(#7) Probe Jev score/noul; escape option on mode #14
+89
@@ -5,6 +5,7 @@ import (
|
||||
"encoding/json"
|
||||
"net/http"
|
||||
"net/http/httptest"
|
||||
"os"
|
||||
"strings"
|
||||
"sync/atomic"
|
||||
"testing"
|
||||
@@ -90,3 +91,91 @@ func TestNoJevKeyMeansNoClassifier(t *testing.T) {
|
||||
t.Fatal("expected no classifier without a key")
|
||||
}
|
||||
}
|
||||
|
||||
var probeQuestions = map[string]any{
|
||||
"kind": map[string]any{
|
||||
"type": "choice",
|
||||
"instructions": "What kind of message is this?",
|
||||
"criteria": map[string]string{"ask": "Asks for work", "say": "Ordinary talk", "none": "Neither"},
|
||||
},
|
||||
"severity": map[string]any{
|
||||
"type": "score",
|
||||
"instructions": "How severe is the problem in this message?",
|
||||
"criteria": []string{"No problem", "Annoying", "Blocking"},
|
||||
},
|
||||
"is_request": map[string]any{
|
||||
"type": "noul",
|
||||
"instructions": "Is the sender asking someone to do something?",
|
||||
},
|
||||
}
|
||||
|
||||
func TestParseMixedAnswers(t *testing.T) {
|
||||
srv := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
||||
json.NewEncoder(w).Encode(map[string]any{
|
||||
"model": "jev-1.13.0",
|
||||
"answers": map[string]any{
|
||||
"kind": map[string]any{"type": "choice", "choice": "ask", "confidence": 0.82},
|
||||
"severity": map[string]any{"type": "score", "score": 1.4, "confidence": 0.6, "legend": map[string]string{"0": "No problem", "1": "Annoying", "2": "Blocking"}, "probabilities": map[string]float64{"0": 0.1, "1": 0.5, "2": 0.4}},
|
||||
"is_request": map[string]any{"type": "noul", "noul": 0.95},
|
||||
},
|
||||
"usage": map[string]int{"input_tokens": 100, "output_tokens": 20},
|
||||
})
|
||||
}))
|
||||
t.Cleanup(srv.Close)
|
||||
j := &JevClient{url: srv.URL, key: "tk", model: "jev-latest", client: srv.Client()}
|
||||
a, tokens, model, err := j.SystemOne(context.Background(), "state text", probeQuestions)
|
||||
if err != nil {
|
||||
t.Fatal(err)
|
||||
}
|
||||
if tokens != 120 || model != "jev-1.13.0" {
|
||||
t.Errorf("got tokens=%d model=%s", tokens, model)
|
||||
}
|
||||
if a["kind"].Choice != "ask" || a["kind"].Confidence == nil || *a["kind"].Confidence != 0.82 {
|
||||
t.Errorf("choice answer wrong: %+v", a["kind"])
|
||||
}
|
||||
if a["severity"].Score == nil || *a["severity"].Score != 1.4 || a["severity"].Legend["2"] != "Blocking" {
|
||||
t.Errorf("score answer wrong: %+v", a["severity"])
|
||||
}
|
||||
if a["is_request"].Noul == nil || *a["is_request"].Noul != 0.95 {
|
||||
t.Errorf("noul answer wrong: %+v", a["is_request"])
|
||||
}
|
||||
if a["is_request"].Confidence != nil {
|
||||
t.Errorf("noul must have no confidence, got %v", *a["is_request"].Confidence)
|
||||
}
|
||||
}
|
||||
|
||||
func TestModeChoiceHasEscapeOption(t *testing.T) {
|
||||
mode := jevQuestions["mode"].(map[string]any)["criteria"].(map[string]string)
|
||||
if _, ok := mode["other"]; !ok {
|
||||
t.Fatalf("mode has no escape option: %v", mode)
|
||||
}
|
||||
}
|
||||
|
||||
func TestClassifySurvivesOtherMode(t *testing.T) {
|
||||
tr, _ := triageWith(t, fakeJev(t, 200, "other", 0.95, "analysis", 0.9))
|
||||
if tr.Mode != "solo" {
|
||||
t.Fatalf("Jev's other mode must normalize to solo, got %q", tr.Mode)
|
||||
}
|
||||
}
|
||||
|
||||
func TestProbeScoreAndNoulAgainstLiveAPI(t *testing.T) {
|
||||
if os.Getenv("TYPESAFE_API_KEY") == "" {
|
||||
t.Skip("no TYPESAFE_API_KEY")
|
||||
}
|
||||
j := NewJev()
|
||||
a, tokens, model, err := j.SystemOne(context.Background(), "@Signals, classify these drops. The export crashes on Safari.", probeQuestions)
|
||||
if err != nil {
|
||||
t.Fatalf("live probe failed: %v", err)
|
||||
}
|
||||
dump, _ := json.MarshalIndent(a, "", " ")
|
||||
t.Logf("model=%s tokens=%d\n%s", model, tokens, dump)
|
||||
if a["kind"].Choice == "" {
|
||||
t.Errorf("choice answer missing: %+v", a["kind"])
|
||||
}
|
||||
if a["severity"].Score == nil {
|
||||
t.Errorf("score answer missing: %+v", a["severity"])
|
||||
}
|
||||
if a["is_request"].Noul == nil {
|
||||
t.Errorf("noul answer missing: %+v", a["is_request"])
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user