Attach independent validation and completion markers for all ablations and reference probes

README.md

@@ -51,3 +51,8 @@ Lower MAE/MSE/RMSE and higher R²/correlations are better. Error units are log10
 [CombiSolv source](https://zenodo.org/records/5970538) · [Prepared dataset](https://huggingface.co/datasets/Queimo/combisolv-qm-rem3di-multitask) · [Checkpoints](https://huggingface.co/Queimo/rem3di-cosmo10-combisolv-multitask) · [Probe W&B](https://wandb.ai/queimo/rem3di-reproduction/runs/u1n6si45) · [Training W&B](https://wandb.ai/queimo/rem3di-reproduction/runs/09d22kwn)
 
 [Monroe UW source](https://github.com/blazejba/Monroe/blob/de075eabe98e4adbbedbc86ca9e877f4bfd9b629/monroe/mtl/UW.py) · [Monroe covariance source](https://github.com/blazejba/Monroe/blob/de075eabe98e4adbbedbc86ca9e877f4bfd9b629/monroe/train/loss.py) · [Solvaformer Table 1](https://arxiv.org/html/2511.09774v1#S3.T1)
+
+
+## Component ablations and reference probes
+
+[Three-epoch ablations, epoch-zero initialization, published Rem3Di, Monroe and all eight Solvaformer Table 1 baselines](ablations/REPORT.md). All measured probe labels, test row IDs, six metrics and checkpoint identities were independently verified.

ablations/complete.json

@@ -0,0 +1,11 @@
+{
+  "completed": true,
+  "variants": [
+    "no_denoising",
+    "no_decorrelation"
+  ],
+  "epochs_each": 3,
+  "probes_each": 3,
+  "revision": "b24915a0f186d23af38850edd551d9d3fb77a37c",
+  "report": "https://huggingface.co/Queimo/rem3di-cosmo10-combisolv-multitask/blob/b24915a0f186d23af38850edd551d9d3fb77a37c/ablations/REPORT.md"
+}

ablations/references/complete.json

@@ -0,0 +1,28 @@
+{
+  "completed": true,
+  "comparison_rows": 22,
+  "published_result": {
+    "representation": "published_rem3di",
+    "method": "frozen_concat_kumo_large",
+    "n_train": 75731,
+    "n_test": 9686,
+    "raw_columns": 1281,
+    "context_blocks": 1,
+    "num_estimators": 4,
+    "seed": 42,
+    "seconds": 125.9508656619946,
+    "rmse": 0.7578431326387834,
+    "mae": 0.5958265571132529,
+    "r2": 0.5575473621174143,
+    "mse": 0.5743262136877647,
+    "pearson": 0.758446579233233,
+    "spearman": 0.7322712013157006
+  },
+  "hub": {
+    "repository": "Queimo/rem3di-cosmo10-combisolv-multitask",
+    "revision": "ff86641eb441ffe038c8aca1df1ba32c02f74cb7",
+    "verified": true,
+    "report": "https://huggingface.co/Queimo/rem3di-cosmo10-combisolv-multitask/blob/ff86641eb441ffe038c8aca1df1ba32c02f74cb7/ablations/REPORT.md"
+  },
+  "completed_unix": 1793262754.1265547
+}

ablations/references/independent_validation.json

@@ -0,0 +1,53 @@
+{
+  "passed": true,
+  "repository": "Queimo/rem3di-cosmo10-combisolv-multitask",
+  "revision": "ff86641eb441ffe038c8aca1df1ba32c02f74cb7",
+  "comparison_rows": 22,
+  "checked": [
+    {
+      "model": "published_rem3di",
+      "all_six_metrics_recomputed": true,
+      "row_ids_exact": true,
+      "label_tolerance": 1e-14,
+      "metrics": {
+        "mae": 0.5958265571132529,
+        "mse": 0.5743262136877647,
+        "rmse": 0.7578431326387834,
+        "r2": 0.5575473621174143,
+        "pearson": 0.7584465792332326,
+        "spearman": 0.7322712013157006
+      }
+    },
+    {
+      "model": "cosmo_epoch10",
+      "all_six_metrics_recomputed": true,
+      "row_ids_exact": true,
+      "label_tolerance": 1e-14,
+      "metrics": {
+        "mae": 0.5949363804813839,
+        "mse": 0.6140966524414051,
+        "rmse": 0.7836431920468684,
+        "r2": 0.5269087892699237,
+        "pearson": 0.7404267461665239,
+        "spearman": 0.7325760677773465
+      }
+    },
+    {
+      "model": "monroe",
+      "all_six_metrics_recomputed": true,
+      "row_ids_exact": true,
+      "label_tolerance": 1e-14,
+      "metrics": {
+        "mae": 0.5773536702967546,
+        "mse": 0.5555317769192204,
+        "rmse": 0.745340041135065,
+        "r2": 0.5720263253400136,
+        "pearson": 0.7696691693136138,
+        "spearman": 0.7593848432185831
+      }
+    }
+  ],
+  "published_checkpoint_hub_hashes_verified": true,
+  "eight_literature_rows_verified": true,
+  "all_eleven_continuation_results_preserved": true
+}

ablations/references/status.json

@@ -1,6 +1,6 @@
 {
-  "phase": "validating_prior_results",
+  "phase": "complete",
   "pid": 54699,
-  "updated_unix": 1793262742.7413404,
-  "wandb_url": "https://wandb.ai/queimo/rem3di-reproduction/runs/2u7ica7w"
+  "updated_unix": 1793262754.126772,
+  "report": "https://huggingface.co/Queimo/rem3di-cosmo10-combisolv-multitask/blob/ff86641eb441ffe038c8aca1df1ba32c02f74cb7/ablations/REPORT.md"
 }