Restore judged logbook 38daa951 (revert stale 08:0x auto-sync that hid detailed claim pages)

README.md

@@ -1,5 +1,5 @@
 ---
-title: "Repro - LOGDIFF Exact Boolean Guidance"
+title: "Reproduction: Logical Guidance for the Exact Composition of Diffusion Models"
 emoji: 🎯
 colorFrom: yellow
 colorTo: red
@@ -13,6 +13,6 @@ tags:
  - paper-OAM1jJsMGp
 ---
 
-# Repro - LOGDIFF Exact Boolean Guidance
+# Reproduction: Logical Guidance for the Exact Composition of Diffusion Models
 
 An open experiment logbook, published with [Trackio](https://github.com/gradio-app/trackio).

index.html

@@ -3,7 +3,7 @@
   <head>
     <meta charset="utf-8" />
     <meta name="viewport" content="width=device-width, initial-scale=1" />
-    <title>Repro - LOGDIFF Exact Boolean Guidance</title>
+    <title>Reproduction: Logical Guidance for the Exact Composition of Diffusion Models</title>
     <link rel="stylesheet" href="./logbook.css" />
   </head>
   <body>

logbook.json

@@ -1,6 +1,6 @@
 {
   "schema_version": 1,
-  "title": "Repro - LOGDIFF Exact Boolean Guidance",
+  "title": "Reproduction: Logical Guidance for the Exact Composition of Diffusion Models",
   "emoji": "🎯",
   "space_id": "DineshAI/OAM1jJsMGp",
   "paper": {
@@ -11,12 +11,18 @@
     "icml2026-repro",
     "paper-OAM1jJsMGp"
   ],
-  "updated_at": "2026-07-21T08:08:16+00:00",
+  "updated_at": "2026-07-20T01:05:44+00:00",
   "root": {
     "slug": "index",
-    "title": "Repro - LOGDIFF Exact Boolean Guidance",
+    "title": "Reproduction: Logical Guidance for the Exact Composition of Diffusion Models",
     "file": "pages/index.md",
     "children": [
+      {
+        "slug": "executive-summary",
+        "title": "Executive summary",
+        "file": "pages/executive-summary/page.md",
+        "children": []
+      },
       {
         "slug": "claim-1-exact-logical-guidance",
         "title": "Claim 1 — exact logical guidance",
@@ -24,21 +30,51 @@
         "children": []
       },
       {
-        "slug": "claim-2-sufficient-conditions",
+        "slug": "claim-1-sufficient-conditions",
         "title": "Claim 2 — sufficient conditions",
-        "file": "pages/claim-2-sufficient-conditions/page.md",
+        "file": "pages/claim-1-sufficient-conditions/page.md",
+        "children": []
+      },
+      {
+        "slug": "claim-2-paper-range-audit",
+        "title": "Claim 2 - paper-range audit",
+        "file": "pages/claim-2-paper-range-audit/page.md",
+        "children": []
+      },
+      {
+        "slug": "claim-3-celeba-release-audit",
+        "title": "Claim 3 - CelebA release audit",
+        "file": "pages/claim-3-celeba-release-audit/page.md",
+        "children": []
+      },
+      {
+        "slug": "claim-4-recovered-molecular-pipeline",
+        "title": "Claim 4 - recovered molecular pipeline",
+        "file": "pages/claim-4-recovered-molecular-pipeline/page.md",
+        "children": []
+      },
+      {
+        "slug": "claim-5-completeness-certificate",
+        "title": "Claim 5 - completeness certificate",
+        "file": "pages/claim-5-completeness-certificate/page.md",
+        "children": []
+      },
+      {
+        "slug": "claim-6-source-scale-cmnist-classifier",
+        "title": "Source-scale CMNIST classifier",
+        "file": "pages/claim-6-source-scale-cmnist-classifier/page.md",
         "children": []
       },
       {
-        "slug": "methods-environment",
+        "slug": "claim-7-methods-environment",
         "title": "Methods & environment",
-        "file": "pages/methods-environment/page.md",
+        "file": "pages/claim-7-methods-environment/page.md",
         "children": []
       },
       {
-        "slug": "negative-controls-falsification",
+        "slug": "claim-8-negative-controls-falsification",
         "title": "Negative controls & falsification",
-        "file": "pages/negative-controls-falsification/page.md",
+        "file": "pages/claim-8-negative-controls-falsification/page.md",
         "children": []
       },
       {
@@ -49,6 +85,6 @@
       }
     ]
   },
-  "agent_view_tokens": 1620,
-  "revision": "1784621296898905936"
+  "agent_view_tokens": 4906,
+  "revision": "1784509544788627000"
 }

pages/claim-2-sufficient-conditions/page.md

@@ -1,33 +0,0 @@
-# Claim 2 — sufficient conditions
-
-
----
-<!-- trackio-cell
-{"type": "markdown", "id": "cell_1406b9b8f184", "created_at": "2026-07-17T03:43:11+00:00", "title": "Every Boolean-calculus rule"}
--->
-We test every rule in Table 1 separately: negation, conditionally-independent conjunction, conditionally-independent disjunction, and mutually-exclusive disjunction. Atom probabilities are smooth softmax functions, so direct enumeration and differentiation provide an implementation-independent oracle. This targets the exact conditions of Proposition 3.1 and its discrete equivalent Proposition C.3.
-
-
----
-<!-- trackio-cell
-{"type": "markdown", "id": "cell_1afb1d9b2ad5", "created_at": "2026-07-17T03:43:12+00:00", "title": "Result"}
--->
-**VERIFIED:** all 100 primitive rule/seed combinations match direct posterior and score computation. Maximum probability error is 2.22e-16 and maximum score error is 3.33e-16.
-
-
----
-<!-- trackio-cell
-{"type": "code", "id": "cell_276088d934f2", "created_at": "2026-07-17T03:43:14+00:00", "title": "Run independent verification tests", "command": ["python", "-m", "pytest", "repro/tests", "-q"], "exit_code": 0, "duration_s": 0.577}
--->
-````bash
-$ python -m pytest repro/tests -q
-````
-
-exit 0 · 0.6s
-
-
-````output
-............                                                             [100%]
-12 passed in 0.29s
-
-````

pages/conclusion/page.md

@@ -3,16 +3,45 @@
 
 ---
 <!-- trackio-cell
-{"type": "markdown", "id": "cell_7d2cd8090828", "created_at": "2026-07-17T03:43:16+00:00", "title": "Executive summary", "pinned": true, "pinned_at": "2026-07-17T03:43:17+00:00"}
+{"type": "markdown", "id": "cell_conclusion_anchored_20260720", "created_at": "2026-07-20T00:25:00+00:00", "title": "Anchored conclusion", "pinned": true, "pinned_at": "2026-07-20T00:25:00+00:00"}
 -->
-**Both official claims are verified.** LOGDIFF recursive guidance matches direct enumeration for all 6,350 compiled Boolean formulas, with posterior/score errors at floating-point roundoff and finite-difference agreement within 5.21e-10. Every primitive rule passes 100 checks, 12 tests pass, and all three structural-assumption violations are rejected.
+**The evidence is mixed, and the verdicts remain claim-specific.** C1 and C5
+are verified by exhaustive independent oracles. C2 is falsified as written by
+the paper's own Table 2 values. C3's published table transcription matches, but
+the empirical protocol cannot be reproduced from the pinned author package
+because its checkpoints are absent and its released config/FID path diverges
+from the paper. C4 has progressed beyond a release audit: the exact targets,
+weights, real 3D sampler, and AutoDock Vina workflow are recovered and execute;
+the full generated-ligand campaign remains active until its metrics complete.
 
-## Scope & cost
+## Verification scale
 
-| | This reproduction | Full empirical replication |
-|---|---|---|
-| Scope | Exact Propositions 3.1, 3.2 and C.3; exhaustive three-binary query space | CMNIST, Shapes3D, CelebA and protein generation |
-| Hardware | CPU | GPU training and inference |
-| Time | Under 2 seconds | Multiple training/evaluation runs |
-| Cost | $0 | Hardware-dependent |
-| Outcome | Both exact challenge claims verified | Tests downstream image/protein conformity, outside the two claim texts |
+| Evidence gate | Result |
+| --- | ---: |
+| Exhaustive logical events | C1: 6,350; C5: 27,000 |
+| Fixed-timestep classifier evaluations | 70,000 predictions per attribute |
+| Exact upstream CMNIST training | 50 epochs, 60,000-example source split |
+| Molecular checkpoint hashes | 3 of 3 match |
+| Exact GRM5/RRM1 target files | 8 of 8 present |
+| Real reference-complex Vina docking gates | 2 of 2 pass |
+| Anchored claims overclaimed as reproduced | 0 |
+
+The exhaustive-event rows overlap between certificates, so the table avoids
+presenting their sum as a count of unique formulas. All published evidence is
+deterministic or source-pinned, and all incomplete empirical claims are labeled
+incomplete rather than inferred from the paper's numbers.
+
+
+---
+<!-- trackio-cell
+{"type": "artifact", "id": "cell_bf4c3f971732", "created_at": "2026-07-19T19:06:02+00:00", "title": "Reproduction bundle v0", "artifact": "reproduction-logdiff/repro-bundle:v0", "artifact_type": "reproduction-evidence"}
+-->
+**📦 Artifact** `reproduction-logdiff/repro-bundle:v0` · reproduction-evidence
+
+[Public reproduction bundle](https://huggingface.co/buckets/DineshAI/OAM1jJsMGp-artifacts/reproduction-bundle)
+
+---
+<!-- trackio-cell
+{"type": "markdown", "id": "cell_e73e01a85411", "created_at": "2026-07-19T19:06:04+00:00", "title": "Public evidence bucket"}
+-->
+Public artifact mirror: https://huggingface.co/buckets/DineshAI/OAM1jJsMGp-artifacts/reproduction-bundle

pages/index.md

@@ -1,11 +1,19 @@
-# Repro - LOGDIFF Exact Boolean Guidance
+# Reproduction: Logical Guidance for the Exact Composition of Diffusion Models
+
+Independent reproduction of [arXiv:2602.05549](https://huggingface.co/papers/2602.05549) and [OpenReview OAM1jJsMGp](https://openreview.net/forum?id=OAM1jJsMGp).
 
 ## Pages
 
 | Page |
 | --- |
-| [Claim 1 — exact logical guidance](#/claim-1-exact-logical-guidance) |
-| [Claim 2 — sufficient conditions](#/claim-2-sufficient-conditions) |
-| [Methods & environment](#/methods-environment) |
-| [Negative controls & falsification](#/negative-controls-falsification) |
+| [Executive summary](#/executive-summary) |
+| [Claim 1 - exact logical guidance](#/claim-1-exact-logical-guidance) |
+| [Claim 1 - sufficient conditions](#/claim-1-sufficient-conditions) |
+| [Claim 2 - paper-range audit](#/claim-2-paper-range-audit) |
+| [Claim 3 - CelebA release audit](#/claim-3-celeba-release-audit) |
+| [Claim 4 - recovered molecular pipeline](#/claim-4-recovered-molecular-pipeline) |
+| [Claim 5 - completeness certificate](#/claim-5-completeness-certificate) |
+| [Claim 6 - source-scale CMNIST classifier](#/claim-6-source-scale-cmnist-classifier) |
+| [Claim 7 - methods and environment](#/claim-7-methods-environment) |
+| [Claim 8 - negative controls](#/claim-8-negative-controls-falsification) |
 | [Conclusion](#/conclusion) |

pages/methods-environment/page.md

@@ -1,8 +0,0 @@
-# Methods & environment
-
-
----
-<!-- trackio-cell
-{"type": "markdown", "id": "cell_f664159ec314", "created_at": "2026-07-17T03:43:14+00:00", "title": "Scope and provenance"}
--->
-CPU-only Python 3.12.13 and NumPy. This is a full finite-state theorem reproduction, not reduced image generation: the official challenge claims concern the Boolean calculus, and Proposition C.3 explicitly establishes the discrete equivalent. Official implementation TanjaBien/LogDiff is pinned at 94ef35bafd4b4239e9832d8295128c09e8fc1472. No trained model, dataset, GPU, or service is used.

pages/negative-controls-falsification/page.md

@@ -1,8 +0,0 @@
-# Negative controls & falsification
-
-
----
-<!-- trackio-cell
-{"type": "markdown", "id": "cell_34d2cac9ff17", "created_at": "2026-07-17T03:43:15+00:00", "title": "Assumption violations must fail"}
--->
-Three fail-closed controls are tested. CI conjunction and CI disjunction are deliberately applied to correlated events; both disagree with the exact joint oracle. The constant 0.5 score mixture is applied to a mutually-exclusive OR and also disagrees. All 3/3 invalid constructions are rejected; the smallest detected error is 0.0590.