Compile real-world Claude Code and Codex trajectories into verified, tradable post-training assets.
-
Updated
Nov 1, 2026 - Python
Compile real-world Claude Code and Codex trajectories into verified, tradable post-training assets.
Procedural data generators for verifiable reasoning, synthetic pretraining, post-training, evaluation, and RL.
Context & Guide For Reinforcement Learning with Verifiable Rewards with Large Language Models
Score the trustworthiness of outputs from any LLM in real-time
A curated list of rubrics, checklists, criteria sets, principles, and scoring guides used to score, rank, verify, filter, or train modern generative models.
🐀 Fuzz your verifier before an RL agent does. Static + dynamic LLM security auditor to detect reward-hacking in RL post-training environments (OpenEnv, verifiers-spec, Gymnasium).
Write loops, not prompts. The loop is easy — the verifier is the whole game. VCN night one at Network School.
An RL Enviorment for AES Inversion
Open Arena: design and run benchmarks for model and harness evals on what good looks like.
Gymnasium environment where AI agents solve SPaRC grid puzzles one move per step, with optional backtracking. Comes as a pip package, a Prime Intellect verifiers env and an OpenEnv server. The best model solves 16.0%, humans 98.0%.
Adversarial QA for LLM-RL environments: find out what reward an empty answer earns. Model-free, zero API cost.
A verifiers RLM environment for testing whether adaptive recursive search outperforms brittle manual RAG choreography on long synthetic corpora.
Probe RL environment verifiers for false passes and false fails
Verifiable RL environments for corporate law & governance — deterministic reward, no LLM judge, fully synthetic worlds.
Typed asset shapes + visual + headless views for AI agents. One asset definition. Three rendering targets (HTML / Markdown / Text).
Reproducible verifier audits, datasheets, agreement metrics, and release gates
An open reinforcement-learning (RL) environment that trains LLM agents to use the current fact, not the stale one — verifiable reward for temporal fact-currency, built on verifiers / prime-rl (GRPO, LoRA).
Train your own support-triage model with RL and a reward computed in code — recipe, evaluator, simulator, trainer. No judge model.
Research proposal for verifier-gated on-policy distillation with explicit evaluation and claim boundaries
Verifiers hello world repo
To associate your repository with the verifiers topic, visit your repo's landing page and select "manage topics."