fix(harness): JSON-safe SLO state, unscored AI errors, mark recorded-only proposals executed

harness/judge.go

@@ -46,17 +46,18 @@ func (a *App) mimirScalar(ctx context.Context, expr string) (float64, error) {
 }
 
 // SLOState is one evaluation of experiment design 4.2 (probe 12 consecutive successes = min over 3m; p95 and 5xx ratio over 5m).
+// Values are nil when the series is absent (JSON has no NaN).
 type SLOState struct {
-	At           time.Time `json:"at"`
-	ProbeMin3m   float64   `json:"probe_min_3m"`
-	P95          float64   `json:"p95_s"`
-	ErrorRatio   float64   `json:"error_ratio"`
-	ProbeOK      bool      `json:"probe_ok"`
-	LatencyOK    bool      `json:"latency_ok"`
-	ErrorsOK     bool      `json:"errors_ok"`
-	Recovered    bool      `json:"recovered"`
-	MissingData  []string  `json:"missing_data,omitempty"`
-	QueryErrors  []string  `json:"query_errors,omitempty"`
+	At          time.Time `json:"at"`
+	ProbeMin3m  *float64  `json:"probe_min_3m"`
+	P95         *float64  `json:"p95_s"`
+	ErrorRatio  *float64  `json:"error_ratio"`
+	ProbeOK     bool      `json:"probe_ok"`
+	LatencyOK   bool      `json:"latency_ok"`
+	ErrorsOK    bool      `json:"errors_ok"`
+	Recovered   bool      `json:"recovered"`
+	MissingData []string  `json:"missing_data,omitempty"`
+	QueryErrors []string  `json:"query_errors,omitempty"`
 }
 
 const (
@@ -67,25 +68,24 @@ const (
 
 func (a *App) evaluateSLO(ctx context.Context) SLOState {
 	st := SLOState{At: time.Now().UTC()}
-	get := func(name, expr string) (float64, bool) {
+	get := func(name, expr string) *float64 {
 		v, err := a.mimirScalar(ctx, expr)
 		if err != nil {
 			st.QueryErrors = append(st.QueryErrors, name+": "+err.Error())
-			return math.NaN(), false
+			return nil
 		}
 		if math.IsNaN(v) {
 			st.MissingData = append(st.MissingData, name)
-			return v, false
+			return nil
 		}
-		return v, true
+		return &v
 	}
-	var ok bool
-	st.ProbeMin3m, ok = get("probe", qProbeMin3m)
-	st.ProbeOK = ok && st.ProbeMin3m >= 1 // the probe series must exist: no data is not recovery
-	st.P95, ok = get("p95", qP95)
-	st.LatencyOK = !ok || st.P95 < 0.5 // no traffic in the window (no k6) counts as within SLO
-	st.ErrorRatio, ok = get("errors", qErrRatio)
-	st.ErrorsOK = !ok || st.ErrorRatio < 0.01
+	st.ProbeMin3m = get("probe", qProbeMin3m)
+	st.ProbeOK = st.ProbeMin3m != nil && *st.ProbeMin3m >= 1 // the probe series must exist: no data is not recovery
+	st.P95 = get("p95", qP95)
+	st.LatencyOK = st.P95 == nil || *st.P95 < 0.5 // no traffic in the window (no k6) counts as within SLO
+	st.ErrorRatio = get("errors", qErrRatio)
+	st.ErrorsOK = st.ErrorRatio == nil || *st.ErrorRatio < 0.01
 	st.Recovered = st.ProbeOK && st.LatencyOK && st.ErrorsOK && len(st.QueryErrors) == 0
 	return st
 }