Keep only the paper dataset release

Remove abandoned training-time caps and document the single dataset revision used by the HAM-Net paper.

README.md

@@ -52,26 +52,17 @@ A simplified MIL record looks like:
 `func_label` is present only when it can be derived from patch or line-level
 annotations. It is not available for PROMISE.
 
-## Published versions
+## Reproducibility release
 
-The repository keeps the original JSONL filenames in every version. Select a version
-through the Hugging Face revision or Git tag:
-
-| Version | Revision | Description |
-|---|---|---|
-| v1 | `v1.0.0` | Reproducibility release for the original HAM-Net CPDP protocol: preserves the original bag identities and order, retains original source text and comments, removes the build-time function-count and 800-node AST limits, and includes the legacy CPDP splits. |
-| v2.0.0 | `v2.0.0` | Reconstructed PROMISE, Defactors, BugsInPy, and Big-Vul from normalized source. Comments/docstrings and non-literal blank lines are removed; Python docstring-only suites are replaced with synthetic `pass` so every v1 bag/function remains. Labels are frozen and AST graphs are regenerated. |
-| v2.0.1 | `v2.0.1` | Reconstructed the same four datasets without synthetic code. Python docstring-only functions/classes are deleted; bags that become empty, or positive bags with no remaining positive function, are deleted. Remaining bag/function labels are frozen and AST graphs are regenerated. |
-
-For example:
+The repository contains the dataset revision used by the HAM-Net paper's fixed
+CPDP experiments. The `main` branch and the immutable `v1.0.0` tag contain the
+same four JSONL files and matching split files. To pin the paper revision:
 
 ```bash
 hf download Scream9371/hamnet-datasets promise_java.jsonl --revision v1.0.0
-hf download Scream9371/hamnet-datasets promise_java.jsonl --revision v2.0.0
-hf download Scream9371/hamnet-datasets promise_java.jsonl --revision v2.0.1
 ```
 
-### v1.0.0 reproducibility contract
+### Reproducibility contract
 
 `v1.0.0` is pinned to the four JSONL files at repository root and the fixed
 CPDP files under `splits/cpdp/`. The JSONL files preserve the bag identity and
@@ -79,10 +70,8 @@ order used by the original local HAM-Net experiments, while each bag contains
 the complete parseable function list. Source text retains comments and AST
 graphs are not truncated at 800 nodes.
 
-Matching deterministic training-time caps are under `caps/v1.0.0/`
-(`ast_topk_v1`, `K=16`, `top_m=12`). Do not reuse caps generated for the
-earlier function-capped bags: their indices are not valid for the unbounded
-PROMISE and Defactors bags.
+The paper experiments use the complete function bags. No deterministic
+training-time caps are provided or used by this release.
 
 The BugsInPy CPDP split intentionally contains the 2,416 bags used by the
 original HAM-Net experiment repository.
@@ -166,8 +155,8 @@ The following rules define the published MIL samples.
   sampling quota targets files in the same directory or top-level module when such
   candidates exist; remaining candidates are sampled from the other files.
 - Function sampling: use seed 42, discard files with no parseable function, and
-  retain the complete parseable function list in the published full-function version.
-  Both releases also exclude bags that contain no function definition: older
+  retain the complete parseable function list. This release also excludes bags
+  that contain no function definition: older
   construction artifacts represented them as a single ``<file>`` placeholder,
   which is not a valid MIL instance.
 
@@ -188,7 +177,7 @@ The following rules define the published MIL samples.
 - Function sampling: discard files with fewer than three parseable functions and
   retain all remaining functions in the full-function version.
 
-### AST and version rules
+### AST normalization rules
 
 - To make ASTs from Java, Python, C, and C++ comparable, language-specific node
   types are mapped to a shared set of structural roles, such as declarations,
@@ -197,18 +186,11 @@ The following rules define the published MIL samples.
   language-specific lexical details. Identifiers, punctuation, comments, and
   documentation text are not retained as structural AST nodes. The normalized
   graph contains parent-child structure, reverse edges, and self-loops.
-- In v2.0.0 and v2.0.1, comments are removed with language-aware lexical
-  scanning while string and character literals are retained. Function labels
-  are derived before source normalization; no defect label is recomputed from
-  normalized source positions.
-- v2.0.0 replaces a Python function/class body consisting solely of a docstring
-  with synthetic `pass`, retaining all v1 bags and functions.
-- v2.0.1 instead deletes Python docstring-only functions and classes. It omits
-  a bag if no functions remain, and omits a positive bag if no remaining
-  function has its frozen positive `func_label`; all retained labels and
-  function order are preserved.
 
 ## Files
 
-- Both versions: `promise_java.jsonl`, `defactors_python.jsonl`,
-  `bugsinpy_python.jsonl`, and `bigvul_c.jsonl`.
+- `promise_java.jsonl`
+- `defactors_python.jsonl`
+- `bugsinpy_python.jsonl`
+- `bigvul_c.jsonl`
+- `splits/cpdp/`: fixed project-disjoint splits used by the paper