diff --git a/DEVLOG.md b/DEVLOG.md
index 6f1dd40..bb6e518 100644
--- a/DEVLOG.md
+++ b/DEVLOG.md
@@ -80,6 +80,32 @@
 
 ---
 
+## 2026-09-23 (夜)：兩條學習曲線與 OOS 消融完成
+
+### 本次工作 / 執行摘要
+- 協定凍結後（lr 5e-5、S_min 400、5 epochs）依序單獨執行，每條以程式印出的完成行為證據（不看離開碼）：
+  - `make curve MODEL=bert`：`completed 18/18 encoder points (bert)`，18:26 到 19:24。k=1 到 50 的 15 點新訓練；k=100 的 3 點重用 AC2（判定為等價 run，archive 完整）。
+  - `make curve MODEL=modernbert`：`completed 18/18 encoder points (modernbert)`，19:25 到 21:47，18 點全部新訓練。
+  - `make oos-ablation`：`completed 3/3 ablation points`，21:48 到 22:47（ModernBERT、k=100、OOS 訓練 0 筆 × 3 seeds）。
+- `make verify-logits`：75 個 archive 全部與 manifest SHA-256 一致（AC2 3 + 基準 36 + BERT 新 15 + ModernBERT 18 + 消融 3）。
+- 曲線權重全部刪除，只保留 logits。
+
+### 核心發現 / 數據
+- (本條只記錄執行完成；數字分析在步驟 4 由分析程式從 logits 產生，不在此手抄)
+
+### Blockers / 遇到的問題
+- 第一次開跑（17:22）的 shell 迴圈在 zsh 下沒拆 `make $t`，兩條曲線沒跑卻 exit 0。PR #11 加上完成性檢查與 Makefile 無目標防護（`make "curve MODEL=bert"` 現在 exit 2），並有測試對應這次事故。
+
+### Next
+- [ ] logits 附到 GitHub Release
+- [ ] 步驟 4：重跑 Haiku（需 `ANTHROPIC_API_KEY`，預算 ≤ US$5）→ 不確定性分析、fallback、oracle
+
+### Files / Budget
+- `results/curves/{bert,modernbert,oos-ablation}.json`、`results/runs/*.json`、`results/logits-manifest.json`
+- API 花費：US$0
+
+---
+
 ## 2026-09-23 (晚 2)：S_min pilot（validation only），協定凍結
 
 ### 本次工作 / 執行摘要
diff --git a/results/curves/bert.json b/results/curves/bert.json
new file mode 100644
index 0000000..03d4d89
--- /dev/null
+++ b/results/curves/bert.json
@@ -0,0 +1,315 @@
+{
+  "name": "bert",
+  "model_name": "google-bert/bert-base-uncased",
+  "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+  "learning_rate": 5e-05,
+  "min_train_steps": 400,
+  "points": [
+    {
+      "k": 1,
+      "seed": 42,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k1-seed42",
+      "logits_file": "bert-base-uncased-k1-seed42.npz",
+      "logits_sha256": "33464fd83ba26e6de38f2df4a347659dc396c57d98ce9fca16b78ab31101da2d",
+      "train_rows": 153,
+      "oos_train_rows": 3,
+      "train_sample_sha256": "8f04aca1ffa32368a0e811795f70067f9d2b6d1551e10a3be5e08f5c954db166",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 25,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 1,
+      "seed": 43,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k1-seed43",
+      "logits_file": "bert-base-uncased-k1-seed43.npz",
+      "logits_sha256": "de95f070e64b443b4761f3ca786da624343fe8a5dc20e3f341220333e586cff3",
+      "train_rows": 153,
+      "oos_train_rows": 3,
+      "train_sample_sha256": "4c659defaf9b9c6f28f4d5e67ef92b466627cdb200623552ba46ced1d2750c9c",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 25,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 1,
+      "seed": 44,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k1-seed44",
+      "logits_file": "bert-base-uncased-k1-seed44.npz",
+      "logits_sha256": "db5595875d86d5b3476689bb5932f72bdb7651effb3686d8636cb308e8279881",
+      "train_rows": 153,
+      "oos_train_rows": 3,
+      "train_sample_sha256": "d69635e70893d70759a46fd0fa54561eae262653eca95898076348b9c660b96a",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 25,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 5,
+      "seed": 42,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k5-seed42",
+      "logits_file": "bert-base-uncased-k5-seed42.npz",
+      "logits_sha256": "12e3d723ecbdeda888a687bd3b539ef5b8bc07515bd72f28293b0dfc4314f94d",
+      "train_rows": 763,
+      "oos_train_rows": 13,
+      "train_sample_sha256": "bfabf08971d3a16a502e25b2d5a7f1ce204504a526d5b3cb34fec9346dd8caa7",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 120,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 5,
+      "seed": 43,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k5-seed43",
+      "logits_file": "bert-base-uncased-k5-seed43.npz",
+      "logits_sha256": "78dd86d8b8a700aa8986c7bf23fef4a99720a8f226df3e6d0d21d896cae3aa8a",
+      "train_rows": 763,
+      "oos_train_rows": 13,
+      "train_sample_sha256": "4e289193d9669c26c40d083246f6c34426bca5ff041ac526796a5f976ca9c638",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 120,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 5,
+      "seed": 44,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k5-seed44",
+      "logits_file": "bert-base-uncased-k5-seed44.npz",
+      "logits_sha256": "175a274248600f6f85681b2e8a824dd9bf17f281782044110b1c3facb10f1da5",
+      "train_rows": 763,
+      "oos_train_rows": 13,
+      "train_sample_sha256": "ba57c52c7c416c611ba215bb10a393fba470ec511aaebad2b416914ca674988d",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 120,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 10,
+      "seed": 42,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k10-seed42",
+      "logits_file": "bert-base-uncased-k10-seed42.npz",
+      "logits_sha256": "51df91b493b56ae7d49fdd3195bc1d53d55099ec5b1cf011dfa9579a8a6320a3",
+      "train_rows": 1525,
+      "oos_train_rows": 25,
+      "train_sample_sha256": "2606cc878d85759a5885e1b9be35205023b156e6cf8a4eea85bd635c3b492e74",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 240,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 10,
+      "seed": 43,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k10-seed43",
+      "logits_file": "bert-base-uncased-k10-seed43.npz",
+      "logits_sha256": "113c11c8b72a8f20ac46e70a1c8492ee1a4d0571b1b86caba1b583cc85d0faf8",
+      "train_rows": 1525,
+      "oos_train_rows": 25,
+      "train_sample_sha256": "072190bdfc028a2e0bc6ec85d39e26ae1b27854281c6226be57b8f8b9d20e165",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 240,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 10,
+      "seed": 44,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k10-seed44",
+      "logits_file": "bert-base-uncased-k10-seed44.npz",
+      "logits_sha256": "b60ca5253fc8ff28108fcfa32d4aa79139357e9f5639d187b2270a266c594adb",
+      "train_rows": 1525,
+      "oos_train_rows": 25,
+      "train_sample_sha256": "31a30630bb8f711a55173e8837eaf4500c2c5dcd5efe461b561a04e843cddd55",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 240,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 25,
+      "seed": 42,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k25-seed42",
+      "logits_file": "bert-base-uncased-k25-seed42.npz",
+      "logits_sha256": "5397a2237935d24210a1b8e47ed837f1807db8beb0a4c18f31d24d0681bef1cc",
+      "train_rows": 3813,
+      "oos_train_rows": 63,
+      "train_sample_sha256": "51184b580bdbe58ab700d2545887a30c442950389b98497c3b7edaa7ca47174c",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 600,
+      "epoch_steps": 600,
+      "decided_by": "epochs",
+      "global_step": 600,
+      "reused_from": null
+    },
+    {
+      "k": 25,
+      "seed": 43,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k25-seed43",
+      "logits_file": "bert-base-uncased-k25-seed43.npz",
+      "logits_sha256": "cf40024e86a42cfe34703b4b04e58696d93a03f3408a44f820960b11d554d626",
+      "train_rows": 3813,
+      "oos_train_rows": 63,
+      "train_sample_sha256": "92f91be6e0948dbc01e82b81485c72829dd232443955fa0ae214b1451a0a704f",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 600,
+      "epoch_steps": 600,
+      "decided_by": "epochs",
+      "global_step": 600,
+      "reused_from": null
+    },
+    {
+      "k": 25,
+      "seed": 44,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k25-seed44",
+      "logits_file": "bert-base-uncased-k25-seed44.npz",
+      "logits_sha256": "c4de18e8e6d5e96011ab20178ec669bd2462c6aae44cd2fd40ae8ba3ce077f30",
+      "train_rows": 3813,
+      "oos_train_rows": 63,
+      "train_sample_sha256": "790cd7cf9eb24143276f6c59c7bac29c4a0b42f14bef38e9634e96253e6315fb",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 600,
+      "epoch_steps": 600,
+      "decided_by": "epochs",
+      "global_step": 600,
+      "reused_from": null
+    },
+    {
+      "k": 50,
+      "seed": 42,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k50-seed42",
+      "logits_file": "bert-base-uncased-k50-seed42.npz",
+      "logits_sha256": "42bd57f63c3f37264962334df25861a73c5279f7559302c9a467d5fadd246b7d",
+      "train_rows": 7625,
+      "oos_train_rows": 125,
+      "train_sample_sha256": "8a9a52f8ae41506d9847bd9ceba987edadc6bc3304487be22e664a95e1b9b18f",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 1195,
+      "epoch_steps": 1195,
+      "decided_by": "epochs",
+      "global_step": 1195,
+      "reused_from": null
+    },
+    {
+      "k": 50,
+      "seed": 43,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k50-seed43",
+      "logits_file": "bert-base-uncased-k50-seed43.npz",
+      "logits_sha256": "21bb8489d1c181fa6982ff89fb8a587505a5fb00ad53031246c0bdecb16bd949",
+      "train_rows": 7625,
+      "oos_train_rows": 125,
+      "train_sample_sha256": "836ad6186d05a042ae00b3ab45efc560eb45f774644bf5e4f9b13a6e66d82982",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 1195,
+      "epoch_steps": 1195,
+      "decided_by": "epochs",
+      "global_step": 1195,
+      "reused_from": null
+    },
+    {
+      "k": 50,
+      "seed": 44,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-k50-seed44",
+      "logits_file": "bert-base-uncased-k50-seed44.npz",
+      "logits_sha256": "e335e894b2eb54e9f4074b861f202c478078700f308503c82ecc34ec7fc1c6fa",
+      "train_rows": 7625,
+      "oos_train_rows": 125,
+      "train_sample_sha256": "5db5ce1ae1864ba63a4164a799b09a7c1d783a4acbd55efc587613d986752013",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 1195,
+      "epoch_steps": 1195,
+      "decided_by": "epochs",
+      "global_step": 1195,
+      "reused_from": null
+    },
+    {
+      "k": 100,
+      "seed": 42,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-full-seed42",
+      "logits_file": "bert-base-uncased-full-seed42.npz",
+      "logits_sha256": "673a4c11200dd61ddd6310f174ab9fb27580cf5b512c7ba3b5db19088958bab2",
+      "train_rows": 15250,
+      "oos_train_rows": 250,
+      "train_sample_sha256": "7a3ebd2cd2f985a26e815c229cc54add6763f783bee572c5e7948636c66c7553",
+      "train_sample_sha256_source": "computed at index time from curve_sample(k, seed); the run predates the field",
+      "planned_steps": 2385,
+      "epoch_steps": 2385,
+      "decided_by": "epochs",
+      "global_step": 2385,
+      "reused_from": "bert-base-uncased-full-seed42"
+    },
+    {
+      "k": 100,
+      "seed": 43,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-full-seed43",
+      "logits_file": "bert-base-uncased-full-seed43.npz",
+      "logits_sha256": "b6963da29f5886b915d2c5626391721fbd18c7f54a68b7720bf2d10d6cc2c58f",
+      "train_rows": 15250,
+      "oos_train_rows": 250,
+      "train_sample_sha256": "7a3ebd2cd2f985a26e815c229cc54add6763f783bee572c5e7948636c66c7553",
+      "train_sample_sha256_source": "computed at index time from curve_sample(k, seed); the run predates the field",
+      "planned_steps": 2385,
+      "epoch_steps": 2385,
+      "decided_by": "epochs",
+      "global_step": 2385,
+      "reused_from": "bert-base-uncased-full-seed43"
+    },
+    {
+      "k": 100,
+      "seed": 44,
+      "oos_train": null,
+      "run_name": "bert-base-uncased-full-seed44",
+      "logits_file": "bert-base-uncased-full-seed44.npz",
+      "logits_sha256": "0324b20d4736ca9ad913b97269100b398475ab06a2bd130f653af55de239c277",
+      "train_rows": 15250,
+      "oos_train_rows": 250,
+      "train_sample_sha256": "7a3ebd2cd2f985a26e815c229cc54add6763f783bee572c5e7948636c66c7553",
+      "train_sample_sha256_source": "computed at index time from curve_sample(k, seed); the run predates the field",
+      "planned_steps": 2385,
+      "epoch_steps": 2385,
+      "decided_by": "epochs",
+      "global_step": 2385,
+      "reused_from": "bert-base-uncased-full-seed44"
+    }
+  ]
+}
diff --git a/results/curves/modernbert.json b/results/curves/modernbert.json
new file mode 100644
index 0000000..40cec5d
--- /dev/null
+++ b/results/curves/modernbert.json
@@ -0,0 +1,315 @@
+{
+  "name": "modernbert",
+  "model_name": "answerdotai/ModernBERT-base",
+  "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+  "learning_rate": 5e-05,
+  "min_train_steps": 400,
+  "points": [
+    {
+      "k": 1,
+      "seed": 42,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k1-seed42",
+      "logits_file": "ModernBERT-base-k1-seed42.npz",
+      "logits_sha256": "aaa5c7cc5e068eed5da1dc4a55a91587761246e273c254564fd6bf69a27c2ae2",
+      "train_rows": 153,
+      "oos_train_rows": 3,
+      "train_sample_sha256": "8f04aca1ffa32368a0e811795f70067f9d2b6d1551e10a3be5e08f5c954db166",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 25,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 1,
+      "seed": 43,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k1-seed43",
+      "logits_file": "ModernBERT-base-k1-seed43.npz",
+      "logits_sha256": "c2eae04634a5224685fed57cf0600993eb975a269a9ef61ec97f33598705786b",
+      "train_rows": 153,
+      "oos_train_rows": 3,
+      "train_sample_sha256": "4c659defaf9b9c6f28f4d5e67ef92b466627cdb200623552ba46ced1d2750c9c",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 25,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 1,
+      "seed": 44,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k1-seed44",
+      "logits_file": "ModernBERT-base-k1-seed44.npz",
+      "logits_sha256": "84067b0f3e6ad7ab14ed6b9b919b1b4b24ed1d0c3399dd81573f325adf2021b4",
+      "train_rows": 153,
+      "oos_train_rows": 3,
+      "train_sample_sha256": "d69635e70893d70759a46fd0fa54561eae262653eca95898076348b9c660b96a",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 25,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 5,
+      "seed": 42,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k5-seed42",
+      "logits_file": "ModernBERT-base-k5-seed42.npz",
+      "logits_sha256": "8ffbbf00d37abfc1fc2e3601e96ed26c71c6cb2452c8c88d32375674f22996c1",
+      "train_rows": 763,
+      "oos_train_rows": 13,
+      "train_sample_sha256": "bfabf08971d3a16a502e25b2d5a7f1ce204504a526d5b3cb34fec9346dd8caa7",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 120,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 5,
+      "seed": 43,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k5-seed43",
+      "logits_file": "ModernBERT-base-k5-seed43.npz",
+      "logits_sha256": "561ef2079fd97de9a221e029b2589c50ccd17a579423fa9503ef95fe2944730c",
+      "train_rows": 763,
+      "oos_train_rows": 13,
+      "train_sample_sha256": "4e289193d9669c26c40d083246f6c34426bca5ff041ac526796a5f976ca9c638",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 120,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 5,
+      "seed": 44,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k5-seed44",
+      "logits_file": "ModernBERT-base-k5-seed44.npz",
+      "logits_sha256": "9c6b23ca81db2a58e2f444a13533edcf9687f6d4020ba6ec6ecc52915f26e784",
+      "train_rows": 763,
+      "oos_train_rows": 13,
+      "train_sample_sha256": "ba57c52c7c416c611ba215bb10a393fba470ec511aaebad2b416914ca674988d",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 120,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 10,
+      "seed": 42,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k10-seed42",
+      "logits_file": "ModernBERT-base-k10-seed42.npz",
+      "logits_sha256": "6fff9cd41ff6f9c828b9337b3c190f6e4cb1e41dff62667329f5e0647f81afba",
+      "train_rows": 1525,
+      "oos_train_rows": 25,
+      "train_sample_sha256": "2606cc878d85759a5885e1b9be35205023b156e6cf8a4eea85bd635c3b492e74",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 240,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 10,
+      "seed": 43,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k10-seed43",
+      "logits_file": "ModernBERT-base-k10-seed43.npz",
+      "logits_sha256": "f08779235eb7af011ab4c6402e86fe393d5cb9cc0d341ecfb01992fe97d52004",
+      "train_rows": 1525,
+      "oos_train_rows": 25,
+      "train_sample_sha256": "072190bdfc028a2e0bc6ec85d39e26ae1b27854281c6226be57b8f8b9d20e165",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 240,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 10,
+      "seed": 44,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k10-seed44",
+      "logits_file": "ModernBERT-base-k10-seed44.npz",
+      "logits_sha256": "6c20ced25748c4ed64688269d0ab4fe74f0b32ae9af565c8ccd4cf6ed72140d9",
+      "train_rows": 1525,
+      "oos_train_rows": 25,
+      "train_sample_sha256": "31a30630bb8f711a55173e8837eaf4500c2c5dcd5efe461b561a04e843cddd55",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 400,
+      "epoch_steps": 240,
+      "decided_by": "min_train_steps",
+      "global_step": 400,
+      "reused_from": null
+    },
+    {
+      "k": 25,
+      "seed": 42,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k25-seed42",
+      "logits_file": "ModernBERT-base-k25-seed42.npz",
+      "logits_sha256": "da2c8b7a2fcefcda47211bec4d2248e7ca5288eda05cc2572ade0945abea36ff",
+      "train_rows": 3813,
+      "oos_train_rows": 63,
+      "train_sample_sha256": "51184b580bdbe58ab700d2545887a30c442950389b98497c3b7edaa7ca47174c",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 600,
+      "epoch_steps": 600,
+      "decided_by": "epochs",
+      "global_step": 600,
+      "reused_from": null
+    },
+    {
+      "k": 25,
+      "seed": 43,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k25-seed43",
+      "logits_file": "ModernBERT-base-k25-seed43.npz",
+      "logits_sha256": "ac640bffd502ea71d7dd5b5f9cc18d7068b6b8b1b950acbf3a2fd58a3c50f1b6",
+      "train_rows": 3813,
+      "oos_train_rows": 63,
+      "train_sample_sha256": "92f91be6e0948dbc01e82b81485c72829dd232443955fa0ae214b1451a0a704f",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 600,
+      "epoch_steps": 600,
+      "decided_by": "epochs",
+      "global_step": 600,
+      "reused_from": null
+    },
+    {
+      "k": 25,
+      "seed": 44,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k25-seed44",
+      "logits_file": "ModernBERT-base-k25-seed44.npz",
+      "logits_sha256": "cd9b8473e1b5b0d4597feb76bdf99d27464910b85b536eb7169193676c94ce68",
+      "train_rows": 3813,
+      "oos_train_rows": 63,
+      "train_sample_sha256": "790cd7cf9eb24143276f6c59c7bac29c4a0b42f14bef38e9634e96253e6315fb",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 600,
+      "epoch_steps": 600,
+      "decided_by": "epochs",
+      "global_step": 600,
+      "reused_from": null
+    },
+    {
+      "k": 50,
+      "seed": 42,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k50-seed42",
+      "logits_file": "ModernBERT-base-k50-seed42.npz",
+      "logits_sha256": "86bcc65f899f519154881c6e1ccb10ba468e055b7a26049fc11727fc74347a4c",
+      "train_rows": 7625,
+      "oos_train_rows": 125,
+      "train_sample_sha256": "8a9a52f8ae41506d9847bd9ceba987edadc6bc3304487be22e664a95e1b9b18f",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 1195,
+      "epoch_steps": 1195,
+      "decided_by": "epochs",
+      "global_step": 1195,
+      "reused_from": null
+    },
+    {
+      "k": 50,
+      "seed": 43,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k50-seed43",
+      "logits_file": "ModernBERT-base-k50-seed43.npz",
+      "logits_sha256": "0f72b3ebe5b54dcee8b7857d339de335b31230f6dd06d4f43297730e19d955ee",
+      "train_rows": 7625,
+      "oos_train_rows": 125,
+      "train_sample_sha256": "836ad6186d05a042ae00b3ab45efc560eb45f774644bf5e4f9b13a6e66d82982",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 1195,
+      "epoch_steps": 1195,
+      "decided_by": "epochs",
+      "global_step": 1195,
+      "reused_from": null
+    },
+    {
+      "k": 50,
+      "seed": 44,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k50-seed44",
+      "logits_file": "ModernBERT-base-k50-seed44.npz",
+      "logits_sha256": "c7427794ce3f205347c4a1bc703409bfece96251974823dcb075bfc2dd3d6da3",
+      "train_rows": 7625,
+      "oos_train_rows": 125,
+      "train_sample_sha256": "5db5ce1ae1864ba63a4164a799b09a7c1d783a4acbd55efc587613d986752013",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 1195,
+      "epoch_steps": 1195,
+      "decided_by": "epochs",
+      "global_step": 1195,
+      "reused_from": null
+    },
+    {
+      "k": 100,
+      "seed": 42,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k100-seed42",
+      "logits_file": "ModernBERT-base-k100-seed42.npz",
+      "logits_sha256": "208f372ab9096fa4f95a006328c8ef766cb65144005d9df8f4bad1c391a100a9",
+      "train_rows": 15250,
+      "oos_train_rows": 250,
+      "train_sample_sha256": "7a3ebd2cd2f985a26e815c229cc54add6763f783bee572c5e7948636c66c7553",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 2385,
+      "epoch_steps": 2385,
+      "decided_by": "epochs",
+      "global_step": 2385,
+      "reused_from": null
+    },
+    {
+      "k": 100,
+      "seed": 43,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k100-seed43",
+      "logits_file": "ModernBERT-base-k100-seed43.npz",
+      "logits_sha256": "58c3bd83838784f23d7e9977921b4a021797799b9a514b3c4cd449c641cd5257",
+      "train_rows": 15250,
+      "oos_train_rows": 250,
+      "train_sample_sha256": "7a3ebd2cd2f985a26e815c229cc54add6763f783bee572c5e7948636c66c7553",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 2385,
+      "epoch_steps": 2385,
+      "decided_by": "epochs",
+      "global_step": 2385,
+      "reused_from": null
+    },
+    {
+      "k": 100,
+      "seed": 44,
+      "oos_train": null,
+      "run_name": "ModernBERT-base-k100-seed44",
+      "logits_file": "ModernBERT-base-k100-seed44.npz",
+      "logits_sha256": "37ebd645a97a654308f9f50cf4501b6b1c5bfed5bbd064fe591d8587a4873966",
+      "train_rows": 15250,
+      "oos_train_rows": 250,
+      "train_sample_sha256": "7a3ebd2cd2f985a26e815c229cc54add6763f783bee572c5e7948636c66c7553",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 2385,
+      "epoch_steps": 2385,
+      "decided_by": "epochs",
+      "global_step": 2385,
+      "reused_from": null
+    }
+  ]
+}
diff --git a/results/curves/oos-ablation.json b/results/curves/oos-ablation.json
new file mode 100644
index 0000000..260c49d
--- /dev/null
+++ b/results/curves/oos-ablation.json
@@ -0,0 +1,60 @@
+{
+  "name": "oos-ablation",
+  "model_name": "answerdotai/ModernBERT-base",
+  "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+  "learning_rate": 5e-05,
+  "min_train_steps": 400,
+  "points": [
+    {
+      "k": 100,
+      "seed": 42,
+      "oos_train": 0,
+      "run_name": "ModernBERT-base-k100-oos0-seed42",
+      "logits_file": "ModernBERT-base-k100-oos0-seed42.npz",
+      "logits_sha256": "c83e3c569facc09d5eb14cc3f837883fb59b0d5d68583257e9366e3750066be6",
+      "train_rows": 15000,
+      "oos_train_rows": 0,
+      "train_sample_sha256": "88418dcc1027e9ccb3d1bf8230bbf9471a749d252302ee65f154e0228fc9b81d",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 2345,
+      "epoch_steps": 2345,
+      "decided_by": "epochs",
+      "global_step": 2345,
+      "reused_from": null
+    },
+    {
+      "k": 100,
+      "seed": 43,
+      "oos_train": 0,
+      "run_name": "ModernBERT-base-k100-oos0-seed43",
+      "logits_file": "ModernBERT-base-k100-oos0-seed43.npz",
+      "logits_sha256": "09029343a7529ac08bbc145fef45edcc1bfb04f6b976dc956473c6c546175343",
+      "train_rows": 15000,
+      "oos_train_rows": 0,
+      "train_sample_sha256": "88418dcc1027e9ccb3d1bf8230bbf9471a749d252302ee65f154e0228fc9b81d",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 2345,
+      "epoch_steps": 2345,
+      "decided_by": "epochs",
+      "global_step": 2345,
+      "reused_from": null
+    },
+    {
+      "k": 100,
+      "seed": 44,
+      "oos_train": 0,
+      "run_name": "ModernBERT-base-k100-oos0-seed44",
+      "logits_file": "ModernBERT-base-k100-oos0-seed44.npz",
+      "logits_sha256": "6edfa2106d6d72ae26449b5598cff759ae601c44c42648de129ceab6e5530a61",
+      "train_rows": 15000,
+      "oos_train_rows": 0,
+      "train_sample_sha256": "88418dcc1027e9ccb3d1bf8230bbf9471a749d252302ee65f154e0228fc9b81d",
+      "train_sample_sha256_source": "recorded by train()",
+      "planned_steps": 2345,
+      "epoch_steps": 2345,
+      "decided_by": "epochs",
+      "global_step": 2345,
+      "reused_from": null
+    }
+  ]
+}
diff --git a/results/logits-manifest.json b/results/logits-manifest.json
index c73aa18..2349aeb 100644
--- a/results/logits-manifest.json
+++ b/results/logits-manifest.json
@@ -1,6 +1,237 @@
 {
   "format_version": 1,
   "files": {
+    "ModernBERT-base-k1-seed42.npz": {
+      "sha256": "aaa5c7cc5e068eed5da1dc4a55a91587761246e273c254564fd6bf69a27c2ae2",
+      "bytes": 4828906,
+      "run_name": "ModernBERT-base-k1-seed42",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T11:28:13+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k1-seed43.npz": {
+      "sha256": "c2eae04634a5224685fed57cf0600993eb975a269a9ef61ec97f33598705786b",
+      "bytes": 4828825,
+      "run_name": "ModernBERT-base-k1-seed43",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T11:31:54+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k1-seed44.npz": {
+      "sha256": "84067b0f3e6ad7ab14ed6b9b919b1b4b24ed1d0c3399dd81573f325adf2021b4",
+      "bytes": 4828958,
+      "run_name": "ModernBERT-base-k1-seed44",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T11:35:39+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k10-seed42.npz": {
+      "sha256": "6fff9cd41ff6f9c828b9337b3c190f6e4cb1e41dff62667329f5e0647f81afba",
+      "bytes": 4832287,
+      "run_name": "ModernBERT-base-k10-seed42",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T11:50:36+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k10-seed43.npz": {
+      "sha256": "f08779235eb7af011ab4c6402e86fe393d5cb9cc0d341ecfb01992fe97d52004",
+      "bytes": 4832661,
+      "run_name": "ModernBERT-base-k10-seed43",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T11:54:23+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k10-seed44.npz": {
+      "sha256": "6c20ced25748c4ed64688269d0ab4fe74f0b32ae9af565c8ccd4cf6ed72140d9",
+      "bytes": 4831998,
+      "run_name": "ModernBERT-base-k10-seed44",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T11:58:06+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k100-oos0-seed42.npz": {
+      "sha256": "c83e3c569facc09d5eb14cc3f837883fb59b0d5d68583257e9366e3750066be6",
+      "bytes": 4833660,
+      "run_name": "ModernBERT-base-k100-oos0-seed42",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T14:07:31+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k100-oos0-seed43.npz": {
+      "sha256": "09029343a7529ac08bbc145fef45edcc1bfb04f6b976dc956473c6c546175343",
+      "bytes": 4833643,
+      "run_name": "ModernBERT-base-k100-oos0-seed43",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T14:27:35+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k100-oos0-seed44.npz": {
+      "sha256": "6edfa2106d6d72ae26449b5598cff759ae601c44c42648de129ceab6e5530a61",
+      "bytes": 4833040,
+      "run_name": "ModernBERT-base-k100-oos0-seed44",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T14:47:39+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k100-seed42.npz": {
+      "sha256": "208f372ab9096fa4f95a006328c8ef766cb65144005d9df8f4bad1c391a100a9",
+      "bytes": 4833253,
+      "run_name": "ModernBERT-base-k100-seed42",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T13:06:17+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k100-seed43.npz": {
+      "sha256": "58c3bd83838784f23d7e9977921b4a021797799b9a514b3c4cd449c641cd5257",
+      "bytes": 4832996,
+      "run_name": "ModernBERT-base-k100-seed43",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T13:26:41+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k100-seed44.npz": {
+      "sha256": "37ebd645a97a654308f9f50cf4501b6b1c5bfed5bbd064fe591d8587a4873966",
+      "bytes": 4833398,
+      "run_name": "ModernBERT-base-k100-seed44",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T13:47:03+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k25-seed42.npz": {
+      "sha256": "da2c8b7a2fcefcda47211bec4d2248e7ca5288eda05cc2572ade0945abea36ff",
+      "bytes": 4833547,
+      "run_name": "ModernBERT-base-k25-seed42",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T12:03:33+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k25-seed43.npz": {
+      "sha256": "ac640bffd502ea71d7dd5b5f9cc18d7068b6b8b1b950acbf3a2fd58a3c50f1b6",
+      "bytes": 4833666,
+      "run_name": "ModernBERT-base-k25-seed43",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T12:09:00+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k25-seed44.npz": {
+      "sha256": "cd9b8473e1b5b0d4597feb76bdf99d27464910b85b536eb7169193676c94ce68",
+      "bytes": 4832954,
+      "run_name": "ModernBERT-base-k25-seed44",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T12:14:29+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k5-seed42.npz": {
+      "sha256": "8ffbbf00d37abfc1fc2e3601e96ed26c71c6cb2452c8c88d32375674f22996c1",
+      "bytes": 4830959,
+      "run_name": "ModernBERT-base-k5-seed42",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T11:39:23+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k5-seed43.npz": {
+      "sha256": "561ef2079fd97de9a221e029b2589c50ccd17a579423fa9503ef95fe2944730c",
+      "bytes": 4831188,
+      "run_name": "ModernBERT-base-k5-seed43",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T11:43:07+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k5-seed44.npz": {
+      "sha256": "9c6b23ca81db2a58e2f444a13533edcf9687f6d4020ba6ec6ecc52915f26e784",
+      "bytes": 4831385,
+      "run_name": "ModernBERT-base-k5-seed44",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T11:46:51+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k50-seed42.npz": {
+      "sha256": "86bcc65f899f519154881c6e1ccb10ba468e055b7a26049fc11727fc74347a4c",
+      "bytes": 4833586,
+      "run_name": "ModernBERT-base-k50-seed42",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T12:24:57+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k50-seed43.npz": {
+      "sha256": "0f72b3ebe5b54dcee8b7857d339de335b31230f6dd06d4f43297730e19d955ee",
+      "bytes": 4833495,
+      "run_name": "ModernBERT-base-k50-seed43",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T12:35:24+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "ModernBERT-base-k50-seed44.npz": {
+      "sha256": "c7427794ce3f205347c4a1bc703409bfece96251974823dcb075bfc2dd3d6da3",
+      "bytes": 4833374,
+      "run_name": "ModernBERT-base-k50-seed44",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T12:45:49+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
     "bert-base-uncased-full-seed42.npz": {
       "sha256": "673a4c11200dd61ddd6310f174ab9fb27580cf5b512c7ba3b5db19088958bab2",
       "bytes": 4824207,
@@ -34,6 +265,171 @@
         "test": 5500
       }
     },
+    "bert-base-uncased-k1-seed42.npz": {
+      "sha256": "33464fd83ba26e6de38f2df4a347659dc396c57d98ce9fca16b78ab31101da2d",
+      "bytes": 4820569,
+      "run_name": "bert-base-uncased-k1-seed42",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T10:31:21+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "bert-base-uncased-k1-seed43.npz": {
+      "sha256": "de95f070e64b443b4761f3ca786da624343fe8a5dc20e3f341220333e586cff3",
+      "bytes": 4823550,
+      "run_name": "bert-base-uncased-k1-seed43",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T10:33:52+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "bert-base-uncased-k1-seed44.npz": {
+      "sha256": "db5595875d86d5b3476689bb5932f72bdb7651effb3686d8636cb308e8279881",
+      "bytes": 4821951,
+      "run_name": "bert-base-uncased-k1-seed44",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T10:36:27+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "bert-base-uncased-k10-seed42.npz": {
+      "sha256": "51df91b493b56ae7d49fdd3195bc1d53d55099ec5b1cf011dfa9579a8a6320a3",
+      "bytes": 4820339,
+      "run_name": "bert-base-uncased-k10-seed42",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T10:46:35+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "bert-base-uncased-k10-seed43.npz": {
+      "sha256": "113c11c8b72a8f20ac46e70a1c8492ee1a4d0571b1b86caba1b583cc85d0faf8",
+      "bytes": 4824784,
+      "run_name": "bert-base-uncased-k10-seed43",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T10:49:08+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "bert-base-uncased-k10-seed44.npz": {
+      "sha256": "b60ca5253fc8ff28108fcfa32d4aa79139357e9f5639d187b2270a266c594adb",
+      "bytes": 4818953,
+      "run_name": "bert-base-uncased-k10-seed44",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T10:51:40+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "bert-base-uncased-k25-seed42.npz": {
+      "sha256": "5397a2237935d24210a1b8e47ed837f1807db8beb0a4c18f31d24d0681bef1cc",
+      "bytes": 4817631,
+      "run_name": "bert-base-uncased-k25-seed42",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T10:55:22+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "bert-base-uncased-k25-seed43.npz": {
+      "sha256": "cf40024e86a42cfe34703b4b04e58696d93a03f3408a44f820960b11d554d626",
+      "bytes": 4820374,
+      "run_name": "bert-base-uncased-k25-seed43",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T10:59:05+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "bert-base-uncased-k25-seed44.npz": {
+      "sha256": "c4de18e8e6d5e96011ab20178ec669bd2462c6aae44cd2fd40ae8ba3ce077f30",
+      "bytes": 4818898,
+      "run_name": "bert-base-uncased-k25-seed44",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T11:02:47+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "bert-base-uncased-k5-seed42.npz": {
+      "sha256": "12e3d723ecbdeda888a687bd3b539ef5b8bc07515bd72f28293b0dfc4314f94d",
+      "bytes": 4820512,
+      "run_name": "bert-base-uncased-k5-seed42",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T10:38:58+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "bert-base-uncased-k5-seed43.npz": {
+      "sha256": "78dd86d8b8a700aa8986c7bf23fef4a99720a8f226df3e6d0d21d896cae3aa8a",
+      "bytes": 4826522,
+      "run_name": "bert-base-uncased-k5-seed43",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T10:41:31+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "bert-base-uncased-k5-seed44.npz": {
+      "sha256": "175a274248600f6f85681b2e8a824dd9bf17f281782044110b1c3facb10f1da5",
+      "bytes": 4818879,
+      "run_name": "bert-base-uncased-k5-seed44",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T10:44:02+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "bert-base-uncased-k50-seed42.npz": {
+      "sha256": "42bd57f63c3f37264962334df25861a73c5279f7559302c9a467d5fadd246b7d",
+      "bytes": 4819519,
+      "run_name": "bert-base-uncased-k50-seed42",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T11:09:55+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "bert-base-uncased-k50-seed43.npz": {
+      "sha256": "21bb8489d1c181fa6982ff89fb8a587505a5fb00ad53031246c0bdecb16bd949",
+      "bytes": 4819533,
+      "run_name": "bert-base-uncased-k50-seed43",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T11:17:02+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
+    "bert-base-uncased-k50-seed44.npz": {
+      "sha256": "e335e894b2eb54e9f4074b861f202c478078700f308503c82ecc34ec7fc1c6fa",
+      "bytes": 4821765,
+      "run_name": "bert-base-uncased-k50-seed44",
+      "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+      "created_at": "2026-09-23T11:24:07+00:00",
+      "rows": {
+        "validation": 3100,
+        "test": 5500
+      }
+    },
     "majority-k1-seed42.npz": {
       "sha256": "ad9e1099ed7d88354021da1cda7ae310d1313af9d7d4654eafd03fabb8318e34",
       "bytes": 24098,
diff --git a/results/runs/ModernBERT-base-k1-seed42.json b/results/runs/ModernBERT-base-k1-seed42.json
new file mode 100644
index 0000000..99e2aaa
--- /dev/null
+++ b/results/runs/ModernBERT-base-k1-seed42.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k1-seed42",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 42,
+    "per_intent": null,
+    "k_shot": 1,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k1-seed42",
+    "seed": 42,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 42,
+      "per_intent": null,
+      "k_shot": 1,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 153,
+    "oos_train_rows": 3,
+    "per_intent": null,
+    "k_shot": 1,
+    "train_sample_sha256": "8f04aca1ffa32368a0e811795f70067f9d2b6d1551e10a3be5e08f5c954db166",
+    "step_plan": {
+      "epoch_steps": 25,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 0.42569934903369355,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 193.341,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 4044521472,
+      "tensor_allocated_bytes": 2422683904
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k1-seed42.npz",
+    "sha256": "aaa5c7cc5e068eed5da1dc4a55a91587761246e273c254564fd6bf69a27c2ae2",
+    "bytes": 4828906
+  },
+  "metrics": {
+    "temperature": 1.3383029824789503,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.46806451612903227,
+        "oos_recall_8": 0.17,
+        "accuracy_8_summed": 0.5241935483870968,
+        "oos_recall_8_summed": 0.09,
+        "in_scope_accuracy_150": 0.20133333333333334,
+        "oos_recall_151": 0.17,
+        "ece_151": 0.08235584183780623,
+        "ece_8_summed": 0.07263813087663859,
+        "nll_151": 4.033748009045811,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.46806451612903227,
+        "oos_recall_8": 0.17,
+        "accuracy_8_summed": 0.5180645161290323,
+        "oos_recall_8_summed": 0.03,
+        "in_scope_accuracy_150": 0.20133333333333334,
+        "oos_recall_151": 0.17,
+        "ece_151": 0.03435128089867157,
+        "ece_8_summed": 0.1547749604730463,
+        "nll_151": 3.923818374286287,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.39890909090909094,
+        "oos_recall_8": 0.107,
+        "accuracy_8_summed": 0.44545454545454544,
+        "oos_recall_8_summed": 0.057,
+        "in_scope_accuracy_150": 0.18511111111111112,
+        "oos_recall_151": 0.107,
+        "ece_151": 0.09475986128113657,
+        "ece_8_summed": 0.02938045048744911,
+        "nll_151": 4.172640216188356,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.39890909090909094,
+        "oos_recall_8": 0.107,
+        "accuracy_8_summed": 0.43763636363636366,
+        "oos_recall_8_summed": 0.025,
+        "in_scope_accuracy_150": 0.18511111111111112,
+        "oos_recall_151": 0.107,
+        "ece_151": 0.014707356045282846,
+        "ece_8_summed": 0.09970366106366622,
+        "nll_151": 4.043753809051077,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k1-seed43.json b/results/runs/ModernBERT-base-k1-seed43.json
new file mode 100644
index 0000000..c0aa1bc
--- /dev/null
+++ b/results/runs/ModernBERT-base-k1-seed43.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k1-seed43",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 43,
+    "per_intent": null,
+    "k_shot": 1,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k1-seed43",
+    "seed": 43,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 43,
+      "per_intent": null,
+      "k_shot": 1,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 153,
+    "oos_train_rows": 3,
+    "per_intent": null,
+    "k_shot": 1,
+    "train_sample_sha256": "4c659defaf9b9c6f28f4d5e67ef92b466627cdb200623552ba46ced1d2750c9c",
+    "step_plan": {
+      "epoch_steps": 25,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 0.4478936505210004,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 196.976,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 4044603392,
+      "tensor_allocated_bytes": 2429958400
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k1-seed43.npz",
+    "sha256": "c2eae04634a5224685fed57cf0600993eb975a269a9ef61ec97f33598705786b",
+    "bytes": 4828825
+  },
+  "metrics": {
+    "temperature": 1.4297384079855429,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.4529032258064516,
+        "oos_recall_8": 0.09,
+        "accuracy_8_summed": 0.5174193548387097,
+        "oos_recall_8_summed": 0.07,
+        "in_scope_accuracy_150": 0.18933333333333333,
+        "oos_recall_151": 0.09,
+        "ece_151": 0.08042337066770214,
+        "ece_8_summed": 0.08773313474000721,
+        "nll_151": 4.241811247899917,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.4529032258064516,
+        "oos_recall_8": 0.09,
+        "accuracy_8_summed": 0.506774193548387,
+        "oos_recall_8_summed": 0.02,
+        "in_scope_accuracy_150": 0.18933333333333333,
+        "oos_recall_151": 0.09,
+        "ece_151": 0.04550875513927487,
+        "ece_8_summed": 0.16999072812642502,
+        "nll_151": 4.0878691551323305,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.37527272727272726,
+        "oos_recall_8": 0.051,
+        "accuracy_8_summed": 0.42818181818181816,
+        "oos_recall_8_summed": 0.026,
+        "in_scope_accuracy_150": 0.16422222222222221,
+        "oos_recall_151": 0.051,
+        "ece_151": 0.11054610593526615,
+        "ece_8_summed": 0.051745458410637914,
+        "nll_151": 4.533081620803697,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.37527272727272726,
+        "oos_recall_8": 0.051,
+        "accuracy_8_summed": 0.424,
+        "oos_recall_8_summed": 0.011,
+        "in_scope_accuracy_150": 0.16422222222222221,
+        "oos_recall_151": 0.051,
+        "ece_151": 0.019607968650674757,
+        "ece_8_summed": 0.12424620694073861,
+        "nll_151": 4.306438625379609,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k1-seed44.json b/results/runs/ModernBERT-base-k1-seed44.json
new file mode 100644
index 0000000..2970013
--- /dev/null
+++ b/results/runs/ModernBERT-base-k1-seed44.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k1-seed44",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 44,
+    "per_intent": null,
+    "k_shot": 1,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k1-seed44",
+    "seed": 44,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 44,
+      "per_intent": null,
+      "k_shot": 1,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 153,
+    "oos_train_rows": 3,
+    "per_intent": null,
+    "k_shot": 1,
+    "train_sample_sha256": "d69635e70893d70759a46fd0fa54561eae262653eca95898076348b9c660b96a",
+    "step_plan": {
+      "epoch_steps": 25,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 0.467102060693569,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 199.408,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 5185536000,
+      "tensor_allocated_bytes": 2437267200
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k1-seed44.npz",
+    "sha256": "84067b0f3e6ad7ab14ed6b9b919b1b4b24ed1d0c3399dd81573f325adf2021b4",
+    "bytes": 4828958
+  },
+  "metrics": {
+    "temperature": 1.5023981918584055,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.4270967741935484,
+        "oos_recall_8": 0.1,
+        "accuracy_8_summed": 0.4770967741935484,
+        "oos_recall_8_summed": 0.06,
+        "in_scope_accuracy_150": 0.171,
+        "oos_recall_151": 0.1,
+        "ece_151": 0.11029143348135177,
+        "ece_8_summed": 0.040958900101696546,
+        "nll_151": 4.347105806113383,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.4270967741935484,
+        "oos_recall_8": 0.1,
+        "accuracy_8_summed": 0.46838709677419355,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.171,
+        "oos_recall_151": 0.1,
+        "ece_151": 0.03365758444549643,
+        "ece_8_summed": 0.14001109240267726,
+        "nll_151": 4.1448123672829436,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.3770909090909091,
+        "oos_recall_8": 0.084,
+        "accuracy_8_summed": 0.4038181818181818,
+        "oos_recall_8_summed": 0.043,
+        "in_scope_accuracy_150": 0.18133333333333335,
+        "oos_recall_151": 0.084,
+        "ece_151": 0.10927512014704013,
+        "ece_8_summed": 0.03627912490058788,
+        "nll_151": 4.319911567451611,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.3770909090909091,
+        "oos_recall_8": 0.084,
+        "accuracy_8_summed": 0.3896363636363636,
+        "oos_recall_8_summed": 0.016,
+        "in_scope_accuracy_150": 0.18133333333333335,
+        "oos_recall_151": 0.084,
+        "ece_151": 0.034503462554386806,
+        "ece_8_summed": 0.08803586981985514,
+        "nll_151": 4.1356272849718625,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k10-seed42.json b/results/runs/ModernBERT-base-k10-seed42.json
new file mode 100644
index 0000000..83c547c
--- /dev/null
+++ b/results/runs/ModernBERT-base-k10-seed42.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k10-seed42",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 42,
+    "per_intent": null,
+    "k_shot": 10,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k10-seed42",
+    "seed": 42,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 42,
+      "per_intent": null,
+      "k_shot": 10,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 1525,
+    "oos_train_rows": 25,
+    "per_intent": null,
+    "k_shot": 10,
+    "train_sample_sha256": "2606cc878d85759a5885e1b9be35205023b156e6cf8a4eea85bd635c3b492e74",
+    "step_plan": {
+      "epoch_steps": 240,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 0.9079872934462037,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 200.03,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 6267830272,
+      "tensor_allocated_bytes": 2458468608
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k10-seed42.npz",
+    "sha256": "6fff9cd41ff6f9c828b9337b3c190f6e4cb1e41dff62667329f5e0647f81afba",
+    "bytes": 4832287
+  },
+  "metrics": {
+    "temperature": 0.9184197849109466,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9293548387096774,
+        "oos_recall_8": 0.47,
+        "accuracy_8_summed": 0.9309677419354838,
+        "oos_recall_8_summed": 0.35,
+        "in_scope_accuracy_150": 0.8373333333333334,
+        "oos_recall_151": 0.47,
+        "ece_151": 0.024376725580847615,
+        "ece_8_summed": 0.04589626481926541,
+        "nll_151": 0.7291367410591377,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9293548387096774,
+        "oos_recall_8": 0.47,
+        "accuracy_8_summed": 0.9316129032258065,
+        "oos_recall_8_summed": 0.37,
+        "in_scope_accuracy_150": 0.8373333333333334,
+        "oos_recall_151": 0.47,
+        "ece_151": 0.014097506481825579,
+        "ece_8_summed": 0.031418664511382595,
+        "nll_151": 0.7234925471869094,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8325454545454546,
+        "oos_recall_8": 0.35,
+        "accuracy_8_summed": 0.8196363636363636,
+        "oos_recall_8_summed": 0.247,
+        "in_scope_accuracy_150": 0.8415555555555555,
+        "oos_recall_151": 0.35,
+        "ece_151": 0.024724677270693933,
+        "ece_8_summed": 0.020672699270695406,
+        "nll_151": 1.0895860546368958,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8325454545454546,
+        "oos_recall_8": 0.35,
+        "accuracy_8_summed": 0.8229090909090909,
+        "oos_recall_8_summed": 0.267,
+        "in_scope_accuracy_150": 0.8415555555555555,
+        "oos_recall_151": 0.35,
+        "ece_151": 0.025777792483893867,
+        "ece_8_summed": 0.020549552560237067,
+        "nll_151": 1.0907244963426552,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k10-seed43.json b/results/runs/ModernBERT-base-k10-seed43.json
new file mode 100644
index 0000000..3bce32d
--- /dev/null
+++ b/results/runs/ModernBERT-base-k10-seed43.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k10-seed43",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 43,
+    "per_intent": null,
+    "k_shot": 10,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k10-seed43",
+    "seed": 43,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 43,
+      "per_intent": null,
+      "k_shot": 10,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 1525,
+    "oos_train_rows": 25,
+    "per_intent": null,
+    "k_shot": 10,
+    "train_sample_sha256": "072190bdfc028a2e0bc6ec85d39e26ae1b27854281c6226be57b8f8b9d20e165",
+    "step_plan": {
+      "epoch_steps": 240,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 1.0230998934477975,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 201.128,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 6469484544,
+      "tensor_allocated_bytes": 2452242176
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k10-seed43.npz",
+    "sha256": "f08779235eb7af011ab4c6402e86fe393d5cb9cc0d341ecfb01992fe97d52004",
+    "bytes": 4832661
+  },
+  "metrics": {
+    "temperature": 0.8845124929504956,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9283870967741935,
+        "oos_recall_8": 0.34,
+        "accuracy_8_summed": 0.9319354838709677,
+        "oos_recall_8_summed": 0.29,
+        "in_scope_accuracy_150": 0.8496666666666667,
+        "oos_recall_151": 0.34,
+        "ece_151": 0.03587922351400995,
+        "ece_8_summed": 0.04840708799679498,
+        "nll_151": 0.6964777905638856,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9283870967741935,
+        "oos_recall_8": 0.34,
+        "accuracy_8_summed": 0.932258064516129,
+        "oos_recall_8_summed": 0.31,
+        "in_scope_accuracy_150": 0.8496666666666667,
+        "oos_recall_151": 0.34,
+        "ece_151": 0.022249355396746755,
+        "ece_8_summed": 0.025771430974333435,
+        "nll_151": 0.6854858611258435,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8156363636363636,
+        "oos_recall_8": 0.252,
+        "accuracy_8_summed": 0.8090909090909091,
+        "oos_recall_8_summed": 0.191,
+        "in_scope_accuracy_150": 0.8475555555555555,
+        "oos_recall_151": 0.252,
+        "ece_151": 0.017374373135857372,
+        "ece_8_summed": 0.03038335393988119,
+        "nll_151": 1.1779487828345685,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8156363636363636,
+        "oos_recall_8": 0.252,
+        "accuracy_8_summed": 0.8123636363636364,
+        "oos_recall_8_summed": 0.206,
+        "in_scope_accuracy_150": 0.8475555555555555,
+        "oos_recall_151": 0.252,
+        "ece_151": 0.031486124405980795,
+        "ece_8_summed": 0.0377950181860128,
+        "nll_151": 1.1940117285601222,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k10-seed44.json b/results/runs/ModernBERT-base-k10-seed44.json
new file mode 100644
index 0000000..33d82f1
--- /dev/null
+++ b/results/runs/ModernBERT-base-k10-seed44.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k10-seed44",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 44,
+    "per_intent": null,
+    "k_shot": 10,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k10-seed44",
+    "seed": 44,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 44,
+      "per_intent": null,
+      "k_shot": 10,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 1525,
+    "oos_train_rows": 25,
+    "per_intent": null,
+    "k_shot": 10,
+    "train_sample_sha256": "31a30630bb8f711a55173e8837eaf4500c2c5dcd5efe461b561a04e843cddd55",
+    "step_plan": {
+      "epoch_steps": 240,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 0.9484245905611897,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 198.757,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 6226051072,
+      "tensor_allocated_bytes": 2447162624
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k10-seed44.npz",
+    "sha256": "6c20ced25748c4ed64688269d0ab4fe74f0b32ae9af565c8ccd4cf6ed72140d9",
+    "bytes": 4831998
+  },
+  "metrics": {
+    "temperature": 0.8746924987303141,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9045161290322581,
+        "oos_recall_8": 0.34,
+        "accuracy_8_summed": 0.9125806451612903,
+        "oos_recall_8_summed": 0.26,
+        "in_scope_accuracy_150": 0.8126666666666666,
+        "oos_recall_151": 0.34,
+        "ece_151": 0.04137772422081867,
+        "ece_8_summed": 0.057719482411671165,
+        "nll_151": 0.8347197641741616,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9045161290322581,
+        "oos_recall_8": 0.34,
+        "accuracy_8_summed": 0.9106451612903226,
+        "oos_recall_8_summed": 0.26,
+        "in_scope_accuracy_150": 0.8126666666666666,
+        "oos_recall_151": 0.34,
+        "ece_151": 0.014429802278052075,
+        "ece_8_summed": 0.02655580763688327,
+        "nll_151": 0.819808676977611,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8116363636363636,
+        "oos_recall_8": 0.262,
+        "accuracy_8_summed": 0.8074545454545454,
+        "oos_recall_8_summed": 0.202,
+        "in_scope_accuracy_150": 0.8168888888888889,
+        "oos_recall_151": 0.262,
+        "ece_151": 0.025000392629662055,
+        "ece_8_summed": 0.03440883544604676,
+        "nll_151": 1.2684326128759011,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8116363636363636,
+        "oos_recall_8": 0.262,
+        "accuracy_8_summed": 0.8098181818181818,
+        "oos_recall_8_summed": 0.221,
+        "in_scope_accuracy_150": 0.8168888888888889,
+        "oos_recall_151": 0.262,
+        "ece_151": 0.02704130975384563,
+        "ece_8_summed": 0.024878763018165655,
+        "nll_151": 1.2772988998728334,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k100-oos0-seed42.json b/results/runs/ModernBERT-base-k100-oos0-seed42.json
new file mode 100644
index 0000000..78a25e4
--- /dev/null
+++ b/results/runs/ModernBERT-base-k100-oos0-seed42.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k100-oos0-seed42",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 42,
+    "per_intent": null,
+    "k_shot": 100,
+    "oos_train": 0,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k100-oos0-seed42",
+    "seed": 42,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 42,
+      "per_intent": null,
+      "k_shot": 100,
+      "oos_train": 0,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 15000,
+    "oos_train_rows": 0,
+    "per_intent": null,
+    "k_shot": 100,
+    "train_sample_sha256": "88418dcc1027e9ccb3d1bf8230bbf9471a749d252302ee65f154e0228fc9b81d",
+    "step_plan": {
+      "epoch_steps": 2345,
+      "min_train_steps": 400,
+      "planned_steps": 2345,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 2345,
+    "train_loss": 0.3709846070326213,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 1177.504,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 4692,
+      "driver_allocated_bytes": 6209683456,
+      "tensor_allocated_bytes": 2446374144
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k100-oos0-seed42.npz",
+    "sha256": "c83e3c569facc09d5eb14cc3f837883fb59b0d5d68583257e9366e3750066be6",
+    "bytes": 4833660
+  },
+  "metrics": {
+    "temperature": 1.5191609932496655,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9593548387096774,
+        "oos_recall_8": 0.0,
+        "accuracy_8_summed": 0.9596774193548387,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.975,
+        "oos_recall_151": 0.0,
+        "ece_151": 0.022891050344983923,
+        "ece_8_summed": 0.020053385402747814,
+        "nll_151": 0.43463425453049326,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9593548387096774,
+        "oos_recall_8": 0.0,
+        "accuracy_8_summed": 0.96,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.975,
+        "oos_recall_151": 0.0,
+        "ece_151": 0.017233740711870685,
+        "ece_8_summed": 0.018894850761206207,
+        "nll_151": 0.36378383907237877,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8087272727272727,
+        "oos_recall_8": 0.0,
+        "accuracy_8_summed": 0.8089090909090909,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.9697777777777777,
+        "oos_recall_151": 0.0,
+        "ece_151": 0.10350165789833915,
+        "ece_8_summed": 0.11837414562368441,
+        "nll_151": 1.7415790493604466,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8087272727272727,
+        "oos_recall_8": 0.0,
+        "accuracy_8_summed": 0.8090909090909091,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.9697777777777777,
+        "oos_recall_151": 0.0,
+        "ece_151": 0.04063383226384718,
+        "ece_8_summed": 0.07069625594861015,
+        "nll_151": 1.3528937375097447,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k100-oos0-seed43.json b/results/runs/ModernBERT-base-k100-oos0-seed43.json
new file mode 100644
index 0000000..70836da
--- /dev/null
+++ b/results/runs/ModernBERT-base-k100-oos0-seed43.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k100-oos0-seed43",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 43,
+    "per_intent": null,
+    "k_shot": 100,
+    "oos_train": 0,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k100-oos0-seed43",
+    "seed": 43,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 43,
+      "per_intent": null,
+      "k_shot": 100,
+      "oos_train": 0,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 15000,
+    "oos_train_rows": 0,
+    "per_intent": null,
+    "k_shot": 100,
+    "train_sample_sha256": "88418dcc1027e9ccb3d1bf8230bbf9471a749d252302ee65f154e0228fc9b81d",
+    "step_plan": {
+      "epoch_steps": 2345,
+      "min_train_steps": 400,
+      "planned_steps": 2345,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 2345,
+    "train_loss": 0.38205325560274916,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 1178.345,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 4692,
+      "driver_allocated_bytes": 7275118592,
+      "tensor_allocated_bytes": 2459188992
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k100-oos0-seed43.npz",
+    "sha256": "09029343a7529ac08bbc145fef45edcc1bfb04f6b976dc956473c6c546175343",
+    "bytes": 4833643
+  },
+  "metrics": {
+    "temperature": 1.5145881679192652,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9609677419354838,
+        "oos_recall_8": 0.0,
+        "accuracy_8_summed": 0.9612903225806452,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.9743333333333334,
+        "oos_recall_151": 0.0,
+        "ece_151": 0.024003477742831326,
+        "ece_8_summed": 0.019987309702905943,
+        "nll_151": 0.43790645504935893,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9609677419354838,
+        "oos_recall_8": 0.0,
+        "accuracy_8_summed": 0.9616129032258065,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.9743333333333334,
+        "oos_recall_151": 0.0,
+        "ece_151": 0.016942668556257024,
+        "ece_8_summed": 0.018241076128710808,
+        "nll_151": 0.36715523886889634,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8094545454545454,
+        "oos_recall_8": 0.0,
+        "accuracy_8_summed": 0.8098181818181818,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.9684444444444444,
+        "oos_recall_151": 0.0,
+        "ece_151": 0.10065328409915852,
+        "ece_8_summed": 0.1163848034076638,
+        "nll_151": 1.9320825344462031,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8094545454545454,
+        "oos_recall_8": 0.0,
+        "accuracy_8_summed": 0.8098181818181818,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.9684444444444444,
+        "oos_recall_151": 0.0,
+        "ece_151": 0.04091992365834156,
+        "ece_8_summed": 0.07135066773389567,
+        "nll_151": 1.4829090615333034,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k100-oos0-seed44.json b/results/runs/ModernBERT-base-k100-oos0-seed44.json
new file mode 100644
index 0000000..f3ed56d
--- /dev/null
+++ b/results/runs/ModernBERT-base-k100-oos0-seed44.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k100-oos0-seed44",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 44,
+    "per_intent": null,
+    "k_shot": 100,
+    "oos_train": 0,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k100-oos0-seed44",
+    "seed": 44,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 44,
+      "per_intent": null,
+      "k_shot": 100,
+      "oos_train": 0,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 15000,
+    "oos_train_rows": 0,
+    "per_intent": null,
+    "k_shot": 100,
+    "train_sample_sha256": "88418dcc1027e9ccb3d1bf8230bbf9471a749d252302ee65f154e0228fc9b81d",
+    "step_plan": {
+      "epoch_steps": 2345,
+      "min_train_steps": 400,
+      "planned_steps": 2345,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 2345,
+    "train_loss": 0.37920121051101036,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 1178.26,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 4692,
+      "driver_allocated_bytes": 6268403712,
+      "tensor_allocated_bytes": 2443689728
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k100-oos0-seed44.npz",
+    "sha256": "6edfa2106d6d72ae26449b5598cff759ae601c44c42648de129ceab6e5530a61",
+    "bytes": 4833040
+  },
+  "metrics": {
+    "temperature": 1.4831388158547218,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.96,
+        "oos_recall_8": 0.0,
+        "accuracy_8_summed": 0.9596774193548387,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.978,
+        "oos_recall_151": 0.0,
+        "ece_151": 0.01954443203730747,
+        "ece_8_summed": 0.021091077665633076,
+        "nll_151": 0.3997870011025945,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.96,
+        "oos_recall_8": 0.0,
+        "accuracy_8_summed": 0.9593548387096774,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.978,
+        "oos_recall_151": 0.0,
+        "ece_151": 0.01650891817020664,
+        "ece_8_summed": 0.01599695954118026,
+        "nll_151": 0.3409452760300024,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8090909090909091,
+        "oos_recall_8": 0.0,
+        "accuracy_8_summed": 0.8092727272727273,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.9686666666666667,
+        "oos_recall_151": 0.0,
+        "ece_151": 0.1006785674476782,
+        "ece_8_summed": 0.11579578712335917,
+        "nll_151": 1.7920862841665879,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8090909090909091,
+        "oos_recall_8": 0.0,
+        "accuracy_8_summed": 0.8105454545454546,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.9686666666666667,
+        "oos_recall_151": 0.0,
+        "ece_151": 0.04362623909763135,
+        "ece_8_summed": 0.07143823736678498,
+        "nll_151": 1.4051037104075679,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k100-seed42.json b/results/runs/ModernBERT-base-k100-seed42.json
new file mode 100644
index 0000000..a9d10c3
--- /dev/null
+++ b/results/runs/ModernBERT-base-k100-seed42.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k100-seed42",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 42,
+    "per_intent": null,
+    "k_shot": 100,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k100-seed42",
+    "seed": 42,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 42,
+      "per_intent": null,
+      "k_shot": 100,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 15250,
+    "oos_train_rows": 250,
+    "per_intent": null,
+    "k_shot": 100,
+    "train_sample_sha256": "7a3ebd2cd2f985a26e815c229cc54add6763f783bee572c5e7948636c66c7553",
+    "step_plan": {
+      "epoch_steps": 2385,
+      "min_train_steps": 400,
+      "planned_steps": 2385,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 2385,
+    "train_loss": 0.4063441063872923,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 1199.467,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 4772,
+      "driver_allocated_bytes": 7442808832,
+      "tensor_allocated_bytes": 2441854720
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k100-seed42.npz",
+    "sha256": "208f372ab9096fa4f95a006328c8ef766cb65144005d9df8f4bad1c391a100a9",
+    "bytes": 4833253
+  },
+  "metrics": {
+    "temperature": 1.2527106150776763,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9809677419354839,
+        "oos_recall_8": 0.76,
+        "accuracy_8_summed": 0.9809677419354839,
+        "oos_recall_8_summed": 0.75,
+        "in_scope_accuracy_150": 0.9706666666666667,
+        "oos_recall_151": 0.76,
+        "ece_151": 0.014950898738364006,
+        "ece_8_summed": 0.007918813716323289,
+        "nll_151": 0.1788101014396808,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9809677419354839,
+        "oos_recall_8": 0.76,
+        "accuracy_8_summed": 0.9803225806451613,
+        "oos_recall_8_summed": 0.75,
+        "in_scope_accuracy_150": 0.9706666666666667,
+        "oos_recall_151": 0.76,
+        "ece_151": 0.00970896775556529,
+        "ece_8_summed": 0.004053248818184231,
+        "nll_151": 0.16802976904688888,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.9172727272727272,
+        "oos_recall_8": 0.609,
+        "accuracy_8_summed": 0.9152727272727272,
+        "oos_recall_8_summed": 0.598,
+        "in_scope_accuracy_150": 0.9673333333333334,
+        "oos_recall_151": 0.609,
+        "ece_151": 0.038343289989175966,
+        "ece_8_summed": 0.039993248954056995,
+        "nll_151": 0.45495326852688894,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.9172727272727272,
+        "oos_recall_8": 0.609,
+        "accuracy_8_summed": 0.9114545454545454,
+        "oos_recall_8_summed": 0.575,
+        "in_scope_accuracy_150": 0.9673333333333334,
+        "oos_recall_151": 0.609,
+        "ece_151": 0.02261743361564421,
+        "ece_8_summed": 0.02515016457563663,
+        "nll_151": 0.4245353050815765,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k100-seed43.json b/results/runs/ModernBERT-base-k100-seed43.json
new file mode 100644
index 0000000..f49239f
--- /dev/null
+++ b/results/runs/ModernBERT-base-k100-seed43.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k100-seed43",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 43,
+    "per_intent": null,
+    "k_shot": 100,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k100-seed43",
+    "seed": 43,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 43,
+      "per_intent": null,
+      "k_shot": 100,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 15250,
+    "oos_train_rows": 250,
+    "per_intent": null,
+    "k_shot": 100,
+    "train_sample_sha256": "7a3ebd2cd2f985a26e815c229cc54add6763f783bee572c5e7948636c66c7553",
+    "step_plan": {
+      "epoch_steps": 2385,
+      "min_train_steps": 400,
+      "planned_steps": 2385,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 2385,
+    "train_loss": 0.3931701684148057,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 1198.802,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 4772,
+      "driver_allocated_bytes": 6411010048,
+      "tensor_allocated_bytes": 2442641152
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k100-seed43.npz",
+    "sha256": "58c3bd83838784f23d7e9977921b4a021797799b9a514b3c4cd449c641cd5257",
+    "bytes": 4832996
+  },
+  "metrics": {
+    "temperature": 1.2591147402079235,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9864516129032258,
+        "oos_recall_8": 0.77,
+        "accuracy_8_summed": 0.9854838709677419,
+        "oos_recall_8_summed": 0.77,
+        "in_scope_accuracy_150": 0.9753333333333334,
+        "oos_recall_151": 0.77,
+        "ece_151": 0.015325338732523154,
+        "ece_8_summed": 0.006167844156927173,
+        "nll_151": 0.17141198149905962,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9864516129032258,
+        "oos_recall_8": 0.77,
+        "accuracy_8_summed": 0.9848387096774194,
+        "oos_recall_8_summed": 0.74,
+        "in_scope_accuracy_150": 0.9753333333333334,
+        "oos_recall_151": 0.77,
+        "ece_151": 0.012448287862137173,
+        "ece_8_summed": 0.005487181110093419,
+        "nll_151": 0.16054260582058097,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.9192727272727272,
+        "oos_recall_8": 0.608,
+        "accuracy_8_summed": 0.9150909090909091,
+        "oos_recall_8_summed": 0.584,
+        "in_scope_accuracy_150": 0.9648888888888889,
+        "oos_recall_151": 0.608,
+        "ece_151": 0.04181985921853623,
+        "ece_8_summed": 0.04218867534445946,
+        "nll_151": 0.48089972845448903,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.9192727272727272,
+        "oos_recall_8": 0.608,
+        "accuracy_8_summed": 0.912,
+        "oos_recall_8_summed": 0.564,
+        "in_scope_accuracy_150": 0.9648888888888889,
+        "oos_recall_151": 0.608,
+        "ece_151": 0.027798525084175747,
+        "ece_8_summed": 0.026145046810305425,
+        "nll_151": 0.44377613396361754,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k100-seed44.json b/results/runs/ModernBERT-base-k100-seed44.json
new file mode 100644
index 0000000..faa6a76
--- /dev/null
+++ b/results/runs/ModernBERT-base-k100-seed44.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k100-seed44",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 44,
+    "per_intent": null,
+    "k_shot": 100,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k100-seed44",
+    "seed": 44,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 44,
+      "per_intent": null,
+      "k_shot": 100,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 15250,
+    "oos_train_rows": 250,
+    "per_intent": null,
+    "k_shot": 100,
+    "train_sample_sha256": "7a3ebd2cd2f985a26e815c229cc54add6763f783bee572c5e7948636c66c7553",
+    "step_plan": {
+      "epoch_steps": 2385,
+      "min_train_steps": 400,
+      "planned_steps": 2385,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 2385,
+    "train_loss": 0.39723677745155317,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 1196.633,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 4772,
+      "driver_allocated_bytes": 6260097024,
+      "tensor_allocated_bytes": 2442804992
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k100-seed44.npz",
+    "sha256": "37ebd645a97a654308f9f50cf4501b6b1c5bfed5bbd064fe591d8587a4873966",
+    "bytes": 4833398
+  },
+  "metrics": {
+    "temperature": 1.2448123342775337,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9829032258064516,
+        "oos_recall_8": 0.82,
+        "accuracy_8_summed": 0.9835483870967742,
+        "oos_recall_8_summed": 0.81,
+        "in_scope_accuracy_150": 0.969,
+        "oos_recall_151": 0.82,
+        "ece_151": 0.016481689215991076,
+        "ece_8_summed": 0.007245674975629916,
+        "nll_151": 0.16857021339137326,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9829032258064516,
+        "oos_recall_8": 0.82,
+        "accuracy_8_summed": 0.9829032258064516,
+        "oos_recall_8_summed": 0.8,
+        "in_scope_accuracy_150": 0.969,
+        "oos_recall_151": 0.82,
+        "ece_151": 0.01099529572537165,
+        "ece_8_summed": 0.006647299679882751,
+        "nll_151": 0.1585533549934568,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.9196363636363636,
+        "oos_recall_8": 0.615,
+        "accuracy_8_summed": 0.9149090909090909,
+        "oos_recall_8_summed": 0.586,
+        "in_scope_accuracy_150": 0.9675555555555555,
+        "oos_recall_151": 0.615,
+        "ece_151": 0.03825491648948749,
+        "ece_8_summed": 0.04046695668417978,
+        "nll_151": 0.46220334957256515,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.9196363636363636,
+        "oos_recall_8": 0.615,
+        "accuracy_8_summed": 0.9114545454545454,
+        "oos_recall_8_summed": 0.563,
+        "in_scope_accuracy_150": 0.9675555555555555,
+        "oos_recall_151": 0.615,
+        "ece_151": 0.023076559095941623,
+        "ece_8_summed": 0.025331000073251158,
+        "nll_151": 0.4320285879525438,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k25-seed42.json b/results/runs/ModernBERT-base-k25-seed42.json
new file mode 100644
index 0000000..449e1b4
--- /dev/null
+++ b/results/runs/ModernBERT-base-k25-seed42.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k25-seed42",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 42,
+    "per_intent": null,
+    "k_shot": 25,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k25-seed42",
+    "seed": 42,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 42,
+      "per_intent": null,
+      "k_shot": 25,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 3813,
+    "oos_train_rows": 63,
+    "per_intent": null,
+    "k_shot": 25,
+    "train_sample_sha256": "51184b580bdbe58ab700d2545887a30c442950389b98497c3b7edaa7ca47174c",
+    "step_plan": {
+      "epoch_steps": 600,
+      "min_train_steps": 400,
+      "planned_steps": 600,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 600,
+    "train_loss": 0.9201174671451251,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 301.341,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 1202,
+      "driver_allocated_bytes": 5227806720,
+      "tensor_allocated_bytes": 2445590272
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k25-seed42.npz",
+    "sha256": "da2c8b7a2fcefcda47211bec4d2248e7ca5288eda05cc2572ade0945abea36ff",
+    "bytes": 4833547
+  },
+  "metrics": {
+    "temperature": 0.9846314527038342,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9596774193548387,
+        "oos_recall_8": 0.58,
+        "accuracy_8_summed": 0.9609677419354838,
+        "oos_recall_8_summed": 0.53,
+        "in_scope_accuracy_150": 0.9316666666666666,
+        "oos_recall_151": 0.58,
+        "ece_151": 0.008817676065802993,
+        "ece_8_summed": 0.010362237455127148,
+        "nll_151": 0.3254486185610647,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9596774193548387,
+        "oos_recall_8": 0.58,
+        "accuracy_8_summed": 0.9609677419354838,
+        "oos_recall_8_summed": 0.53,
+        "in_scope_accuracy_150": 0.9316666666666666,
+        "oos_recall_151": 0.58,
+        "ece_151": 0.0107072798818727,
+        "ece_8_summed": 0.009979343530069424,
+        "nll_151": 0.3253482737774197,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8776363636363637,
+        "oos_recall_8": 0.453,
+        "accuracy_8_summed": 0.868,
+        "oos_recall_8_summed": 0.391,
+        "in_scope_accuracy_150": 0.9228888888888889,
+        "oos_recall_151": 0.453,
+        "ece_151": 0.02407467251354046,
+        "ece_8_summed": 0.033702748197960174,
+        "nll_151": 0.7089586131205532,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8776363636363637,
+        "oos_recall_8": 0.453,
+        "accuracy_8_summed": 0.8687272727272727,
+        "oos_recall_8_summed": 0.394,
+        "in_scope_accuracy_150": 0.9228888888888889,
+        "oos_recall_151": 0.453,
+        "ece_151": 0.025322825629980027,
+        "ece_8_summed": 0.03523535261128395,
+        "nll_151": 0.7106040357631259,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k25-seed43.json b/results/runs/ModernBERT-base-k25-seed43.json
new file mode 100644
index 0000000..950f435
--- /dev/null
+++ b/results/runs/ModernBERT-base-k25-seed43.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k25-seed43",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 43,
+    "per_intent": null,
+    "k_shot": 25,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k25-seed43",
+    "seed": 43,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 43,
+      "per_intent": null,
+      "k_shot": 25,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 3813,
+    "oos_train_rows": 63,
+    "per_intent": null,
+    "k_shot": 25,
+    "train_sample_sha256": "92f91be6e0948dbc01e82b81485c72829dd232443955fa0ae214b1451a0a704f",
+    "step_plan": {
+      "epoch_steps": 600,
+      "min_train_steps": 400,
+      "planned_steps": 600,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 600,
+    "train_loss": 1.1246477735042573,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 301.325,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 1202,
+      "driver_allocated_bytes": 5261443072,
+      "tensor_allocated_bytes": 2450603776
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k25-seed43.npz",
+    "sha256": "ac640bffd502ea71d7dd5b5f9cc18d7068b6b8b1b950acbf3a2fd58a3c50f1b6",
+    "bytes": 4833666
+  },
+  "metrics": {
+    "temperature": 0.9479710330394893,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9548387096774194,
+        "oos_recall_8": 0.41,
+        "accuracy_8_summed": 0.954516129032258,
+        "oos_recall_8_summed": 0.36,
+        "in_scope_accuracy_150": 0.9036666666666666,
+        "oos_recall_151": 0.41,
+        "ece_151": 0.010969482942210248,
+        "ece_8_summed": 0.019063760727659744,
+        "nll_151": 0.41825109788820103,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9548387096774194,
+        "oos_recall_8": 0.41,
+        "accuracy_8_summed": 0.9548387096774194,
+        "oos_recall_8_summed": 0.37,
+        "in_scope_accuracy_150": 0.9036666666666666,
+        "oos_recall_151": 0.41,
+        "ece_151": 0.013912018072524134,
+        "ece_8_summed": 0.015554030062465345,
+        "nll_151": 0.41686162189056986,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8592727272727273,
+        "oos_recall_8": 0.365,
+        "accuracy_8_summed": 0.8494545454545455,
+        "oos_recall_8_summed": 0.305,
+        "in_scope_accuracy_150": 0.9064444444444445,
+        "oos_recall_151": 0.365,
+        "ece_151": 0.01728390469011367,
+        "ece_8_summed": 0.03300547658995442,
+        "nll_151": 0.8325383350650202,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8592727272727273,
+        "oos_recall_8": 0.365,
+        "accuracy_8_summed": 0.8501818181818181,
+        "oos_recall_8_summed": 0.311,
+        "in_scope_accuracy_150": 0.9064444444444445,
+        "oos_recall_151": 0.365,
+        "ece_151": 0.026375631657722408,
+        "ece_8_summed": 0.04085960590320895,
+        "nll_151": 0.8389104013528332,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k25-seed44.json b/results/runs/ModernBERT-base-k25-seed44.json
new file mode 100644
index 0000000..f050992
--- /dev/null
+++ b/results/runs/ModernBERT-base-k25-seed44.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k25-seed44",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 44,
+    "per_intent": null,
+    "k_shot": 25,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k25-seed44",
+    "seed": 44,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 44,
+      "per_intent": null,
+      "k_shot": 25,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 3813,
+    "oos_train_rows": 63,
+    "per_intent": null,
+    "k_shot": 25,
+    "train_sample_sha256": "790cd7cf9eb24143276f6c59c7bac29c4a0b42f14bef38e9634e96253e6315fb",
+    "step_plan": {
+      "epoch_steps": 600,
+      "min_train_steps": 400,
+      "planned_steps": 600,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 600,
+    "train_loss": 0.860421137685577,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 303.429,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 1202,
+      "driver_allocated_bytes": 6276546560,
+      "tensor_allocated_bytes": 2449686272
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k25-seed44.npz",
+    "sha256": "cd9b8473e1b5b0d4597feb76bdf99d27464910b85b536eb7169193676c94ce68",
+    "bytes": 4832954
+  },
+  "metrics": {
+    "temperature": 0.9525033395582145,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.957741935483871,
+        "oos_recall_8": 0.54,
+        "accuracy_8_summed": 0.9593548387096774,
+        "oos_recall_8_summed": 0.46,
+        "in_scope_accuracy_150": 0.9313333333333333,
+        "oos_recall_151": 0.54,
+        "ece_151": 0.013387174137220294,
+        "ece_8_summed": 0.010594975333864338,
+        "nll_151": 0.33733990364866845,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.957741935483871,
+        "oos_recall_8": 0.54,
+        "accuracy_8_summed": 0.9593548387096774,
+        "oos_recall_8_summed": 0.47,
+        "in_scope_accuracy_150": 0.9313333333333333,
+        "oos_recall_151": 0.54,
+        "ece_151": 0.013027089357715637,
+        "ece_8_summed": 0.006646744076316116,
+        "nll_151": 0.33637794541587396,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.866,
+        "oos_recall_8": 0.378,
+        "accuracy_8_summed": 0.8596363636363636,
+        "oos_recall_8_summed": 0.33,
+        "in_scope_accuracy_150": 0.9268888888888889,
+        "oos_recall_151": 0.378,
+        "ece_151": 0.02538309806937753,
+        "ece_8_summed": 0.03842383351409838,
+        "nll_151": 0.7504001688655524,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.866,
+        "oos_recall_8": 0.378,
+        "accuracy_8_summed": 0.8601818181818182,
+        "oos_recall_8_summed": 0.334,
+        "in_scope_accuracy_150": 0.9268888888888889,
+        "oos_recall_151": 0.378,
+        "ece_151": 0.032481646351823996,
+        "ece_8_summed": 0.04485471997743555,
+        "nll_151": 0.7572399543907828,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k5-seed42.json b/results/runs/ModernBERT-base-k5-seed42.json
new file mode 100644
index 0000000..6bc8088
--- /dev/null
+++ b/results/runs/ModernBERT-base-k5-seed42.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k5-seed42",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 42,
+    "per_intent": null,
+    "k_shot": 5,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k5-seed42",
+    "seed": 42,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 42,
+      "per_intent": null,
+      "k_shot": 5,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 763,
+    "oos_train_rows": 13,
+    "per_intent": null,
+    "k_shot": 5,
+    "train_sample_sha256": "bfabf08971d3a16a502e25b2d5a7f1ce204504a526d5b3cb34fec9346dd8caa7",
+    "step_plan": {
+      "epoch_steps": 120,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 0.74402411722338,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 199.008,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 5060034560,
+      "tensor_allocated_bytes": 2433893120
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k5-seed42.npz",
+    "sha256": "8ffbbf00d37abfc1fc2e3601e96ed26c71c6cb2452c8c88d32375674f22996c1",
+    "bytes": 4830959
+  },
+  "metrics": {
+    "temperature": 0.9099450771752746,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.8416129032258064,
+        "oos_recall_8": 0.24,
+        "accuracy_8_summed": 0.8587096774193549,
+        "oos_recall_8_summed": 0.17,
+        "in_scope_accuracy_150": 0.675,
+        "oos_recall_151": 0.24,
+        "ece_151": 0.062161651989533856,
+        "ece_8_summed": 0.11015694113657264,
+        "nll_151": 1.4819869794603935,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.8416129032258064,
+        "oos_recall_8": 0.24,
+        "accuracy_8_summed": 0.8570967741935483,
+        "oos_recall_8_summed": 0.18,
+        "in_scope_accuracy_150": 0.675,
+        "oos_recall_151": 0.24,
+        "ece_151": 0.031062836801666285,
+        "ece_8_summed": 0.07725423582198379,
+        "nll_151": 1.4710546213166127,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.7363636363636363,
+        "oos_recall_8": 0.19,
+        "accuracy_8_summed": 0.7423636363636363,
+        "oos_recall_8_summed": 0.145,
+        "in_scope_accuracy_150": 0.6735555555555556,
+        "oos_recall_151": 0.19,
+        "ece_151": 0.03440568139122288,
+        "ece_8_summed": 0.06359228552946722,
+        "nll_151": 1.867806133896425,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.7363636363636363,
+        "oos_recall_8": 0.19,
+        "accuracy_8_summed": 0.7427272727272727,
+        "oos_recall_8_summed": 0.156,
+        "in_scope_accuracy_150": 0.6735555555555556,
+        "oos_recall_151": 0.19,
+        "ece_151": 0.021633810470886253,
+        "ece_8_summed": 0.03644023952357199,
+        "nll_151": 1.8724236388567417,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k5-seed43.json b/results/runs/ModernBERT-base-k5-seed43.json
new file mode 100644
index 0000000..12dca2f
--- /dev/null
+++ b/results/runs/ModernBERT-base-k5-seed43.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k5-seed43",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 43,
+    "per_intent": null,
+    "k_shot": 5,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k5-seed43",
+    "seed": 43,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 43,
+      "per_intent": null,
+      "k_shot": 5,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 763,
+    "oos_train_rows": 13,
+    "per_intent": null,
+    "k_shot": 5,
+    "train_sample_sha256": "4e289193d9669c26c40d083246f6c34426bca5ff041ac526796a5f976ca9c638",
+    "step_plan": {
+      "epoch_steps": 120,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 0.7925192982531007,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 199.147,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 5185781760,
+      "tensor_allocated_bytes": 2433042688
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k5-seed43.npz",
+    "sha256": "561ef2079fd97de9a221e029b2589c50ccd17a579423fa9503ef95fe2944730c",
+    "bytes": 4831188
+  },
+  "metrics": {
+    "temperature": 0.9592186111654715,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.7683870967741936,
+        "oos_recall_8": 0.18,
+        "accuracy_8_summed": 0.7906451612903226,
+        "oos_recall_8_summed": 0.16,
+        "in_scope_accuracy_150": 0.5643333333333334,
+        "oos_recall_151": 0.18,
+        "ece_151": 0.04534183612251056,
+        "ece_8_summed": 0.1164177289507522,
+        "nll_151": 2.00203982718905,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.7683870967741936,
+        "oos_recall_8": 0.18,
+        "accuracy_8_summed": 0.7880645161290323,
+        "oos_recall_8_summed": 0.16,
+        "in_scope_accuracy_150": 0.5643333333333334,
+        "oos_recall_151": 0.18,
+        "ece_151": 0.0348051747813108,
+        "ece_8_summed": 0.09846275343267964,
+        "nll_151": 1.9996422962274394,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.6552727272727272,
+        "oos_recall_8": 0.105,
+        "accuracy_8_summed": 0.6736363636363636,
+        "oos_recall_8_summed": 0.065,
+        "in_scope_accuracy_150": 0.5782222222222222,
+        "oos_recall_151": 0.105,
+        "ece_151": 0.0270418416547968,
+        "ece_8_summed": 0.06212333171143692,
+        "nll_151": 2.3877260790976638,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.6552727272727272,
+        "oos_recall_8": 0.105,
+        "accuracy_8_summed": 0.6749090909090909,
+        "oos_recall_8_summed": 0.071,
+        "in_scope_accuracy_150": 0.5782222222222222,
+        "oos_recall_151": 0.105,
+        "ece_151": 0.021417415634296718,
+        "ece_8_summed": 0.04896126734410877,
+        "nll_151": 2.3937863979657847,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k5-seed44.json b/results/runs/ModernBERT-base-k5-seed44.json
new file mode 100644
index 0000000..145e5be
--- /dev/null
+++ b/results/runs/ModernBERT-base-k5-seed44.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k5-seed44",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 44,
+    "per_intent": null,
+    "k_shot": 5,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k5-seed44",
+    "seed": 44,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 44,
+      "per_intent": null,
+      "k_shot": 5,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 763,
+    "oos_train_rows": 13,
+    "per_intent": null,
+    "k_shot": 5,
+    "train_sample_sha256": "ba57c52c7c416c611ba215bb10a393fba470ec511aaebad2b416914ca674988d",
+    "step_plan": {
+      "epoch_steps": 120,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 0.9212492994784407,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 200.396,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 6100303872,
+      "tensor_allocated_bytes": 2426519296
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k5-seed44.npz",
+    "sha256": "9c6b23ca81db2a58e2f444a13533edcf9687f6d4020ba6ec6ecc52915f26e784",
+    "bytes": 4831385
+  },
+  "metrics": {
+    "temperature": 0.87849637852735,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.8254838709677419,
+        "oos_recall_8": 0.23,
+        "accuracy_8_summed": 0.844516129032258,
+        "oos_recall_8_summed": 0.19,
+        "in_scope_accuracy_150": 0.656,
+        "oos_recall_151": 0.23,
+        "ece_151": 0.06026716941311009,
+        "ece_8_summed": 0.11740008671807475,
+        "nll_151": 1.5078675452929309,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.8254838709677419,
+        "oos_recall_8": 0.23,
+        "accuracy_8_summed": 0.8429032258064516,
+        "oos_recall_8_summed": 0.19,
+        "in_scope_accuracy_150": 0.656,
+        "oos_recall_151": 0.23,
+        "ece_151": 0.033669443693063,
+        "ece_8_summed": 0.07118383052399721,
+        "nll_151": 1.4871205021259406,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.7067272727272728,
+        "oos_recall_8": 0.147,
+        "accuracy_8_summed": 0.7152727272727273,
+        "oos_recall_8_summed": 0.095,
+        "in_scope_accuracy_150": 0.6513333333333333,
+        "oos_recall_151": 0.147,
+        "ece_151": 0.022416913453683854,
+        "ece_8_summed": 0.05911494603705631,
+        "nll_151": 2.0156527924089267,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.7067272727272728,
+        "oos_recall_8": 0.147,
+        "accuracy_8_summed": 0.7121818181818181,
+        "oos_recall_8_summed": 0.103,
+        "in_scope_accuracy_150": 0.6513333333333333,
+        "oos_recall_151": 0.147,
+        "ece_151": 0.04148688977876454,
+        "ece_8_summed": 0.027790426101590188,
+        "nll_151": 2.0387874164737303,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k50-seed42.json b/results/runs/ModernBERT-base-k50-seed42.json
new file mode 100644
index 0000000..2e2ea72
--- /dev/null
+++ b/results/runs/ModernBERT-base-k50-seed42.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k50-seed42",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 42,
+    "per_intent": null,
+    "k_shot": 50,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k50-seed42",
+    "seed": 42,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 42,
+      "per_intent": null,
+      "k_shot": 50,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 7625,
+    "oos_train_rows": 125,
+    "per_intent": null,
+    "k_shot": 50,
+    "train_sample_sha256": "8a9a52f8ae41506d9847bd9ceba987edadc6bc3304487be22e664a95e1b9b18f",
+    "step_plan": {
+      "epoch_steps": 1195,
+      "min_train_steps": 400,
+      "planned_steps": 1195,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 1195,
+    "train_loss": 0.5208185849828202,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 602.167,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 2392,
+      "driver_allocated_bytes": 7451279360,
+      "tensor_allocated_bytes": 2440411904
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k50-seed42.npz",
+    "sha256": "86bcc65f899f519154881c6e1ccb10ba468e055b7a26049fc11727fc74347a4c",
+    "bytes": 4833586
+  },
+  "metrics": {
+    "temperature": 1.1256634063399562,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9770967741935483,
+        "oos_recall_8": 0.7,
+        "accuracy_8_summed": 0.9761290322580645,
+        "oos_recall_8_summed": 0.65,
+        "in_scope_accuracy_150": 0.9606666666666667,
+        "oos_recall_151": 0.7,
+        "ece_151": 0.015290190388898471,
+        "ece_8_summed": 0.003587238381429791,
+        "nll_151": 0.2280515118955254,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9770967741935483,
+        "oos_recall_8": 0.7,
+        "accuracy_8_summed": 0.9761290322580645,
+        "oos_recall_8_summed": 0.65,
+        "in_scope_accuracy_150": 0.9606666666666667,
+        "oos_recall_151": 0.7,
+        "ece_151": 0.011158989492944088,
+        "ece_8_summed": 0.0057768564231399444,
+        "nll_151": 0.22406311200443582,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.908,
+        "oos_recall_8": 0.565,
+        "accuracy_8_summed": 0.9027272727272727,
+        "oos_recall_8_summed": 0.533,
+        "in_scope_accuracy_150": 0.9577777777777777,
+        "oos_recall_151": 0.565,
+        "ece_151": 0.02936344299761908,
+        "ece_8_summed": 0.035375104622989394,
+        "nll_151": 0.5180205893201404,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.908,
+        "oos_recall_8": 0.565,
+        "accuracy_8_summed": 0.8996363636363637,
+        "oos_recall_8_summed": 0.516,
+        "in_scope_accuracy_150": 0.9577777777777777,
+        "oos_recall_151": 0.565,
+        "ece_151": 0.021282499196595698,
+        "ece_8_summed": 0.026226123667256497,
+        "nll_151": 0.5045254307139664,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k50-seed43.json b/results/runs/ModernBERT-base-k50-seed43.json
new file mode 100644
index 0000000..b86ea60
--- /dev/null
+++ b/results/runs/ModernBERT-base-k50-seed43.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k50-seed43",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 43,
+    "per_intent": null,
+    "k_shot": 50,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k50-seed43",
+    "seed": 43,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 43,
+      "per_intent": null,
+      "k_shot": 50,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 7625,
+    "oos_train_rows": 125,
+    "per_intent": null,
+    "k_shot": 50,
+    "train_sample_sha256": "836ad6186d05a042ae00b3ab45efc560eb45f774644bf5e4f9b13a6e66d82982",
+    "step_plan": {
+      "epoch_steps": 1195,
+      "min_train_steps": 400,
+      "planned_steps": 1195,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 1195,
+    "train_loss": 0.5623688727492568,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 600.849,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 2392,
+      "driver_allocated_bytes": 6695976960,
+      "tensor_allocated_bytes": 2445225728
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k50-seed43.npz",
+    "sha256": "0f72b3ebe5b54dcee8b7857d339de335b31230f6dd06d4f43297730e19d955ee",
+    "bytes": 4833495
+  },
+  "metrics": {
+    "temperature": 1.1102088153968748,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9803225806451613,
+        "oos_recall_8": 0.74,
+        "accuracy_8_summed": 0.9787096774193549,
+        "oos_recall_8_summed": 0.69,
+        "in_scope_accuracy_150": 0.9586666666666667,
+        "oos_recall_151": 0.74,
+        "ece_151": 0.014282126187099314,
+        "ece_8_summed": 0.005079268613441623,
+        "nll_151": 0.21872437128681166,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9803225806451613,
+        "oos_recall_8": 0.74,
+        "accuracy_8_summed": 0.9783870967741936,
+        "oos_recall_8_summed": 0.68,
+        "in_scope_accuracy_150": 0.9586666666666667,
+        "oos_recall_151": 0.74,
+        "ece_151": 0.012104636733870716,
+        "ece_8_summed": 0.0057285856521321525,
+        "nll_151": 0.2157544655805519,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.9032727272727272,
+        "oos_recall_8": 0.53,
+        "accuracy_8_summed": 0.8998181818181819,
+        "oos_recall_8_summed": 0.505,
+        "in_scope_accuracy_150": 0.9517777777777777,
+        "oos_recall_151": 0.53,
+        "ece_151": 0.0415650212444922,
+        "ece_8_summed": 0.04009929407836817,
+        "nll_151": 0.5720288148212682,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.9032727272727272,
+        "oos_recall_8": 0.53,
+        "accuracy_8_summed": 0.897090909090909,
+        "oos_recall_8_summed": 0.49,
+        "in_scope_accuracy_150": 0.9517777777777777,
+        "oos_recall_151": 0.53,
+        "ece_151": 0.03239504142595282,
+        "ece_8_summed": 0.031664815416184926,
+        "nll_151": 0.5549542486811021,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/ModernBERT-base-k50-seed44.json b/results/runs/ModernBERT-base-k50-seed44.json
new file mode 100644
index 0000000..f2c7e37
--- /dev/null
+++ b/results/runs/ModernBERT-base-k50-seed44.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "ModernBERT-base-k50-seed44",
+  "config": {
+    "model_name": "answerdotai/ModernBERT-base",
+    "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+    "seed": 44,
+    "per_intent": null,
+    "k_shot": 50,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "ModernBERT-base-k50-seed44",
+    "seed": 44,
+    "config": {
+      "model_name": "answerdotai/ModernBERT-base",
+      "model_revision": "8949b909ec900327062f0ebf497f51aef5e6f0c8",
+      "seed": 44,
+      "per_intent": null,
+      "k_shot": 50,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 7625,
+    "oos_train_rows": 125,
+    "per_intent": null,
+    "k_shot": 50,
+    "train_sample_sha256": "5db5ce1ae1864ba63a4164a799b09a7c1d783a4acbd55efc587613d986752013",
+    "step_plan": {
+      "epoch_steps": 1195,
+      "min_train_steps": 400,
+      "planned_steps": 1195,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 1195,
+    "train_loss": 0.5822031815181716,
+    "parameters": {
+      "total": 149720983,
+      "trainable": 149720983
+    },
+    "train_wall_seconds": 599.716,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 2392,
+      "driver_allocated_bytes": 6226214912,
+      "tensor_allocated_bytes": 2447128320
+    }
+  },
+  "logits": {
+    "file": "ModernBERT-base-k50-seed44.npz",
+    "sha256": "c7427794ce3f205347c4a1bc703409bfece96251974823dcb075bfc2dd3d6da3",
+    "bytes": 4833374
+  },
+  "metrics": {
+    "temperature": 1.0645875580879014,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9790322580645161,
+        "oos_recall_8": 0.78,
+        "accuracy_8_summed": 0.9790322580645161,
+        "oos_recall_8_summed": 0.74,
+        "in_scope_accuracy_150": 0.964,
+        "oos_recall_151": 0.78,
+        "ece_151": 0.010001836395241447,
+        "ece_8_summed": 0.0040118116153775715,
+        "nll_151": 0.18350928526323293,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9790322580645161,
+        "oos_recall_8": 0.78,
+        "accuracy_8_summed": 0.9787096774193549,
+        "oos_recall_8_summed": 0.73,
+        "in_scope_accuracy_150": 0.964,
+        "oos_recall_151": 0.78,
+        "ece_151": 0.00950883124531918,
+        "ece_8_summed": 0.0054248816221562305,
+        "nll_151": 0.18261486875509572,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.9005454545454545,
+        "oos_recall_8": 0.521,
+        "accuracy_8_summed": 0.8956363636363637,
+        "oos_recall_8_summed": 0.487,
+        "in_scope_accuracy_150": 0.9575555555555556,
+        "oos_recall_151": 0.521,
+        "ece_151": 0.03751367481767784,
+        "ece_8_summed": 0.04328220484627914,
+        "nll_151": 0.5400900159144765,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.9005454545454545,
+        "oos_recall_8": 0.521,
+        "accuracy_8_summed": 0.895090909090909,
+        "oos_recall_8_summed": 0.485,
+        "in_scope_accuracy_150": 0.9575555555555556,
+        "oos_recall_151": 0.521,
+        "ece_151": 0.030455764621388284,
+        "ece_8_summed": 0.03686340123695944,
+        "nll_151": 0.5301356052706606,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k1-seed42.json b/results/runs/bert-base-uncased-k1-seed42.json
new file mode 100644
index 0000000..5faf4c5
--- /dev/null
+++ b/results/runs/bert-base-uncased-k1-seed42.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k1-seed42",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 42,
+    "per_intent": null,
+    "k_shot": 1,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k1-seed42",
+    "seed": 42,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 42,
+      "per_intent": null,
+      "k_shot": 1,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 153,
+    "oos_train_rows": 3,
+    "per_intent": null,
+    "k_shot": 1,
+    "train_sample_sha256": "8f04aca1ffa32368a0e811795f70067f9d2b6d1551e10a3be5e08f5c954db166",
+    "step_plan": {
+      "epoch_steps": 25,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 2.5258790248632432,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 133.46,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 3369058304,
+      "tensor_allocated_bytes": 1781053696
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k1-seed42.npz",
+    "sha256": "33464fd83ba26e6de38f2df4a347659dc396c57d98ce9fca16b78ab31101da2d",
+    "bytes": 4820569
+  },
+  "metrics": {
+    "temperature": 0.5034190526029269,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.5954838709677419,
+        "oos_recall_8": 0.35,
+        "accuracy_8_summed": 0.46838709677419355,
+        "oos_recall_8_summed": 0.02,
+        "in_scope_accuracy_150": 0.2946666666666667,
+        "oos_recall_151": 0.35,
+        "ece_151": 0.22927929427372284,
+        "ece_8_summed": 0.1906097039619262,
+        "nll_151": 3.843036767804298,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.5954838709677419,
+        "oos_recall_8": 0.35,
+        "accuracy_8_summed": 0.6216129032258064,
+        "oos_recall_8_summed": 0.23,
+        "in_scope_accuracy_150": 0.2946666666666667,
+        "oos_recall_151": 0.35,
+        "ece_151": 0.040060994715560794,
+        "ece_8_summed": 0.18109482452372203,
+        "nll_151": 3.445326471611791,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.5336363636363637,
+        "oos_recall_8": 0.238,
+        "accuracy_8_summed": 0.3870909090909091,
+        "oos_recall_8_summed": 0.015,
+        "in_scope_accuracy_150": 0.2833333333333333,
+        "oos_recall_151": 0.238,
+        "ece_151": 0.21294848226527868,
+        "ece_8_summed": 0.16876418276817265,
+        "nll_151": 3.9935479114913166,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.5336363636363637,
+        "oos_recall_8": 0.238,
+        "accuracy_8_summed": 0.5396363636363637,
+        "oos_recall_8_summed": 0.129,
+        "in_scope_accuracy_150": 0.2833333333333333,
+        "oos_recall_151": 0.238,
+        "ece_151": 0.043020086741883956,
+        "ece_8_summed": 0.12567983850717687,
+        "nll_151": 3.695140880695411,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k1-seed43.json b/results/runs/bert-base-uncased-k1-seed43.json
new file mode 100644
index 0000000..e028c7a
--- /dev/null
+++ b/results/runs/bert-base-uncased-k1-seed43.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k1-seed43",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 43,
+    "per_intent": null,
+    "k_shot": 1,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k1-seed43",
+    "seed": 43,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 43,
+      "per_intent": null,
+      "k_shot": 1,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 153,
+    "oos_train_rows": 3,
+    "per_intent": null,
+    "k_shot": 1,
+    "train_sample_sha256": "4c659defaf9b9c6f28f4d5e67ef92b466627cdb200623552ba46ced1d2750c9c",
+    "step_plan": {
+      "epoch_steps": 25,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 2.566484009027481,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 136.608,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 3549331456,
+      "tensor_allocated_bytes": 1778498560
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k1-seed43.npz",
+    "sha256": "de95f070e64b443b4761f3ca786da624343fe8a5dc20e3f341220333e586cff3",
+    "bytes": 4823550
+  },
+  "metrics": {
+    "temperature": 0.4579855609174441,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.6151612903225806,
+        "oos_recall_8": 0.18,
+        "accuracy_8_summed": 0.4638709677419355,
+        "oos_recall_8_summed": 0.04,
+        "in_scope_accuracy_150": 0.3396666666666667,
+        "oos_recall_151": 0.18,
+        "ece_151": 0.2700096700377473,
+        "ece_8_summed": 0.19553895220475095,
+        "nll_151": 3.768915380620551,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.6151612903225806,
+        "oos_recall_8": 0.18,
+        "accuracy_8_summed": 0.6493548387096775,
+        "oos_recall_8_summed": 0.11,
+        "in_scope_accuracy_150": 0.3396666666666667,
+        "oos_recall_151": 0.18,
+        "ece_151": 0.03870215399407613,
+        "ece_8_summed": 0.16696347047292984,
+        "nll_151": 3.2415185997877214,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.5463636363636364,
+        "oos_recall_8": 0.205,
+        "accuracy_8_summed": 0.384,
+        "oos_recall_8_summed": 0.02,
+        "in_scope_accuracy_150": 0.3277777777777778,
+        "oos_recall_151": 0.205,
+        "ece_151": 0.24385252710380353,
+        "ece_8_summed": 0.15476146703499763,
+        "nll_151": 3.896224661181068,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.5463636363636364,
+        "oos_recall_8": 0.205,
+        "accuracy_8_summed": 0.5643636363636364,
+        "oos_recall_8_summed": 0.14,
+        "in_scope_accuracy_150": 0.3277777777777778,
+        "oos_recall_151": 0.205,
+        "ece_151": 0.028556943654340878,
+        "ece_8_summed": 0.09775900530787661,
+        "nll_151": 3.4779455197828066,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k1-seed44.json b/results/runs/bert-base-uncased-k1-seed44.json
new file mode 100644
index 0000000..abdb6ef
--- /dev/null
+++ b/results/runs/bert-base-uncased-k1-seed44.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k1-seed44",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 44,
+    "per_intent": null,
+    "k_shot": 1,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k1-seed44",
+    "seed": 44,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 44,
+      "per_intent": null,
+      "k_shot": 1,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 153,
+    "oos_train_rows": 3,
+    "per_intent": null,
+    "k_shot": 1,
+    "train_sample_sha256": "d69635e70893d70759a46fd0fa54561eae262653eca95898076348b9c660b96a",
+    "step_plan": {
+      "epoch_steps": 25,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 2.574421035349369,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 140.036,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 4581130240,
+      "tensor_allocated_bytes": 1804772352
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k1-seed44.npz",
+    "sha256": "db5595875d86d5b3476689bb5932f72bdb7651effb3686d8636cb308e8279881",
+    "bytes": 4821951
+  },
+  "metrics": {
+    "temperature": 0.44264955435197817,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.6061290322580645,
+        "oos_recall_8": 0.11,
+        "accuracy_8_summed": 0.4635483870967742,
+        "oos_recall_8_summed": 0.01,
+        "in_scope_accuracy_150": 0.31966666666666665,
+        "oos_recall_151": 0.11,
+        "ece_151": 0.25649720347456095,
+        "ece_8_summed": 0.18134762552998304,
+        "nll_151": 3.8452775528093452,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.6061290322580645,
+        "oos_recall_8": 0.11,
+        "accuracy_8_summed": 0.6432258064516129,
+        "oos_recall_8_summed": 0.09,
+        "in_scope_accuracy_150": 0.31966666666666665,
+        "oos_recall_151": 0.11,
+        "ece_151": 0.044029782936651336,
+        "ece_8_summed": 0.1672088911590039,
+        "nll_151": 3.3066050487502494,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.5216363636363637,
+        "oos_recall_8": 0.074,
+        "accuracy_8_summed": 0.39036363636363636,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.32711111111111113,
+        "oos_recall_151": 0.074,
+        "ece_151": 0.2262759928913901,
+        "ece_8_summed": 0.14042133286934327,
+        "nll_151": 4.060788945026958,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.5216363636363637,
+        "oos_recall_8": 0.074,
+        "accuracy_8_summed": 0.5438181818181819,
+        "oos_recall_8_summed": 0.028,
+        "in_scope_accuracy_150": 0.32711111111111113,
+        "oos_recall_151": 0.074,
+        "ece_151": 0.021102029795381574,
+        "ece_8_summed": 0.08679343673213227,
+        "nll_151": 3.774391459119817,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k10-seed42.json b/results/runs/bert-base-uncased-k10-seed42.json
new file mode 100644
index 0000000..3f4df78
--- /dev/null
+++ b/results/runs/bert-base-uncased-k10-seed42.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k10-seed42",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 42,
+    "per_intent": null,
+    "k_shot": 10,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k10-seed42",
+    "seed": 42,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 42,
+      "per_intent": null,
+      "k_shot": 10,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 1525,
+    "oos_train_rows": 25,
+    "per_intent": null,
+    "k_shot": 10,
+    "train_sample_sha256": "2606cc878d85759a5885e1b9be35205023b156e6cf8a4eea85bd635c3b492e74",
+    "step_plan": {
+      "epoch_steps": 240,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 3.499563281536102,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 137.968,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 4623564800,
+      "tensor_allocated_bytes": 1814937344
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k10-seed42.npz",
+    "sha256": "51df91b493b56ae7d49fdd3195bc1d53d55099ec5b1cf011dfa9579a8a6320a3",
+    "bytes": 4820339
+  },
+  "metrics": {
+    "temperature": 0.2687193947564064,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9290322580645162,
+        "oos_recall_8": 0.39,
+        "accuracy_8_summed": 0.7867741935483871,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.7953333333333333,
+        "oos_recall_151": 0.39,
+        "ece_151": 0.6857741701240475,
+        "ece_8_summed": 0.43720340998164003,
+        "nll_151": 2.5978301896441924,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9290322580645162,
+        "oos_recall_8": 0.39,
+        "accuracy_8_summed": 0.9312903225806451,
+        "oos_recall_8_summed": 0.33,
+        "in_scope_accuracy_150": 0.7953333333333333,
+        "oos_recall_151": 0.39,
+        "ece_151": 0.03443225818005568,
+        "ece_8_summed": 0.016065020421327382,
+        "nll_151": 0.881133832082484,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8425454545454546,
+        "oos_recall_8": 0.345,
+        "accuracy_8_summed": 0.6627272727272727,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.7984444444444444,
+        "oos_recall_151": 0.345,
+        "ece_151": 0.6285557973645203,
+        "ece_8_summed": 0.33098499000716164,
+        "nll_151": 2.8420446669257067,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8425454545454546,
+        "oos_recall_8": 0.345,
+        "accuracy_8_summed": 0.8325454545454546,
+        "oos_recall_8_summed": 0.287,
+        "in_scope_accuracy_150": 0.7984444444444444,
+        "oos_recall_151": 0.345,
+        "ece_151": 0.02345066031891548,
+        "ece_8_summed": 0.03766315891690147,
+        "nll_151": 1.4055155931475403,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k10-seed43.json b/results/runs/bert-base-uncased-k10-seed43.json
new file mode 100644
index 0000000..cca2e34
--- /dev/null
+++ b/results/runs/bert-base-uncased-k10-seed43.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k10-seed43",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 43,
+    "per_intent": null,
+    "k_shot": 10,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k10-seed43",
+    "seed": 43,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 43,
+      "per_intent": null,
+      "k_shot": 10,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 1525,
+    "oos_train_rows": 25,
+    "per_intent": null,
+    "k_shot": 10,
+    "train_sample_sha256": "072190bdfc028a2e0bc6ec85d39e26ae1b27854281c6226be57b8f8b9d20e165",
+    "step_plan": {
+      "epoch_steps": 240,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 3.5259680235385895,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 137.792,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 3658956800,
+      "tensor_allocated_bytes": 1811726848
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k10-seed43.npz",
+    "sha256": "113c11c8b72a8f20ac46e70a1c8492ee1a4d0571b1b86caba1b583cc85d0faf8",
+    "bytes": 4824784
+  },
+  "metrics": {
+    "temperature": 0.2583952940375509,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9306451612903226,
+        "oos_recall_8": 0.34,
+        "accuracy_8_summed": 0.7838709677419354,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.8163333333333334,
+        "oos_recall_151": 0.34,
+        "ece_151": 0.7041825441664477,
+        "ece_8_summed": 0.4472515722331045,
+        "nll_151": 2.5692356737788766,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9306451612903226,
+        "oos_recall_8": 0.34,
+        "accuracy_8_summed": 0.9270967741935484,
+        "oos_recall_8_summed": 0.24,
+        "in_scope_accuracy_150": 0.8163333333333334,
+        "oos_recall_151": 0.34,
+        "ece_151": 0.01835228039016208,
+        "ece_8_summed": 0.014544923839084204,
+        "nll_151": 0.7620569245556964,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8136363636363636,
+        "oos_recall_8": 0.23,
+        "accuracy_8_summed": 0.6574545454545454,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.8026666666666666,
+        "oos_recall_151": 0.23,
+        "ece_151": 0.6110469136169548,
+        "ece_8_summed": 0.33409434668954063,
+        "nll_151": 2.8658147227581376,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8136363636363636,
+        "oos_recall_8": 0.23,
+        "accuracy_8_summed": 0.8067272727272727,
+        "oos_recall_8_summed": 0.19,
+        "in_scope_accuracy_150": 0.8026666666666666,
+        "oos_recall_151": 0.23,
+        "ece_151": 0.03172331292981914,
+        "ece_8_summed": 0.06349641837862197,
+        "nll_151": 1.4924731857196547,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k10-seed44.json b/results/runs/bert-base-uncased-k10-seed44.json
new file mode 100644
index 0000000..1f7ac17
--- /dev/null
+++ b/results/runs/bert-base-uncased-k10-seed44.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k10-seed44",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 44,
+    "per_intent": null,
+    "k_shot": 10,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k10-seed44",
+    "seed": 44,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 44,
+      "per_intent": null,
+      "k_shot": 10,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 1525,
+    "oos_train_rows": 25,
+    "per_intent": null,
+    "k_shot": 10,
+    "train_sample_sha256": "31a30630bb8f711a55173e8837eaf4500c2c5dcd5efe461b561a04e843cddd55",
+    "step_plan": {
+      "epoch_steps": 240,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 3.567430211305618,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 136.475,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 3650322432,
+      "tensor_allocated_bytes": 1793966592
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k10-seed44.npz",
+    "sha256": "b60ca5253fc8ff28108fcfa32d4aa79139357e9f5639d187b2270a266c594adb",
+    "bytes": 4818953
+  },
+  "metrics": {
+    "temperature": 0.25407926732066993,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9306451612903226,
+        "oos_recall_8": 0.33,
+        "accuracy_8_summed": 0.8041935483870968,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.803,
+        "oos_recall_151": 0.33,
+        "ece_151": 0.6988700419864706,
+        "ece_8_summed": 0.4729820445654247,
+        "nll_151": 2.6670983827667443,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9306451612903226,
+        "oos_recall_8": 0.33,
+        "accuracy_8_summed": 0.9303225806451613,
+        "oos_recall_8_summed": 0.28,
+        "in_scope_accuracy_150": 0.803,
+        "oos_recall_151": 0.33,
+        "ece_151": 0.02727029922061202,
+        "ece_8_summed": 0.017651469240006754,
+        "nll_151": 0.8517544973183014,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8107272727272727,
+        "oos_recall_8": 0.205,
+        "accuracy_8_summed": 0.6772727272727272,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.7906666666666666,
+        "oos_recall_151": 0.205,
+        "ece_151": 0.6029255807398551,
+        "ece_8_summed": 0.3610062755704072,
+        "nll_151": 2.9651457602687357,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8107272727272727,
+        "oos_recall_8": 0.205,
+        "accuracy_8_summed": 0.8049090909090909,
+        "oos_recall_8_summed": 0.158,
+        "in_scope_accuracy_150": 0.7906666666666666,
+        "oos_recall_151": 0.205,
+        "ece_151": 0.03779134385609432,
+        "ece_8_summed": 0.06813118305970127,
+        "nll_151": 1.6461227954309618,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k25-seed42.json b/results/runs/bert-base-uncased-k25-seed42.json
new file mode 100644
index 0000000..1187f10
--- /dev/null
+++ b/results/runs/bert-base-uncased-k25-seed42.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k25-seed42",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 42,
+    "per_intent": null,
+    "k_shot": 25,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k25-seed42",
+    "seed": 42,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 42,
+      "per_intent": null,
+      "k_shot": 25,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 3813,
+    "oos_train_rows": 63,
+    "per_intent": null,
+    "k_shot": 25,
+    "train_sample_sha256": "51184b580bdbe58ab700d2545887a30c442950389b98497c3b7edaa7ca47174c",
+    "step_plan": {
+      "epoch_steps": 600,
+      "min_train_steps": 400,
+      "planned_steps": 600,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 600,
+    "train_loss": 2.8445407867431642,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 207.001,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 1202,
+      "driver_allocated_bytes": 3608379392,
+      "tensor_allocated_bytes": 1790657024
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k25-seed42.npz",
+    "sha256": "5397a2237935d24210a1b8e47ed837f1807db8beb0a4c18f31d24d0681bef1cc",
+    "bytes": 4817631
+  },
+  "metrics": {
+    "temperature": 0.3732579618909568,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9583870967741935,
+        "oos_recall_8": 0.55,
+        "accuracy_8_summed": 0.9370967741935484,
+        "oos_recall_8_summed": 0.22,
+        "in_scope_accuracy_150": 0.9003333333333333,
+        "oos_recall_151": 0.55,
+        "ece_151": 0.6226428392330094,
+        "ece_8_summed": 0.4327191294899506,
+        "nll_151": 1.5332406261149025,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9583870967741935,
+        "oos_recall_8": 0.55,
+        "accuracy_8_summed": 0.9570967741935484,
+        "oos_recall_8_summed": 0.55,
+        "in_scope_accuracy_150": 0.9003333333333333,
+        "oos_recall_151": 0.55,
+        "ece_151": 0.02944408128890611,
+        "ece_8_summed": 0.012184974440626427,
+        "nll_151": 0.4708538536740342,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8714545454545455,
+        "oos_recall_8": 0.415,
+        "accuracy_8_summed": 0.8187272727272727,
+        "oos_recall_8_summed": 0.175,
+        "in_scope_accuracy_150": 0.8964444444444445,
+        "oos_recall_151": 0.415,
+        "ece_151": 0.5629702992098558,
+        "ece_8_summed": 0.35577814780967265,
+        "nll_151": 1.8596286509949456,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8714545454545455,
+        "oos_recall_8": 0.415,
+        "accuracy_8_summed": 0.8650909090909091,
+        "oos_recall_8_summed": 0.387,
+        "in_scope_accuracy_150": 0.8964444444444445,
+        "oos_recall_151": 0.415,
+        "ece_151": 0.02348073597498302,
+        "ece_8_summed": 0.05922496794804483,
+        "nll_151": 1.0598236452803298,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k25-seed43.json b/results/runs/bert-base-uncased-k25-seed43.json
new file mode 100644
index 0000000..ff35c52
--- /dev/null
+++ b/results/runs/bert-base-uncased-k25-seed43.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k25-seed43",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 43,
+    "per_intent": null,
+    "k_shot": 25,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k25-seed43",
+    "seed": 43,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 43,
+      "per_intent": null,
+      "k_shot": 25,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 3813,
+    "oos_train_rows": 63,
+    "per_intent": null,
+    "k_shot": 25,
+    "train_sample_sha256": "92f91be6e0948dbc01e82b81485c72829dd232443955fa0ae214b1451a0a704f",
+    "step_plan": {
+      "epoch_steps": 600,
+      "min_train_steps": 400,
+      "planned_steps": 600,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 600,
+    "train_loss": 2.8801915995279948,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 206.269,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 1202,
+      "driver_allocated_bytes": 3583213568,
+      "tensor_allocated_bytes": 1815069184
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k25-seed43.npz",
+    "sha256": "cf40024e86a42cfe34703b4b04e58696d93a03f3408a44f820960b11d554d626",
+    "bytes": 4820374
+  },
+  "metrics": {
+    "temperature": 0.3776942358099553,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9564516129032258,
+        "oos_recall_8": 0.48,
+        "accuracy_8_summed": 0.942258064516129,
+        "oos_recall_8_summed": 0.14,
+        "in_scope_accuracy_150": 0.9003333333333333,
+        "oos_recall_151": 0.48,
+        "ece_151": 0.6202851616232495,
+        "ece_8_summed": 0.44475103559051127,
+        "nll_151": 1.5417622685817294,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9564516129032258,
+        "oos_recall_8": 0.48,
+        "accuracy_8_summed": 0.954516129032258,
+        "oos_recall_8_summed": 0.43,
+        "in_scope_accuracy_150": 0.9003333333333333,
+        "oos_recall_151": 0.48,
+        "ece_151": 0.02211881973175207,
+        "ece_8_summed": 0.00888893521401323,
+        "nll_151": 0.48363611586607536,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8585454545454545,
+        "oos_recall_8": 0.372,
+        "accuracy_8_summed": 0.8063636363636364,
+        "oos_recall_8_summed": 0.097,
+        "in_scope_accuracy_150": 0.8953333333333333,
+        "oos_recall_151": 0.372,
+        "ece_151": 0.5587474586312459,
+        "ece_8_summed": 0.35250582047879936,
+        "nll_151": 1.8793379056708166,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8585454545454545,
+        "oos_recall_8": 0.372,
+        "accuracy_8_summed": 0.8512727272727273,
+        "oos_recall_8_summed": 0.322,
+        "in_scope_accuracy_150": 0.8953333333333333,
+        "oos_recall_151": 0.372,
+        "ece_151": 0.029437965167064824,
+        "ece_8_summed": 0.06757927487820457,
+        "nll_151": 1.045268868487479,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k25-seed44.json b/results/runs/bert-base-uncased-k25-seed44.json
new file mode 100644
index 0000000..4e51e97
--- /dev/null
+++ b/results/runs/bert-base-uncased-k25-seed44.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k25-seed44",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 44,
+    "per_intent": null,
+    "k_shot": 25,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k25-seed44",
+    "seed": 44,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 44,
+      "per_intent": null,
+      "k_shot": 25,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 3813,
+    "oos_train_rows": 63,
+    "per_intent": null,
+    "k_shot": 25,
+    "train_sample_sha256": "790cd7cf9eb24143276f6c59c7bac29c4a0b42f14bef38e9634e96253e6315fb",
+    "step_plan": {
+      "epoch_steps": 600,
+      "min_train_steps": 400,
+      "planned_steps": 600,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 600,
+    "train_loss": 2.8521534983317056,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 206.632,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 1202,
+      "driver_allocated_bytes": 4648648704,
+      "tensor_allocated_bytes": 1809169408
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k25-seed44.npz",
+    "sha256": "c4de18e8e6d5e96011ab20178ec669bd2462c6aae44cd2fd40ae8ba3ce077f30",
+    "bytes": 4818898
+  },
+  "metrics": {
+    "temperature": 0.35758235214673606,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9658064516129032,
+        "oos_recall_8": 0.54,
+        "accuracy_8_summed": 0.9474193548387096,
+        "oos_recall_8_summed": 0.07,
+        "in_scope_accuracy_150": 0.9203333333333333,
+        "oos_recall_151": 0.54,
+        "ece_151": 0.643871313760336,
+        "ece_8_summed": 0.44914044797815217,
+        "nll_151": 1.5195780894850688,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9658064516129032,
+        "oos_recall_8": 0.54,
+        "accuracy_8_summed": 0.9641935483870968,
+        "oos_recall_8_summed": 0.49,
+        "in_scope_accuracy_150": 0.9203333333333333,
+        "oos_recall_151": 0.54,
+        "ece_151": 0.02477818643648947,
+        "ece_8_summed": 0.007214920474403505,
+        "nll_151": 0.4080676107434186,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8654545454545455,
+        "oos_recall_8": 0.353,
+        "accuracy_8_summed": 0.8069090909090909,
+        "oos_recall_8_summed": 0.034,
+        "in_scope_accuracy_150": 0.9144444444444444,
+        "oos_recall_151": 0.353,
+        "ece_151": 0.5731156997259947,
+        "ece_8_summed": 0.36150164273348634,
+        "nll_151": 1.8975552673941638,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8654545454545455,
+        "oos_recall_8": 0.353,
+        "accuracy_8_summed": 0.862,
+        "oos_recall_8_summed": 0.324,
+        "in_scope_accuracy_150": 0.9144444444444444,
+        "oos_recall_151": 0.353,
+        "ece_151": 0.02885174629406659,
+        "ece_8_summed": 0.0659482101510376,
+        "nll_151": 1.0970270922714995,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k5-seed42.json b/results/runs/bert-base-uncased-k5-seed42.json
new file mode 100644
index 0000000..b9ab2ad
--- /dev/null
+++ b/results/runs/bert-base-uncased-k5-seed42.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k5-seed42",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 42,
+    "per_intent": null,
+    "k_shot": 5,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k5-seed42",
+    "seed": 42,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 42,
+      "per_intent": null,
+      "k_shot": 5,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 763,
+    "oos_train_rows": 13,
+    "per_intent": null,
+    "k_shot": 5,
+    "train_sample_sha256": "bfabf08971d3a16a502e25b2d5a7f1ce204504a526d5b3cb34fec9346dd8caa7",
+    "step_plan": {
+      "epoch_steps": 120,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 3.3099954000115392,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 136.13,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 3549577216,
+      "tensor_allocated_bytes": 1778237184
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k5-seed42.npz",
+    "sha256": "12e3d723ecbdeda888a687bd3b539ef5b8bc07515bd72f28293b0dfc4314f94d",
+    "bytes": 4820512
+  },
+  "metrics": {
+    "temperature": 0.30579748711761595,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9048387096774193,
+        "oos_recall_8": 0.39,
+        "accuracy_8_summed": 0.7535483870967742,
+        "oos_recall_8_summed": 0.04,
+        "in_scope_accuracy_150": 0.7546666666666667,
+        "oos_recall_151": 0.39,
+        "ece_151": 0.6473592132295444,
+        "ece_8_summed": 0.42539970983096514,
+        "nll_151": 2.70382154174019,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9048387096774193,
+        "oos_recall_8": 0.39,
+        "accuracy_8_summed": 0.9138709677419354,
+        "oos_recall_8_summed": 0.33,
+        "in_scope_accuracy_150": 0.7546666666666667,
+        "oos_recall_151": 0.39,
+        "ece_151": 0.0282079241617613,
+        "ece_8_summed": 0.04843608907202708,
+        "nll_151": 1.148816568466163,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8038181818181818,
+        "oos_recall_8": 0.277,
+        "accuracy_8_summed": 0.647090909090909,
+        "oos_recall_8_summed": 0.04,
+        "in_scope_accuracy_150": 0.746,
+        "oos_recall_151": 0.277,
+        "ece_151": 0.572152224784998,
+        "ece_8_summed": 0.33089548603173474,
+        "nll_151": 2.9778920864048275,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8038181818181818,
+        "oos_recall_8": 0.277,
+        "accuracy_8_summed": 0.8030909090909091,
+        "oos_recall_8_summed": 0.243,
+        "in_scope_accuracy_150": 0.746,
+        "oos_recall_151": 0.277,
+        "ece_151": 0.0194948633771942,
+        "ece_8_summed": 0.03929560764610777,
+        "nll_151": 1.7990380464826812,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k5-seed43.json b/results/runs/bert-base-uncased-k5-seed43.json
new file mode 100644
index 0000000..daa1a1d
--- /dev/null
+++ b/results/runs/bert-base-uncased-k5-seed43.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k5-seed43",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 43,
+    "per_intent": null,
+    "k_shot": 5,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k5-seed43",
+    "seed": 43,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 43,
+      "per_intent": null,
+      "k_shot": 5,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 763,
+    "oos_train_rows": 13,
+    "per_intent": null,
+    "k_shot": 5,
+    "train_sample_sha256": "4e289193d9669c26c40d083246f6c34426bca5ff041ac526796a5f976ca9c638",
+    "step_plan": {
+      "epoch_steps": 120,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 3.3666005605459213,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 137.435,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 3499409408,
+      "tensor_allocated_bytes": 1786099200
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k5-seed43.npz",
+    "sha256": "78dd86d8b8a700aa8986c7bf23fef4a99720a8f226df3e6d0d21d896cae3aa8a",
+    "bytes": 4826522
+  },
+  "metrics": {
+    "temperature": 0.2828802903337913,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9029032258064517,
+        "oos_recall_8": 0.34,
+        "accuracy_8_summed": 0.747741935483871,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.781,
+        "oos_recall_151": 0.34,
+        "ece_151": 0.6734704969926137,
+        "ece_8_summed": 0.4217254548961758,
+        "nll_151": 2.6819227052093977,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9029032258064517,
+        "oos_recall_8": 0.34,
+        "accuracy_8_summed": 0.9041935483870968,
+        "oos_recall_8_summed": 0.25,
+        "in_scope_accuracy_150": 0.781,
+        "oos_recall_151": 0.34,
+        "ece_151": 0.02318892484588278,
+        "ece_8_summed": 0.03351708920570171,
+        "nll_151": 0.9866900230402638,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.7945454545454546,
+        "oos_recall_8": 0.26,
+        "accuracy_8_summed": 0.6285454545454545,
+        "oos_recall_8_summed": 0.003,
+        "in_scope_accuracy_150": 0.7668888888888888,
+        "oos_recall_151": 0.26,
+        "ece_151": 0.5888314799816248,
+        "ece_8_summed": 0.31545154808543346,
+        "nll_151": 2.938763327304599,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.7945454545454546,
+        "oos_recall_8": 0.26,
+        "accuracy_8_summed": 0.7947272727272727,
+        "oos_recall_8_summed": 0.218,
+        "in_scope_accuracy_150": 0.7668888888888888,
+        "oos_recall_151": 0.26,
+        "ece_151": 0.034201923690586396,
+        "ece_8_summed": 0.038090746466780595,
+        "nll_151": 1.5938205196216466,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k5-seed44.json b/results/runs/bert-base-uncased-k5-seed44.json
new file mode 100644
index 0000000..d7d336c
--- /dev/null
+++ b/results/runs/bert-base-uncased-k5-seed44.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k5-seed44",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 44,
+    "per_intent": null,
+    "k_shot": 5,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k5-seed44",
+    "seed": 44,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 44,
+      "per_intent": null,
+      "k_shot": 5,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 763,
+    "oos_train_rows": 13,
+    "per_intent": null,
+    "k_shot": 5,
+    "train_sample_sha256": "ba57c52c7c416c611ba215bb10a393fba470ec511aaebad2b416914ca674988d",
+    "step_plan": {
+      "epoch_steps": 120,
+      "min_train_steps": 400,
+      "planned_steps": 400,
+      "decided_by": "min_train_steps",
+      "max_steps_arg": 400
+    },
+    "global_step": 400,
+    "train_loss": 3.4177334290742873,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 136.019,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 802,
+      "driver_allocated_bytes": 3574906880,
+      "tensor_allocated_bytes": 1788788480
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k5-seed44.npz",
+    "sha256": "175a274248600f6f85681b2e8a824dd9bf17f281782044110b1c3facb10f1da5",
+    "bytes": 4818879
+  },
+  "metrics": {
+    "temperature": 0.28149418436722007,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.8980645161290323,
+        "oos_recall_8": 0.22,
+        "accuracy_8_summed": 0.7609677419354839,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.7413333333333333,
+        "oos_recall_151": 0.22,
+        "ece_151": 0.6423710499094285,
+        "ece_8_summed": 0.4358271579214899,
+        "nll_151": 2.8263357895563788,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.8980645161290323,
+        "oos_recall_8": 0.22,
+        "accuracy_8_summed": 0.9048387096774193,
+        "oos_recall_8_summed": 0.15,
+        "in_scope_accuracy_150": 0.7413333333333333,
+        "oos_recall_151": 0.22,
+        "ece_151": 0.02474239717994434,
+        "ece_8_summed": 0.04348543877433324,
+        "nll_151": 1.168143794334556,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.766,
+        "oos_recall_8": 0.145,
+        "accuracy_8_summed": 0.6392727272727273,
+        "oos_recall_8_summed": 0.0,
+        "in_scope_accuracy_150": 0.7233333333333334,
+        "oos_recall_151": 0.145,
+        "ece_151": 0.5418609528935713,
+        "ece_8_summed": 0.3242014929965551,
+        "nll_151": 3.1254537433963057,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.766,
+        "oos_recall_8": 0.145,
+        "accuracy_8_summed": 0.772,
+        "oos_recall_8_summed": 0.112,
+        "in_scope_accuracy_150": 0.7233333333333334,
+        "oos_recall_151": 0.145,
+        "ece_151": 0.041523099220243755,
+        "ece_8_summed": 0.04549406836544095,
+        "nll_151": 1.9796040227078606,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k50-seed42.json b/results/runs/bert-base-uncased-k50-seed42.json
new file mode 100644
index 0000000..a81b3f5
--- /dev/null
+++ b/results/runs/bert-base-uncased-k50-seed42.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k50-seed42",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 42,
+    "per_intent": null,
+    "k_shot": 50,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k50-seed42",
+    "seed": 42,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 42,
+      "per_intent": null,
+      "k_shot": 50,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 7625,
+    "oos_train_rows": 125,
+    "per_intent": null,
+    "k_shot": 50,
+    "train_sample_sha256": "8a9a52f8ae41506d9847bd9ceba987edadc6bc3304487be22e664a95e1b9b18f",
+    "step_plan": {
+      "epoch_steps": 1195,
+      "min_train_steps": 400,
+      "planned_steps": 1195,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 1195,
+    "train_loss": 1.5251224821082718,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 412.406,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 2392,
+      "driver_allocated_bytes": 4682612736,
+      "tensor_allocated_bytes": 1827780864
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k50-seed42.npz",
+    "sha256": "42bd57f63c3f37264962334df25861a73c5279f7559302c9a467d5fadd246b7d",
+    "bytes": 4819519
+  },
+  "metrics": {
+    "temperature": 0.7557682510218567,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9706451612903226,
+        "oos_recall_8": 0.56,
+        "accuracy_8_summed": 0.9696774193548388,
+        "oos_recall_8_summed": 0.49,
+        "in_scope_accuracy_150": 0.9553333333333334,
+        "oos_recall_151": 0.56,
+        "ece_151": 0.09483870567248122,
+        "ece_8_summed": 0.06304654970125087,
+        "nll_151": 0.3535677227171782,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9706451612903226,
+        "oos_recall_8": 0.56,
+        "accuracy_8_summed": 0.9709677419354839,
+        "oos_recall_8_summed": 0.52,
+        "in_scope_accuracy_150": 0.9553333333333334,
+        "oos_recall_151": 0.56,
+        "ece_151": 0.014923405200318995,
+        "ece_8_summed": 0.007255170584746151,
+        "nll_151": 0.2839660848141631,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8874545454545455,
+        "oos_recall_8": 0.448,
+        "accuracy_8_summed": 0.8821818181818182,
+        "oos_recall_8_summed": 0.413,
+        "in_scope_accuracy_150": 0.9526666666666667,
+        "oos_recall_151": 0.448,
+        "ece_151": 0.07106428295886225,
+        "ece_8_summed": 0.059034919457277665,
+        "nll_151": 0.7815750641131792,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8874545454545455,
+        "oos_recall_8": 0.448,
+        "accuracy_8_summed": 0.885090909090909,
+        "oos_recall_8_summed": 0.431,
+        "in_scope_accuracy_150": 0.9526666666666667,
+        "oos_recall_151": 0.448,
+        "ece_151": 0.04079916316534721,
+        "ece_8_summed": 0.056157615948682654,
+        "nll_151": 0.7785008501634858,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k50-seed43.json b/results/runs/bert-base-uncased-k50-seed43.json
new file mode 100644
index 0000000..a5e6971
--- /dev/null
+++ b/results/runs/bert-base-uncased-k50-seed43.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k50-seed43",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 43,
+    "per_intent": null,
+    "k_shot": 50,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k50-seed43",
+    "seed": 43,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 43,
+      "per_intent": null,
+      "k_shot": 50,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 7625,
+    "oos_train_rows": 125,
+    "per_intent": null,
+    "k_shot": 50,
+    "train_sample_sha256": "836ad6186d05a042ae00b3ab45efc560eb45f774644bf5e4f9b13a6e66d82982",
+    "step_plan": {
+      "epoch_steps": 1195,
+      "min_train_steps": 400,
+      "planned_steps": 1195,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 1195,
+    "train_loss": 1.5310048378661087,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 410.787,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 2392,
+      "driver_allocated_bytes": 4816666624,
+      "tensor_allocated_bytes": 1833615872
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k50-seed43.npz",
+    "sha256": "21bb8489d1c181fa6982ff89fb8a587505a5fb00ad53031246c0bdecb16bd949",
+    "bytes": 4819533
+  },
+  "metrics": {
+    "temperature": 0.7445813686242696,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9719354838709677,
+        "oos_recall_8": 0.58,
+        "accuracy_8_summed": 0.9716129032258064,
+        "oos_recall_8_summed": 0.55,
+        "in_scope_accuracy_150": 0.9573333333333334,
+        "oos_recall_151": 0.58,
+        "ece_151": 0.1003710316840679,
+        "ece_8_summed": 0.06582087117659888,
+        "nll_151": 0.3464624695725995,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9719354838709677,
+        "oos_recall_8": 0.58,
+        "accuracy_8_summed": 0.9725806451612903,
+        "oos_recall_8_summed": 0.57,
+        "in_scope_accuracy_150": 0.9573333333333334,
+        "oos_recall_151": 0.58,
+        "ece_151": 0.012984647533506592,
+        "ece_8_summed": 0.004612353109791956,
+        "nll_151": 0.2715979968423024,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8952727272727272,
+        "oos_recall_8": 0.488,
+        "accuracy_8_summed": 0.8847272727272727,
+        "oos_recall_8_summed": 0.427,
+        "in_scope_accuracy_150": 0.9531111111111111,
+        "oos_recall_151": 0.488,
+        "ece_151": 0.07594617798340388,
+        "ece_8_summed": 0.05601624769619422,
+        "nll_151": 0.7407453046375219,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8952727272727272,
+        "oos_recall_8": 0.488,
+        "accuracy_8_summed": 0.8907272727272727,
+        "oos_recall_8_summed": 0.461,
+        "in_scope_accuracy_150": 0.9531111111111111,
+        "oos_recall_151": 0.488,
+        "ece_151": 0.04040421351787878,
+        "ece_8_summed": 0.052333279758769086,
+        "nll_151": 0.7294778579471436,
+        "n": 5500
+      }
+    }
+  }
+}
diff --git a/results/runs/bert-base-uncased-k50-seed44.json b/results/runs/bert-base-uncased-k50-seed44.json
new file mode 100644
index 0000000..0590289
--- /dev/null
+++ b/results/runs/bert-base-uncased-k50-seed44.json
@@ -0,0 +1,150 @@
+{
+  "run_name": "bert-base-uncased-k50-seed44",
+  "config": {
+    "model_name": "google-bert/bert-base-uncased",
+    "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+    "seed": 44,
+    "per_intent": null,
+    "k_shot": 50,
+    "oos_train": null,
+    "max_length": 64,
+    "learning_rate": 5e-05,
+    "weight_decay": 0.01,
+    "warmup_ratio": 0.1,
+    "num_train_epochs": 5,
+    "max_steps": -1,
+    "min_train_steps": 400,
+    "train_batch_size": 32,
+    "eval_batch_size": 128,
+    "device": "auto",
+    "replace_classifier_head": false,
+    "checkpoint_root": "checkpoints",
+    "results_root": "results",
+    "eval_per_intent": null
+  },
+  "environment": {
+    "python": "3.12.12",
+    "torch": "2.14.0",
+    "transformers": "5.17.0",
+    "platform": "macOS-26.6.2-arm64-arm-64bit",
+    "device": "mps",
+    "device_name": "Apple M4",
+    "deterministic_algorithms": false,
+    "dataset_revision": "155b9c710419136e17307b80d0a13e68cd46b4ec"
+  },
+  "training": {
+    "run_name": "bert-base-uncased-k50-seed44",
+    "seed": 44,
+    "config": {
+      "model_name": "google-bert/bert-base-uncased",
+      "model_revision": "86b5e0934494bd15c9632b12f734a8a67f723594",
+      "seed": 44,
+      "per_intent": null,
+      "k_shot": 50,
+      "oos_train": null,
+      "max_length": 64,
+      "learning_rate": 5e-05,
+      "weight_decay": 0.01,
+      "warmup_ratio": 0.1,
+      "num_train_epochs": 5,
+      "max_steps": -1,
+      "min_train_steps": 400,
+      "train_batch_size": 32,
+      "eval_batch_size": 128,
+      "device": "auto",
+      "replace_classifier_head": false,
+      "checkpoint_root": "checkpoints",
+      "results_root": "results",
+      "eval_per_intent": null
+    },
+    "git_commit": "f19592f99a9e3f5738435ce45b822812bfb8a13d",
+    "git_dirty": false,
+    "device": "mps",
+    "train_rows": 7625,
+    "oos_train_rows": 125,
+    "per_intent": null,
+    "k_shot": 50,
+    "train_sample_sha256": "5db5ce1ae1864ba63a4164a799b09a7c1d783a4acbd55efc587613d986752013",
+    "step_plan": {
+      "epoch_steps": 1195,
+      "min_train_steps": 400,
+      "planned_steps": 1195,
+      "decided_by": "epochs",
+      "max_steps_arg": -1
+    },
+    "global_step": 1195,
+    "train_loss": 1.528021754181036,
+    "parameters": {
+      "total": 109598359,
+      "trainable": 109598359
+    },
+    "train_wall_seconds": 410.099,
+    "peak_memory": {
+      "device": "mps",
+      "measures": "sampled after each backward and optimizer step: torch.mps.driver_allocated_memory (Metal driver, includes allocator cache; headline, usually at or above the tensor peak but not guaranteed) and current_allocated_memory (live tensors; lower bound)",
+      "samples": 2392,
+      "driver_allocated_bytes": 3734372352,
+      "tensor_allocated_bytes": 1826144768
+    }
+  },
+  "logits": {
+    "file": "bert-base-uncased-k50-seed44.npz",
+    "sha256": "e335e894b2eb54e9f4074b861f202c478078700f308503c82ecc34ec7fc1c6fa",
+    "bytes": 4821765
+  },
+  "metrics": {
+    "temperature": 0.7471188445106913,
+    "validation": {
+      "raw": {
+        "accuracy_8": 0.9693548387096774,
+        "oos_recall_8": 0.57,
+        "accuracy_8_summed": 0.9687096774193549,
+        "oos_recall_8_summed": 0.51,
+        "in_scope_accuracy_150": 0.9566666666666667,
+        "oos_recall_151": 0.57,
+        "ece_151": 0.09335938594501544,
+        "ece_8_summed": 0.06305904733100268,
+        "nll_151": 0.3348679749657449,
+        "n": 3100
+      },
+      "calibrated": {
+        "accuracy_8": 0.9693548387096774,
+        "oos_recall_8": 0.57,
+        "accuracy_8_summed": 0.9690322580645161,
+        "oos_recall_8_summed": 0.53,
+        "in_scope_accuracy_150": 0.9566666666666667,
+        "oos_recall_151": 0.57,
+        "ece_151": 0.01577312201656912,
+        "ece_8_summed": 0.012457932817165913,
+        "nll_151": 0.26291174176244136,
+        "n": 3100
+      }
+    },
+    "test": {
+      "raw": {
+        "accuracy_8": 0.8896363636363637,
+        "oos_recall_8": 0.453,
+        "accuracy_8_summed": 0.8809090909090909,
+        "oos_recall_8_summed": 0.4,
+        "in_scope_accuracy_150": 0.9542222222222222,
+        "oos_recall_151": 0.453,
+        "ece_151": 0.06863477125045468,
+        "ece_8_summed": 0.05745325047739194,
+        "nll_151": 0.8042390815133198,
+        "n": 5500
+      },
+      "calibrated": {
+        "accuracy_8": 0.8896363636363637,
+        "oos_recall_8": 0.453,
+        "accuracy_8_summed": 0.8847272727272727,
+        "oos_recall_8_summed": 0.425,
+        "in_scope_accuracy_150": 0.9542222222222222,
+        "oos_recall_151": 0.453,
+        "ece_151": 0.041007999249223,
+        "ece_8_summed": 0.055990101420774874,
+        "nll_151": 0.8111096671309723,
+        "n": 5500
+      }
+    }
+  }
+}