v2.1 card: document cwes_predicted (cascade-arm column)

README.md

@@ -143,6 +143,7 @@ The full source analysis is documented in the
 | `cwes` | list[string] | CWE assignments, e.g. `CWE-502 Deserialization of Untrusted Data` — v2 |
 | `affected_products` | list[string] | `vendor product` pairs from the CVE record — v2 |
 | `cpes` | list[string] | CPE identifiers — v2 |
+| `cwes_predicted` | list[string] | Top-1 prediction of [CIRCL/cwe-parent-vulnerability-classification-roberta-base](https://huggingface.co/CIRCL/cwe-parent-vulnerability-classification-roberta-base) on the description, same display format as `cwes` — v2.1 |
 
 ### Structured metadata columns (v2, added 2026-08-06)
 
@@ -165,6 +166,18 @@ by `label_sources` when using these columns as model inputs:
 
 CVSS versions among the 869 vectors: 677 × v3.1, 173 × v3.0, 18 × v4.0, 1 × v2.0.
 
+### Predicted CWE column (v2.1, added 2026-08-08)
+
+`cwes_predicted` holds the top-1 output of the deployed CIRCL CWE guesser
+([CIRCL/cwe-parent-vulnerability-classification-roberta-base](https://huggingface.co/CIRCL/cwe-parent-vulnerability-classification-roberta-base),
+parent-level, 303 classes) run on each row's title + description. Coverage
+is 100% by construction; agreement with the gold `cwes` column (ancestor
+level, on the 814 rows whose gold entry carries a parseable CWE id) is
+27.3% top-1. The column exists to measure the *cascade cost* of replacing
+gold CWE input with a model prediction in downstream CVE→ATT&CK
+classifiers; it is a model output, not curated ground truth — do not use
+it as labels. v1/v2 columns are unchanged (strictly additive update).
+
 ## Label statistics
 
 192 distinct techniques; 66 with at least 5 examples. Most CVEs carry 1–3