Update logbook: Reproduction: Logical Guidance for the Exact Composition of Diffusion Models

logbook.json

@@ -11,7 +11,7 @@
     "icml2026-repro",
     "paper-OAM1jJsMGp"
   ],
-  "updated_at": "2026-07-20T01:05:29+00:00",
+  "updated_at": "2026-07-20T01:05:44+00:00",
   "root": {
     "slug": "index",
     "title": "Reproduction: Logical Guidance for the Exact Composition of Diffusion Models",
@@ -85,6 +85,6 @@
       }
     ]
   },
-  "agent_view_tokens": 4521,
-  "revision": "1784509529440969000"
+  "agent_view_tokens": 4906,
+  "revision": "1784509544788627000"
 }

pages/claim-4-recovered-molecular-pipeline/page.md

@@ -363,3 +363,16 @@ if __name__ == "__main__":
 }
 
 ````
+
+
+---
+<!-- trackio-cell
+{"type": "markdown", "id": "cell_e179e0a0b124", "created_at": "2026-07-20T01:05:39+00:00", "title": "Interim interpretation and scope boundary"}
+-->
+## Interim full-step molecular evidence (4/32)
+
+This is a scientific checkpoint, not full Claim 4 verification. Four independent AND ligands completed the recovered TargetDiff protocol at 1,000 denoising steps, beta 2.0, 23 atoms, and one posterior sample. All four trajectories were finite, all four reconstructed to valid and distinct SMILES, and all eight generated-ligand AutoDock Vina evaluations succeeded at exhaustiveness 32.
+
+The four-ligand means are GRM5 -8.0038 kcal/mol, RRM1 -8.0668 kcal/mol, AND product 64.5327, diversity 0.8871, valid-unique fraction 1.0, and paper-defined quality fraction 0.5. Same-run reference docking was -5.866 for GRM5 and -7.640 for RRM1. The paper reports LOGDIFF AND product 73.20 +/- 3.18 and DualDiff 71.87 +/- 3.33, so this four-ligand checkpoint is below both reported means and is not treated as confirming the headline magnitude.
+
+The AND-NOT separation metric is intentionally not inferred from these AND samples; it requires the separate logdiff_and_not campaign. The active 32-ligand AND campaign continues with restartable shards, and full Claim 4 assessment remains pending both complete AND and AND-NOT evidence.