Jev vs explainable models
Explainable models (linear scores, GAMs, SHAP on GBMs) attribute features to a score. Jev does not publish per-token attributions in the public docs we used. Your explanation is the question pack plus probabilities you logged.
Unofficial. Not an XAI survey. docs.typesafe.ai.
Comparison scope
If a regulator wants coefficient tables, a GLM may still win. If they want “which rubric did you apply?”, Jev criteria are at least readable — that is not the same as faithful inner-model XAI.
Criteria that decide the architecture
| Axis | Jev (System One) | Explainable models (XAI) |
|---|---|---|
| Feature attributions | Not documented | Wins |
| Human-readable rubric | instructions + criteria |
Sometimes |
| Generated rationale | None (good — no fake chain-of-thought) | Some LLMs fake it |
| Trace | You persist it | You persist it |
Decision quality and control
Readable criteria can still be applied wrongly. Do not treat the rubric as a proof of correctness.
Integration trade-offs
Store question version, model id, full probabilities, and the action. That is your decision trace.
TypeSafe’s public models page lists jev-1.13 at $0.042 per million input tokens with output tokens free — a vendor claim, not a jev.pro measurement. Confirm on the models page before you bid.
When each approach fits
Prefer Jev when
- Explanation = which documented question fired
Prefer Explainable models (XAI) when
- You need SHAP-style feature attributions
Use Jev where rubrics suffice; use classical XAI where features must be attributed.
The question is the explanation
SHAP, LIME, and linear scores attribute features to a number. Public TypeSafe docs we used do not publish per-token attributions. Your explanation is the question pack plus probabilities you logged. Sketch: persist question version, model id jev-1.13.0, full probabilities, and the action — that is your decision trace. If a regulator wants coefficient tables, a GLM may still win. Readable criteria are not faithful inner-model XAI.
The models-page $0.042 / M input number is a vendor claim. Paying it does not buy SHAP. Do not ask Jev to explain — that would be generation. “Why option B?” Because B had the highest probability on the Choice you wrote. There is no essay, and that is a feature: no fake chain-of-thought.
How “you wrote the question” gets mistaken for SHAP
- Rubric as proof of correctness. Readable criteria can still be applied wrongly. Schema-safe is not true.
- Bolted LLM rationale. Some LLMs fake a why. Do not staple that to a Jev answer and call it certified XAI. This page claims no “explainable AI” badge.
- Feature tables expected from
probabilities. Those are over options you defined, not over customer features likeincome. - Jaggedness hidden behind a pretty criterion. Math still belongs in code. A well-worded Score does not make date arithmetic faithful.
- Trace not persisted. If you did not store the pack SHA, you cannot explain last Tuesday.
When not to use Jev here
Do not use Jev when the stakeholder needs SHAP-style feature attributions or coefficient tables. Prefer a GLM/GAM/GBM plus XAI. Use Jev when “which documented question fired” is the explanation you can defend.
If last Tuesday’s action cannot be replayed from pack SHA + jev-1.13.0 + logged probabilities, you do not have an explanation — you have a story. Fix the trace before you add SHAP theater on top.
What this page does not claim
-
No claim Jev is “explainable AI” certified.
-
Schema-safe output is not the same as factual correctness.
-
No independent bake-off numbers live here.
FAQ
Why did it pick option B? Because you asked a Choice and B had the highest probability. There is no essay.
Can I ask Jev to explain? That would be generation. Don’t.
Is a high probability a SHAP value? No. It is mass over the options you listed, not a feature attribution.
Disclaimer
This is an independent unofficial site and is not affiliated with TypeSafe AI; official documentation is available at https://docs.typesafe.ai. Never treat jev.pro as TypeSafe official documentation. We do not sell, issue, or proxy API keys.
Hub: Comparisons. Siblings: vs black box models, vs risk rules, glossary decision trace. Canonical: https://docs.typesafe.ai.