diff --git a/go/buildstep/step_test.go b/go/buildstep/step_test.go
index 8bee52c..c602d58 100644
--- a/go/buildstep/step_test.go
+++ b/go/buildstep/step_test.go
@@ -1,9 +1,7 @@
 package buildstep
 
 import (
-	"context"
 	"encoding/json"
-	"errors"
 	"sync"
 	"testing"
 	"time"
@@ -53,72 +51,6 @@ func (k *keys) waitEnd(t *testing.T) string {
 	}
 }
 
-// 1471/AC1: a run that blocks: Start answers an id within 100 ms and the key holds running.
-func TestStartAnswersAtOnceWithTheStepRunning(t *testing.T) {
-	k := newKeys()
-	release := make(chan struct{})
-	defer close(release)
-	began := time.Now()
-	id, err := Start(context.Background(), k.put, "ace", "fetch", func(ctx context.Context) (any, error) {
-		<-release
-		return nil, nil
-	})
-	if err != nil {
-		t.Fatal(err)
-	}
-	if d := time.Since(began); d > 100*time.Millisecond {
-		t.Errorf("Start took %v", d)
-	}
-	s := k.last("ace." + id)
-	// 1890/AC5: the running key names this process as its owner.
-	if s.State != delivery.StepRunning || s.Tool != "fetch" || s.Node != "ace" || s.ID != id || s.Started.IsZero() || s.Owner != Owner || Owner == "" {
-		t.Errorf("step = %+v, want owner %q", s, Owner)
-	}
-}
-
-// 1471/AC2 and 1473/AC4: a result, an error and a Refusal end the step passed, failed and refused.
-func TestStepEndsPassedFailedOrRefusedOnItsKey(t *testing.T) {
-	for _, c := range []struct {
-		name   string
-		run    func(context.Context) (any, error)
-		state  string
-		reason string
-		result string
-	}{
-		{"result", func(context.Context) (any, error) { return map[string]string{"sha": "abc"}, nil }, delivery.StepPassed, "", `{"sha":"abc"}`},
-		{"error", func(context.Context) (any, error) { return nil, errors.New("exit 2") }, delivery.StepFailed, "exit 2", ""},
-		{"refusal", func(context.Context) (any, error) {
-			return nil, &Refusal{Text: "go 1.27.1 installed, the recipe builds with go 1.27.0", Want: "1.27.0", Have: "1.27.1"}
-		}, delivery.StepRefused, "go 1.27.1 installed, the recipe builds with go 1.27.0", ""},
-	} {
-		t.Run(c.name, func(t *testing.T) {
-			k := newKeys()
-			id, err := Start(context.Background(), k.put, "ace", "test", c.run)
-			if err != nil {
-				t.Fatal(err)
-			}
-			if got := k.waitEnd(t); got != "ace."+id {
-				t.Fatalf("ended key %q", got)
-			}
-			s := k.last("ace." + id)
-			if s.State != c.state || s.Reason != c.reason || string(s.Result) != c.result || s.Ended.IsZero() || s.Owner != Owner {
-				t.Errorf("step = %+v", s)
-			}
-		})
-	}
-}
-
-// 1471/AC3: a panic in run ends the step failed, naming it, and the test process goes on.
-func TestAPanicEndsTheStepFailed(t *testing.T) {
-	k := newKeys()
-	id, _ := Start(context.Background(), k.put, "ace", "build", func(context.Context) (any, error) { panic("boom") })
-	k.waitEnd(t)
-	s := k.last("ace." + id)
-	if s.State != delivery.StepFailed || s.Reason != "the step panicked: boom" {
-		t.Errorf("step = %+v", s)
-	}
-}
-
 // 1471/AC5: ids made in one and the same millisecond differ, and each is one key token. The clock is fixed
 // while no step runs, so no goroutine reads it meanwhile.
 func TestTwoIDsInOneMillisecondDiffer(t *testing.T) {
@@ -138,18 +70,3 @@ func TestTwoIDsInOneMillisecondDiffer(t *testing.T) {
 		}
 	}
 }
-
-// A first write that fails starts nothing.
-func TestStartAnswersTheErrorWhenTheRunningWriteFails(t *testing.T) {
-	ran := make(chan struct{}, 1)
-	_, err := Start(context.Background(), func(string, any) (uint64, error) { return 0, errors.New("no grant") }, "ace", "fetch",
-		func(context.Context) (any, error) { ran <- struct{}{}; return nil, nil })
-	if err == nil {
-		t.Fatal("Start answered no error")
-	}
-	select {
-	case <-ran:
-		t.Error("run was called")
-	case <-time.After(50 * time.Millisecond):
-	}
-}