Benchmark Report · Protocol 3.0.1 · September 2026

Jev vs. Machine Learning

An empirical benchmark comparing Jev with conventional machine-learning pipelines across text and tabular classification tasks.

Balanced Accuracy · 3 Training Seeds (2027, 2028, 2029) · Fixed Test Holdout (Seed 20260920)
Benchmark Map 8 Datasets · 2 Domains
JEV IMDb 96.3% SMS 96.1% News 87.5% Bank 59.7% Shop 51.8% Diag 88.4% ← TEXT DOMAIN TABULAR DOMAIN →
08 Datasets (4 Text · 4 Tabular)
11 Classical Model Families
03 Training Seeds (2027, 2028, 2029)
01 Fixed Holdout (Seed 20260920)
02 Decision Panels (Raw · Calibrated)

Where does Jev differ from classical ML?

Balanced accuracy on the fixed test holdout under Protocol 3.0.1. Inspect both raw default rules and calibrated policy thresholds.

Decision Rule
Domain
Primary Visual · Range Comparison

Performance Range Comparison

Balanced Accuracy (%) on test holdout · One row per dataset · Direct signed Δ displacement

Sort By:
Jev Zero-Shot Best Classical ML Range Span
40%50%60%70%80%90%100%BALANCED ACCURACY SCALESIGNED Δ (pp)IMDbTEXT · 2 CL88.4%96.3%+7.9 ppLargest Jev LeadSMS SpamTEXT · 2 CLJev 96.1%ML 95.0%+1.1 ppNear ParityAG NewsTEXT · 4 CLJev 87.5%ML 88.4%-0.9 ppNear ParityIrisTABULAR · 3 CLJev 97.0%ML 100.0%-3.0 ppBanking77TEXT · 77 CL89.7%78.9%-10.8 ppOnline ShoppersTABULAR · 2 CL69.1%51.4%-17.7 ppBank MarketingTABULAR · 2 CL71.8%53.4%-18.4 ppBreast CancerTABULAR · 2 CL100.0%61.0%-39.0 ppLargest ML Lead
Secondary Visual · Signed Effect Size

Per-Dataset Signed Difference (Δ)

Zero-centered displacement (Jev zero-shot minus best classical pipeline in percentage points)

Classical Lead (< 0) 0.0 Parity Jev Lead (> 0)
← CLASSICAL ADVANTAGEJEV ADVANTAGE →-40 pp-30 pp-20 pp-10 pp0.0 (Parity)+10 ppIMDbTEXT+7.9 ppSMS SpamTEXT+1.1 ppAG NewsTEXT-0.9 ppIrisTABULAR-3.0 ppBanking77TEXT-10.8 ppOnline ShoppersTABULAR-17.7 ppBank MarketingTABULAR-18.4 ppBreast CancerTABULAR-39.0 pp
Text · 4 cl

AG News

-0.9%
0% 50% 100%
Jev Zero 87.5%
Best ML (SVM) 88.4%
Text · 77 cl

Banking77

-10.8%
0% 50% 100%
Jev Zero 78.9%
Best ML (SVM) 89.7%
Text · 2 cl

SMS Spam

+1.1%
0% 50% 100%
Jev Zero 96.1%
Best ML (Naive Bayes) 95.0%
Text · 2 cl

IMDb

+7.9%
0% 50% 100%
Jev Zero 96.3%
Best ML (Logistic reg) 88.4%
Tabular · 2 cl

Bank Marketing

-18.4%
0% 50% 100%
Jev Zero 53.4%
Best ML (SVM) 71.8%
Tabular · 2 cl

Online Shoppers

-17.7%
0% 50% 100%
Jev Zero 51.4%
Best ML (SVM) 69.1%
Tabular · 2 cl

Breast Cancer

-39.0%
0% 50% 100%
Jev Zero 61.0%
Best ML (SVM, k-NN) 100.0%
Tabular · 3 cl

Iris

-3.0%
0% 50% 100%
Jev Zero 97.0%
Best ML (Logistic reg) 100.0%
← Swipe horizontally to view all model columns →
Main benchmark comparison: Jev vs best classical pipeline by dataset
Dataset ↕Domain ↕Classes ↕Test N ↕Jev Zero-shot ↕Jev Few-shot ↕Best Classical ↕Δ vs Best ↕Inspect

Key Empirical Observations

Data-grounded findings across domain types, policy thresholds, and conditioning strategies.

01 · Text Polarity

IMDb Sentiment Standout

On IMDb, zero-shot Jev reaches 96.3% balanced accuracy vs. 88.4% for TF-IDF Logistic Regression (+7.9 pp lead). The advantage persists under policy calibration (96.1% vs 88.3%).

02 · Threshold Shift

Thresholding Changes SMS Spam

Under raw decisions, Jev leads Naive Bayes (96.1% vs 95.0%). Following policy threshold optimization, Naive Bayes reaches 96.3% against Jev's 95.9%.

03 · Tabular Ensembles

Tabular Business Tasks

On Bank Marketing, adjusted zero-shot Jev reaches 59.7% vs. 73.3% for the voting ensemble. On Online Shoppers, Jev scores 51.8% vs. 71.2%.

04 · Few-Shot Scope

Selective Few-Shot Efficacy

One example per class lifts raw Breast Cancer from 61.0% to 88.8% and Banking77 from 78.9% to 81.9%, while slightly reducing means on AG News and IMDb.

Interactive Methodology Pipeline

Protocol 3.0.1 strictly isolates candidate selection, decision-threshold calibration, and final holdout evaluation. Click any stage to inspect.

01 · Data Split & Holdout Isolation

A dedicated test holdout is isolated before any model exploration (Seed 20260920). It remains completely untouched during feature extraction, hyperparameter tuning, and threshold selection.

Train Partition Up to 8,000 rows Model parameter learning
Validation Split Up to 1,000 rows Hyperparameter candidate tuning
Policy Split Up to 500 rows Threshold search & calibration
Test Holdout Shared & Frozen Final evaluation across 3 seeds

Model Architecture Matrix

11 classical model families, 2 Jev prompt formulations, and a majority-class reference baseline.

Model Family Feature Representation Backend Runtime GPU Support Tuning Budget
Jev Zero-Shot Structured text / feature prompts Jev API (jev-1.13.0) ● (Cloud) 1 config
Jev Few-Shot In-context 1-example per class Jev API (jev-1.13.0) ● (Cloud) 1 config
Logistic Regression TF-IDF / StandardScaler cuML / scikit-learn 4 candidates
Linear / RBF SVM TF-IDF / Dense Scaled cuML / scikit-learn 4 candidates
Random Forest Dense / Sparse Scaled cuML / scikit-learn 4 candidates (max 150 trees)
Extra Trees Dense / Sparse Scaled scikit-learn 4 candidates (max 150 trees)
k-Nearest Neighbors TruncatedSVD (300d) / Dense cuML / scikit-learn 4 candidates
Naive Bayes Count / TF-IDF / Gaussian scikit-learn 4 candidates
Hist Gradient Boosting Binned Numerical / Dense scikit-learn 4 candidates (max 60 iters)
XGBoost Tree Features / DMatrix CUDA / CPU 4 candidates
CatBoost Native Categorical & Numerical CUDA / CPU 4 candidates
Voting Ensemble Probability Aggregation CPU / Mixed Composite soft voting
Majority Baseline Constant class assignment Reference 1 config

Verification & Execution

Execute offline validation tests or reproduce the benchmark in a CUDA GPU environment.

# Run complete offline validation test suite (zero GPU requirement) $ python validate_all.py ✓ validate_benchmark_package.py · Modular benchmark.* package contracts ✓ validate_v3_artifact.py · Source hashes, complete final panels, CSVs & manifest ✓ validate_v3_backends.py · Backend estimator routing matrix & constructors ✓ validate_gpu_process.py · Multi-GPU subprocess device masking & isolation ✓ validate_sampling.py · Proportional capping & stratified split checks ======================================================== Summary: 5/5 checks passed in 15.47s All offline checks PASSED. Ready for evaluation.

Analytical Boundaries

Explicit documentation of boundary conditions, statistical assumptions, and non-claims.

Statistical Inference

Reported ± values represent sample standard deviations across 3 random training seeds on the same test set. They do not constitute formal confidence intervals.

Cached Zero-Shot Calls

Identical successful API requests are cached and reused across seeds. Zero variance reflects deterministic prompt caching rather than independent API replications.

Banking77 API Failures

Requests yielding unparseable or out-of-vocabulary label predictions were assigned failure code -1 and counted as incorrect.

Sample Size Constraints

Iris (n=30) and Breast Cancer (n=114) feature compact holdouts. Near-perfect scores on these samples should not be generalized to larger distributions.

Classical Baseline Scope

Classical pipelines operate under a strict computational budget (max 150 trees, 60 hist iters). They do not represent fine-tuned deep transformers.

Latency & Cost Scope

The benchmark logged 38,922 request attempts with an estimated ~$4.19 input cost. This is an empirical run estimate, not an SLA or billing quote.