Publish LedgerBench-100 v3.1 realism release

README.md

@@ -37,7 +37,7 @@ appears in the reward path.
 
 - Tasks: 100 across 22 families: anomaly_triage (1), bank_rec (4), business_brief (3), business_brief_fb (3), cash_app (3), cash_forecast (2), close_mgmt (6), collections_ops (1), cross_system (7), erp_qa (11), erp_qa_fb (5), erpbench (10), expense_audit (4), finance_qa (10), finance_qa_fb (2), fpna (5), journal_entry (2), payment_proposal (2), payment_run (3), pbc (4), threeway_match (5), vendor_master (7)
 - Oracle walk length: min 30 / median 34 / max 132 MCP calls (4334 total); required distributed evidence reads are 19-24 per task
-- Executable checks: 434 answer + 857 trace + 764 state, expanded into 24-30 exact public criteria per task
+- Executable checks: 434 answer + 857 trace + 764 state, expanded into 42-48 exact public criteria per task
 - Inspectable assets: 28 agent-visible native files per task, including valid XLSX, PDF, EML, CSV, JSON, Markdown, and text records; gold and oracle recipes are excluded from this tree
 - Reference diversity: 100/100 distinct raw server/tool sequences and 100/100 distinct semantic action graphs
 - Escalated variants: 30 tasks are controls-review follow-ups whose governing policy must be found among seeded adjacent documents; every follow-up now has its own human request and policy-discovery trajectory

reports/build.json

@@ -29,9 +29,9 @@
     "unique_sha256": 2836
   },
   "criteria_per_task": {
-    "max": 30,
-    "median": 27,
-    "min": 24
+    "max": 48,
+    "median": 45,
+    "min": 42
   },
   "decision_options_per_task": 3,
   "escalated_variant_pairs": 30,
@@ -55,6 +55,7 @@
   "quality_gates": {
     "all_assets_content_unique": true,
     "all_native_assets_parse": true,
+    "at_least_forty_specific_public_causal_criteria": true,
     "deep_evidence_intersection": true,
     "deterministic_verifier": true,
     "evidence_depth_varies": true,
@@ -69,7 +70,6 @@
     "real_native_formats_present": true,
     "reference_sequence_similarity_below_limit": true,
     "semantic_action_graph_similarity_below_limit": true,
-    "specific_public_causal_criteria": true,
     "ten_negative_controls": true,
     "three_options_one_selected": true,
     "twenty_eight_generated_assets_per_task": true,
@@ -137,7 +137,7 @@
     "random_calls": 0,
     "wall_clock_reads_in_reward_path": 0
   },
-  "version": "3.0.0",
+  "version": "3.1.0",
   "walk_len": {
     "max": 132,
     "median": 34,