Complete reproducibility package: data, scripts, results, proofs, paper

README.md

@@ -23,26 +23,27 @@ Reproducibility data for **"Why Additive Span Extraction Heads Cannot Be Improve
 | Method | EM | ε* median | Δε* |
 |---|---|---|---|
 | Additive baseline | 59.8% | 2.17 | — |
-| CE-only fine-tune | 64.0% | 2.20 | +1% (n.s.) |
-| **CBCT (RoBERTa)** | **65.2%** | **3.44** | **+58%** |
+| Elsayed (global margin) | 65.6% | 2.96 | +36% |
+| **CBCT (RoBERTa)** | **66.0%** | **3.31** | **+52%** |
 | CBCT (BERT-large) | 62.0% | 4.71 | +113% |
 | CBCT (DistilBERT) | 57.8% | 2.96 | +45% |
 | Biaffine spectral | 65.4% | 0.93 | -57% |
-| Sign-safe coupling | 63.7% | 1.55 | -29% |
+
+Statistical tests: CBCT vs baseline Wilcoxon p = 10⁻⁴², CBCT vs Elsayed p = 10⁻⁶.
 
 ## Contents
 
-- **data/logits/**: 14,863 examples × 10 extraction heads × 6 configs. Full start/end logits, margins, certified radii.
-- **data/hidden_states/**: Hidden state vectors for 119 SQuAD examples (RoBERTa-base-SQuAD2).
-- **data/results/**: All experiment outputs as JSON (sign coupling, spectral coupling, baseline comparison, multi-model CBCT, cross-dataset).
-- **verify_paper_data.py**: Checks every number in the paper against the data files.
+- **data/logits/**: 14,863 examples × 10 extraction heads × 6 configs
+- **data/hidden_states/**: Hidden state vectors for SQuAD examples
+- **data/results/**: All experiment outputs (baseline comparison, multi-model CBCT, paired tests, theory verification, hyperparameter sensitivity, bucket analysis, cross-dataset)
+- **verify_all.py**: Checks every number in the paper against data files
 
 Full code release on GitHub after acceptance.
 
 ## Verification
 
 ```python
-python verify_paper_data.py
+python verify_all.py
 ```
 
 ## Citation

data/results/high_impact_checkpoint.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:29c4cd41a54a8b3fac5f03b8f229dd488f9fcc76f360cefa996072aa1fe04e66
+size 7851

data/results/high_impact_results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:29c4cd41a54a8b3fac5f03b8f229dd488f9fcc76f360cefa996072aa1fe04e66
+size 7851

data/results/hyperparam_checkpoint.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:57c162251d7a0536e01453563c02843a2937d214c29bf3151fce4b18c5fd00ea
+size 2238

data/results/nq_results.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0991913129a2330e39955c715f6e9ee3b2df00b857379842529c2bf0e5227b6b
+size 182

data/results/paired_tests.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0008a59d569e30bf294b6d5838de09e71c99cdb73bcc043d053b7c5edf2249eb
+size 862

data/results/remaining_data.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00f29d3ece2821971963cec40a5a9ee0732c46ccf26c1bd5331bd423c0c39321
+size 4012

data/results/remaining_data_checkpoint.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00f29d3ece2821971963cec40a5a9ee0732c46ccf26c1bd5331bd423c0c39321
+size 4012

data/results/theory_verification.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f2b0867d35ba75b28cd1f0e9c0b991eb880021741138a9c44652eb80ff950bc
+size 2721

data/results/theory_verification_checkpoint.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0f2b0867d35ba75b28cd1f0e9c0b991eb880021741138a9c44652eb80ff950bc
+size 2721