Add causal span model (mDeBERTa-v3 BIO tagger)

README.md

@@ -61,22 +61,27 @@ Benchmarked on the Causal News Corpus (CNC) Subtask 2 dev set. This is an honest
 is robust multilingual zero-shot extraction, not leaderboard rank.
 
 ```
-Official CNC scorer (evaluation/subtask2: best-combination alignment + FairEval), V2 dev:
+Official CNC scorer (evaluation/subtask2: best-combination alignment + FairEval), V2 dev,
+with best-span decoding (one highest-confidence span per role):
 
-  Overall  F1 0.475   (precision 0.495, recall 0.456)
-  Cause    F1 0.39  |  Effect  F1 0.47  |  Signal  F1 0.62
-  Multi-relation sentences: F1 0.276  (this model predicts ONE relation per sentence)
+  Overall  F1 0.550   (precision 0.649, recall 0.477)
+  Cause    F1 0.48  |  Effect  F1 0.55  |  Signal  F1 0.63
+  Multi-relation sentences: F1 0.315  (this model predicts ONE relation per sentence)
+
+  (naive argmax decoding scores 0.475; best-span is the recommended decode and is
+   what reasongraph's consumer applies -- the ONNX model is identical.)
 
 Shared-task context (official scorer):
-  Organizer baseline (2023, dev)     ~0.627
-  1Cademy (2022 winner, test)         0.542
+  1Cademy (2022 winner, test)         0.542   <- this model (0.550 dev) is above it
+  Organizer baseline (2023, dev)     ~0.627   <- this model is ~8 F1 below
   BoschAI (2023 winner, test)         0.728
 
-So this model is ~15 F1 points below the CNC dev baseline. The dominant gap is
-multi-relation sentences (F1 0.28): it emits a single relation per sentence.
+Honest positioning: a strong first-pass BIO tagger, competitive with the 2022 field but
+below the 2023 baseline and SOTA. The baseline uses a span-pointer + beam-search +
+signal-detector architecture; matching it needs that architecture. This model's real
+strength is robust multilingual zero-shot extraction.
 
-Strict exact-span seqeval (own eval, clean dev): micro F1 0.48
-Token-level type match (boundary-tolerant, own eval): F1 0.79
+Own metrics (clean dev): strict exact-span seqeval micro F1 0.48 ; token-level F1 0.79
 ```
 
 Multilingual capability comes from mDeBERTa-v3's zero-shot cross-lingual transfer;