remove ambiguous moderation rows, replace with clear-cut examples

replace ambiguous salary issue with date format fix

remove ambiguous LR fix — identify-only, any valid LR works

fix moderation issue row collisions and verify all data

add content moderation task with real OpenAI Moderation data

add toxic/biased response issue to alignment task

replace ambiguous fixes with deterministic ones across all tasks

make alignment issues subtler to challenge frontier models

use real NVIDIA HelpSteer data for alignment task

improve alignment task: replace label swaps with real contamination

use real Stanford Alpaca data for alignment task

add alignment data QA task: 12 issues in LLM instruction-tuning data

expand datasets to include harder real-world scenarios

expand datasets

add fix stage+demo

fixes v1: add per step reward

init