Automated README summary
Automatically summarized from the README on 19 Oct 2026; not a run by jevbooks. Run instructions and author-reported outcomes still need verification.
What it is
DC-Jev-RL studies using Jev (Kev-4B) as a calibrated reward judge in PPO for MuJoCo control and GRPO for math and code, reporting measured results including where the judge did not help.
How it uses Jev
Kev-4B judges short motion windows as a PPO reward on five MuJoCo tasks, and scores or gates solutions in GRPO for MATH and code repair; its probabilities are also used to rank unlabelled samples and choose tree-rollout prefixes.
Primitives:scorechoice
Technique worth stealing
Gating a verifier reward with a calibrated judge score, and sharing one Qwen3.5-4B backbone between policy and judge to save memory.
Try it
bash scripts/setup_colab.sh, then bash scripts/run_validation.sh; see README for part scripts.
Model judgments from README excerpts
Project signals use repository metadata, an existing summary and the first 700 README characters, not the full project. “Not found” means not detected in that material; it does not establish absence. These are the original judgments with recorded models and dates, not source audits or run results.
Models & evaluation dates
- Inclusion gate
jev-1.13.0- Project signals
jev-1.13.0- Category
jev-1.13.0- Pattern tags
jev-1.13.0
Evidence
These signals use repository metadata, an existing summary and at most 700 cleaned README characters. ≥ 0.60 means reported in that excerpt; ≤ 0.40 means not found there. The middle band is inconclusive. None is a source-code audit or a test result. The runnable-tool question excludes research experiments; the recommendation question asks for a learning shortlist, not a quality verdict.
- Jev is core to the projectFound in material0.73
- Shows a System One patternnot found in the excerpt0.17
- Describes handling uncertaintynot found in the excerpt0.06
- Reports measured numbersnot found in the excerpt0.20
- Is a runnable toolnot found in the excerpt0.07
- Top-10 learning shortlist signalnot found in the excerpt0.24
- Is a replica of the modelunclear0.60
- Solves a common problem (0–2)score on a 0–2 scale1.00 / 2
- About JevFound in material0.66
no-judgment band (0.40–0.60)
no-judgment band (0.40–0.60)
The Jev model and date for each judgment stage are listed above. Card written by DeepSeek V4.1 Flash from the README on 19 Oct 2026.