Overview
daf-jev 0.7.2 provides typed native decision clients, provider-neutral backends, pure composition, exact discrete graphical inference, independently labeled dataset preparation, frozen benchmark protocols, decimal cost accounting, and reproducible manuscript evidence. This release adds an explicitly bounded native OpenRouter execution contract for TypeSafe Jev, a separately defined Perplexity Decisions contract, and prospective interleaved native-model comparison protocols. The selected native hosted evidence contains 12 successful capability calls, including complete 77-way BANKING77 and 151-way CLINC option vocabularies, with reported billing of USD 0.000556920. These capability checks establish endpoint and task support; they do not establish dataset accuracy or calibrated probability meaning. Larger native hosted studies are ongoing and remain outside this release's completed research claims. Historical unknown billing and partial local results remain explicitly identified. Two native Mac Python profiles passed all 2,436 unit cases with combined coverage above 92%; package and manuscript reproduction were independently reviewed. The seven release assets contain the source, wheel, sdist, 30-page manuscript, selected evidence archive, release manifest, and checksums. Historical synthetic security-test canaries are retained as fixtures; real credentials, private journals, model weights, and raw corpora are excluded. Source build, artifact, and main commit identities are recorded separately in the release manifest.
Overview source: Curated paper metadata.
Methods and contributions
Read the source for the full argument, qualifications, and evidence.
Findings and contributions
- Batching questions into one call gave wall-time speedups of 3.98, 8.77, and 18.55 across the configured batch widths.
- At the widest batch width the sequential strategy used 4.22 times the tokens of the batched call, so batching was both faster and cheaper.
- End-to-end decision pipelines had median wall times of 0.133 s (composite scoring) and 0.129 s (intent routing).
- Under a self-consistency proxy, the calibration benchmark reported ECE 0.0730 and Brier 0.0252; the author notes this shows consistency with proxy labels, not correctness about the world.
- The unit test suite reached 93.08 measured coverage.
Methods
- daf-jev layered Python client for the System One endpoint — Implements typed builders (noul, choice, score), sync/async clients with a pure retry policy, and a typed exception hierarchy over one HTTP endpoint.
- Pure-function decision patterns and a concurrent batch Evaluator — composite_score, confidence_gate, and route/pick turn typed answers into decisions locally; the Evaluator runs a fixed question set over many states without aborting on failures.
- Live-API batching benchmark at batch widths 5, 10, and 20 — Compares one batched call against sequential single-question calls on the jev-latest model, recording wall time and token totals.
- Self-consistency calibration benchmark (6 states × 5 repeats) — Scores choice answers against the modal choice across repeats and noul answers by pairwise stability, computing ECE and Brier score without external ground truth.
- No-mock test suite and hash-manifested documentation snapshot — Unit tests drive the real transport against a local HTTP server; model claims are grounded in a 108-page hashed snapshot of the vendor documentation.
Key concepts
- paired publication
Summary sources: Paper metadata and evidence · Learning guide · Extracted source text.
Document downloads
PDFs are hosted here; supporting sources may use pinned archives.
- daf-jev-0.7.2-manuscript.pdf PDF · 1.75 MiB
- daf-jev_combined.pdf PDF · 0.71 MiB · extracted-text source
These files may be different versions or companion documents. The archive does not identify a latest edition.
Citation
Citation metadata follows the unified bibliography.
Catalog details and resources
Related works
Ranked by recorded methods, topics, contributors and associated software, with contributor roles and domain-only matches labelled. Metadata overlap is not a citation or a scientific equivalence.