Automated README summary
Automatically summarized from the README on 3 Nov 2026; not a run by jevbooks. Run instructions and author-reported outcomes still need verification.
What it is
MicroLLM is a local, CPU-only probabilistic decision engine for AI agents, replacing generative System-2 calls for routing, policy checks, severity scoring, injection filtering and RAG groundedness with calibrated System-1 judgments.
How it uses Jev
It exposes Jev-style primitives (noul, choice, score, multi) over a state string or JSON via POST /v1/systemone. Decisions are calibrated with per-K temperature scaling, two-stage OpenJev-RLCD² training and split conformal prediction, emitting an x_abstain signal when options remain plausible.
Primitives:choicescorenoul
Technique worth stealing
Split conformal prediction plus two-stage calibrate-then-reinforce training to emit certified abstention instead of guessing.
Try it
Run `microllm benchmark --all-suites` (or `--fast`) to reproduce benchmarks on CPU.
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 material0.86
- Shows a System One patternunclear0.54
- Describes handling uncertaintynot found in the excerpt0.26
- Reports measured numbersnot found in the excerpt0.34
- Is a runnable toolFound in material0.83
- Top-10 learning shortlist signalnot found in the excerpt0.14
- Is a replica of the modelFound in material0.75
- Solves a common problem (0–2)score on a 0–2 scale1.61 / 2
- About JevFound in material0.86
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 3 Nov 2026.