doormile_backend/internal/ai/outcomes/rules_test.go

package outcomes

import (
	"testing"
	"time"
)

func at(h int) time.Time {
	return time.Date(2026, 10, 5, h, 0, 0, 0, time.UTC)
}

func tp(t time.Time) *time.Time { return &t }

const window = 48 * time.Hour

// ─── Too soon to judge ─────────────────────────────────────────────────────

func TestYoungDecisionIsLeftPending(t *testing.T) {
	decided := at(10)
	for _, dt := range []string{"stall_response", "assignment_failure", "something_new"} {
		_, ok := judge(dt, bookingFacts{DecidedAt: decided}, decided.Add(5*time.Minute), window)
		if ok {
			t.Errorf("%s: judged a 5-minute-old decision; the booking is still in flight", dt)
		}
	}
}

func TestPendingWhileInsideTheWindow(t *testing.T) {
	decided := at(10)
	// An hour in: no delivery yet, but the window has not closed. Recording
	// failure here would blame the decision for a parcel still on its way.
	if _, ok := judge("stall_response", bookingFacts{DecidedAt: decided}, decided.Add(time.Hour), window); ok {
		t.Error("stall_response judged before its window closed")
	}
	if _, ok := judge("assignment_failure", bookingFacts{DecidedAt: decided}, decided.Add(time.Hour), window); ok {
		t.Error("assignment_failure judged before its window closed")
	}
}

// ─── stall_response ────────────────────────────────────────────────────────

func TestStallDeliveredOnTimeIsSuccess(t *testing.T) {
	decided := at(10)
	f := bookingFacts{
		DecidedAt:   decided,
		DeliveredAt: tp(decided.Add(3 * time.Hour)),
		SLADueAt:    tp(decided.Add(12 * time.Hour)),
	}
	got, ok := judge("stall_response", f, decided.Add(4*time.Hour), window)
	if !ok || got != OutcomeSuccess {
		t.Errorf("got %q ok=%v, want success", got, ok)
	}
}

// The case that matters most: counting any eventual delivery as success would
// teach the model that the action is always vindicated, which is exactly the
// wrong lesson.
func TestStallDeliveredLateIsFailure(t *testing.T) {
	decided := at(10)
	f := bookingFacts{
		DecidedAt:   decided,
		DeliveredAt: tp(decided.Add(20 * time.Hour)),
		SLADueAt:    tp(decided.Add(12 * time.Hour)),
	}
	got, ok := judge("stall_response", f, decided.Add(21*time.Hour), window)
	if !ok || got != OutcomeFailure {
		t.Errorf("got %q ok=%v, want failure — delivered 8h past SLA", got, ok)
	}
}

func TestStallDeliveredWithNoSLAIsSuccess(t *testing.T) {
	decided := at(10)
	// No sladueat recorded (an older row). Delivery is the best evidence
	// available and there is nothing to call it late against.
	f := bookingFacts{DecidedAt: decided, DeliveredAt: tp(decided.Add(5 * time.Hour))}
	got, ok := judge("stall_response", f, decided.Add(6*time.Hour), window)
	if !ok || got != OutcomeSuccess {
		t.Errorf("got %q ok=%v, want success", got, ok)
	}
}

func TestStallNeverDeliveredAfterWindowIsFailure(t *testing.T) {
	decided := at(10)
	got, ok := judge("stall_response", bookingFacts{DecidedAt: decided}, decided.Add(window+time.Hour), window)
	if !ok || got != OutcomeFailure {
		t.Errorf("got %q ok=%v, want failure", got, ok)
	}
}

func TestStallCancelledIsFailureImmediately(t *testing.T) {
	decided := at(10)
	f := bookingFacts{DecidedAt: decided, Cancelled: true}
	// Cancellation is conclusive — no need to wait out the window.
	got, ok := judge("stall_response", f, decided.Add(time.Hour), window)
	if !ok || got != OutcomeFailure {
		t.Errorf("got %q ok=%v, want failure on cancellation", got, ok)
	}
}

// ─── assignment_failure ────────────────────────────────────────────────────

func TestAssignmentLaterAssignedIsSuccess(t *testing.T) {
	decided := at(10)
	f := bookingFacts{
		DecidedAt:  decided,
		Assigned:   true,
		AssignedAt: tp(decided.Add(2 * time.Hour)),
	}
	got, ok := judge("assignment_failure", f, decided.Add(3*time.Hour), window)
	if !ok || got != OutcomeSuccess {
		t.Errorf("got %q ok=%v, want success", got, ok)
	}
}

// A rider assigned BEFORE the decision is not evidence the decision worked —
// it is the assignment that was already there. Without the ordering check,
// every reassignment decision would read as an instant success.
func TestAssignmentPredatingTheDecisionIsNotSuccess(t *testing.T) {
	decided := at(10)
	f := bookingFacts{
		DecidedAt:  decided,
		Assigned:   true,
		AssignedAt: tp(decided.Add(-2 * time.Hour)),
	}
	got, ok := judge("assignment_failure", f, decided.Add(time.Hour), window)
	if ok && got == OutcomeSuccess {
		t.Error("counted a pre-existing assignment as the decision's success")
	}
}

func TestAssignmentNeverAssignedAfterWindowIsFailure(t *testing.T) {
	decided := at(10)
	got, ok := judge("assignment_failure", bookingFacts{DecidedAt: decided}, decided.Add(window+time.Hour), window)
	if !ok || got != OutcomeFailure {
		t.Errorf("got %q ok=%v, want failure", got, ok)
	}
}

func TestAssignmentCancelledIsFailure(t *testing.T) {
	decided := at(10)
	f := bookingFacts{DecidedAt: decided, Cancelled: true}
	got, ok := judge("assignment_failure", f, decided.Add(time.Hour), window)
	if !ok || got != OutcomeFailure {
		t.Errorf("got %q ok=%v, want failure", got, ok)
	}
}

// ─── Unknown types are never guessed ───────────────────────────────────────

func TestUnknownTypeIsNeverSuccessOrFailure(t *testing.T) {
	decided := at(10)
	// Inside the window: pending.
	if _, ok := judge("a_new_decision_type", bookingFacts{DecidedAt: decided}, decided.Add(time.Hour), window); ok {
		t.Error("judged an unknown decision type inside its window")
	}
	// Past it: unknown, so it stops being rescanned — but never a label that
	// would teach the model something nobody defined.
	got, ok := judge("a_new_decision_type", bookingFacts{DecidedAt: decided}, decided.Add(window+time.Hour), window)
	if !ok || got != OutcomeUnknown {
		t.Errorf("got %q ok=%v, want unknown", got, ok)
	}
}

// A delivered unknown type must still not be called a success: the rule for
// what success means for that type does not exist yet.
func TestUnknownTypeWithDeliveryIsStillUnknown(t *testing.T) {
	decided := at(10)
	f := bookingFacts{DecidedAt: decided, DeliveredAt: tp(decided.Add(time.Hour))}
	got, ok := judge("a_new_decision_type", f, decided.Add(window+time.Hour), window)
	if !ok || got != OutcomeUnknown {
		t.Errorf("got %q ok=%v, want unknown", got, ok)
	}
}

// ─── Configuration ─────────────────────────────────────────────────────────

func TestOutcomeWindowDefaultAndOverride(t *testing.T) {
	t.Setenv("AGENT_OUTCOME_WINDOW_HOURS", "")
	if got := OutcomeWindow(); got != defaultOutcomeWindowHours*time.Hour {
		t.Errorf("default window = %v, want %v", got, defaultOutcomeWindowHours*time.Hour)
	}
	t.Setenv("AGENT_OUTCOME_WINDOW_HOURS", "12")
	if got := OutcomeWindow(); got != 12*time.Hour {
		t.Errorf("window = %v, want 12h", got)
	}
	// Garbage falls back rather than producing a zero window, which would
	// judge every decision the instant it passed minDecisionAge.
	t.Setenv("AGENT_OUTCOME_WINDOW_HOURS", "not-a-number")
	if got := OutcomeWindow(); got != defaultOutcomeWindowHours*time.Hour {
		t.Errorf("window on garbage = %v, want the default", got)
	}
	t.Setenv("AGENT_OUTCOME_WINDOW_HOURS", "0")
	if got := OutcomeWindow(); got != defaultOutcomeWindowHours*time.Hour {
		t.Errorf("window on 0 = %v, want the default", got)
	}
}

func TestRetentionDaysDefaultAndOverride(t *testing.T) {
	t.Setenv("AGENT_DECISION_RETENTION_DAYS", "")
	if got := RetentionDays(); got != 180 {
		t.Errorf("default retention = %d, want 180", got)
	}
	t.Setenv("AGENT_DECISION_RETENTION_DAYS", "90")
	if got := RetentionDays(); got != 90 {
		t.Errorf("retention = %d, want 90", got)
	}
	// Retention must be longer than aiagentruns' 30 days, because old
	// precedent is the point of this table. Not enforced in code — asserted
	// here so a future change to the default has to confront it.
	t.Setenv("AGENT_DECISION_RETENTION_DAYS", "")
	if RetentionDays() <= 30 {
		t.Error("decision retention is no longer than telemetry retention; precedent will be purged before it is useful")
	}
}