Automated README summary
Automatically summarized from the README on 30 Oct 2026; not a run by jevbooks. Run instructions and author-reported outcomes still need verification.
What it is
AgentJev-0.6B is a framework-independent decision service for AI agent loops. It takes unstructured state (diffs, traces, logs) and typed questions, returning calibrated probability distributions for gates, routes, or scores.
How it uses Jev
Jev makes decisions by encoding the state once with a Qwen3-0.6B backbone (LM head removed) and scoring candidate options via a permutation-equivariant head. It answers Boolean, Choice, and Score questions, returning full distributions. Results are used in code by thresholding probabilities or comparing margins.
Primitives:choicescorenoul
Technique worth stealing
Tree-encoded forward pass merges candidate paths into a token trie, encoding state once for all questions and cutting latency.
Try it
Run via HTTP API or Python client; see README for commands.
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-1.14.0- Project signals
jev-1.14.0- Category
jev-1.14.0- Pattern tags
jev-1.14.0
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 material0.74
- Shows a System One patternFound in material0.85
- Describes handling uncertaintynot found in the excerpt0.03
- Reports measured numbersunclear0.53
- Is a runnable toolFound in material0.87
- Top-10 learning shortlist signalnot found in the excerpt0.18
- Is a replica of the modelFound in material0.91
- Solves a common problem (0–2)score on a 0–2 scale1.70 / 2
- About JevFound in material0.82
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 30 Oct 2026.