Spaces:
Sleeping
Sleeping
Commit History
-
Sanitize empty finding descriptions
e3e5d23 -
Harden reward parsing for truncated JSON
fd84f8b
-
Round2 hardening, campaign fixes, training quickstart, and CI gate
92ead25 -
Round 2 Implementation: Multi-period campaign, regulatory shocks, adversarial grading, and GRPO training infrastructure
52f5c27
-
Emergency fix: drop scoring limit to 2 decimal points and 0.01 / 0.99 to pass deep verification
57d984d
-
Fix phase 2 deep validation grader scoring out-of-bounds error
140ce9d -
Refactor graders.py: add explicit partial credit comments, strict duplicate handling, and weighted false negative tracking
e43c1bc -
Fix inference stdout formatting and update graders tests for strictly clamped reward constraint bounds
21fc032 -
fix: increase score epsilon to 0.01 for validator safety
b4d5e6a
-
feat: v2.0 — fraud detection task, severity grading, investigation mode, security hardening, 78 tests
- Added Task 4: Fraud Pattern Detection (Expert) with 10 fraud types - Fixed datetime.now() reproducibility bug (REFERENCE_DATE) - Added noise/red herring entries to all data generators - Scaled up GST task (40+ entries, lower error density) - Added cumulative_breach and cascading_total errors - Enhanced grading: severity-weighted F1, partial credit, confusion matrix, risk scoring - Added investigation mode (opt-in multi-step data exploration) - Added adaptive difficulty tracking - Security: rate limiter TTL cleanup, request IDs, CORS env toggle, ground truth leakage prevention - Added /leaderboard, /metrics, /session, /adaptive-difficulty endpoints - Session-based multi-tenancy with global fallback - Comprehensive test suite: 78 tests covering all components126bdbd