fix to readme

README.md

@@ -69,14 +69,11 @@ Each example includes: the NLQ, database identifier, a canonical dataset id, the
 
 `synthetic_train` is produced via our **auto-annotation pipeline**, which aligns SQL-referenced entities to NLQ spans using string-similarity candidates (Jaccard 3-gram / Levenshtein) and a **non-overlapping ILP** selection objective. See **Annotation** below.
 
-### `human_gold_50` — database-scoped gold standard (evaluation only)
-
-A 50-question gold standard for evaluating **database-aware** (closed-world) DB-ER systems. Unlike the other splits, annotators labeled these questions with **read-only access to the actual database instance** (BIRD-train and Spider SQLite databases), so labels are grounded in the concrete tables, columns, and stored values of each database — not only in the schema and paired SQL.
+`human_gold_50` 
+A 50-question gold standard. Unlike the other splits, annotators labeled these questions with **access to the actual database instance** (BIRD-train and Spider SQLite databases), so labels are grounded in the concrete tables, columns, and stored values of each database, not only in the schema and paired SQL.
 
 - 50 questions over **45 distinct databases**; 750 tokens, **232 entity tokens** (68 Table / 105 Column / 59 Value), **190 IOB2 mentions**.
-- **Disjoint from every other split** (by `dber_id`); never used for training or hyperparameter tuning.
-- Label reliability: an independent blind re-annotation of all 50 questions yielded **Cohen's κ = 0.90** (4-class; 95% CI [0.86, 0.93]) and 95.1% raw agreement.
-- Intended use: evaluation only.
+- Reliability: an independent blind re-annotation of all 50 questions yielded **Cohen's κ = 0.90** (4-class; 95% CI [0.86, 0.93]) and 95.1% raw agreement.
 
 ---