, Python and compiler;
+- all source/public/installed contract outcomes;
+- each case's role, exam count, conflict count, query count, SAT/UNSAT counts,
+  selected automatic engine and policy reason;
+- all five paired `auto_elapsed / raw_elapsed` ratios per holdout;
+- paired-median ratio and pass/fail against 1.15;
+- every execution or audit failure, including resource failures.
+
+A useful independent report explicitly says whether the bytes and conclusions
+match the project-owned CI artifact. A disagreement is a valid replication result
+and should not be hidden or relabelled as an environment problem without evidence.

From d294631d8063438d42e318b5167fc6e3808eff9f Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 14:59:30 -0400
Subject: [PATCH 21/24] Link the independent exam-repair replication procedure

---
 docs/EXAM_REPAIR.md | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/docs/EXAM_REPAIR.md b/docs/EXAM_REPAIR.md
index 304b0b3..340333b 100644
--- a/docs/EXAM_REPAIR.md
+++ b/docs/EXAM_REPAIR.md
@@ -60,3 +60,8 @@ The suite rotates measured engine order, preserves every failed cell, and gates
 only the predeclared auto/raw overhead on holdout cases.  Passing that suite is
 application-transfer evidence, not an operator deployment or full ITC-2007
 optimization result.
+
+An outside runner can reproduce the source, installed-wheel, freeze, execution and
+offline-audit boundaries using the
+[independent replication guide](../experiments/compiled_query/ITC_EXAM_REPLICATION.md).
+Negative or conflicting replications are explicitly valid results.

From ed011fc2280b1da05a6e0fc8ecfd7639cf2a735e Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 15:03:38 -0400
Subject: [PATCH 22/24] Publish prospective multi-instance ITC exam-repair
 results

---
 .../ITC_EXAM_SUITE_RESULTS_20261016.md        | 164 ++++++++++++++++++
 1 file changed, 164 insertions(+)
 create mode 100644 experiments/compiled_query/ITC_EXAM_SUITE_RESULTS_20261016.md

diff --git a/experiments/compiled_query/ITC_EXAM_SUITE_RESULTS_20261016.md b/experiments/compiled_query/ITC_EXAM_SUITE_RESULTS_20261016.md
new file mode 100644
index 0000000..592fa15
--- /dev/null
+++ b/experiments/compiled_query/ITC_EXAM_SUITE_RESULTS_20261016.md
@@ -0,0 +1,164 @@
+# Prospective ITC exam-repair suite — 2026-10-16
+
+## Result
+
+The prospectively fixed bounded exam-repair suite completed on all three declared
+environments. Both execution holdouts passed the unchanged paired-median
+automatic/raw guardrail of **1.15**. Every admitted result was independently
+replayed from the original pinned ITC bytes, and no failed cell or output was
+removed.
+
+This is application-transfer evidence for the binary-domain exam-conflict core. It
+is **not** a full ITC-2007 timetabling result, an operator deployment, outside-team
+replication, universal performance evidence, or an algorithmic-priority claim.
+
+## Frozen boundary
+
+| Item | Value |
+|---|---|
+| Measured SPECTRA commit | `ff77d802e9a2bb3c1d97c5acf70d8871dc114f7c` |
+| Published implementation commit | `3036a3c97cf5706d60ab9e277708081b4f241954` |
+| Upstream ITC repository commit | `ed2d4c0cf5302b77a8b321cd0380de0bb579d538` |
+| Selector policy | `structural-v2` |
+| Requests per case | 1,024 |
+| Warmup | one untimed call per mode |
+| Measured repetitions | five |
+| Mode order | rotated each repetition |
+| Holdout gate | paired median of `auto_elapsed / raw_elapsed <= 1.15` |
+| Development case | `set9` |
+| Execution holdouts | `set10`, `set12` |
+
+The implementation, policy, case roles, event generator, repetition count and gate
+were committed before any holdout execution. The automatic engine used packet
+structure only. All three cases triggered `high_pin_density` and selected `raw`.
+
+## Functional outcomes
+
+The decisions and counts were identical on x86 Python 3.11, x86 Python 3.13 and
+native ARM64 Python 3.13.
+
+| Case | Role | Exams | Periods | Enrolments | Conflicts | Explicit exclusions | SAT | UNSAT | Pins |
+|---|---|---:|---:|---:|---:|---:|---:|---:|---:|
+| `set9` | development | 169 | 25 | 2,532 | 1,114 | 1 | 268 | 756 | 9,698 |
+| `set10` | holdout | 214 | 32 | 7,853 | 1,133 | 0 | 327 | 697 | 9,311 |
+| `set12` | holdout | 78 | 12 | 3,685 | 560 | 6 | 24 | 1,000 | 10,361 |
+
+For `set10`, 178 UNSAT answers used an empty-domain proof and 519 used a
+forced-opposites path proof. For `set12`, those counts were 269 and 731. The native
+engines agreed on every SAT/UNSAT decision, and the standard-library audit checked
+every complete assignment or contradiction proof against the original fixed graph
+and current restrictions.
+
+## Prospective guardrail
+
+Values below are the median of the five paired repetition ratios, not a ratio of
+independently aggregated medians.
+
+| Environment | `set9` development | `set10` holdout | `set12` holdout | Holdout gate |
+|---|---:|---:|---:|---|
+| Linux x86, Python 3.11 | 0.995442 | **0.999355** | **1.003860** | PASS |
+| Linux x86, Python 3.13 | 1.073799 | **1.019084** | **1.026623** | PASS |
+| Linux ARM64, Python 3.13 | 1.017422 | **0.999036** | **1.000705** | PASS |
+
+The largest holdout paired-median ratio was **1.026623**, leaving approximately
+10.7 percentage points of margin to the fixed 1.15 limit.
+
+Individual repetitions were intentionally retained. In particular, the first
+`set12` repetition on x86 Python 3.13 had ratio **1.168387**, and the first
+`set10` repetition on x86 Python 3.11 had ratio **1.158769**. Both exceed the gate
+when viewed alone; the predeclared gate is the paired median over five rotating
+repetitions, so these are negative retained observations rather than deleted or
+retried cells.
+
+### Paired ratios, all repetitions
+
+| Environment | Case | Five `auto/raw` ratios |
+|---|---|---|
+| x86 / Python 3.11 | `set10` | 1.158769, 0.999355, 1.000486, 0.984425, 0.995438 |
+| x86 / Python 3.11 | `set12` | 1.123582, 1.004558, 1.000192, 1.003860, 0.997922 |
+| x86 / Python 3.13 | `set10` | 1.098272, 1.066995, 1.018905, 0.998810, 1.019084 |
+| x86 / Python 3.13 | `set12` | 1.168387, 1.012663, 1.026623, 0.978225, 1.040015 |
+| ARM64 / Python 3.13 | `set10` | 1.112409, 0.999036, 0.987604, 1.004978, 0.985442 |
+| ARM64 / Python 3.13 | `set12` | 1.099958, 0.985422, 0.994436, 1.000705, 1.030142 |
+
+The automatic and explicit raw modes execute the same native engine after policy
+selection. Ratios close to one therefore measure selection/dispatch noise and host
+variation rather than an algorithmic speedup. The purpose of this gate is to stop
+the automatic policy from imposing material overhead on unseen application cases.
+
+## Complete-call medians
+
+Elapsed time includes native construction, all 1,024 decisions, complete witnesses
+or portable proofs, independent native checking, `SCQ2` construction, disposal and
+hashing. Candidate generation, packet creation, compilation and process startup are
+outside the measured call.
+
+| Environment | Case | Auto | Raw | SCC | Route |
+|---|---|---:|---:|---:|---:|
+| x86 / Python 3.11 | `set9` | 1.286 ms | 1.292 ms | 1.288 ms | 1.657 ms |
+| x86 / Python 3.11 | `set10` | 1.163 ms | 1.161 ms | 1.144 ms | 1.358 ms |
+| x86 / Python 3.11 | `set12` | 0.888 ms | 0.887 ms | 0.987 ms | 1.199 ms |
+| x86 / Python 3.13 | `set9` | 0.823 ms | 0.765 ms | 0.788 ms | 0.952 ms |
+| x86 / Python 3.13 | `set10` | 0.697 ms | 0.666 ms | 0.693 ms | 0.795 ms |
+| x86 / Python 3.13 | `set12` | 0.536 ms | 0.523 ms | 0.605 ms | 0.725 ms |
+| ARM64 / Python 3.13 | `set9` | 1.114 ms | 1.095 ms | 1.052 ms | 1.402 ms |
+| ARM64 / Python 3.13 | `set10` | 1.004 ms | 1.004 ms | 0.972 ms | 1.172 ms |
+| ARM64 / Python 3.13 | `set12` | 0.772 ms | 0.771 ms | 0.824 ms | 1.050 ms |
+
+SCC remains slightly faster than raw on some cells. Selector-v2 deliberately
+prioritises a conservative no-material-overhead rule for heavily pinned application
+banks rather than claiming a per-cell optimum. The development result does not
+justify retuning after seeing the holdouts.
+
+## CI and artifacts
+
+### Development transfer
+
+Workflow run: `38951373279`
+
+- Linux x86 / Python 3.11: PASS
+- Linux x86 / Python 3.13: PASS
+- Native Linux ARM64 / Python 3.13: PASS
+
+Artifacts:
+
+| Environment | Artifact ID | Artifact SHA-256 |
+|---|---:|---|
+| x86 / Python 3.11 | `12068467826` | `3afc63b70f8499c03fe97df78eff288c8316accaf714bace3585be938cab1fe4` |
+| x86 / Python 3.13 | `12068077883` | `32823ece7589b6f76901469c141210ca3796d489f8a90eaa29aaac0e933d69d1` |
+| ARM64 / Python 3.13 | `12068222627` | `d5de0a96b0a8347df58296e6bad01e4cd24e49cb908e952c47056609f2acf762` |
+
+### Prospective suite
+
+Workflow run: `38951398495`
+
+- Linux x86 / Python 3.11: PASS
+- Linux x86 / Python 3.13: PASS
+- Native Linux ARM64 / Python 3.13: PASS
+- Source contracts: PASS on all cells
+- Installed-wheel exam-repair execution: PASS on all cells
+- Freeze-before-execution: PASS on all cells
+- Independent source-only audit: PASS on all cells
+
+Artifacts:
+
+| Environment | Artifact ID | Artifact SHA-256 |
+|---|---:|---|
+| x86 / Python 3.11 | `12068498169` | `fb100a67983b02c7969809efb680deaf5b4002c6f2242122d8cca8ff4a415c23` |
+| x86 / Python 3.13 | `12068552843` | `8f86935fa6e6811c56fc4156d8dc9ea422c4a14b5ef0cba356b496d3ef4208e7` |
+| ARM64 / Python 3.13 | `12068023095` | `35247cef687fc26ccbfec69d6903ed421a4a3fc9df8682b7143afa0600b055f4` |
+
+## Preserved failed attempts
+
+The first transfer run (`38951148893`) and first suite run (`38951172493`) failed
+during test collection because the dependency-light public-test command loaded the
+repository-wide ML `tests/conftest.py` without NumPy. They stopped before evidence
+freezing or any holdout execution. The correction isolated public tests with
+`--confcutdir=tests/public`; it did not change the selector, cases, event bank,
+repetitions, timing order or guardrail. The failed runs remain visible.
+
+## Replication
+
+Follow [`ITC_EXAM_REPLICATION.md`](ITC_EXAM_REPLICATION.md) from a clean checkout.
+An independent disagreement or negative result is valid evidence and should be
+published with the complete artifacts rather than normalised away.

From 008e6d68290e96706b586b8a33634f507f73a66d Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 15:03:45 -0400
Subject: [PATCH 23/24] Link the completed prospective ITC suite results

---
 docs/EXAM_REPAIR.md | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/docs/EXAM_REPAIR.md b/docs/EXAM_REPAIR.md
index 340333b..812b150 100644
--- a/docs/EXAM_REPAIR.md
+++ b/docs/EXAM_REPAIR.md
@@ -56,6 +56,9 @@ The research transfer protocol is documented in
 
 The selector-v2 policy and its `set10`/`set12` execution holdouts are fixed in
 [`ITC_EXAM_SUITE_PROTOCOL.md`](../experiments/compiled_query/ITC_EXAM_SUITE_PROTOCOL.md).
+The completed cross-platform measurements, retained negative repetitions and
+artifact digests are published in the
+[`2026-10-16 suite results`](../experiments/compiled_query/ITC_EXAM_SUITE_RESULTS_20261016.md).
 The suite rotates measured engine order, preserves every failed cell, and gates
 only the predeclared auto/raw overhead on holdout cases.  Passing that suite is
 application-transfer evidence, not an operator deployment or full ITC-2007

From 13d1d4192e97a4e372ac4aac7743a88b55e814bc Mon Sep 17 00:00:00 2001
From: Tugrap <106766396+tugrapaydiner@users.noreply.github.com>
Date: Fri, 16 Oct 2026 15:05:34 -0400
Subject: [PATCH 24/24] Add self-verifying recovery artifacts for the 100-point
 blitz checkpoint

---
 .github/workflows/checkpoint-recovery.yml | 59 +++++++++++++++++++++++
 1 file changed, 59 insertions(+)
 create mode 100644 .github/workflows/checkpoint-recovery.yml

diff --git a/.github/workflows/checkpoint-recovery.yml b/.github/workflows/checkpoint-recovery.yml
new file mode 100644
index 0000000..20e9302
--- /dev/null
+++ b/.github/workflows/checkpoint-recovery.yml
@@ -0,0 +1,59 @@
+name: checkpoint-recovery
+on:
+  push:
+    branches: [research/100-blitz-20261016]
+    paths: ['.github/workflows/checkpoint-recovery.yml']
+  workflow_dispatch:
+permissions:
+  contents: read
+concurrency:
+  group: checkpoint-recovery-${{ github.ref }}
+  cancel-in-progress: false
+jobs:
+  recover:
+    runs-on: ubuntu-24.04
+    timeout-minutes: 15
+    steps:
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+          persist-credentials: false
+      - name: Verify repository and create exact recovery materials
+        env:
+          REVIEW_BASE: 575eb1e657fc8c5d06fd56a519efd59f95dc0a2d
+        run: |
+          set -euo pipefail
+          mkdir -p checkpoint-recovery
+          git merge-base --is-ancestor "$REVIEW_BASE" HEAD
+          git diff --check "$REVIEW_BASE"..HEAD
+          git fsck --strict
+          git rev-parse HEAD > checkpoint-recovery/source_commit.txt
+          git rev-parse HEAD^{tree} > checkpoint-recovery/source_tree.txt
+          printf '%s\n' "$REVIEW_BASE" > checkpoint-recovery/review_base.txt
+          git log --format='%H %T %s' "$REVIEW_BASE"..HEAD > checkpoint-recovery/commit_ledger.txt
+          git diff --binary "$REVIEW_BASE"..HEAD > checkpoint-recovery/review.patch
+          git format-patch --stdout --binary "$REVIEW_BASE"..HEAD > checkpoint-recovery/review-series.patch
+          git archive --format=tar.gz --output=checkpoint-recovery/source.tar.gz HEAD
+          git bundle create checkpoint-recovery/source.bundle HEAD
+          git bundle verify checkpoint-recovery/source.bundle > checkpoint-recovery/bundle-verify.txt 2>&1
+          python - <<'PY'
+          import hashlib, json
+          from pathlib import Path
+          root = Path('checkpoint-recovery')
+          files = {}
+          for path in sorted(root.iterdir()):
+              if path.is_file() and path.name != 'SHA256SUMS.json':
+                  files[path.name] = {
+                      'bytes': path.stat().st_size,
+                      'sha256': hashlib.sha256(path.read_bytes()).hexdigest(),
+                  }
+          (root / 'SHA256SUMS.json').write_text(
+              json.dumps(files, indent=2, sort_keys=True) + '\n', encoding='utf-8')
+          PY
+      - uses: actions/upload-artifact@v4
+        with:
+          name: spectra-100-blitz-recovery-${{ github.sha }}
+          path: checkpoint-recovery/
+          include-hidden-files: true
+          retention-days: 90
+          if-no-files-found: error