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
test(evals): assert the directory refusal by leaf name, not the runner's temp path
Aashish808288bf168d
diff --git a/tests/test_evidence.py b/tests/test_evidence.pyindex e6ec3aab..9d9a92fb 100644--- a/tests/test_evidence.py+++ b/tests/test_evidence.py@@ -266,7 +266,7 @@ check("missing checkpoint dir -> usage error 2", rc_bad, 2)import contextlibimport io-_dir_report = os.path.join(tmp, "a-directory")+_dir_report = os.path.join(tmp, "adir")os.makedirs(_dir_report)_not_object = os.path.join(tmp, "not-an-object.json")with open(_not_object, "w", encoding="utf-8") as f:@@ -282,9 +282,16 @@ os.makedirs(_bad_config_ckpt)with open(os.path.join(_bad_config_ckpt, "rl_agent_config.json"), "w", encoding="utf-8") as f:f.write("{nope")-for _label, _argv, _path in (+# The last tuple element is the string the refusal must contain. For four rows that is the whole+# path, because `laya/evidence.py` builds those messages from the argument the caller passed. The+# directory row is the exception: its refusal is the operating system's own `OSError` text, and the+# Windows CI runner reports that path in 8.3 form (`C:\Users\RUNNER~1\...` where the test holds+# `C:\Users\runneradmin\...`), so an exact full-path match there tests the runner's temp layout+# rather than the message. Its fixture is named `adir` -- short enough to be its own 8.3 name --+# and the leaf is what the row asserts, because naming the file the caller passed is the property.+for _label, _argv, _expect in (("--report is a directory", ["evidence", "--checkpoint", ckpt, "--report", _dir_report],- _dir_report),+ os.path.basename(_dir_report)),("--report is JSON but not an object",["evidence", "--checkpoint", ckpt, "--report", _not_object], _not_object),("--report is not JSON", ["evidence", "--checkpoint", ckpt, "--report", _not_json],@@ -299,7 +306,7 @@ for _label, _argv, _path in (_rc = evals_cli.main(_argv)check("evidence CLI: %s -> usage error 2" % _label, _rc, 2)check_true("evidence CLI: %s prints no traceback" % _label, "Traceback" not in _err.getvalue())- check_true("evidence CLI: %s names the path" % _label, _path in _err.getvalue())+ check_true("evidence CLI: %s names the path" % _label, _expect in _err.getvalue())# ... and on the artifact the research harness actually writes: identity at the document's top# level, no `config` key. Read through `inspect_checkpoint`/`format_summary`, which is the whole of