Update project name: judgebench -> judgecliff

README.md

@@ -4,20 +4,20 @@ pipeline_tag: image-feature-extraction
 tags:
 - reward-model
 - brand-fidelity
-- judgebench
+- judgecliff
 ---
 
-# judgebench SigLIP brand judge v1
+# judgecliff SigLIP brand judge v1
 
-Part of [judgebench](https://github.com/amargupta0428/judgebench): *which image-QC judges survive optimization pressure?*
+Part of [judgecliff](https://github.com/amargupta0428/judgecliff): *which image-QC judges survive optimization pressure?*
 A brand-fidelity judge for **rhode** (beauty brand). A judge = these SigLIP weights
 + `calibration.json` (rhode train-split centroid + Platt scaling params) in this repo.
 
 **Base:** `google/siglip-so400m-patch14-384` (full `SiglipModel`, contrastively fine-tuned).
 
-**Training:** SupCon contrastive fine-tune, rhode positives vs competitor-brand negatives (Glossier, ILIA and others), judgebench train split. No violation negatives.
+**Training:** SupCon contrastive fine-tune, rhode positives vs competitor-brand negatives (Glossier, ILIA and others), judgecliff train split. No violation negatives.
 
-**Report card (judgebench Phase 1, 2,622-item test set):** brand AUC **0.99**, logo-masking delta **0.00** (style reader, not a wordmark reader), brand-dial Spearman 0.24, near-zero violation detection.
+**Report card (judgecliff Phase 1, 2,622-item test set):** brand AUC **0.99**, logo-masking delta **0.00** (style reader, not a wordmark reader), brand-dial Spearman 0.24, near-zero violation detection.
 
 v1 learned the brand's *center*, not its *boundaries*: it is violation-blind, and under SRPO gradient pressure it was fully exploited (hacked images scored 0.84). See v2 (+violation negatives) and v3 (hardened) for the ablations.
 
@@ -45,4 +45,4 @@ cos = float(v @ centroid)
 score = 1 / (1 + np.exp(-(params["platt_a"] * cos + params["platt_b"])))  # calibrated on-brand prob
 ```
 
-Full evaluation protocol, test-set construction, and findings: [https://github.com/amargupta0428/judgebench](https://github.com/amargupta0428/judgebench).
+Full evaluation protocol, test-set construction, and findings: [https://github.com/amargupta0428/judgecliff](https://github.com/amargupta0428/judgecliff).