Non-autoregressive System 1 decision engine. Typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request.
Python calibrationclassificationdecision-modelhuggingfacejevmodernbertmultilingualnlppythonpytorchroutingtyped-decisionszero-shot
docs(http-api): strict JEV answers carry `type`, the pages did not say so
Aashish5dc7b742b2f8
diff --git a/docs/http-api.md b/docs/http-api.mdindex 31ce234b..bc6b580a 100644--- a/docs/http-api.md+++ b/docs/http-api.md@@ -258,9 +258,9 @@ rejects the full payload, so `LAYA_JEV_STRICT=1` projects the response onto theanswering, on both `/v1/systemone` and `/v1/systemone/batch`:- the root keeps `model`, `answers` and `usage` only; `routing` is not sent;-- a `choice` answer keeps `choice`, `probabilities` and `confidence`;-- a `score` answer keeps `score`, `probabilities`, `confidence` and `legend`;-- a `noul` answer keeps `noul` only;+- a `choice` answer keeps `type`, `choice`, `probabilities` and `confidence`;+- a `score` answer keeps `type`, `score`, `probabilities`, `confidence` and `legend`;+- a `noul` answer keeps `type` and `noul`;- `usage` keeps `input_tokens` and `output_tokens`; the truncation facts and the collapsed-options ceiling are not sent.diff --git a/laya/serve.py b/laya/serve.pyindex f5a432c5..cde2ee09 100644--- a/laya/serve.py+++ b/laya/serve.py@@ -163,10 +163,10 @@ def _project_jev_strict(result: Dict[str, Any]) -> Dict[str, Any]:"""Project a result onto the strict Jev wire contract (`LAYA_JEV_STRICT`).The Jev `/v1/systemone` response defines exactly three top-level fields- (`model`, `answers`, `usage`), and each answer carries only its type's fields:- choice = `choice` + `probabilities` + `confidence`, score = `score` +- `probabilities` + `confidence` + `legend`, noul = `noul` only, and `usage` the- two token counts. Laya's full payload adds more: a root `routing` report, a+ (`model`, `answers`, `usage`), and each answer carries its discriminator plus its type's+ fields: choice = `type` + `choice` + `probabilities` + `confidence`, score = `type` ++ `score` + `probabilities` + `confidence` + `legend`, noul = `type` + `noul`, and `usage`+ the two token counts. Laya's full payload adds more: a root `routing` report, aper-answer `action` head plus the calibrated `answer_confidence`, a`confidence` on noul answers, and a usage report extended with the truncationfacts and the collapsed-options ceiling. Those additions are what a clientdiff --git a/tests/test_serve.py b/tests/test_serve.pyindex 016703cd..f50bedc0 100644--- a/tests/test_serve.py+++ b/tests/test_serve.py@@ -3288,3 +3288,63 @@ def test_idle_unload_retries_failure_and_stops_at_shutdown(monkeypatch):router.loaded = ["english"]router.unloaded.clear()assert not router.unloaded.wait(0.15)+++def test_jev_strict_projection_page_names_the_projected_keys():+ """docs/http-api.md's four strict bullets must be the key sets `_project_jev_strict` returns.++ The page is the second copy of the contract, after the handler's docstring. Both used to+ claim ``noul = `noul` only``, but the projection emits ``{"type": "noul", "noul": ...}``,+ and the pre-existing ``test_jev_strict_projects_the_full_payload`` asserts that two-key+ shape verbatim -- so the prose and the pinned test disagreed, and a client that validated+ the "no fields other than `noul`" sentence rejected every strict noul answer the flag+ exists to sanitize.++ The gate calls the real projection on a fixture that carries all three answer types plus+ the additions a strict client must not see, then reads the four bullet lines from+ `docs/http-api.md` and asserts each bullet's backticked key set equals the corresponding+ projected key set. The pre-fix wording is banned as a direct substring so the gate fails+ if the page reverts.+ """+ import re+ from pathlib import Path++ from laya.serve import _project_jev_strict++ fixture = {+ "model": "laya-rl-agent",+ "answers": {+ "queue": {"type": "choice", "choice": "billing",+ "probabilities": {"billing": 0.95, "tech": 0.05},+ "confidence": 0.80, "action": "answer",+ "answer_confidence": 0.75},+ "urgency": {"type": "score", "score": 1.70, "confidence": 0.19,+ "probabilities": {"0": 0.02, "1": 0.65, "2": 0.33},+ "legend": {"0": "calm", "1": "firm", "2": "angry"},+ "action": "answer"},+ "threat": {"type": "noul", "noul": 0.9148, "confidence": 0.85,+ "answer_confidence": 0.70, "action": "answer"},+ },+ "usage": {"input_tokens": 83, "output_tokens": 0,+ "windows": 1, "collapsed_options": 0},+ "routing": {"model": "typed-decisions", "reason": "typed workflow"},+ }+ projected = _project_jev_strict(fixture)++ page = (Path(__file__).resolve().parents[1] / "docs" / "http-api.md").read_text(encoding="utf-8")++ def bullet_keys(label):+ for line in page.splitlines():+ if line.startswith("- a `%s` answer keeps" % label):+ back = line.split("keeps", 1)[1]+ return set(re.findall(r"`([A-Za-z_][A-Za-z_0-9]*)`", back))+ raise AssertionError("bullet for %r not found" % label)++ assert set(projected["answers"]["queue"]) == bullet_keys("choice"), \+ "doc says %r, handler emits %r" % (sorted(bullet_keys("choice")),+ sorted(projected["answers"]["queue"]))+ assert set(projected["answers"]["urgency"]) == bullet_keys("score")+ assert set(projected["answers"]["threat"]) == bullet_keys("noul")++ # The noul bullet must not shrink the discriminator away.+ assert "noul only" not in page, "pre-fix 'noul only' wording is still on the page"