Why should I reach for Discern instead of using Effect Decision directly?

Direct Decision code works for plain boolean branching, but it rounds uncertain evidence down to false. Discern adds ordered rules, miss bands, and an explicit onUncertain path so a mid-range score routes to human review instead.

Does policy batching actually reduce the number of model calls?

Yes. Discern collects the distinct decisions a policy needs and answers them in one DecisionModel call no matter how many cases reference them, and the headline example is compiled so it cannot drift from the library.

What happens to a borderline regression-risk score like 0.62?

It never reaches ship. A 0.62 score sits between the 0.8 threshold and the 0.5 miss band, so it is not a miss and the program takes the onUncertain branch to human review rather than rounding down.

Does the procedure router send my full diff just to pick a handler?

No. The registry routes on a projection such as the request question string, so the diff never reaches the routing prompt. That keeps token use down and the routing answer cacheable per question.

Run npm install @doeixd/discern effect@rc; effect is a peer dependency (>=4.0.0-rc.116 <5), and effect@latest is still 3.x, so v4 must be requested by tag or the library will not work.

Can I try it without spending on model calls?

Yes. A runnable tour in examples/walkthrough.mjs (npm run example) uses a stub provider, so it is deterministic and costs nothing while exercising uncertainty, recording, replay, budgets, and routing end to end.

Is it stable enough to build on?

Not yet. It is an early pre-1.0 build targeting Effect v4 release candidates and effect/unstable/ai, so both the underlying Effect APIs and Discern's own surface can change between minor versions.

CommunityJev Integrations & Adapters

Discern

Q: Can I try it without spending on model calls?

Yes. A runnable tour in examples/walkthrough.mjs (npm run example) uses a stub provider, so it is deterministic and costs nothing while exercising uncertainty, recording, replay, budgets, and routing end to end.

Q: Is it stable enough to build on?

Not yet. It is an early pre-1.0 build targeting Effect v4 release candidates and effect/unstable/ai, so both the underlying Effect APIs and Discern's own surface can change between minor versions.

Uncertainty-aware semantic control flow and routable procedures for Effect, with type-safe patterns and policies over Effect Decision and DecisionModel where uncertain answers stay explicit branches.

See highlights

Open github.com

Stars: 15
Language: TypeScript
License: MIT
Published by: Community
Author: doeixd
Last verified: Oct 2026

doeixd/discern

User repository on GitHub

View on GitHub

Craft Type-Safe Uncertainty-aware semantic pattern matching, control flow, and smart procedures for Effect DecisionModel and Jev

Forks: 3
Last pushed: Updated Sep 2026

Repo stats from the GitHub API, cached Oct 2026.Homepage

About this project

Uncertainty-aware, smart semantic control flow and procedures for Effect. Type-safe patterns, policies and routable procedures over Effect's Decision and DecisionModel — where maybe is a branch you handle, not a rounding error. Provider-neutral: runs on TypeSafe's Jev, or any other DecisionModel. Watch the full tutorial — 2:44, silent, with on-screen explanations. Follow one code change through Jev → Effect Decision / DecisionModel → Discern → Procedures → nested registries.

Quoted from the project's README. Read the full README.

Tagscommunity jevbest catalog-import awesomejev

Works forJev for Classification

Used inJev for Software & DevTools

Languages: TypeScript

Highlights

Ordered policy rules over classification and probability evidence, with miss bands and onUncertain branches.
Distinct decisions batched into one DecisionModel call however many policy cases reference them.
Named procedures plus registries that route on intent alone, with uncertain routes reported, not guessed.
Provider-neutral design running on Jev or any DecisionModel, with a stub-provider walkthrough that costs nothing.

Watch out

Early and unstable pre-1.0 build targeting Effect v4 release candidates, so both the underlying Effect APIs and Discern's own surface can shift between minors. Copy reviewed against the repository README on 2026-10-04, not from running it.

Questions about this entry

Short answers drawn from the linked project or page.

More Jev resources like this

system-one

Provider-neutral TypeScript runtime for System One decisions: write choice, noul, and score code once, then run it on TypeSafe Jev, local Reflex, any compatible /v1/systemone endpoint, or a custom provider.