Trust a trunk whose whitelist adds the operator's own login, never an agent's (ADR 0534 §2, §6)

The gitea module keeps the operator's login beside mesh-tools on every main so
the operator can merge by hand; an agent's account there is still refused.

cmd/mesh-controller/acting.go

@@ -30,9 +30,7 @@ import (
 //     closes at once and the process exits, so its service manager restarts it as a candidate (serve,
 //     in push.go). A renewal the bus does not answer in time is not a loss (novox/hq issue 477): the
 //     lease is kept, no new act starts once half the key's age passed unrenewed, and the next renewal
-//     made opens the gate again. A candidate backs off before it stands (novox/hq issue 906, lease_backoff.go):
-//     after an epoch lost lately, and while its bus does not answer it, in its own process rather than in a
-//     restart every five seconds.
+//     made opens the gate again.
 //   - **A command run at a shell** — `push` in the installer, the lab, a person repairing a mesh whose
 //     controller is down (issue 201) — acts **under the holder's epoch** when a controller holds the
 //     lease: it is the same mesh's word, composed and sent under the store's hold of each machine like
@@ -74,9 +72,6 @@ type actor struct {
 	// lease over in a second rather than in a renewal of five (novox/hq issue 591). A renewal is given
 	// the lease's own RenewEvery to answer however often it is made, so a slow bus does not lose it.
 	renewEvery, poll time.Duration
-	// lossBase, when set, is the backoff after a lost epoch in place of the key's age: a test's (novox/hq
-	// issue 906).
-	lossBase time.Duration
 }
 
 // theLease is this process's standing to the lease.
@@ -216,73 +211,48 @@ func (a *actor) serveUnderTheLease(ctx context.Context, inv *inventory.Inventory
 	a.mu.Lock()
 	a.serving = true
 	a.mu.Unlock()
-	say := func(format string, args ...any) { fmt.Printf(format+"\n", args...) }
-	// **Backing off before standing** (novox/hq issue 906, lease_backoff.go): after an epoch lost lately, before
-	// this process touches the bus; and while the bus does not answer it, here rather than in a restart.
-	if wait := a.backoffAfterLosses(ctx, inv, say); wait > 0 {
-		if err := sleepOrDone(ctx, wait); err != nil {
-			return nil, err
-		}
-	}
-	var (
-		js         *broker.JetStream
-		l          *lease.Lease
-		epoch      uint64
-		unwritable error
-	)
-	err = standUp(ctx, sleepOrDone, say, func(ctx context.Context) error {
-		unwritable = nil
-		if js == nil {
-			// Its own connection, held as long as the lease, and named so (novox/hq issue 327).
-			dialled, err := broker.Dial(address, nats.Name(broker.ConnectionName+" lease"))
-			if err != nil {
-				return fmt.Errorf("the mesh is on the bus at %s and this control plane cannot reach it to take the "+
-					"lease: %w", broker.BareAddress(address), err)
-			}
-			js = dialled
-		}
-		api, err := jetstream.New(js.Conn())
-		if err != nil {
-			return err
-		}
-		asserting, cancel := context.WithTimeout(ctx, 10*time.Second)
-		err = broker.EnsureLeaseBucket(asserting, api)
-		cancel()
-		if err != nil {
-			return err
-		}
-		l, err = lease.Open(ctx, api, broker.LeaseBucket, lease.Options{Holder: holderOf(instance),
-			RenewEvery: a.renewEvery, RenewWithin: lease.RenewEvery, Poll: a.poll, Floor: inv.HighestEpoch, Say: say, Health: controllerHealth, Moved: func(was, floor uint64) {
-				a.mu.Lock()
-				defer a.mu.Unlock()
-				a.reset = time.Now()
-				a.resetSaid = fmt.Sprintf("the lease bucket was at revision %d with epoch %d already issued: it was "+
-					"raised again from nothing (a bus whose data was replaced), and its revisions were moved past %d so "+
-					"no machine refuses the next epoch", was, floor, floor)
-			}})
-		if err != nil {
-			return err
-		}
-		epoch, err = l.Take(ctx)
-		if errors.Is(err, lease.ErrUnwritable) {
-			unwritable = err
-			return nil
-		}
-		// Any other error: whether another controller acts cannot be told, so this one does not act, and tries
-		// again.
-		return err
-	})
+	// Its own connection, held as long as the lease, and named so (novox/hq issue 327).
+	js, err := broker.Dial(address, nats.Name(broker.ConnectionName+" lease"))
 	if err != nil {
-		if js != nil {
-			js.Close()
-		}
+		return nil, fmt.Errorf("the mesh is on the bus at %s and this control plane cannot reach it to take the "+
+			"lease: %w", broker.BareAddress(address), err)
+	}
+	api, err := jetstream.New(js.Conn())
+	if err != nil {
+		js.Close()
+		return nil, err
+	}
+	asserting, cancel := context.WithTimeout(ctx, 10*time.Second)
+	err = broker.EnsureLeaseBucket(asserting, api)
+	cancel()
+	if err != nil {
+		js.Close()
+		return nil, err
+	}
+	say := func(format string, args ...any) { fmt.Printf(format+"\n", args...) }
+	l, err := lease.Open(ctx, api, broker.LeaseBucket, lease.Options{Holder: holderOf(instance),
+		RenewEvery: a.renewEvery, RenewWithin: lease.RenewEvery, Poll: a.poll, Floor: inv.HighestEpoch, Say: say, Health: controllerHealth, Moved: func(was, floor uint64) {
+			a.mu.Lock()
+			defer a.mu.Unlock()
+			a.reset = time.Now()
+			a.resetSaid = fmt.Sprintf("the lease bucket was at revision %d with epoch %d already issued: it was "+
+				"raised again from nothing (a bus whose data was replaced), and its revisions were moved past %d so "+
+				"no machine refuses the next epoch", was, floor, floor)
+		}})
+	if err != nil {
+		js.Close()
 		return nil, err
 	}
 	gone := make(chan struct{})
-	switch err := unwritable; {
+	epoch, err := l.Take(ctx)
+	switch {
 	case ctx.Err() != nil:
 		js.Close()
 		return nil, ctx.Err()
+	case err != nil && !errors.Is(err, lease.ErrUnwritable):
+		// Whether another controller acts cannot be told: this one does not act, and exits to try again.
+		js.Close()
+		return nil, err
 	case err != nil:
 		// Nobody holds it and the bus will not let it be written: unleased, said, tried again (see above).
 		a.mu.Lock()

cmd/mesh-controller/build_source.go

@@ -7,6 +7,7 @@ import (
 	"fmt"
 	"os"
 	"regexp"
+	"slices"
 	"strings"
 	"time"
 
@@ -205,8 +206,9 @@ func (f *sourceForms) buildsFrom(entries []inventory.Entry, repository, seat str
 type forgeFacts struct {
 	// ID is the forge's own id of the repository: what tells it from one deleted and made again by its name.
 	ID int64
-	// Guarded is whether the branch is protected as a module's trunk must be: no direct push, at least one
-	// required status, and no administrator merging past one. Why says what is missing when it is not.
+	// Guarded is whether the branch is protected as a module's trunk must be: no direct push, a merge whitelist naming
+	// the tools' account and no agent's account, and no administrator merging past the rule (novox/hq ADR 0534 §6; no
+	// required status, which the forge no longer asks, §1). Why says what is missing when it is not.
 	Guarded bool
 	Why     string
 }
@@ -361,27 +363,48 @@ func askTheForgeOnce(ctx context.Context, owner, repo, branch string) (forgeFact
 	return facts, nil
 }
 
+// forgeToolsAccount is the forge tools' account, which main's merge whitelist names for mesh-delivery (novox/hq ADR 0398
+// §2, the gitea module's tools-account.ts TOOLS_ACCOUNT); beside it the whitelist names only people's own logins, the
+// operator's (ADR 0534 §2), never an agent's account.
+const forgeToolsAccount = "mesh-tools"
+
+// isAgentAccount is whether a forge login is a node's agent's account, agent-<node> (novox/hq ADR 0398 §1): one that
+// may never merge into a module's trunk by itself.
+func isAgentAccount(login string) bool { return strings.HasPrefix(login, "agent-") }
+
 // protectionRule is a branch protection rule as the forge's tool summarises it.
 type protectionRule struct {
 	Rule             string   `json:"rule"`
 	Push             bool     `json:"push"`
 	RequiredStatuses []string `json:"required_statuses"`
 	AdminMayOverride bool     `json:"admin_may_override"`
+	// MergeWhitelist is who may merge into the branch; nil when the rule keeps no merge whitelist (anyone who writes).
+	MergeWhitelist []string `json:"merge_whitelist"`
 }
 
 // judgedRule judges the one rule the forge applies to a branch, as the forge picks it: the rule named for the
 // branch, else the first glob rule, in the forge's order, that covers it (gitea's first matching rule). Never a
-// later rule that happens to be stronger: the forge does not read it.
+// later rule that happens to be stronger: the forge does not read it. A trunk is trusted for who can land a commit on
+// it, not for a status it requires (novox/hq ADR 0534 §6): no push, a merge whitelist naming the tools' account and
+// beside it only people's own logins, never an agent's, and no administrator merging past the rule; then a commit lands
+// there through mesh-delivery, on a passing mesh/check and the operator's Approve (ADR 0534 §4), or by the operator's
+// own hand at the forge (§2).
 func judgedRule(rules []protectionRule, branch string) forgeFacts {
 	switch r := appliedRule(rules, branch); {
 	case r == nil:
 		return forgeFacts{Why: fmt.Sprintf("no protection rule covers %s", branch)}
 	case r.Push:
 		return forgeFacts{Why: fmt.Sprintf("the rule %s lets a person push to %s directly", r.Rule, branch)}
-	case len(r.RequiredStatuses) == 0:
-		return forgeFacts{Why: fmt.Sprintf("the rule %s requires no status before a merge into %s", r.Rule, branch)}
+	case r.MergeWhitelist == nil || len(r.MergeWhitelist) == 0:
+		return forgeFacts{Why: fmt.Sprintf("the rule %s keeps no merge whitelist naming %s alone for %s", r.Rule, forgeToolsAccount, branch)}
+	case !slices.Contains(r.MergeWhitelist, forgeToolsAccount):
+		return forgeFacts{Why: fmt.Sprintf("the rule %s lets %s merge into %s, and not %s", r.Rule,
+			strings.Join(r.MergeWhitelist, ", "), branch, forgeToolsAccount)}
+	case slices.ContainsFunc(r.MergeWhitelist, isAgentAccount):
+		return forgeFacts{Why: fmt.Sprintf("the rule %s lets an agent's account merge into %s: %s", r.Rule, branch,
+			strings.Join(r.MergeWhitelist, ", "))}
 	case r.AdminMayOverride:
-		return forgeFacts{Why: fmt.Sprintf("the rule %s lets an administrator merge into %s past a status", r.Rule, branch)}
+		return forgeFacts{Why: fmt.Sprintf("the rule %s lets an administrator merge into %s past it", r.Rule, branch)}
 	}
 	return forgeFacts{Guarded: true}
 }

cmd/mesh-controller/build_source_test.go

@@ -102,16 +102,17 @@ func init() {
 // The forge applies one rule to a branch: the rule named for it, else the first glob that covers it. A stronger
 // rule later in the list is not what guards the branch, and is not read as if it were (the confirmation review).
 func TestTheRuleJudgedIsTheOneTheForgeApplies(t *testing.T) {
-	guarded := protectionRule{Rule: "*", RequiredStatuses: []string{"mesh/merge-gate"}}
-	open := protectionRule{Rule: "main", Push: true, RequiredStatuses: []string{"mesh/merge-gate"}}
+	tools := []string{forgeToolsAccount}
+	guarded := protectionRule{Rule: "*", MergeWhitelist: tools}
+	open := protectionRule{Rule: "main", Push: true, MergeWhitelist: tools}
 	if f := judgedRule([]protectionRule{guarded, open}, "main"); f.Guarded {
 		t.Error("an exact rule letting pushes was passed over for a glob that guards")
 	}
-	if f := judgedRule([]protectionRule{{Rule: "ma*", RequiredStatuses: nil}, {Rule: "*", RequiredStatuses: []string{"x"}}},
+	if f := judgedRule([]protectionRule{{Rule: "ma*", MergeWhitelist: nil}, {Rule: "*", MergeWhitelist: tools}},
 		"main"); f.Guarded || !strings.Contains(f.Why, "ma*") {
 		t.Errorf("the first glob covering the branch was not the one judged: %+v", f)
 	}
-	if f := judgedRule([]protectionRule{{Rule: "release/*"}, {Rule: "main", RequiredStatuses: []string{"x"}}}, "main"); !f.Guarded {
+	if f := judgedRule([]protectionRule{{Rule: "release/*"}, {Rule: "main", MergeWhitelist: tools}}, "main"); !f.Guarded {
 		t.Errorf("the rule named for the branch was not judged: %+v", f)
 	}
 	if f := judgedRule(nil, "main"); f.Guarded {
@@ -119,6 +120,39 @@ func TestTheRuleJudgedIsTheOneTheForgeApplies(t *testing.T) {
 	}
 }
 
+// **A module's trunk is trusted when it takes no push, only the tools' account and people's own logins merge into it,
+// and no administrator merges past its rule** (novox/hq ADR 0534 §2, §6): the forge requires no status to merge (§1), so
+// a rule requiring none is trusted; one with no merge whitelist, no tools' account in it, an agent's account in it, a
+// push allowed or an administrator's override is not.
+func TestATrunkIsTrustedForItsMergersNotForAStatus(t *testing.T) {
+	tools := []string{forgeToolsAccount}
+	for _, c := range []struct {
+		name string
+		rule protectionRule
+		want bool
+		why  string
+	}{
+		{"no status, no push, the tools alone merge", protectionRule{Rule: "main", MergeWhitelist: tools}, true, ""},
+		{"a status required as well", protectionRule{Rule: "main", MergeWhitelist: tools, RequiredStatuses: []string{"mesh/check"}}, true, ""},
+		{"no merge whitelist", protectionRule{Rule: "main"}, false, "merge whitelist"},
+		{"an empty merge whitelist", protectionRule{Rule: "main", MergeWhitelist: []string{}}, false, "merge whitelist"},
+		{"the operator's own login beside the tools", protectionRule{Rule: "main", MergeWhitelist: []string{forgeToolsAccount, "jschoubben"}}, true, ""},
+		{"an agent's account in the whitelist", protectionRule{Rule: "main", MergeWhitelist: []string{forgeToolsAccount, "agent-ace"}}, false, "agent-ace"},
+		{"an agent's account alone", protectionRule{Rule: "main", MergeWhitelist: []string{"agent-ace"}}, false, "agent-ace"},
+		{"a person's login without the tools", protectionRule{Rule: "main", MergeWhitelist: []string{"jschoubben"}}, false, forgeToolsAccount},
+		{"a push allowed", protectionRule{Rule: "main", Push: true, MergeWhitelist: tools}, false, "push"},
+		{"an administrator's override", protectionRule{Rule: "main", MergeWhitelist: tools, AdminMayOverride: true}, false, "administrator"},
+	} {
+		f := judgedRule([]protectionRule{c.rule}, "main")
+		if f.Guarded != c.want || !strings.Contains(f.Why, c.why) {
+			t.Errorf("%s: %+v, want guarded %v saying %q", c.name, f, c.want, c.why)
+		}
+		if strings.Contains(f.Why, "requires no status") {
+			t.Errorf("%s: a trunk is still judged on a required status: %s", c.name, f.Why)
+		}
+	}
+}
+
 // An id the forge could not give keeps the id already recorded.
 
 // **The operator's Approve of a reviewed group registers a moved module from its new repository** (novox/hq ADR 0420 §7,

cmd/mesh-controller/lease_backoff.go

@@ -1,117 +0,0 @@
-package main
-
-import (
-	"context"
-	"time"
-
-	"github.com/novox/mesh-controller/internal/broker"
-	"github.com/novox/mesh-controller/internal/inventory"
-)
-
-// Backing off before standing (novox/hq issue 906).
-//
-// A controller loses its lease when its bus stops answering for longer than the key's age: on the control-node,
-// a bus whose disks are saturated has taken 30 to 42 seconds to answer (2026-10-23 05:52). The process then
-// exits, as ADR 0229 rule 4 says, and its service manager starts it again five seconds later. **A start is the
-// heaviest thing a controller asks of the bus** — every bucket, stream and consumer asserted again — and a start
-// that cannot assert the lease's bucket used to exit at once and be started again five seconds later, on the
-// same overloaded bus, nine times in twelve minutes (2026-10-16). So a candidate backs off twice:
-//
-//   - **After a loss**, before it touches the bus: from the newest epoch lost in the last ten minutes (the
-//     controller's store says it, rule 7), it waits one key's age, doubled for every further loss in that
-//     window, at most two minutes. A single loss costs a key's age; a controller losing it again and again
-//     stands less and less often.
-//   - **While the bus does not answer it**, in its own process: asserting the lease's bucket, opening it or
-//     reading who holds it is tried again after 5 s, doubling to at most 60 s, instead of exiting into a restart
-//     every five seconds. Nothing acts meanwhile: the process holds no lease, and the gate refuses every act.
-
-const (
-	// lossWindow is how far back lost epochs count towards the backoff.
-	lossWindow = 10 * time.Minute
-	// lossMost is the longest a candidate waits after losses.
-	lossMost = 2 * time.Minute
-	// standFirst and standMost are the waits between tries of a candidate the bus does not answer.
-	standFirst = 5 * time.Second
-	standMost  = time.Minute
-)
-
-// lossBackoff is how long a candidate waits now before it stands, after the epochs it read: one key's age from
-// the newest loss in the window, doubled for every further loss in it, at most lossMost.
-func lossBackoff(epochs []inventory.Epoch, now time.Time) time.Duration {
-	return lossBackoffFrom(epochs, now, broker.LeaseTTL)
-}
-
-func lossBackoffFrom(epochs []inventory.Epoch, now time.Time, base time.Duration) time.Duration {
-	lost, newest := lostSince(epochs, now)
-	if lost == 0 {
-		return 0
-	}
-	delay := base
-	for i := 1; i < lost && delay < lossMost; i++ {
-		delay *= 2
-	}
-	delay = min(delay, lossMost)
-	return max(newest.Add(delay).Sub(now), 0)
-}
-
-// standUp runs try until it answers nil, waiting standFirst between tries and doubling to standMost; it answers
-// ctx's error once ctx ends.
-func standUp(ctx context.Context, sleep func(context.Context, time.Duration) error, say func(string, ...any),
-	try func(context.Context) error) error {
-	wait := standFirst
-	for {
-		err := try(ctx)
-		if err == nil {
-			return nil
-		}
-		if ctx.Err() != nil {
-			return ctx.Err()
-		}
-		say("this controller cannot stand for the lease yet: %v — it tries again in %s, in this process, rather "+
-			"than start again on a bus that does not answer (novox/hq issue 906)", err, wait)
-		if err := sleep(ctx, wait); err != nil {
-			return err
-		}
-		wait = min(wait*2, standMost)
-	}
-}
-
-// backoffAfterLosses is how long this candidate waits before it stands, read from the epochs its store kept,
-// and said. A store that cannot be read makes it wait nothing: the backoff spares the bus, and never keeps a
-// controller from standing.
-func (a *actor) backoffAfterLosses(ctx context.Context, inv *inventory.Inventory, say func(string, ...any)) time.Duration {
-	now := time.Now()
-	reading, cancel := context.WithTimeout(ctx, 10*time.Second)
-	epochs, err := inv.EpochsSince(reading, now.Add(-lossWindow))
-	cancel()
-	if err != nil {
-		say("whether the lease was lost lately cannot be read, so this controller stands without backing off: %v", err)
-		return 0
-	}
-	base := a.lossBase
-	if base <= 0 {
-		base = broker.LeaseTTL
-	}
-	wait := lossBackoffFrom(epochs, now, base)
-	if wait > 0 {
-		lost, _ := lostSince(epochs, now)
-		say("the controller lease was lost lately (%d time(s) in the last %s): this controller backs off %s "+
-			"before it stands, so its start does not add to the load that lost it (novox/hq issue 906)",
-			lost, lossWindow, wait.Round(time.Second))
-	}
-	return wait
-}
-
-// lostSince is how many epochs were lost in the window before now, and when the newest of them was.
-func lostSince(epochs []inventory.Epoch, now time.Time) (lost int, newest time.Time) {
-	for _, e := range epochs {
-		if e.Ended == nil || e.How != inventory.EpochLost || now.Sub(*e.Ended) > lossWindow {
-			continue
-		}
-		lost++
-		if e.Ended.After(newest) {
-			newest = *e.Ended
-		}
-	}
-	return lost, newest
-}

cmd/mesh-controller/lease_backoff_test.go

@@ -1,97 +0,0 @@
-package main
-
-import (
-	"context"
-	"errors"
-	"slices"
-	"testing"
-	"time"
-
-	"github.com/novox/mesh-controller/internal/inventory"
-)
-
-// A controller that lost its lease to a slow bus backs off before it stands again (novox/hq issue 906): it
-// does not add its start — every bucket and consumer asserted again — to the load that lost it the lease.
-
-func lostAt(epoch uint64, ended time.Time, how string) inventory.Epoch {
-	return inventory.Epoch{Epoch: epoch, Instance: "controller@anchor", Taken: ended.Add(-time.Hour), Ended: &ended,
-		How: how}
-}
-
-// 906/AC1: given epochs lost in the last ten minutes, when a candidate starts, then it waits from the newest
-// loss one key's age, doubled for each further loss, at most two minutes; nothing else makes it wait.
-func TestACandidateAfterALostLeaseBacksOff(t *testing.T) {
-	now := time.Date(2026, 10, 23, 5, 52, 40, 0, time.UTC)
-	for _, c := range []struct {
-		name   string
-		epochs []inventory.Epoch
-		want   time.Duration
-	}{
-		{"nothing lost", nil, 0},
-		{"one held, one given back", []inventory.Epoch{
-			{Epoch: 3, Taken: now.Add(-time.Minute)},
-			lostAt(2, now.Add(-2*time.Second), inventory.EpochReleased),
-		}, 0},
-		{"one lost five seconds ago", []inventory.Epoch{lostAt(7, now.Add(-5*time.Second), inventory.EpochLost)},
-			10 * time.Second},
-		{"one lost long ago enough", []inventory.Epoch{lostAt(7, now.Add(-20*time.Second), inventory.EpochLost)}, 0},
-		{"three lost, the newest a second ago", []inventory.Epoch{
-			lostAt(9, now.Add(-time.Second), inventory.EpochLost),
-			lostAt(8, now.Add(-3*time.Minute), inventory.EpochLost),
-			lostAt(7, now.Add(-6*time.Minute), inventory.EpochLost),
-		}, 59 * time.Second},
-		{"six lost: at most two minutes", []inventory.Epoch{
-			lostAt(12, now, inventory.EpochLost), lostAt(11, now.Add(-time.Minute), inventory.EpochLost),
-			lostAt(10, now.Add(-2*time.Minute), inventory.EpochLost), lostAt(9, now.Add(-3*time.Minute), inventory.EpochLost),
-			lostAt(8, now.Add(-4*time.Minute), inventory.EpochLost), lostAt(7, now.Add(-5*time.Minute), inventory.EpochLost),
-		}, 2 * time.Minute},
-		{"a loss older than ten minutes does not count", []inventory.Epoch{
-			lostAt(9, now.Add(-time.Second), inventory.EpochLost),
-			lostAt(8, now.Add(-11*time.Minute), inventory.EpochLost),
-		}, 14 * time.Second},
-		{"expired is not lost", []inventory.Epoch{lostAt(9, now.Add(-time.Second), inventory.EpochExpired)}, 0},
-	} {
-		t.Run(c.name, func(t *testing.T) {
-			if got := lossBackoff(c.epochs, now); got != c.want {
-				t.Fatalf("a candidate after %d epoch(s) waits %s, want %s", len(c.epochs), got, c.want)
-			}
-		})
-	}
-}
-
-// 906/AC2: given a bus that does not answer a candidate, when it asserts the lease's bucket or reads who holds
-// it, then the candidate tries again in its own process after 5 s, doubling to at most 60 s, instead of exiting
-// to be restarted by its service manager every five seconds.
-func TestACandidateTheBusDoesNotAnswerTriesAgainWithBackoff(t *testing.T) {
-	var waited []time.Duration
-	sleep := func(ctx context.Context, d time.Duration) error { waited = append(waited, d); return nil }
-	tries := 0
-	slow := errors.New("asserting bucket mesh-controller_lease: context deadline exceeded")
-	err := standUp(t.Context(), sleep, func(string, ...any) {}, func(context.Context) error {
-		tries++
-		if tries < 7 {
-			return slow
-		}
-		return nil
-	})
-	if err != nil {
-		t.Fatalf("a candidate whose bus answered at the seventh try gave up: %v", err)
-	}
-	want := []time.Duration{5 * time.Second, 10 * time.Second, 20 * time.Second, 40 * time.Second, time.Minute,
-		time.Minute}
-	if tries != 7 || !slices.Equal(waited, want) {
-		t.Fatalf("a candidate tried %d time(s), waiting %v between; want 7, waiting %v", tries, waited, want)
-	}
-}
-
-// 906/AC2: and a candidate stopping while it waits stops, with why.
-func TestACandidateStoppingWhileItBacksOffStops(t *testing.T) {
-	ctx, cancel := context.WithCancel(t.Context())
-	cancel()
-	err := standUp(ctx, sleepOrDone, func(string, ...any) {}, func(context.Context) error {
-		return errors.New("who holds the lease cannot be read: context deadline exceeded")
-	})
-	if !errors.Is(err, context.Canceled) {
-		t.Fatalf("a candidate stopped while backing off answered %v", err)
-	}
-}

cmd/mesh-controller/lease_integration_test.go

@@ -152,33 +152,3 @@ func TestTwoControllersOneActs(t *testing.T) {
 		t.Fatalf("B's epoch does not say it was lost: %+v", ended)
 	}
 }
-
-// 906/AC1, on a real bus and store: a controller starting right after an epoch was lost backs off before it
-// stands — it takes the lease no sooner than the backoff after the loss, and then holds it as usual.
-func TestAControllerAfterALostLeaseBacksOffBeforeItStands(t *testing.T) {
-	url, _ := aBusForTheLease(t)
-	inv := inventory.ForTest(t)
-	ctx := t.Context()
-	// The controller before it lost epoch 5, and said so in the store, as `took` does on a loss.
-	if _, err := inv.TookEpoch(ctx, inventory.Epoch{Epoch: 5, Instance: "controller@anchor pid 1", Host: "anchor",
-		Taken: time.Now()}); err != nil {
-		t.Fatal(err)
-	}
-	if err := inv.EndEpoch(ctx, 5, inventory.EpochLost); err != nil {
-		t.Fatal(err)
-	}
-	// The key's age stands in as a backoff of two seconds, so the case runs in seconds (issue 591).
-	const backoff = 2 * time.Second
-	a := &actor{renewEvery: lease.RenewEvery / 5, poll: lease.Poll / 10, lossBase: backoff}
-	start := time.Now()
-	if _, err := a.serveUnderTheLease(ctx, inv, url); err != nil {
-		t.Fatal(err)
-	}
-	defer a.release()
-	if waited := time.Since(start); waited < backoff-300*time.Millisecond {
-		t.Fatalf("a controller took the lease %s after an epoch was lost, without backing off %s", waited, backoff)
-	}
-	if epoch, err := a.epoch(ctx); err != nil || epoch <= 5 {
-		t.Fatalf("after backing off the controller holds epoch %d (%v), want one above the lost 5", epoch, err)
-	}
-}

cmd/mesh-controller/plan_retry.go

@@ -218,6 +218,15 @@ func retryRefusal(p inventory.Plan, plans []inventory.Plan, onPath map[string]st
 		}
 		return oneWalkAtATime(p, plans, onPath)
 	}
+	// **A walk lost to silence is resumed** (novox/hq ADR 0531, issue 1886): its tier was sent whole and only the gate after it failed it, every
+	// machine it waited on silent; once they reported, it goes on from that gate.
+	if len(stopped) == 0 && lostToSilence(p) {
+		if q, found := newerOpenPlan(p, plans); found {
+			return fmt.Errorf("%s supersedes it: a newer merge of %s (%s at %s) is open, and resuming %s would send "+
+				"what that one replaced", q.ID, q.Repository, q.ID, short(q.Commit), p.ID)
+		}
+		return oneWalkAtATime(p, plans, onPath)
+	}
 	if len(stopped) == 0 {
 		return fmt.Errorf("nothing in tier %d of %s failed to build or stopped rolling out — it stopped at: %s",
 			p.Tier, p.ID, p.Note)
@@ -412,6 +421,20 @@ func retryPlan(ctx context.Context, open *stores, id string) (string, error) {
 		}
 		return said, nil
 	}
+	if len(failed) == 0 && lostToSilence(p) {
+		reports, err := inv.LastReports(ctx)
+		if err != nil {
+			return "", err
+		}
+		said, err := resumeLost(&p, silentNodes(p, reports, time.Now().UTC(), provLookup(ctx)))
+		if err != nil {
+			return "", err
+		}
+		if err := inv.SavePlan(ctx, &p); err != nil {
+			return "", err
+		}
+		return said, nil
+	}
 	if len(failed) == 0 {
 		return retryRollouts(ctx, open, &p)
 	}
@@ -693,3 +716,51 @@ func resumeStopped(p *inventory.Plan) string {
 	return fmt.Sprintf("%s resumed at tier %d of %d, stopped by %s: %s; this verb asks nothing — the next look carries it "+
 		"on as any walk", p.ID, p.Tier+1, len(p.Tiers), by, waits)
 }
+
+// silentGateWords say, in a plan's note, that the gate between two tiers found every machine it waits on silent past
+// ApplySilentAfter, and failed the plan as lost (issue 1160, ADR 0531): one text for the writer and the reader (lostToSilence).
+const silentGateWords = "every machine it waits on said nothing on its send within"
+
+// lostToSilence is whether a plan failed only at the gate after its tier, every machine it waited on silent (ADR 0531,
+// issue 1886): nothing in its tier failed to build, stopped rolling out or is still to ask — the look had sent the whole tier.
+func lostToSilence(p inventory.Plan) bool {
+	if p.State != inventory.PlanFailed || p.Tier >= len(p.Tiers) || !strings.Contains(p.Note, silentGateWords) ||
+		len(failedIn(p)) > 0 || len(stoppedRollouts(p)) > 0 {
+		return false
+	}
+	asked, _, unasked := waitingIn(p)
+	return asked == 0 && unasked == 0
+}
+
+// resumeLost sets a walk lost to silence building again at the tier it failed after (ADR 0531, issue 1886), for the next look to
+// judge its gate again on the machines' reports, unless every machine its tier was sent to is still silent: then
+// nothing it needs is delivered anywhere yet, and the retry is refused as the gate would fail it again.
+func resumeLost(p *inventory.Plan, silent map[string]bool) (string, error) {
+	var sent, still []string
+	for _, m := range p.Tiers[p.Tier] {
+		if s := p.Modules[m]; s != nil {
+			for n := range s.Rest {
+				if !slices.Contains(sent, n) {
+					sent = append(sent, n)
+				}
+			}
+		}
+	}
+	for _, n := range sent {
+		if silent[n] {
+			still = append(still, n)
+		}
+	}
+	sort.Strings(still)
+	if len(sent) > 0 && len(still) == len(sent) {
+		return "", fmt.Errorf("%s failed because every machine its tier %d waits on said nothing on its send, and %s "+
+			"still said nothing: retry once it reported", p.ID, p.Tier+1, strings.Join(still, ", "))
+	}
+	was := p.Note
+	p.State = inventory.PlanBuilding
+	goesOnAgain(p)
+	p.Note = fmt.Sprintf("tier %d resumed by hand after it failed as lost (%s): its gate is judged again on the "+
+		"machines' reports", p.Tier, was)
+	return fmt.Sprintf("%s resumed at tier %d of %d, failed as lost to silence (%s); this verb asks and sends nothing — "+
+		"the next look judges its gate again and carries it on as any walk", p.ID, p.Tier+1, len(p.Tiers), was), nil
+}

cmd/mesh-controller/release_plan.go

@@ -260,7 +260,7 @@ type gateNeed struct {
 }
 
 // silentNodes is the machines a walk sent and heard nothing from past ApplySilentAfter, on whichever tier it sent
-// them (novox/hq ADR 0360 §5, issue 1160): the one rule both the gates between tiers and the walk's end read, through
+// them (novox/hq ADR 0360 §7, issue 1160): the one rule both the gates between tiers and the walk's end read, through
 // walkReports. A machine that reports on its send is not silent from then on.
 func silentNodes(p inventory.Plan, reports []inventory.Reported, now time.Time, prov inventory.ProvLookup) map[string]bool {
 	_, _, silent := walkReports(p, reports, now, prov)
@@ -272,7 +272,7 @@ func silentNodes(p inventory.Plan, reports []inventory.Reported, now time.Time,
 }
 
 // withoutSilent is the machines of running that are not silent: a later tier is not sent to a machine the walk heard
-// nothing from, which keeps its want of the plan's builds for the send it gets when it is heard again (ADR 0360 §5).
+// nothing from, which keeps its want of the plan's builds for the send it gets when it is heard again (ADR 0360 §7).
 func withoutSilent(running []string, silent map[string]bool) []string {
 	return slices.DeleteFunc(slices.Clone(running), func(n string) bool { return silent[n] })
 }
@@ -1129,7 +1129,7 @@ func advanceOnce(ctx context.Context, open *stores, p *inventory.Plan,
 	// being behind. Now every module of the tier whose first machine is the same is sent there in one
 	// send, judged there by one gate — kept on the first of them, the others gated by it — and the rest
 	// of the tier's machines are sent once, together, when the gates have passed.
-	// **A machine silent past the bound is left as it was, on every tier** (novox/hq ADR 0360 §5, issue 1160): the
+	// **A machine silent past the bound is left as it was, on every tier** (novox/hq ADR 0360 §7, issue 1160): the
 	// walk's end names it, and the gates between tiers and the sends of the tiers after do not wait for it or go to it.
 	allReports, err := inv.LastReports(ctx)
 	if err != nil {
@@ -1457,8 +1457,7 @@ func advanceOnce(ctx context.Context, open *stores, p *inventory.Plan,
 			// Every machine the next tier waits on is silent: nothing it needs was delivered anywhere, so the plan does
 			// not go on as if it had been (issue 1160).
 			p.State = inventory.PlanFailed
-			p.Note = fmt.Sprintf("tier %d built; every machine it waits on said nothing on its send within %s: %s",
-				p.Tier, applySilentAfter, strings.Join(quiet, ", "))
+			p.Note = fmt.Sprintf("tier %d built; %s %s: %s", p.Tier, silentGateWords, applySilentAfter, strings.Join(quiet, ", "))
 			fmt.Printf("%s: %s\n", p.ID, p.Note)
 			return true, nil
 		}
@@ -1476,7 +1475,7 @@ func advanceOnce(ctx context.Context, open *stores, p *inventory.Plan,
 			return changed, nil
 		}
 	}
-	// **A walk ends on its machines' reports** (novox/hq ADR 0360 §4, §5): after its last tier, every machine it sent
+	// **A walk ends on its machines' reports** (novox/hq ADR 0360 §4, §7): after its last tier, every machine it sent
 	// has reported on that send, or is silent past ApplySilentAfter; never a health reading.
 	end := ""
 	if p.Tier == len(p.Tiers)-1 {
@@ -1543,50 +1542,62 @@ func walkReports(p inventory.Plan, reports []inventory.Reported, now time.Time,
 			carried[node] = append(carried[node], m)
 		}
 	}
-	seen := map[string]bool{}
+	// **Each machine on the walk's last send to it** (novox/hq ADR 0531, issue 1886): a send carries the machine's whole declaration (ADR
+	// 0221), so a report on the last one answers every earlier one, and the bound runs from it. Judged on every send
+	// it had made, a walk whose tiers were sent hours apart — R4e, stopped at tier 1 and resumed — held tier 0's send of
+	// 00:23 against tier 1's templates and named its machines silent 35 seconds after tier 1 was sent.
+	last := map[string]inventory.SentDeclaration{}
+	lastAt := map[string]time.Time{}
 	for _, s := range p.Modules {
 		if s == nil {
 			continue
 		}
 		for node, sent := range s.Rest {
-			sent = sent.Only(carried[node]...)
-			if seen[node] {
-				continue
-			}
-			r := byNode[node]
 			at := sent.At
 			if at.IsZero() && s.SentAt != nil {
 				at = *s.SentAt
 			}
-			// A module handed to the provisioner is applied only once its client said so too (ADR 0362 §1-5).
-			outcome, reported := provisionedOutcome(node, sent, r, r.At != nil && sent.ReportsOn(r), prov)
-			switch {
-			case reported:
-				seen[node] = true
-				switch {
-				case outcome == inventory.OutcomeFailed || outcome == inventory.OutcomeRefused:
-					failed = append(failed, node)
-				case strings.HasPrefix(outcome, inventory.OutcomeProvisioningFailed):
-					failed = append(failed, node+" ("+outcome+")")
-				}
-			case !at.IsZero() && now.Sub(at) > applySilentAfter:
-				seen[node] = true
-				silent = append(silent, node)
-			default:
-				if !slices.Contains(waiting, node) {
-					waiting = append(waiting, node)
-				}
+			was, ok := last[node]
+			if !ok || laterSend(sent, at, was, lastAt[node]) {
+				last[node], lastAt[node] = sent, at
 			}
 		}
 	}
-	// A machine one module's send was reported on and another's not yet is the same send: reported.
-	waiting = slices.DeleteFunc(waiting, func(n string) bool { return seen[n] })
+	for node, sent := range last {
+		at := lastAt[node]
+		sent = sent.Only(carried[node]...)
+		r := byNode[node]
+		// A module handed to the provisioner is applied only once its client said so too (ADR 0362 §1-5).
+		outcome, reported := provisionedOutcome(node, sent, r, r.At != nil && sent.ReportsOn(r), prov)
+		switch {
+		case reported:
+			switch {
+			case outcome == inventory.OutcomeFailed || outcome == inventory.OutcomeRefused:
+				failed = append(failed, node)
+			case strings.HasPrefix(outcome, inventory.OutcomeProvisioningFailed):
+				failed = append(failed, node+" ("+outcome+")")
+			}
+		case !at.IsZero() && now.Sub(at) > applySilentAfter:
+			silent = append(silent, node)
+		default:
+			waiting = append(waiting, node)
+		}
+	}
 	sort.Strings(waiting)
 	sort.Strings(failed)
 	sort.Strings(silent)
 	return waiting, failed, silent
 }
 
+// laterSend is whether a, sent at aAt, is a later send to a machine than b, sent at bAt: by its sequence when both carry
+// one, else by when it went.
+func laterSend(a inventory.SentDeclaration, aAt time.Time, b inventory.SentDeclaration, bAt time.Time) bool {
+	if a.Sequence > 0 && b.Sequence > 0 && a.Sequence != b.Sequence {
+		return a.Sequence > b.Sequence
+	}
+	return aAt.After(bAt)
+}
+
 // restOf is, for one module, every machine of its rest that the send reached and what it carried there.
 func restOf(to, sent []string, what map[string]inventory.SentDeclaration) map[string]inventory.SentDeclaration {
 	out := map[string]inventory.SentDeclaration{}

cmd/mesh-controller/walk_resumed_silence_1886_test.go

@@ -0,0 +1,173 @@
+package main
+
+import (
+	"slices"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/novox/mesh-controller/internal/inventory"
+)
+
+// Issue 1886, the resume of R4e (plan-1792707174627079192): tier 0 was sent at 00:23:28, the walk was stopped at
+// 02:01 and resumed at 07:43, and tier 1 was sent at 07:49:54. 35 seconds later the gate named ace silent "within
+// 15m0s", and at 07:51:30 shanks, in the same second shanks reported it applied: the plan failed as lost. The walk
+// judged each machine on every send it had made it, tier 0's of 00:23 too, narrowed to every module the walk carried
+// there — so tier 0's send was held against the templates of tier 1's modules it never carried, and silence was
+// measured from 00:23. A machine is judged on the walk's last send to it: that send carries its whole declaration.
+
+func r4eResumed() (inventory.Plan, []inventory.Reported, inventory.ProvLookup, time.Time) {
+	tier0 := time.Date(2026, 10, 22, 22, 23, 28, 0, time.UTC)
+	tier1 := time.Date(2026, 10, 23, 5, 49, 54, 0, time.UTC)
+	now := time.Date(2026, 10, 23, 5, 51, 30, 0, time.UTC)
+	t0, t1 := tier0, tier1
+	p := inventory.Plan{ID: "plan-1792707174627079192", State: inventory.PlanRolling, Tier: 1,
+		Tiers: [][]string{{"pacman"}, {"mesh-build-image"}, {"gitea"}},
+		Modules: map[string]*inventory.PlanModule{
+			"pacman": {State: "built", SentAt: &t0, Rest: map[string]inventory.SentDeclaration{
+				"shanks": {Digest: "d0-shanks", Sequence: 900, At: tier0,
+					Provisioned: map[string]string{"pacman": "pac-1", "mesh-build-image": "img-old"}},
+				"ace": {Digest: "d0-ace", Sequence: 1100, At: tier0,
+					Provisioned: map[string]string{"pacman": "pac-1", "mesh-build-image": "img-old"}}}},
+			"mesh-build-image": {State: "built", SentAt: &t1, Rest: map[string]inventory.SentDeclaration{
+				"shanks": {Digest: "d1-shanks", Sequence: 971, At: tier1,
+					Provisioned: map[string]string{"pacman": "pac-1", "mesh-build-image": "img-new"}},
+				"ace": {Digest: "d1-ace", Sequence: 1212, At: tier1,
+					Provisioned: map[string]string{"pacman": "pac-1", "mesh-build-image": "img-new"}}}},
+			"gitea": {}}}
+	shanksAt := now
+	aceAt := time.Date(2026, 10, 23, 0, 57, 50, 0, time.UTC)
+	reports := []inventory.Reported{
+		// shanks reported on tier 1's send in the second the gate looked.
+		{Node: "shanks", Outcome: inventory.OutcomeApplied, At: &shanksAt, Declared: "d1-shanks", Current: true,
+			ReportedSequence: 971},
+		// ace's last report is from before tier 1's send, on a declaration sequenced after tier 0's.
+		{Node: "ace", Outcome: inventory.OutcomeApplied, At: &aceAt, Declared: "d-ace-0257", ReportedSequence: 1150},
+	}
+	says := map[string]inventory.ProvReport{
+		"shanks/pacman":           {Status: inventory.ProvApplied, Templates: "pac-1", At: now},
+		"shanks/mesh-build-image": {Status: inventory.ProvApplied, Templates: "img-new", At: now},
+		"ace/pacman":              {Status: inventory.ProvApplied, Templates: "pac-1", At: aceAt},
+		// ace's client had not placed the old image's files (issue 1884), and has not yet the new one's.
+		"ace/mesh-build-image": {Status: inventory.ProvWaiting, Templates: "img-old", At: aceAt},
+	}
+	prov := func(node, module string) (inventory.ProvReport, bool) { r, ok := says[node+"/"+module]; return r, ok }
+	return p, reports, prov, now
+}
+
+func TestAResumedWalkJudgesEachMachineOnItsLastSend(t *testing.T) {
+	p, reports, prov, now := r4eResumed()
+	waiting, failed, silent := walkReports(p, reports, now, prov)
+	if len(silent) != 0 {
+		t.Fatalf("1886: %v named silent within %s of tier 1's send, from tier 0's send of 00:23", silent, applySilentAfter)
+	}
+	if len(failed) != 0 || !slices.Equal(waiting, []string{"ace"}) {
+		t.Fatalf("1886: waiting %v, failed %v; want ace waited for (sent 96 s ago) and shanks reported", waiting, failed)
+	}
+	// The gate the plan failed at: mesh-build-image runs on ace and shanks, which tier 2 needs.
+	needs := []gateNeed{{module: "mesh-build-image", running: []string{"ace", "shanks"},
+		since: func(string) time.Time { return *p.Modules["mesh-build-image"].SentAt }}}
+	w, quiet, lost := settleTierGate(needs, reports, silentNodes(p, reports, now, prov))
+	if lost || len(quiet) != 0 || !slices.Equal(w, []string{"mesh-build-image on ace"}) {
+		t.Fatalf("1886: the gate says waiting %v, silent %v, lost %v; want mesh-build-image on ace waited for", w, quiet, lost)
+	}
+}
+
+// The machine silent on the walk's last send to it is still silent once the bound has passed since that send.
+func TestAMachineSilentOnItsLastSendIsStillSilent(t *testing.T) {
+	p, reports, prov, now := r4eResumed()
+	later := now.Add(applySilentAfter + time.Minute)
+	_, _, silent := walkReports(p, reports, later, prov)
+	if !slices.Equal(silent, []string{"ace"}) {
+		t.Fatalf("1886: silent %v past the bound from tier 1's send; want ace", silent)
+	}
+}
+
+// A walk failed only because every machine its gate waited on was silent, whose tiers are all sent, is carried on by
+// `plans retry` once those machines reported (issue 1886): it failed on no build and no rollout, so the retry of failed
+// builds and of stopped rollouts refused it with "nothing in tier 1 … failed to build or stopped rolling out".
+func r4eLost() inventory.Plan {
+	p, _, _, _ := r4eResumed()
+	p.State = inventory.PlanFailed
+	p.Note = "tier 1 built; " + silentGateWords + " 15m0s: shanks"
+	// mesh-tools is built and recorded, never sent, as R4e's was.
+	p.Tiers[1] = append(p.Tiers[1], "mesh-tools")
+	p.Modules["mesh-tools"] = &inventory.PlanModule{State: "built"}
+	return p
+}
+
+func TestRetryCarriesOnAWalkLostToSilenceOnceItsMachinesReported(t *testing.T) {
+	p, reports, prov, now := r4eResumed()
+	p = r4eLost()
+	if !lostToSilence(p) {
+		t.Fatalf("1886: the plan R4e failed as is not read as lost to silence: %q", p.Note)
+	}
+	if err := retryRefusal(p, []inventory.Plan{p}, ownPathOf(nil)); err != nil {
+		t.Fatalf("1886: a walk lost to silence was refused a retry: %v", err)
+	}
+	reports[1].Current, reports[1].Declared, reports[1].ReportedSequence = true, "d1-ace", 1212
+	reports[1].At = &now
+	says := map[string]inventory.ProvReport{}
+	for _, n := range []string{"ace", "shanks"} {
+		says[n+"/pacman"] = inventory.ProvReport{Status: inventory.ProvApplied, Templates: "pac-1", At: now}
+		says[n+"/mesh-build-image"] = inventory.ProvReport{Status: inventory.ProvApplied, Templates: "img-new", At: now}
+	}
+	prov = func(node, module string) (inventory.ProvReport, bool) { r, ok := says[node+"/"+module]; return r, ok }
+	said, err := resumeLost(&p, silentNodes(p, reports, now.Add(time.Hour), prov))
+	if err != nil {
+		t.Fatalf("1886: the resume of a walk whose machines reported refused: %v", err)
+	}
+	if p.State != inventory.PlanBuilding || p.Tier != 1 {
+		t.Fatalf("1886: resumed walk is %s at tier %d; want building at tier 1, for the next look to pass its gate", p.State, p.Tier)
+	}
+	if !strings.Contains(p.Note, "resumed by hand") || !strings.Contains(said, "resumed at tier 2 of 3") {
+		t.Fatalf("1886: note %q, said %q", p.Note, said)
+	}
+}
+
+// Every machine its tier was sent to still silent: the gate would fail it again, so the retry says so instead.
+func TestRetryRefusesAWalkLostToSilenceWhileEveryMachineIsStillSilent(t *testing.T) {
+	p := r4eLost()
+	if _, err := resumeLost(&p, map[string]bool{"ace": true, "shanks": true}); err == nil || !strings.Contains(err.Error(), "ace, shanks") {
+		t.Fatalf("1886: a walk whose machine is still silent was resumed: %v", err)
+	}
+	if p.State != inventory.PlanFailed {
+		t.Fatalf("1886: a refused resume changed the plan to %s", p.State)
+	}
+}
+
+// A walk lost to silence with something of its tier never sent is not this: its send is the look's to make.
+func TestALostWalkWithATierUnsentIsNotLostToSilence(t *testing.T) {
+	p := r4eLost()
+	p.Modules["mesh-build-image"].State = "asked"
+	if lostToSilence(p) {
+		t.Fatal("1886: a tier with a module still asked was read as lost to silence")
+	}
+	p = r4eLost()
+	p.Note = "tier 1 built; mesh-build-image failed to build"
+	if lostToSilence(p) {
+		t.Fatal("1886: a plan failed by a build was read as lost to silence")
+	}
+}
+
+// A walk lost to silence is refused its resume beside a newer open plan of its repository and branch, which supersedes
+// it, and beside a related open plan: one plan at a time (ADR 0531).
+func TestRetryRefusesAWalkLostToSilenceBesideANewerOrARelatedOpenPlan(t *testing.T) {
+	p := r4eLost()
+	created := time.Date(2026, 10, 22, 22, 15, 9, 0, time.UTC)
+	p.Repository, p.Branch, p.Commit, p.Created = "novox/mesh-sdk", "main", "7aacbc16", created
+	newerOpen := inventory.Plan{ID: "plan-newer", Repository: "novox/mesh-sdk", Branch: "main", Commit: "d00dfeed",
+		Created: created.Add(time.Hour), State: inventory.PlanBuilding, Tiers: [][]string{{"zsh"}},
+		Modules: map[string]*inventory.PlanModule{"zsh": {State: "asked"}}}
+	if err := retryRefusal(p, []inventory.Plan{p, newerOpen}, ownPathOf(nil)); err == nil ||
+		!strings.Contains(err.Error(), "plan-newer supersedes it") {
+		t.Errorf("1886: a walk lost to silence was resumed under a newer open plan of its branch: %v", err)
+	}
+	beside := inventory.Plan{ID: "plan-beside", Repository: "novox/mesh-sdk", Branch: "release", Commit: "bbadbb2a",
+		Created: created.Add(time.Hour), State: inventory.PlanBuilding, Tiers: [][]string{{"mesh-build-image"}},
+		Modules: map[string]*inventory.PlanModule{"mesh-build-image": {State: "asked"}}}
+	if err := retryRefusal(p, []inventory.Plan{p, beside}, ownPathOf(nil)); err == nil ||
+		!strings.Contains(err.Error(), "one walk at a time") {
+		t.Errorf("1886: a walk lost to silence was resumed beside a related open plan: %v", err)
+	}
+}

internal/identity/fortest.go

@@ -37,8 +37,7 @@ func ForTest(t *testing.T) *Identity {
 	}
 	store.CreateMigratedForTest(t, admin, name, migrations)
 
-	cut := strings.LastIndex(admin, "/")
-	t.Setenv(store.Variable(Name), admin[:cut]+"/"+name+"?sslmode=disable")
+	t.Setenv(store.Variable(Name), store.DatabaseForTest(admin, name))
 
 	ident, err := Open(t.Context())
 	if err != nil {

internal/identity/identity_test.go

@@ -7,61 +7,19 @@ import (
 	"context"
 	"crypto/ed25519"
 	"errors"
-	"fmt"
 	"os"
-	"strings"
 	"sync"
 	"testing"
-	"time"
 
 	"github.com/jackc/pgx/v5"
-	"github.com/novox/mesh-controller/internal/store"
 )
 
+// fresh is the package's own name for ForTest: one fixture for every store test of identity, so the way
+// a test's database is made and addressed is kept true in one place (novox/hq issue 1889: this copy
+// dropped the store address's query and reached no server on the checker's socket-addressed store).
 func fresh(t *testing.T) *Identity {
 	t.Helper()
-	admin := os.Getenv("MESH_TEST_POSTGRES")
-	if admin == "" {
-		t.Skip("no MESH_TEST_POSTGRES; run `make check` to raise one")
-	}
-	name := fmt.Sprintf("ident_%d", time.Now().UnixNano()%10_000_000)
-
-	conn, err := pgx.Connect(t.Context(), admin)
-	if err != nil {
-		t.Fatalf("cannot reach the test PostgreSQL: %v", err)
-	}
-	if _, err := conn.Exec(t.Context(), "create database "+name); err != nil {
-		t.Fatalf("cannot create %s: %v", name, err)
-	}
-	conn.Close(t.Context())
-
-	cut := strings.LastIndex(admin, "/")
-	t.Setenv(store.Variable(Name), admin[:cut]+"/"+name+"?sslmode=disable")
-
-	ident, err := Open(t.Context())
-	if err != nil {
-		t.Fatal(err)
-	}
-	t.Cleanup(func() {
-		ident.Close()
-		c, err := pgx.Connect(context.Background(), admin)
-		if err != nil {
-			return
-		}
-		defer c.Close(context.Background())
-		_, _ = c.Exec(context.Background(), "drop database if exists "+name+" with (force)")
-	})
-	if err := ident.Ready(t.Context(), 20*time.Second); err != nil {
-		t.Fatal(err)
-	}
-	migrations, err := Migrations()
-	if err != nil {
-		t.Fatal(err)
-	}
-	if _, err := ident.store.Migrate(t.Context(), migrations); err != nil {
-		t.Fatal(err)
-	}
-	return ident
+	return ForTest(t)
 }
 
 func TestNoKeyIsAnErrorRatherThanAnEmptyKey(t *testing.T) {

internal/inventory/chain_rollouts.go

@@ -675,6 +675,12 @@ func tierLeftOf(old *Plan, p Plan, now time.Time) (time.Time, *Duration) {
 		since = old.Created
 	}
 	wasOpen := old.State == PlanBuilding || old.State == PlanRolling
+	if !wasOpen && p.Open() {
+		// **A walk set open again enters its tier anew** (issue 1886): a resume or a retry at the tier it ended at. Kept,
+		// the entry of before its stop measured the tier from hours before the resume — R4e read "waiting … for 5h48m53s"
+		// seven minutes after it was resumed — and the watchdog of its progress (S3) found it late at once.
+		return now, nil
+	}
 	if !wasOpen || (old.Tier == p.Tier && p.Open()) {
 		return since, nil // still in the tier, or already ended
 	}

internal/inventory/fortest.go

@@ -42,8 +42,7 @@ func ForTest(t *testing.T) *Inventory {
 	store.CreateMigratedForTest(t, admin, name, migrations)
 	busPasswordCostForTest.Store(int32(bcrypt.MinCost))
 
-	cut := strings.LastIndex(admin, "/")
-	t.Setenv(store.Variable(Name), admin[:cut]+"/"+name+"?sslmode=disable")
+	t.Setenv(store.Variable(Name), store.DatabaseForTest(admin, name))
 
 	inv, err := Open(t.Context())
 	if err != nil {
@@ -113,8 +112,7 @@ func ForTestBefore(t *testing.T, number int) (*Inventory, func()) {
 	}
 	store.CreateMigratedForTest(t, admin, name, before)
 	busPasswordCostForTest.Store(int32(bcrypt.MinCost))
-	cut := strings.LastIndex(admin, "/")
-	t.Setenv(store.Variable(Name), admin[:cut]+"/"+name+"?sslmode=disable")
+	t.Setenv(store.Variable(Name), store.DatabaseForTest(admin, name))
 	inv, err := Open(t.Context())
 	if err != nil {
 		t.Fatal(err)

internal/inventory/tier_entered_resume_1886_test.go

@@ -0,0 +1,32 @@
+package inventory
+
+import (
+	"testing"
+	"time"
+)
+
+// Issue 1886: R4e entered tier 1 at 02:01, was stopped, and was resumed at 07:43 at the same tier. Its save kept the
+// tier's entry of 02:01, so at 07:50 the plan read "waiting for mesh-build-image on shanks to be applied for 5h48m53s"
+// and its progress watchdog measured the tier from before the stop. A walk set open again at the tier it ended at
+// enters that tier anew: its time in the tier runs from the resume.
+func TestAWalkResumedAtItsTierEntersItAnew(t *testing.T) {
+	entered := time.Date(2026, 10, 23, 0, 1, 7, 0, time.UTC)
+	resumed := time.Date(2026, 10, 23, 5, 43, 30, 0, time.UTC)
+	old := &Plan{ID: "plan-r4e", State: PlanFailed, Tier: 1, Tiers: [][]string{{"a"}, {"b"}, {"c"}},
+		Created: entered.Add(-2 * time.Hour), TierEntered: entered}
+	p := *old
+	p.State = PlanBuilding
+	got, d := tierLeftOf(old, p, resumed)
+	if !got.Equal(resumed) || d != nil {
+		t.Fatalf("1886: a walk resumed at tier 1 entered it at %s (duration %v); want %s, no tier left", got, d, resumed)
+	}
+	// A walk still in its tier keeps its entry; one that ended keeps it too.
+	open := *old
+	open.State = PlanRolling
+	if got, _ := tierLeftOf(&open, open, resumed); !got.Equal(entered) {
+		t.Fatalf("an open walk still in its tier was stamped anew: %s", got)
+	}
+	if got, _ := tierLeftOf(old, *old, resumed); !got.Equal(entered) {
+		t.Fatalf("a failed walk saved again was stamped anew: %s", got)
+	}
+}

internal/licences/fortest.go

@@ -40,8 +40,7 @@ func ForTest(t *testing.T) *Licences {
 	}
 	store.CreateMigratedForTest(t, admin, name, migrations)
 
-	cut := strings.LastIndex(admin, "/")
-	t.Setenv(store.Variable(Name), admin[:cut]+"/"+name+"?sslmode=disable")
+	t.Setenv(store.Variable(Name), store.DatabaseForTest(admin, name))
 
 	held, err := Open(t.Context())
 	if err != nil {

internal/store/fortest.go

@@ -110,12 +110,7 @@ func raiseTemplate(ctx context.Context, admin, name string, key int64, migration
 	if _, err := conn.Exec(ctx, "create database "+name); err != nil {
 		return err
 	}
-	cut := strings.LastIndex(admin, "/")
-	rest := ""
-	if q := strings.Index(admin[cut:], "?"); q >= 0 {
-		rest = admin[cut+q:]
-	}
-	pool, err := pgxpool.New(ctx, admin[:cut]+"/"+name+rest)
+	pool, err := pgxpool.New(ctx, DatabaseForTest(admin, name))
 	if err != nil {
 		return err
 	}
@@ -136,3 +131,19 @@ func raiseTemplate(ctx context.Context, admin, name string, key int64, migration
 	}
 	return nil
 }
+
+// DatabaseForTest is the address of the database name on the same test PostgreSQL that admin reaches: admin
+// with its database replaced and its query kept.
+//
+// **The query is kept, never replaced** (novox/hq issue 1889): the checker's integration store answers on a
+// unix socket named in the query (`?host=<socket directory>&sslmode=disable`), and an address that drops it
+// reaches pgx's default socket, where no server is, so every store test waited out its readiness bound and
+// failed. Every store fixture builds its database's address here, so none can drop it again.
+func DatabaseForTest(admin, name string) string {
+	cut := strings.LastIndex(admin, "/")
+	query := ""
+	if q := strings.Index(admin[cut:], "?"); q >= 0 {
+		query = admin[cut+q:]
+	}
+	return admin[:cut] + "/" + name + query
+}

internal/store/fortest_test.go

@@ -0,0 +1,25 @@
+package store
+
+import "testing"
+
+// 1889/AC1: the checker's integration store answers on a unix socket named in the address's query
+// (mesh-build-go's store). A test database's address that drops the query reaches pgx's default socket,
+// where no server is, and every store test waits out its 20 s and fails.
+func TestATestDatabaseIsOnTheSocketItsAdminAddressNames(t *testing.T) {
+	admin := "postgres://postgres@/postgres?host=%2Fwork%2Frun-1%2Fpg%2Fsocket&sslmode=disable"
+	got := DatabaseForTest(admin, "ident_1_x")
+	want := "postgres://postgres@/ident_1_x?host=%2Fwork%2Frun-1%2Fpg%2Fsocket&sslmode=disable"
+	if got != want {
+		t.Fatalf("the test database's address is %q, want %q", got, want)
+	}
+}
+
+// 1889/AC2: the address `make check` gives (TCP, with sslmode) builds what the fixtures always built.
+func TestATestDatabaseOnTCPIsAddressedAsBefore(t *testing.T) {
+	admin := "postgres://postgres:check@127.0.0.1:55566/postgres?sslmode=disable"
+	got := DatabaseForTest(admin, "inv_2_y")
+	want := "postgres://postgres:check@127.0.0.1:55566/inv_2_y?sslmode=disable"
+	if got != want {
+		t.Fatalf("the test database's address is %q, want %q", got, want)
+	}
+}

internal/store/live_test.go

@@ -55,7 +55,7 @@ func freshStore(t *testing.T) *Store {
 	}
 	conn.Close(t.Context())
 
-	t.Setenv(Variable("testing"), replaceDatabase(dsn, name))
+	t.Setenv(Variable("testing"), DatabaseForTest(dsn, name))
 	s, err := Open(t.Context(), "testing")
 	if err != nil {
 		t.Fatal(err)
@@ -75,15 +75,6 @@ func freshStore(t *testing.T) *Store {
 	return s
 }
 
-func replaceDatabase(dsn, name string) string {
-	cut := strings.LastIndex(dsn, "/")
-	rest := ""
-	if q := strings.Index(dsn[cut:], "?"); q >= 0 {
-		rest = dsn[cut+q:]
-	}
-	return dsn[:cut] + "/" + name + rest
-}
-
 func TestTheSchemaIsAppliedAndRecorded(t *testing.T) {
 	s := freshStore(t)
 	migrations := []Migration{{Number: 1, Name: "people", SQL: "create table person (id int)", Checksum: "a"}}