From 04011a880cbd59ee022d566220bf7fd4818205e2 Mon Sep 17 00:00:00 2001
From: YurenHao0426 <Blackhao0426@gmail.com>
Date: Wed, 8 Apr 2026 20:00:22 -0500
Subject: paper v2.34.1: SB/CB training loss decrease values from 3-seed (were
 s42)
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Same bug pattern as v2.33's nudging test: the §4 ¶4 training loss
decrease values for SB+pen (-0.458) and CB+pen (-0.122) were s42
single-seed numbers labeled as part of the "three seeds" framing.
DFA+pen (-0.095 ± 0.007) was actually 3-seed.

Re-aggregated from existing JSONs (no new compute):
  SB+pen: per-seed {0.457, 0.444, 0.439} → 0.447 ± 0.008 (was 0.458)
  CB+pen: per-seed {0.123, 0.118, 0.124} → 0.121 ± 0.003 (was 0.122)
  DFA+pen: per-seed {0.104, 0.088, 0.093} → 0.095 ± 0.007 ✓ (unchanged)

Changes:
- §4 ¶4 training-loss trajectory line now uses 3-seed mean ± std for
  all three methods
- Appendix L paragraph now lists per-seed decreases for all three
- New auditable file: results/training_loss_decrease_3seed.json

Ratios SB ≫ CB ≈ DFA unchanged. The "all three functional metrics
agree on the SB ≫ CB ≈ DFA ordering" claim is unchanged.

Page layout: §1-§7 still 9 pages, refs p10, total 19 pages. 0 overfull.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---
 results/training_loss_decrease_3seed.json | 68 +++++++++++++++++++++++++++++++
 1 file changed, 68 insertions(+)
 create mode 100644 results/training_loss_decrease_3seed.json

(limited to 'results')

diff --git a/results/training_loss_decrease_3seed.json b/results/training_loss_decrease_3seed.json
new file mode 100644
index 0000000..43202b9
--- /dev/null
+++ b/results/training_loss_decrease_3seed.json
@@ -0,0 +1,68 @@
+{
+  "description": "Training loss decrease (loss[ep1] - loss[ep30]) for SB+pen, CB+pen, DFA+pen 3-seed. Used in \u00a74 \u00b64 functional triangulation alongside nudging test.",
+  "per_method": {
+    "state_bridge": {
+      "per_seed": {
+        "42": {
+          "ep1_loss": 2.0474753201293945,
+          "ep30_loss": 1.5901478890609742,
+          "decrease": 0.45732743106842033
+        },
+        "123": {
+          "ep1_loss": 2.0394552979278564,
+          "ep30_loss": 1.5955050112533569,
+          "decrease": 0.44395028667449954
+        },
+        "456": {
+          "ep1_loss": 2.0369125997924806,
+          "ep30_loss": 1.5980215132904052,
+          "decrease": 0.4388910865020754
+        }
+      },
+      "three_seed_mean": 0.44672293474833175,
+      "three_seed_std_ddof0": 0.007777762456655643
+    },
+    "credit_bridge": {
+      "per_seed": {
+        "42": {
+          "ep1_loss": 1.9963660776519776,
+          "ep30_loss": 1.8736286736679078,
+          "decrease": 0.12273740398406985
+        },
+        "123": {
+          "ep1_loss": 1.9903950632476806,
+          "ep30_loss": 1.8727011511993408,
+          "decrease": 0.11769391204833979
+        },
+        "456": {
+          "ep1_loss": 1.9934451489257812,
+          "ep30_loss": 1.8695441330337523,
+          "decrease": 0.12390101589202884
+        }
+      },
+      "three_seed_mean": 0.12144411064147949,
+      "three_seed_std_ddof0": 0.0026940044894884965
+    },
+    "dfa": {
+      "per_seed": {
+        "42": {
+          "ep1_loss": 1.9962164908218383,
+          "ep30_loss": 1.8923273934555054,
+          "decrease": 0.1038890973663329
+        },
+        "123": {
+          "ep1_loss": 1.9917970180892943,
+          "ep30_loss": 1.9041575216674804,
+          "decrease": 0.0876394964218139
+        },
+        "456": {
+          "ep1_loss": 1.9882178707504272,
+          "ep30_loss": 1.8952873357391358,
+          "decrease": 0.09293053501129145
+        }
+      },
+      "three_seed_mean": 0.09481970959981274,
+      "three_seed_std_ddof0": 0.006767033726504009
+    }
+  }
+}
\ No newline at end of file
-- 
cgit v1.2.3