Leave a kept core move out of what a gate judges (issue 201, review 91)

gatedSend took its judged moves from the record, so on a machine kept at
a newer controller the gate judged the controller moving back to the
record, read its own send's kept controller as a later build there, and
was superseded with nothing judged. The kept moves are now left out
there as in ungatedIn (withoutKept). The doc says the bus step keeps
too, and the three store tests are listed.

cmd/mesh-controller/core_downgrade.go

@@ -0,0 +1,215 @@
+package main
+
+import (
+	"context"
+	"fmt"
+	"sort"
+	"time"
+
+	"github.com/novox/mesh-controller/internal/conditions"
+	"github.com/novox/mesh-controller/internal/inventory"
+	"github.com/novox/mesh-controller/internal/lease"
+)
+
+// **No send but a person's `push <node>` and a gate's own put-back moves the mesh's own core to a build older
+// than the one its machine was last sent** (novox/hq issue 201). Every send made through sendToEach keeps it — a
+// plan's, a release's, a healer's, a rotation's, and the bus step too, which is a person's act but composes under
+// sendKeeps like the rest. On 2026-10-02 two merges a minute apart made two plans;
+// the controller's plan rolled the new controller, which widened its seat's row, and the node-engine's plan then
+// sent the control node the controller build before it. The older binary read the row it could not serve and the
+// mesh lost its voice for ten minutes — the thing that would have corrected it was the thing that was down. ADR
+// 0185 took the outage out of it (a controller behind its row serves what it can); the race is closed here,
+// without needing its trigger: whatever made the mesh's record older, the send keeps the build.
+//
+// **Kept, not refused.** The send composes the controller or the node-engine at the build the machine was last
+// sent, as a recorded module is kept (ADR 0245), and carries everything else: no other module, secret, healer or
+// walk waits on it, and nothing is retried. The kept move is no move for a gate either: ungatedIn and gatedSend
+// leave it out (withoutKept), so a gate judges the other moves of its send. A plan whose own build is the older one finds its gate superseded by
+// the build kept there (issue 352) and ends. The mesh's record stays older, and a warning condition says so until
+// a send carries the record — a person's `push <node>` accepts the older build — or the record moves forward: a
+// newer merge, or `rebuild mesh-controller`.
+//
+// Older is by the builds' registration order — when each was asked (novox/hq issue 219) — never by the commits'
+// dates. So a retry or a rebuild of an old commit, asked anew, counts as newer: it is registered as the module's
+// current build by the same order, and the guard does not stop that move. A build that failed its gate is never
+// ahead: the machine was sent it and the record was put back before it (gateFailed), and if the put-back's own
+// send was refused, the next send completes it rather than keeping the failed build there. A gate's own put-back
+// carries its modules back whatever the order (sendScope.puttingBack); the controller's is held to the store's
+// schema already (issue 352).
+
+// kindCoreDowngrade is a core module kept at the build its machine was last sent, over an older record.
+const kindCoreDowngrade = "core-downgrade-kept"
+
+// coreDowngradeSource is the condition's source: the sends, which raise and clear it.
+const coreDowngradeSource = "sends"
+
+// downgradeGuarded is the core components a send may not move backwards: the controller, which reads the
+// store and its seat's row, and the node-engine, which reads what the controller declares.
+var downgradeGuarded = []string{lease.ComponentController, lease.ComponentEngine}
+
+// askedBuild is what the order needs of one build of a module: when it was asked, and whether it failed its gate.
+type askedBuild struct {
+	asked      time.Time
+	failedGate bool
+}
+
+// buildAsked is the newest build of a module from a commit that worked — when it was asked (or recorded, when
+// the request time is not known), and whether it failed its gate — and false when the records hold none. A
+// variable, so a test reads its own order.
+var buildAsked = func(ctx context.Context, inv *inventory.Inventory, module, commit string) (askedBuild, bool, error) {
+	builds, err := inv.Builds(ctx, module, askedLookBack)
+	if err != nil {
+		return askedBuild{}, false, err
+	}
+	for _, b := range builds {
+		if b.Worked() && sameCommit(b.Commit, commit) {
+			failed, err := inv.GateFailed(ctx, b.ID)
+			if err != nil {
+				return askedBuild{}, false, err
+			}
+			return askedBuild{asked: b.AskedOrAt(), failedGate: failed}, true, nil
+		}
+	}
+	return askedBuild{}, false, nil
+}
+
+// askedLookBack is how many of a module's newest builds are read for when a build was asked.
+const askedLookBack = 50
+
+// coreDowngrade is one core module a machine was last sent a build of that is newer than the mesh's record.
+type coreDowngrade struct {
+	module, node, component string
+	// sent is the build the machine was last sent, record the one the mesh holds.
+	sent, record           string
+	sentAsked, recordAsked time.Time
+}
+
+func (d coreDowngrade) String() string {
+	return fmt.Sprintf("%s was last sent %s %s, and the mesh holds the older %s (asked %s, before %s): a send there "+
+		"keeps %s; `push %s` sends the older build, and a newer merge or `rebuild %s` moves the record forward "+
+		"(novox/hq issue 201)", d.node, d.module, short(d.sent), short(d.record),
+		d.recordAsked.UTC().Format(time.RFC3339), d.sentAsked.UTC().Format(time.RFC3339), short(d.sent), d.node,
+		d.module)
+}
+
+// coreDowngrades is, for one machine, each guarded core module whose record is older than the build the machine
+// was last sent, leaving out what a gate's own put-back puts back and a sent build that failed its gate. A move
+// whose either build the records do not hold is not proved older and is left to the gate that judges it. Pure
+// but for asked.
+func coreDowngrades(node string, sent map[string]string, record map[string]string, scope sendScope,
+	asked func(module, commit string) (askedBuild, bool, error)) ([]coreDowngrade, error) {
+	var out []coreDowngrade
+	for module, was := range sent {
+		component := coreComponent(module)
+		now := record[module]
+		if !guarded(component) || was == "" || now == "" || sameCommit(was, now) {
+			continue
+		}
+		if scope.puttingBack && scope.modules[module] {
+			continue
+		}
+		from, known, err := asked(module, was)
+		if err != nil {
+			return nil, fmt.Errorf("when %s's build %s was asked cannot be read: %w", module, short(was), err)
+		}
+		if !known || from.failedGate {
+			continue
+		}
+		to, known, err := asked(module, now)
+		if err != nil {
+			return nil, fmt.Errorf("when %s's build %s was asked cannot be read: %w", module, short(now), err)
+		}
+		if known && to.asked.Before(from.asked) {
+			out = append(out, coreDowngrade{module: module, node: node, component: component, sent: was, record: now,
+				sentAsked: from.asked, recordAsked: to.asked})
+		}
+	}
+	sort.Slice(out, func(i, j int) bool { return out[i].module < out[j].module })
+	return out, nil
+}
+
+func guarded(component string) bool {
+	for _, c := range downgradeGuarded {
+		if c == component {
+			return true
+		}
+	}
+	return false
+}
+
+// coreKept is, for a send that keeps (sendKeeps), each guarded core module of a machine to keep at the build it
+// was last sent, module → commit, each said and raised as a condition.
+func coreKept(ctx context.Context, inv *inventory.Inventory, node string, sent map[string]string,
+	current map[string]inventory.CurrentBuild) (map[string]string, error) {
+	record := map[string]string{}
+	for m, c := range current {
+		record[m] = c.Commit
+	}
+	found, err := coreDowngrades(node, sent, record, scopeOf(ctx), func(module, commit string) (askedBuild, bool, error) {
+		return buildAsked(ctx, inv, module, commit)
+	})
+	if err != nil || len(found) == 0 {
+		return nil, err
+	}
+	out := map[string]string{}
+	for _, d := range found {
+		out[d.module] = d.sent
+		sayCoreDowngrade(ctx, d)
+	}
+	return out, nil
+}
+
+// coreDowngradeObservation is a kept core build as a condition: `core.<component>.<machine>.<kind>`.
+func coreDowngradeObservation(component, node, summary string) conditions.Observation {
+	return conditions.Observation{Scope: conditions.ScopeCore, ID: component + "." + node, Kind: kindCoreDowngrade,
+		Token: kindCoreDowngrade, Machine: node, Severity: conditions.Warning, Source: coreDowngradeSource,
+		Summary: summary}
+}
+
+func sayCoreDowngrade(ctx context.Context, d coreDowngrade) {
+	fmt.Printf("KEPT: %s\n", d)
+	keeper := conditionsFrom
+	if keeper == nil {
+		return
+	}
+	if _, err := keeper.Observe(ctx, coreDowngradeObservation(d.component, d.node, d.String())); err != nil {
+		fmt.Printf("the condition for %s kept on %s could not be raised: %v\n", d.module, d.node, err)
+	}
+}
+
+// settleCoreDowngrades clears, after a send to a machine is away, the condition of each guarded core module the
+// send carried at the mesh's record — a person's push that accepted the older build, or a send after the record
+// moved forward — or did not carry at all. A send that kept the newer build leaves it open.
+func settleCoreDowngrades(ctx context.Context, inv *inventory.Inventory, node string, carried map[string]string) {
+	keeper := conditionsFrom
+	if keeper == nil || carried == nil {
+		return
+	}
+	var open []string
+	for _, c := range downgradeGuarded {
+		if _, found, err := keeper.Get(ctx, coreDowngradeObservation(c, node, "").Key()); err == nil && found {
+			open = append(open, c)
+		}
+	}
+	if len(open) == 0 {
+		return
+	}
+	current, err := inv.CurrentBuilds(ctx)
+	if err != nil {
+		fmt.Printf("whether %s still keeps a core build over the mesh's record cannot be read: %v\n", node, err)
+		return
+	}
+	for module, now := range current {
+		component := coreComponent(module)
+		if !guarded(component) {
+			continue
+		}
+		if was, has := carried[module]; has && !sameCommit(was, now.Commit) {
+			continue
+		}
+		key := coreDowngradeObservation(component, node, "").Key()
+		if _, err := keeper.Clear(ctx, key, node+" was sent the "+component+" the mesh holds"); err != nil {
+			fmt.Printf("the condition %s could not be cleared: %v\n", key, err)
+		}
+	}
+}

cmd/mesh-controller/gate.go

@@ -1039,7 +1039,7 @@ func gateFailed(ctx context.Context, open *stores, p *inventory.Plan, module str
 		}
 		return
 	}
-	sent, err := sendRollout(withScope(ctx, sendScope{modules: map[string]bool{module: true}}), open, g.Machines)
+	sent, err := sendRollout(withScope(ctx, sendScope{modules: map[string]bool{module: true}, puttingBack: true}), open, g.Machines)
 	if err != nil {
 		notBack(fmt.Sprintf("its registered build is back at %s, and sending it to %s was refused: %v — `push %s` "+
 			"sends it", short(previous.Commit), strings.Join(g.Machines, ", "), err, g.Machines[0]))
@@ -1132,7 +1132,7 @@ func sendRollbacks(ctx context.Context, open *stores, p *inventory.Plan, b *roll
 	}
 	inv := open.inventory
 	sort.Strings(b.machines)
-	sent, err := sendRollout(withScope(ctx, sendScope{modules: b.modules}), open, b.machines)
+	sent, err := sendRollout(withScope(ctx, sendScope{modules: b.modules, puttingBack: true}), open, b.machines)
 	var back []string
 	for _, r := range b.pending {
 		if err != nil {

cmd/mesh-controller/issue201_test.go

@@ -0,0 +1,259 @@
+package main
+
+import (
+	"context"
+	"encoding/json"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/novox/mesh-controller/internal/catalogue"
+	"github.com/novox/mesh-controller/internal/inventory"
+	"github.com/novox/mesh-controller/internal/lease"
+)
+
+// novox/hq issue 201: on 2026-10-02 a plan sent the control node a controller older than the one running there,
+// whose seat row the newer one had already widened; the older binary refused the row and the mesh went quiet for
+// ten minutes. No send but a person's push and a gate's own put-back moves the controller or the node-engine back
+// behind the build its machine was last sent: the send keeps it there, and carries the rest.
+
+// askedIn answers a build from a fixed table: module@commit → when it was asked; failed names those that failed
+// their gate.
+func askedIn(table map[string]time.Time, failed ...string) func(module, commit string) (askedBuild, bool, error) {
+	return func(module, commit string) (askedBuild, bool, error) {
+		at, known := table[module+"@"+commit]
+		b := askedBuild{asked: at}
+		for _, f := range failed {
+			if f == module+"@"+commit {
+				b.failedGate = true
+			}
+		}
+		return b, known, nil
+	}
+}
+
+func TestACoreModuleIsKeptOverAnOlderRecord(t *testing.T) {
+	old, now := t0, t0.Add(time.Minute)
+	table := map[string]time.Time{
+		"mesh-controller@c1": old, "mesh-controller@c2": now,
+		"mesh-host@h1": old, "mesh-host@h2": now,
+		"app@a1": old, "app@a2": now,
+	}
+	sent := map[string]string{"mesh-controller": "c2", "mesh-host": "h2", "app": "a2"}
+	record := map[string]string{"mesh-controller": "c1", "mesh-host": "h1", "app": "a1"}
+	found, err := coreDowngrades("anchor", sent, record, sendScope{judged: map[string]bool{"anchor": true}}, askedIn(table))
+	if err != nil {
+		t.Fatal(err)
+	}
+	if len(found) != 2 || found[0].module != "mesh-controller" || found[0].component != lease.ComponentController ||
+		found[0].sent != "c2" || found[1].module != "mesh-host" || found[1].component != lease.ComponentEngine {
+		t.Fatalf("the controller and the node-engine over an older record are kept, and nothing else: %+v", found)
+	}
+	if said := found[0].String(); !strings.Contains(said, "anchor was last sent mesh-controller c2") ||
+		!strings.Contains(said, "`push anchor`") || !strings.Contains(said, "`rebuild mesh-controller`") {
+		t.Fatalf("it says the machine, the build kept and the two ways out: %s", said)
+	}
+
+	// The record forward, the same build, or a build the records do not hold: nothing kept.
+	for _, c := range [][2]string{{"c1", "c2"}, {"c2", "c2"}, {"c2", "c9"}, {"c9", "c1"}} {
+		found, err := coreDowngrades("anchor", map[string]string{"mesh-controller": c[0]},
+			map[string]string{"mesh-controller": c[1]}, sendScope{}, askedIn(table))
+		if err != nil || len(found) != 0 {
+			t.Fatalf("sent %s, record %s: no downgrade proved, nothing kept: %+v %v", c[0], c[1], found, err)
+		}
+	}
+
+	// A sent build that failed its gate is never ahead: the record put back before it is sent.
+	if found, _ := coreDowngrades("anchor", sent, record, sendScope{}, askedIn(table, "mesh-controller@c2")); len(found) != 1 ||
+		found[0].module != "mesh-host" {
+		t.Fatalf("a controller that failed its gate was kept over the build put back: %+v", found)
+	}
+
+	// A gate's own put-back carries its own modules back, and nothing else.
+	putBack := sendScope{modules: map[string]bool{"mesh-controller": true}, puttingBack: true}
+	if found, _ := coreDowngrades("anchor", sent, record, putBack, askedIn(table)); len(found) != 1 || found[0].module != "mesh-host" {
+		t.Fatalf("a put-back of the controller carries it back, and not the node-engine: %+v", found)
+	}
+	if found, _ := coreDowngrades("anchor", sent, record, sendScope{modules: map[string]bool{"mesh-controller": true}},
+		askedIn(table)); len(found) != 2 {
+		t.Fatalf("a send for the controller that is no put-back carries it back: %+v", found)
+	}
+}
+
+// core201 is a mesh whose control node, anchor, was sent the newer controller c2 while the mesh's record of the
+// controller is the build before it, c1, which passed its gate when it rolled out.
+func core201(t *testing.T) *stores {
+	t.Helper()
+	open := aMesh(t)
+	ctx := t.Context()
+	inv := open.inventory
+	old, now := time.Now().Add(-2*time.Hour), time.Now().Add(-time.Minute)
+	record := func(commit string, asked time.Time) inventory.Build {
+		manifest, _ := json.Marshal(catalogue.Manifest{Module: "mesh-controller", Version: "1"})
+		b := inventory.Build{ID: "build-mesh-controller-" + commit, Module: "mesh-controller", Commit: commit,
+			Repository: "novox/mesh-controller", Manifest: manifest, Asked: asked, At: asked,
+			Made: []inventory.Artifact{{Name: "x", Kind: "bundle", Reference: "sha256:" + commit + "0000000000000000"}}}
+		if err := inv.RecordBuild(ctx, b); err != nil {
+			t.Fatal(err)
+		}
+		return b
+	}
+	before := record("c1aaaaaaaaaa", old)
+	if err := inv.RecordGate(ctx, inventory.GateVerdict{Build: before.ID, Module: "mesh-controller", Commit: before.Commit,
+		Machines: []string{"anchor"}, Verdict: inventory.GatePassed}); err != nil {
+		t.Fatal(err)
+	}
+	record("c2bbbbbbbbbb", now)
+	if err := inv.RegisterModule(ctx, catalogue.Manifest{Module: "mesh-controller", Version: "1"}, inventory.Source{
+		Repository: "novox/mesh-controller", BuiltFrom: "c2bbbbbbbbbb", Head: "c2bbbbbbbbbb", Asked: now}); err != nil {
+		t.Fatal(err)
+	}
+	if _, err := inv.Assign(ctx, "anchor", "mesh-controller"); err != nil {
+		t.Fatal(err)
+	}
+	if err := inv.RecordSent(ctx, nodeID(t, open, "anchor"), "d-anchor", map[string]string{"mesh-controller": "c2bbbbbbbbbb"}); err != nil {
+		t.Fatal(err)
+	}
+	if err := inv.RestoreModule(ctx, before); err != nil {
+		t.Fatal(err)
+	}
+	wasMoves := machineMoves
+	machineMoves = func(ctx context.Context, open *stores, f moveFacts, node string, all bool) ([]inventory.CarriedMove, error) {
+		modules, err := open.inventory.Assigned(ctx, node)
+		if err != nil {
+			return nil, err
+		}
+		sent, known, err := open.inventory.SentBuilds(ctx, node)
+		if err != nil {
+			return nil, err
+		}
+		return f.moves(node, modules, sent, known, all), nil
+	}
+	t.Cleanup(func() { machineMoves = wasMoves })
+	return open
+}
+
+// TestReplay201 replays the race on a store: the control node runs the newer controller and the mesh's record is
+// the build before it. A plan's send there — the node-engine's plan of 2026-10-02 — goes, keeps the newer
+// controller, and says so as a condition; the gate's own put-back carries it back; a send that carries the record,
+// as a person's push does, clears the condition.
+func TestReplay201(t *testing.T) {
+	open := core201(t)
+	ctx := t.Context()
+	k, _ := withConditionsInMemory(t)
+
+	judged := withScope(ctx, sendScope{judged: map[string]bool{"anchor": true}})
+	if kept, err := ungatedIn(judged, open, []string{"anchor"}, ""); err != nil || len(kept) != 1 {
+		t.Fatalf("a plan's send to the machine is made, the rest of it with it: %v %v", kept, err)
+	}
+	keeps, err := recordedKept(keepingRecorded(judged), open, "anchor")
+	if err != nil || keeps["mesh-controller"] != "c2bbbbbbbbbb" {
+		t.Fatalf("the send keeps the newer controller there: %v %v", keeps, err)
+	}
+	shelf := map[string]catalogue.Manifest{"mesh-controller": {Module: "mesh-controller", Version: "old"}}
+	if _, err := keepRecorded(keepingRecorded(judged), open, "anchor", shelf); err != nil || shelf["mesh-controller"].Version != "1" {
+		t.Fatalf("the send composes the kept build's manifest: %+v %v", shelf, err)
+	}
+	key := coreDowngradeObservation(lease.ComponentController, "anchor", "").Key()
+	if c, found, err := k.Get(ctx, key); err != nil || !found || !strings.Contains(c.Summary, "c2bbbbbb") {
+		t.Fatalf("the condition %s is raised, saying the build kept: %+v %v %v", key, c, found, err)
+	}
+
+	putBack := keepingRecorded(withScope(ctx, sendScope{modules: map[string]bool{"mesh-controller": true}, puttingBack: true}))
+	if keeps, err := recordedKept(putBack, open, "anchor"); err != nil || len(keeps) != 0 {
+		t.Fatalf("the gate's own put-back of the controller carries it back: %v %v", keeps, err)
+	}
+
+	settleCoreDowngrades(ctx, open.inventory, "anchor", map[string]string{"mesh-controller": "c2bbbbbbbbbb"})
+	if _, found, _ := k.Get(ctx, key); !found {
+		t.Fatal("a send that kept the newer controller cleared the condition")
+	}
+	settleCoreDowngrades(ctx, open.inventory, "anchor", map[string]string{"mesh-controller": "c1aaaaaaaaaa"})
+	if _, found, _ := k.Get(ctx, key); found {
+		t.Fatal("a send carrying the mesh's record — a person's push — leaves the condition open")
+	}
+}
+
+// The review of #273: a gate failed the newer controller, put the record back, and the put-back's own send was
+// refused. The next automatic send — a healer's, which judges nothing — completes the put-back: a build that
+// failed its gate is never kept.
+func TestARollbackWhoseSendWasRefusedIsCompletedByTheNextSend(t *testing.T) {
+	open := core201(t)
+	ctx := t.Context()
+	k, _ := withConditionsInMemory(t)
+	if err := open.inventory.RecordGate(ctx, inventory.GateVerdict{Build: "build-mesh-controller-c2bbbbbbbbbb",
+		Module: "mesh-controller", Commit: "c2bbbbbbbbbb", Previous: "c1aaaaaaaaaa", Machines: []string{"anchor"},
+		Verdict: inventory.GateFailed, Rollback: inventory.NotRolledBack}); err != nil {
+		t.Fatal(err)
+	}
+	healer := withScope(ctx, sendScope{})
+	if kept, err := ungatedIn(healer, open, []string{"anchor"}, ""); err != nil || len(kept) != 1 {
+		t.Fatalf("the healer's send is made: %v %v", kept, err)
+	}
+	if keeps, err := recordedKept(keepingRecorded(healer), open, "anchor"); err != nil || len(keeps) != 0 {
+		t.Fatalf("the failed controller was kept over the build put back: %v %v", keeps, err)
+	}
+	if open, _ := k.Open(ctx); len(open) != 0 {
+		t.Fatalf("a completed put-back raised a condition: %+v", open)
+	}
+}
+
+// Review 91 of #273: a gated send to a machine whose controller is kept over an older record leaves the kept
+// controller out of what its gate judges. Judged as moving back to the record, the gate would read the send's own
+// kept controller as another build sent there after it (issue 352) and be superseded, judging nothing.
+func TestAGateJudgesTheOtherMovesBesideAKeptCore(t *testing.T) {
+	open := core201(t)
+	ctx := t.Context()
+	withConditionsInMemory(t)
+	inv := open.inventory
+	old, now := time.Now().Add(-2*time.Hour), time.Now().Add(-time.Minute)
+	for _, b := range []struct {
+		commit string
+		at     time.Time
+	}{{"a1aaaaaaaaaa", old}, {"a2bbbbbbbbbb", now}} {
+		manifest, _ := json.Marshal(catalogue.Manifest{Module: "app", Version: "1"})
+		if err := inv.RecordBuild(ctx, inventory.Build{ID: "build-app-" + b.commit, Module: "app", Commit: b.commit,
+			Repository: "novox/mesh-catalog", Path: "modules/app", Manifest: manifest, Asked: b.at, At: b.at,
+			Made: []inventory.Artifact{{Name: "x", Kind: "bundle", Reference: "sha256:app-" + b.commit}}}); err != nil {
+			t.Fatal(err)
+		}
+		if err := inv.RegisterModule(ctx, catalogue.Manifest{Module: "app", Version: "1"}, inventory.Source{
+			Repository: "novox/mesh-catalog", Seat: "git", Path: "modules/app", BuiltFrom: b.commit, Head: b.commit,
+			Asked: b.at}); err != nil {
+			t.Fatal(err)
+		}
+	}
+	if _, err := inv.Assign(ctx, "anchor", "app"); err != nil {
+		t.Fatal(err)
+	}
+	if err := inv.RecordSent(ctx, nodeID(t, open, "anchor"), "d-anchor-2",
+		map[string]string{"mesh-controller": "c2bbbbbbbbbb", "app": "a1aaaaaaaaaa"}); err != nil {
+		t.Fatal(err)
+	}
+	// The send as it is made: the controller kept at c2, the app moved to the record.
+	wasSend := sendRollout
+	sendRollout = func(ctx context.Context, open *stores, names []string) ([]string, error) {
+		return names, open.inventory.RecordSent(ctx, nodeID(t, open, "anchor"), "d-anchor-3",
+			map[string]string{"mesh-controller": "c2bbbbbbbbbb", "app": "a2bbbbbbbbbb"})
+	}
+	t.Cleanup(func() { sendRollout = wasSend })
+
+	moves, sent, _, err := gatedSend(ctx, open, "anchor", nil)
+	if err != nil || len(sent) != 1 {
+		t.Fatalf("the gated send is made: %v %v", sent, err)
+	}
+	if len(moves) != 1 || moves[0].Module != "app" || moves[0].To != "a2bbbbbbbbbb" {
+		t.Fatalf("the gate judges the app's move and not the kept controller: %+v", moves)
+	}
+	builds, _, err := inv.SentBuilds(ctx, "anchor")
+	if err != nil {
+		t.Fatal(err)
+	}
+	g := &inventory.PlanGate{Carried: moves}
+	f := gateFacts{commits: judgedCommits(g, nil), sentBuilds: map[string]map[string]string{"anchor": builds}}
+	for _, m := range []string{"app", "mesh-controller"} {
+		if why, superseded := f.supersededOn(m, "anchor"); superseded {
+			t.Fatalf("the gate is superseded by its own send: %s", why)
+		}
+	}
+}

cmd/mesh-controller/plain_words.go

@@ -182,6 +182,20 @@ var plainWordings = map[string]func(conditions.Observation) words{
 				"register it, so it was not made.", module, m),
 			Resolved: "Resolved: " + module + " on " + m + " is settled"}
 	}),
+	// The controller or the node-engine kept at the build a machine was last sent, over an older record (novox/hq
+	// issue 201).
+	kindCoreDowngrade: worded(func(o conditions.Observation) words {
+		m := machineOr(o, "a machine")
+		component := idPart(o, 0)
+		if o.Scope != conditions.ScopeCore || !guarded(component) {
+			component = "core component"
+		}
+		return words{Headline: m + " keeps a newer " + component + " than the mesh holds",
+			Needs: fmt.Sprintf("push %s to take the older build, or a newer merge or rebuild moves the record forward.", m),
+			Explanation: fmt.Sprintf("The mesh's record of its own %s is older than the build %s was last sent, so "+
+				"the mesh's sends keep the newer one there rather than put it back.", component, m),
+			Resolved: "Resolved: " + m + " runs the " + component + " the mesh holds"}
+	}),
 	kindBindingKept: worded(func(o conditions.Observation) words {
 		m := machineOr(o, "a machine")
 		return words{Headline: "A module's data source is held on " + m,

cmd/mesh-controller/plan_ids.go

@@ -0,0 +1,26 @@
+package main
+
+import (
+	"fmt"
+	"regexp"
+)
+
+// planID is every id the mesh gives a plan: a merge's walk or batch (plan-<nanoseconds>) or a release
+// (release-<nanoseconds>).
+var planID = regexp.MustCompile(`^(plan|release)-[0-9]+$`)
+
+// planIDArguments are the arguments of plans that name a plan. wait is among them for when it is declared
+// (novox/hq issue 229); an argument not declared is refused before this is asked.
+var planIDArguments = []string{"id", "wait", "stop", "close", "retry", "go"}
+
+// plainPlanIDs refuses a plans call naming a plan by anything but a plan's id (novox/hq issue 561): checked
+// once, where the call becomes a command line, so nothing else a caller sent reaches a command or the store.
+func plainPlanIDs(given map[string]string) error {
+	for _, arg := range planIDArguments {
+		if id, named := given[arg]; named && !planID.MatchString(id) {
+			return fmt.Errorf("plans %s: %q is not a plan's id (plan-<digits> or release-<digits>, as `plans` "+
+				"lists them); nothing was done", arg, id)
+		}
+	}
+	return nil
+}

cmd/mesh-controller/plan_ids_test.go

@@ -0,0 +1,37 @@
+package main
+
+import (
+	"strings"
+	"testing"
+)
+
+// **A plan's id is checked once, where a call becomes a command line** (novox/hq issue 561): every argument of
+// plans that names a plan is refused unless it is an id the mesh gives a plan, so a quote never reaches a
+// command, nor the store behind it.
+func TestAPlanIdThatIsNoPlansIdIsRefusedBeforeACommandStarts(t *testing.T) {
+	for _, arg := range []string{"id", "stop", "close", "retry", "go"} {
+		for _, bad := range []string{"x' or '1'='1", "plan-1'", "plan-", "plan-1 2", "batch-1", "PLAN-1"} {
+			args := map[string]any{arg: bad}
+			if arg == "stop" || arg == "close" || arg == "go" {
+				args["why"] = "a test"
+			}
+			argv, err := argvFor("plans", args)
+			if err == nil {
+				t.Errorf("plans %s %q was composed: %v", arg, bad, argv)
+				continue
+			}
+			if !strings.Contains(err.Error(), "is not a plan's id") {
+				t.Errorf("plans %s %q was refused for another reason: %v", arg, bad, err)
+			}
+		}
+		for _, good := range []string{"plan-1791813626034291775", "release-1791811938267404408"} {
+			args := map[string]any{arg: good}
+			if arg == "stop" || arg == "close" || arg == "go" {
+				args["why"] = "a test"
+			}
+			if _, err := argvFor("plans", args); err != nil {
+				t.Errorf("plans %s %q was refused: %v", arg, good, err)
+			}
+		}
+	}
+}

cmd/mesh-controller/plans_follow.go

@@ -0,0 +1,158 @@
+package main
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"os"
+	"strconv"
+	"strings"
+	"time"
+
+	"github.com/novox/mesh-controller/internal/inventory"
+	"github.com/novox/mesh-controller/internal/link"
+)
+
+// Following a plan (novox/hq issue 229): `plans wait <id>` holds its answer until the plan moves, reaches its
+// next tier or ends, or until its bound passes, and says which. Before it, an agent following a rollout had
+// `plans` answer once, in prose, and wrote a loop of its own round the mesh's tools to be told the plan moved.
+//
+// **No wait is longer than a call is answered within** (link.LongestAnswer, inside the console's thirty
+// seconds): a longer one would answer "still running" and leave its caller to ask `calls` — the loop again.
+// So a caller following a plan to its end asks again; each call answers the moment the plan moves.
+
+// maxPlanWait is the longest bound a wait takes, and the bound when none is given: what a call is answered
+// within, less the time a verb's command takes to start and read the store.
+var maxPlanWait = link.LongestAnswer - planWaitStart
+
+// planWaitStart is what the verb's window allows beyond the bound: starting the command and opening the stores.
+// A store slow enough to push the read and the answer past it degrades as any long call does: the caller is
+// answered that the call is still running, and `calls` with its id says what came of it.
+const planWaitStart = 5 * time.Second
+
+// planWaitEvery is how often a wait reads the plan again.
+const planWaitEvery = time.Second
+
+// The things a wait waits until.
+var planWaitUntil = []string{"moved", "tier", "done"}
+
+// planWaitAsked checks what a wait is asked for: until (moved when empty) and a bound (a duration, or a number
+// of seconds; maxPlanWait when empty).
+func planWaitAsked(until, bound string) (string, time.Duration, error) {
+	if until == "" {
+		until = "moved"
+	}
+	known := false
+	for _, u := range planWaitUntil {
+		known = known || u == until
+	}
+	if !known {
+		return "", 0, fmt.Errorf("a wait is until %s, not %q", strings.Join(planWaitUntil, ", "), until)
+	}
+	d := maxPlanWait
+	if bound = strings.TrimSpace(bound); bound != "" {
+		var err error
+		if secs, isNumber := strconv.Atoi(bound); isNumber == nil {
+			d = time.Duration(secs) * time.Second
+		} else if d, err = time.ParseDuration(bound); err != nil {
+			return "", 0, fmt.Errorf("the bound %q is not a duration (such as 20s) or a number of seconds", bound)
+		}
+		if d <= 0 || d > maxPlanWait {
+			return "", 0, fmt.Errorf("the bound is at most %s, the longest a call is answered within; %s was asked "+
+				"— to follow a plan longer, ask again: each call answers the moment the plan moves", maxPlanWait, d)
+		}
+	}
+	return until, d, nil
+}
+
+// planEnded is a plan that will not move again: done, failed or superseded. A batch being assembled is not one.
+func planEnded(p inventory.Plan) bool { return !p.Open() && !p.Batch() }
+
+// waitOnPlan reads the plan until what it waits for happens or the bound passes, and answers the plan as last
+// read and what ended the wait: moved (any save of it), tier (its tier changed), done (it ended — which ends
+// every wait, since nothing comes after it), or bound. Moved and tier are judged against the plan as first read.
+func waitOnPlan(ctx context.Context, read func(context.Context) (inventory.Plan, error), until string,
+	bound, every time.Duration) (inventory.Plan, string, error) {
+	first, err := read(ctx)
+	if err != nil {
+		return inventory.Plan{}, "", err
+	}
+	if planEnded(first) {
+		return first, "done", nil
+	}
+	deadline := time.Now().Add(bound)
+	tick := time.NewTicker(every)
+	defer tick.Stop()
+	last := first
+	for {
+		select {
+		case <-ctx.Done():
+			return last, "", ctx.Err()
+		case <-tick.C:
+		}
+		if !time.Now().Before(deadline) {
+			return last, "bound", nil
+		}
+		p, err := read(ctx)
+		if err != nil {
+			return last, "", err
+		}
+		last = p
+		switch {
+		case planEnded(p):
+			return p, "done", nil
+		case until == "moved" && p.Revision != first.Revision:
+			return p, "moved", nil
+		case until == "tier" && p.Tier != first.Tier:
+			return p, "tier", nil
+		}
+	}
+}
+
+// planWaited is what `plans wait` answers: the plan, what ended the wait, and the line a person reads (`text`,
+// which the verb answers as its output and not again in the data).
+type planWaited struct {
+	// Came is what ended the wait: moved, tier, done, or bound when nothing did.
+	Came   string     `json:"came"`
+	Until  string     `json:"until"`
+	Waited string     `json:"waited"`
+	Plan   planStatus `json:"plan"`
+	// Revision is the plan's as read: a caller asking again with the same plan sees whether it moved since.
+	Revision int64  `json:"revision"`
+	Text     string `json:"text"`
+}
+
+func planWaitedDoc(p inventory.Plan, st planStatus, until, came string, waited time.Duration, line string) planWaited {
+	said := map[string]string{
+		"moved": "moved", "tier": "reached another tier", "done": "ended",
+		"bound": "did not " + map[string]string{"moved": "move", "tier": "reach another tier", "done": "end"}[until],
+	}[came]
+	text := fmt.Sprintf("%s %s in %s (waited until %s)\n%s — %s\n", p.ID, said, waited.Round(time.Second), until,
+		p.ID, line)
+	return planWaited{Came: came, Until: until, Waited: waited.Round(time.Millisecond).String(), Plan: st,
+		Revision: p.Revision, Text: text}
+}
+
+// planWaitCommand is `plans wait <id>`: the plan followed until it moves, or the bound passes.
+func planWaitCommand(ctx context.Context, inv *inventory.Inventory, id, untilAsked, boundAsked string) error {
+	until, bound, err := planWaitAsked(untilAsked, boundAsked)
+	if err != nil {
+		return err
+	}
+	began := time.Now()
+	p, came, err := waitOnPlan(ctx, func(ctx context.Context) (inventory.Plan, error) {
+		return inv.PlanByID(ctx, id)
+	}, until, bound, planWaitEvery)
+	if err != nil {
+		return err
+	}
+	// Said as `plans <id>` says its first line, and as `status` says the plan.
+	now, one := time.Now(), []inventory.Plan{p}
+	bounds, pause := readTierBounds(ctx, inv, now), buildSeatPause(ctx, inv, one)
+	line := planLineSaid(p, now, pause, bounds.of(p.Repository), func(module string) []string {
+		on, _ := inv.Running(ctx, module)
+		return on
+	}, waitsOf(ctx, inv, one))
+	doc := planWaitedDoc(p, planStatuses(one, now, pause, bounds)[0], until, came, now.Sub(began), line)
+	return json.NewEncoder(os.Stdout).Encode(doc)
+}

cmd/mesh-controller/plans_follow_test.go

@@ -0,0 +1,196 @@
+package main
+
+import (
+	"context"
+	"encoding/json"
+	"strings"
+	"testing"
+	"time"
+
+	"github.com/novox/mesh-controller/internal/inventory"
+)
+
+// A plan as the store would answer it at each read: the reads are counted, and each answers the next.
+type planReads struct {
+	plans []inventory.Plan
+	n     int
+}
+
+func (r *planReads) read(context.Context) (inventory.Plan, error) {
+	p := r.plans[min(r.n, len(r.plans)-1)]
+	r.n++
+	return p, nil
+}
+
+func followedPlan(rev int64, state string, tier int) inventory.Plan {
+	return inventory.Plan{ID: "plan-1", Repository: "novox/x", Commit: "abc", State: state, Tier: tier,
+		Tiers: [][]string{{"a"}, {"b"}}, Revision: rev}
+}
+
+// **`plans` waits for a plan, and says what ended the wait** (novox/hq issue 229): an agent following a
+// rollout is told when the plan moved, reached its next tier or ended, or that the bound passed — in one
+// call, without a loop of its own.
+func TestAWaitForAPlanEndsWhenWhatItWaitsForHappens(t *testing.T) {
+	cases := []struct {
+		name  string
+		until string
+		reads []inventory.Plan
+		came  string
+		times int
+	}{
+		{"moved: any save", "moved", []inventory.Plan{
+			followedPlan(1, inventory.PlanBuilding, 0), followedPlan(1, inventory.PlanBuilding, 0),
+			followedPlan(2, inventory.PlanBuilding, 0)}, "moved", 3},
+		{"tier: a save inside the tier is not enough", "tier", []inventory.Plan{
+			followedPlan(1, inventory.PlanBuilding, 0), followedPlan(2, inventory.PlanBuilding, 0),
+			followedPlan(3, inventory.PlanRolling, 1)}, "tier", 3},
+		{"tier: a plan that ended has no next tier to wait for", "tier", []inventory.Plan{
+			followedPlan(1, inventory.PlanBuilding, 0), followedPlan(2, inventory.PlanFailed, 0)}, "done", 2},
+		{"done: a new tier is not the end", "done", []inventory.Plan{
+			followedPlan(1, inventory.PlanBuilding, 0), followedPlan(2, inventory.PlanRolling, 1),
+			followedPlan(3, inventory.PlanDone, 1)}, "done", 3},
+		{"done: a plan already ended answers at once", "done", []inventory.Plan{
+			followedPlan(4, inventory.PlanSuperseded, 1)}, "done", 1},
+		{"the bound passes", "moved", []inventory.Plan{followedPlan(1, inventory.PlanBuilding, 0)}, "bound", -1},
+	}
+	for _, c := range cases {
+		t.Run(c.name, func(t *testing.T) {
+			r := &planReads{plans: c.reads}
+			p, came, err := waitOnPlan(context.Background(), r.read, c.until, 50*time.Millisecond, time.Millisecond)
+			if err != nil {
+				t.Fatal(err)
+			}
+			if came != c.came {
+				t.Fatalf("the wait ended as %q, want %q", came, c.came)
+			}
+			if c.times > 0 && r.n != c.times {
+				t.Errorf("read the plan %d times, want %d: it answered before or after what it waited for", r.n, c.times)
+			}
+			if want := c.reads[min(r.n, len(c.reads))-1]; p.Revision != want.Revision {
+				t.Errorf("answered revision %d, not the plan as it was last read (%d)", p.Revision, want.Revision)
+			}
+		})
+	}
+}
+
+// What a wait is asked for is checked before anything is read: what it waits until, and a bound no longer than
+// a call is answered within.
+func TestAWaitIsRefusedWhatItCannotDo(t *testing.T) {
+	for _, bad := range []struct{ until, bound string }{{"landed", ""}, {"", "1h"}, {"", "-3s"}, {"", "soon"}} {
+		if _, _, err := planWaitAsked(bad.until, bad.bound); err == nil {
+			t.Errorf("until %q, bound %q was taken", bad.until, bad.bound)
+		}
+	}
+	until, bound, err := planWaitAsked("", "")
+	if err != nil || until != "moved" || bound != maxPlanWait {
+		t.Fatalf("the defaults are %q %s %v, want moved and %s", until, bound, err, maxPlanWait)
+	}
+	if _, bound, _ := planWaitAsked("tier", "12"); bound != 12*time.Second {
+		t.Errorf("a bare number is seconds; read %s", bound)
+	}
+}
+
+// **The answer is data, with its prose as a field beside it** (novox/hq issue 229): the verb's `answer` carries
+// the plan and what ended the wait, and its output is the line a person reads — never JSON inside a string.
+func TestAWaitForAPlanAnswersAsData(t *testing.T) {
+	p := followedPlan(7, inventory.PlanRolling, 1)
+	st := planStatuses([]inventory.Plan{p}, time.Now(), pauseView{}, tierBounds{})[0]
+	doc := planWaitedDoc(p, st, "tier", "tier", 3*time.Second, "rolling, tier 2 of 2")
+	stdout, _ := json.Marshal(doc)
+	a := answerOf([]string{"plans", "wait", "plan-1"}, stdout, "", true)
+	data, ok := a.Answer.(map[string]any)
+	if !ok {
+		t.Fatalf("the answer is not data: %v", a.Answer)
+	}
+	if data["came"] != "tier" || data["revision"] != float64(7) {
+		t.Errorf("the data does not say what ended the wait: %v", data)
+	}
+	if plan, _ := data["plan"].(map[string]any); plan["state"] != "rolling" || plan["tier"] != float64(1) {
+		t.Errorf("the data does not carry the plan: %v", data["plan"])
+	}
+	if _, text := data["text"]; text {
+		t.Errorf("the prose is in the data too: %v", data)
+	}
+	if strings.Contains(a.Output, "{") || !strings.Contains(a.Output, "plan-1") || !strings.Contains(a.Output, "tier") {
+		t.Errorf("the output is not the line a person reads: %q", a.Output)
+	}
+}
+
+// The verb composes the wait, and asks for a window as long as its bound.
+func TestThePlansVerbComposesAWait(t *testing.T) {
+	argv, err := argvFor("plans", map[string]any{"wait": "plan-1", "until": "done", "bound": "15s"})
+	if err != nil {
+		t.Fatal(err)
+	}
+	if got := strings.Join(argv, " "); got != "plans wait plan-1 --until done --bound 15s" {
+		t.Fatalf("composed %q", got)
+	}
+	if _, err := argvFor("plans", map[string]any{"until": "done"}); err == nil {
+		t.Error("until without wait was taken")
+	}
+	if _, err := argvFor("plans", map[string]any{"wait": "plan-1", "bound": "1h"}); err == nil {
+		t.Error("a bound past what a call can wait was taken")
+	}
+}
+
+// Live, with the store: `plans wait` run as the verb runs it answers when the plan reaches its next tier, as
+// data, and says the bound passed when nothing moved (novox/hq issue 229).
+func TestAPlanFollowedFromTheStoreAnswersWhenItMoves(t *testing.T) {
+	open := aMesh(t)
+	ctx := t.Context()
+	now := time.Now().UTC()
+	walk := inventory.Plan{ID: "plan-229", Repository: "novox/mesh-catalog", Branch: "main", Commit: "c229c229",
+		Created: now, State: inventory.PlanBuilding, Tiers: [][]string{{"a"}, {"b"}},
+		Modules: map[string]*inventory.PlanModule{"a": {}, "b": {}}}
+	if err := open.inventory.SavePlan(ctx, &walk); err != nil {
+		t.Fatal(err)
+	}
+	moved := make(chan error, 1)
+	go func() {
+		time.Sleep(1500 * time.Millisecond)
+		p, err := open.inventory.PlanByID(ctx, walk.ID)
+		if err == nil {
+			p.Tier, p.State = 1, inventory.PlanRolling
+			err = open.inventory.SavePlan(ctx, &p)
+		}
+		moved <- err
+	}()
+	began := time.Now()
+	var got planWaited
+	out := printed(t, func() error {
+		return plansCommand(ctx, []string{"wait", walk.ID, "--until", "tier", "--bound", "15s"})
+	})
+	if err := <-moved; err != nil {
+		t.Fatal(err)
+	}
+	if err := json.Unmarshal([]byte(out), &got); err != nil {
+		t.Fatalf("plans wait did not answer data: %v: %q", err, out)
+	}
+	if got.Came != "tier" || got.Plan.Tier != 1 || got.Plan.State != inventory.PlanRolling {
+		t.Fatalf("the wait answered %+v", got)
+	}
+	if waited := time.Since(began); waited > 10*time.Second {
+		t.Errorf("it answered after %s, not when the plan moved", waited)
+	}
+	out = printed(t, func() error {
+		return plansCommand(ctx, []string{"wait", walk.ID, "--bound", "2"})
+	})
+	if err := json.Unmarshal([]byte(out), &got); err != nil || got.Came != "bound" || got.Until != "moved" {
+		t.Fatalf("a plan that did not move answered %+v (%v)", got, err)
+	}
+}
+
+// The verb's schema says the longest bound in words, and the words are the value (review 75 on #266): the
+// catalogue cannot read maxPlanWait, so this holds them together.
+func TestThePlansSchemaSaysTheLongestBoundThereIs(t *testing.T) {
+	v, ok := controllerVerb("plans")
+	if !ok {
+		t.Fatal("no plans verb")
+	}
+	props, _ := v.Input["properties"].(map[string]any)
+	bound, _ := props["bound"].(map[string]any)
+	said, _ := bound["description"].(string)
+	if !strings.Contains(said, maxPlanWait.String()+", the default, is the longest") {
+		t.Fatalf("the schema says %q; the longest bound is %s", said, maxPlanWait)
+	}
+}

cmd/mesh-controller/push.go

@@ -1027,6 +1027,9 @@ func (b overTheBus) declare(ctx context.Context, s readyNode, body []byte) (stri
 	if err != nil {
 		return "", err
 	}
+	// A send that carried the mesh's record of the controller and the node-engine — a person's push included —
+	// ends the condition of one kept over it (novox/hq issue 201).
+	settleCoreDowngrades(ctx, b.open.inventory, s.node, s.declared.Builds)
 	if len(s.declared.Bindings) > 0 {
 		// And where it bound each consumer of a provision that keeps its data (novox/hq ADR 0232):
 		// what the next resolution keeps it at. On the same outliving context as the send's record.

cmd/mesh-controller/recorded_kept.go

@@ -135,6 +135,17 @@ func recordedKept(ctx context.Context, open *stores, node string) (map[string]st
 		}
 		out[m] = was
 	}
+	// **And the controller and the node-engine at the build the machine was last sent, over an older record**
+	// (novox/hq issue 201, core_downgrade.go). Under keepingEveryBuild they are kept already.
+	if !every {
+		core, err := coreKept(ctx, inv, node, sent, current)
+		if err != nil {
+			return nil, err
+		}
+		for m, was := range core {
+			out[m] = was
+		}
+	}
 	return out, nil
 }
 
@@ -159,6 +170,12 @@ func keepRecorded(ctx context.Context, open *stores, node string, shelf map[stri
 			return nil, fmt.Errorf("%s runs %s's build %s, which the build records no longer hold: the bus's user "+
 				"list cannot be sent there alone with it kept (novox/hq issue 490)", node, m, short(kept[m]))
 		}
+		if !found && guarded(coreComponent(m)) {
+			return nil, fmt.Errorf("%s was last sent %s's build %s, newer than the one the mesh holds, and the build records "+
+				"no longer hold it: this send cannot keep it and does not move it back — `push %s` sends the older build on "+
+				"a person's word, and a newer merge or `rebuild %s` moves the record forward (novox/hq issue 201)",
+				node, m, short(kept[m]), node, m)
+		}
 		if !found {
 			return nil, fmt.Errorf("%s records rather than rolls out, and %s runs its build %s, which the build "+
 				"records no longer hold: this send cannot keep it and does not move it — `push %s` sends the new "+

cmd/mesh-controller/release.go

@@ -51,6 +51,9 @@ type sendScope struct {
 	modules map[string]bool
 	// person is a person's act: it carries what it carries.
 	person bool
+	// puttingBack is a gate's own put-back of its modules: the one send that may carry a core module to a build
+	// older than the one its machine runs (novox/hq issue 201).
+	puttingBack bool
 }
 
 type sendScopeKey struct{}
@@ -237,6 +240,32 @@ var machineMoves = func(ctx context.Context, open *stores, f moveFacts, node str
 	return f.moves(node, modules, sent, known, all), nil
 }
 
+// withoutKept is a machine's moves less those its send keeps at the build the machine was last sent, and so does
+// not make: the grants step keeps every module (novox/hq issue 490), and every send keeps the controller and the
+// node-engine over an older record (issue 201). Read as the send composes it (sendKeeps, set after the checks that
+// call this), so a move it could not keep is still counted.
+func withoutKept(ctx context.Context, open *stores, node string, moves []inventory.CarriedMove) ([]inventory.CarriedMove, error) {
+	if len(moves) == 0 {
+		return moves, nil
+	}
+	keeping := ctx
+	if _, on := keptExcept(ctx); !on {
+		keeping = keepingRecorded(ctx)
+	}
+	keeps, err := recordedKept(keeping, open, node)
+	if err != nil || len(keeps) == 0 {
+		return moves, err
+	}
+	var out []inventory.CarriedMove
+	for _, mv := range moves {
+		if was, kept := keeps[mv.Module]; kept && sameCommit(was, mv.From) {
+			continue
+		}
+		out = append(out, mv)
+	}
+	return out, nil
+}
+
 // ungatedIn refuses a send whose machines would move a build no gate has seen, outside its scope: the
 // machines named, and why; the machine holding the bus, added only for its user list, is left instead.
 func ungatedIn(ctx context.Context, open *stores, names []string, addedHolder string) ([]string, error) {
@@ -254,19 +283,11 @@ func ungatedIn(ctx context.Context, open *stores, names []string, addedHolder st
 		if err != nil {
 			return nil, err
 		}
-		// The grants step composes every module at the build its machine was last sent (novox/hq issue 490):
-		// a move it keeps is not made. Read, not assumed, so a move it could not keep is still refused here.
-		var keeps map[string]string
-		if everyBuildKept(ctx, n) {
-			if keeps, err = recordedKept(ctx, open, n); err != nil {
-				return nil, err
-			}
+		if moves, err = withoutKept(ctx, open, n, moves); err != nil {
+			return nil, err
 		}
 		var waiting []string
 		for _, mv := range moves {
-			if was, kept := keeps[mv.Module]; kept && sameCommit(was, mv.From) {
-				continue
-			}
 			if !scope.judged[n] && !scope.modules[mv.Module] {
 				waiting = append(waiting, fmt.Sprintf("%s %s → %s", mv.Module, short(mv.From), short(mv.To)))
 			}
@@ -308,6 +329,11 @@ func gatedSend(ctx context.Context, open *stores, node string, owns []inventory.
 	if err != nil {
 		return nil, nil, nil, err
 	}
+	// A move the send keeps is not made, and is no move for the gate to judge (novox/hq issue 201): a core module
+	// kept over an older record judged as moving back to it would read the gate as superseded by its own send.
+	if moves, err = withoutKept(ctx, open, node, moves); err != nil {
+		return nil, nil, nil, err
+	}
 	own := func(module string) bool {
 		return slices.ContainsFunc(owns, func(o inventory.CarriedMove) bool { return o.Module == module })
 	}

cmd/mesh-controller/release_plan.go

@@ -1463,6 +1463,10 @@ func plansCommand(ctx context.Context, args []string) error {
 	modules := set.String("modules", "", "or the modules it would change, comma-separated")
 	moduleDirs := set.String("module-dirs", "", "with --paths: the directories holding a module.json at the commit, "+
 		"comma-separated, as the forge's announcer says them (novox/hq issue 278)")
+	// Following a plan (novox/hq issue 229): `plans wait <id>`.
+	until := set.String("until", "", "with wait: what to wait for — moved (any change), tier (its next tier) or done (its end)")
+	bound := set.String("bound", "", "with wait: how long to wait at most, as 15s or seconds (default and at most the "+
+		"longest a call is answered within)")
 	// Ending a plan by hand is a repair, and says why (novox/hq to-be 45 §7).
 	why := addHandActFlags(set)
 	positionals, err := parseAround(set, args)
@@ -1482,6 +1486,9 @@ func plansCommand(ctx context.Context, args []string) error {
 	defer open.Close()
 	inv := open.inventory
 	now := time.Now()
+	if len(positionals) == 2 && positionals[0] == "wait" {
+		return planWaitCommand(ctx, inv, positionals[1], *until, *bound)
+	}
 	if len(positionals) == 1 {
 		p, err := inv.PlanByID(ctx, positionals[0])
 		if err != nil {

cmd/mesh-controller/seatverbs.go

@@ -336,6 +336,9 @@ func (a *verbArguments) commandLine() ([]string, error) {
 		}
 		return argv, nil
 	case "plans":
+		if err := plainPlanIDs(a.given); err != nil {
+			return nil, err
+		}
 		if r := str("repository"); r != "" {
 			argv := []string{"plans", "--what-if", r}
 			if p := str("paths"); p != "" {
@@ -365,6 +368,21 @@ func (a *verbArguments) commandLine() ([]string, error) {
 				return argv, nil
 			}
 		}
+		if id := str("wait"); id != "" {
+			// Following a plan (novox/hq issue 229): checked here too, so a call asking what no wait can do is
+			// refused before a command is started.
+			if _, _, err := planWaitAsked(a.given["until"], a.given["bound"]); err != nil {
+				return nil, err
+			}
+			argv := []string{"plans", "wait", id}
+			if u := str("until"); u != "" {
+				argv = append(argv, "--until", u)
+			}
+			if b := str("bound"); b != "" {
+				argv = append(argv, "--bound", b)
+			}
+			return argv, nil
+		}
 		if id := str("id"); id != "" {
 			return []string{"plans", id}, nil
 		}
@@ -985,12 +1003,14 @@ var jsonVerbs = map[string]bool{"status": true, "seats": true, "plan": true, "co
 	// The delivery's owner's verbs answer JSON where they read (plan, order, check, walks) — novox/hq ADR 0239.
 	"delivery": true,
 	// The build queue, each ask with its position among the waiting (novox/hq issue 455).
-	"queue": true}
+	"queue": true,
+	// A plan followed (novox/hq issue 229): `plans wait` answers JSON; its other forms say text, and so answer no data.
+	"plans": true}
 
 // tableBesideVerbs are the JSON verbs whose document carries the table a person reads as its `text`: the
 // verb answers that table as its output, as it did before it answered data, and the rest as `answer` —
 // both from the one read, so they never disagree (novox/hq issue 455).
-var tableBesideVerbs = map[string]bool{"queue": true}
+var tableBesideVerbs = map[string]bool{"queue": true, "plans": true}
 
 // overviewVerbs are the JSON verbs whose answer is said once, as data, and not again as the text the
 // command printed: the overviews, which grow with the mesh (novox/hq issue 314). Every reader of theirs
@@ -1264,6 +1284,16 @@ func seatToolHandlers() (map[string]link.ToolHandler, []string, error) {
 				// Whatever it did, `status` is composed again once it has.
 				defer statusFrom.nudge()
 			}
+			if verb == "plans" && len(argv) > 1 && argv[1] == "wait" {
+				// A wait holds its answer up to its bound (novox/hq issue 229), and is answered in full inside it.
+				asked := ""
+				if i := slices.Index(argv, "--bound"); i > 0 && i+1 < len(argv) {
+					asked = argv[i+1]
+				}
+				if _, bound, err := planWaitAsked("", asked); err == nil {
+					link.AnswersWithin(ctx, bound+planWaitStart)
+				}
+			}
 			if answer, read := readHere(ctx, argv); read {
 				return answer, nil
 			}

internal/catalogue/verbs.go

@@ -116,9 +116,15 @@ var ControllerVerbs = []Verb{
 			"limit":  "how many builds to list (default 20); not with log",
 		}, nil)},
 	{Name: "plans", Description: "What the last merges produced and where each stands (novox/hq ADR 0162): " +
-		"the tiers, the tier a plan is at, what it waits for and since when; one plan whole, given its id.",
+		"the tiers, the tier a plan is at, what it waits for and since when; one plan whole, given its id; " +
+		"or, given wait, one plan followed until it moves.",
 		Input: schema(map[string]string{
-			"id":    "a plan's id (as `plans` lists them): that plan, tier by tier",
+			"id": "a plan's id (as `plans` lists them): that plan, tier by tier",
+			"wait": "a plan's id: wait until it moves, then answer it as data — the plan, and what ended the wait " +
+				"(moved, tier, done, or bound); ask again to follow it further (novox/hq issue 229)",
+			"until": "with wait: moved (any change, the default), tier (it reaches another tier) or done (it ends)",
+			"bound": "with wait: how long to wait at most, as 15s or seconds — 20s, the default, is the longest, " +
+				"inside what a call is answered within",
 			"stop":  "a plan's id: stop it — what was asked still builds, nothing further is asked",
 			"close": "a plan's id: close a plan that will not move again, as failed by hand (novox/hq issue 254)",
 			"retry": "a failed plan's id: ask its failed builds again under new ids, and carry the plan on from that tier (novox/hq ADR 0219)",

internal/inventory/plan_id_test.go

@@ -0,0 +1,45 @@
+package inventory
+
+import (
+	"strings"
+	"testing"
+	"time"
+)
+
+// **A plan's id is a value, never SQL** (novox/hq issue 561): the id reaches PlanByID from a caller of the
+// controller's plans verb, so an id holding a quote is read as a plain id — no error from the store, and no
+// other plan answered for it.
+func TestAPlanIdHoldingAQuoteIsReadAsAPlainId(t *testing.T) {
+	inv := ForTest(t)
+	ctx := t.Context()
+	p := Plan{ID: "plan-561", Repository: "novox/mesh-controller", Commit: "abc", Created: time.Now().UTC(),
+		State: PlanBuilding, Tiers: [][]string{{"a"}}, Modules: map[string]*PlanModule{"a": {}}}
+	if err := inv.SavePlan(ctx, &p); err != nil {
+		t.Fatal(err)
+	}
+	for _, id := range []string{
+		"x' or '1'='1",            // every plan, when pasted into the query
+		"plan-561' or id like '%", // the plan above, by a pattern
+		"it's",                    // a quote alone: a syntax error, when pasted
+	} {
+		got, err := inv.PlanByID(ctx, id)
+		if err == nil {
+			t.Fatalf("PlanByID(%q) answered %s, a plan of another id", id, got.ID)
+		}
+		if want := "no plan " + id; err.Error() != want {
+			t.Fatalf("PlanByID(%q) failed with %q, not as a plain id no plan has (%q)", id, err, want)
+		}
+	}
+	// A plan whose id holds a quote is found by it, and by nothing else.
+	quoted := p
+	quoted.ID, quoted.Revision = "plan-'561'", 0
+	if err := inv.SavePlan(ctx, &quoted); err != nil {
+		t.Fatal(err)
+	}
+	if got, err := inv.PlanByID(ctx, quoted.ID); err != nil || got.ID != quoted.ID {
+		t.Fatalf("the plan with a quote in its id was not read back: %v %+v", err, got.ID)
+	}
+	if recent, err := inv.RecentPlans(ctx, 1); err != nil || len(recent) != 1 || !strings.HasPrefix(recent[0].ID, "plan-") {
+		t.Fatalf("RecentPlans with a parameter: %v %+v", err, recent)
+	}
+}

internal/inventory/plans.go

@@ -623,7 +623,7 @@ func (i *Inventory) PlansSince(ctx context.Context, since time.Time) ([]Plan, er
 
 // RecentPlans is the last few plans, newest first, open or not — what the overview shows.
 func (i *Inventory) RecentPlans(ctx context.Context, limit int) ([]Plan, error) {
-	return i.plans(ctx, fmt.Sprintf(`order by created desc limit %d`, limit))
+	return i.plans(ctx, `order by created desc limit $1`, limit)
 }
 
 // NewestMergeOf is the plan, in any state, of the newest merge into a repository's branch that the mesh
@@ -638,9 +638,10 @@ func (i *Inventory) NewestMergeOf(ctx context.Context, repository, branch string
 	return plans[0], true, nil
 }
 
-// PlanByID is one plan.
+// PlanByID is one plan. The id is a parameter, never part of the query (novox/hq issue 561): it comes from a
+// caller of the plans verb.
 func (i *Inventory) PlanByID(ctx context.Context, id string) (Plan, error) {
-	plans, err := i.plans(ctx, `where id = '`+id+`'`)
+	plans, err := i.plans(ctx, `where id = $1`, id)
 	if err != nil {
 		return Plan{}, err
 	}

internal/link/calls.go

@@ -601,6 +601,24 @@ func Acknowledge(ctx context.Context) {
 	}
 }
 
+// LongestAnswer is the longest a handler that waits on purpose may hold its caller's one answer
+// (AnswersWithin, novox/hq issue 229): inside the console's thirty seconds with room for the trip back,
+// and inside the bus's window for an answer (broker.ResponseTTL). A variable so a test need not wait.
+var LongestAnswer = 25 * time.Second
+
+type answerWithinKey struct{}
+
+// AnswersWithin asks that the call's caller be answered in full if the handler ends within d of the
+// call's start, rather than within AnswerWithin: for a handler that waits on purpose, with a bound —
+// `plans` waiting for a plan to move (novox/hq issue 229). Past d, or past LongestAnswer whichever is
+// shorter, its caller is told it is running, as for any call. Never shortens the window; without effect
+// outside a served call, and once the caller was answered.
+func AnswersWithin(ctx context.Context, d time.Duration) {
+	if f, ok := ctx.Value(answerWithinKey{}).(func(time.Duration)); ok {
+		f(d)
+	}
+}
+
 // running is the interim answer: what a caller reads when the call has not finished.
 func running(c *Call, within time.Duration, acknowledged bool, follow string) []byte {
 	why := fmt.Sprintf("it has not finished in %s", within)
@@ -645,6 +663,13 @@ func (l *CallLog) serveCallWithin(seat, verb string, args json.RawMessage, reply
 	acknowledged := make(chan struct{})
 	var once sync.Once
 	ctx = context.WithValue(ctx, answerNowKey{}, func() { once.Do(func() { close(acknowledged) }) })
+	longer := make(chan time.Duration, 1)
+	ctx = context.WithValue(ctx, answerWithinKey{}, func(d time.Duration) {
+		select {
+		case longer <- d:
+		default:
+		}
+	})
 
 	type outcome struct {
 		body   []byte
@@ -703,17 +728,29 @@ func (l *CallLog) serveCallWithin(seat, verb string, args json.RawMessage, reply
 			}
 		}
 	}
-	timer := time.NewTimer(AnswerWithin)
+	within, start := AnswerWithin, time.Now()
+	timer := time.NewTimer(within)
 	defer timer.Stop()
-	select {
-	case o := <-done:
-		l.finish(c, o.body, o.failed, false)
-		say(o.body)
-		return
-	case <-acknowledged:
-		say(running(c, AnswerWithin, true, l.Follow))
-	case <-timer.C:
-		say(running(c, AnswerWithin, false, l.Follow))
+waiting:
+	for {
+		select {
+		case o := <-done:
+			l.finish(c, o.body, o.failed, false)
+			say(o.body)
+			return
+		case <-acknowledged:
+			say(running(c, within, true, l.Follow))
+			break waiting
+		case d := <-longer:
+			// A handler waiting on purpose (AnswersWithin): its window is stretched, never past LongestAnswer.
+			if d = min(d, LongestAnswer); d > within {
+				within = d
+				timer.Reset(time.Until(start.Add(d)))
+			}
+		case <-timer.C:
+			say(running(c, within, false, l.Follow))
+			break waiting
+		}
 	}
 	o := <-done
 	l.finish(c, o.body, o.failed, true)

internal/link/calls_wait_test.go

@@ -0,0 +1,55 @@
+package link
+
+import (
+	"context"
+	"encoding/json"
+	"testing"
+	"time"
+)
+
+// **A handler that waits on purpose is answered in full when it ends inside the window it asked for**
+// (novox/hq issue 229): `plans` asked to wait for a plan holds its caller's one answer until the plan
+// moves or its bound passes, rather than being answered "still running" at AnswerWithin.
+func TestAHandlerThatWaitsOnPurposeIsAnsweredInFullInsideItsWindow(t *testing.T) {
+	shortWindow(t, 20*time.Millisecond)
+	l, a := NewCallLog(), newAnswers(t)
+	l.serveCall("mesh-controller", "plans", json.RawMessage(`{"wait":"plan-1"}`), "_INBOX.x.3",
+		func(ctx context.Context, _ json.RawMessage) (any, error) {
+			AnswersWithin(ctx, 200*time.Millisecond)
+			time.Sleep(80 * time.Millisecond) // past AnswerWithin, inside what it asked for
+			return "plan-1 moved", nil
+		}, a.respond, nil)
+	if got := a.only()["result"]; got != "plan-1 moved" {
+		t.Fatalf("a waiting handler inside its window was answered %v, not in full", got)
+	}
+}
+
+// What a handler may ask for is bounded: past LongestAnswer its caller is told it is running, as any
+// call is, so no answer is ever later than the console waits.
+func TestAWaitIsNeverLongerThanTheLongestAnswer(t *testing.T) {
+	shortWindow(t, 10*time.Millisecond)
+	was := LongestAnswer
+	LongestAnswer = 30 * time.Millisecond
+	t.Cleanup(func() { LongestAnswer = was })
+	l, a := NewCallLog(), newAnswers(t)
+	release := make(chan struct{})
+	finished := make(chan struct{})
+	go func() {
+		l.serveCall("mesh-controller", "plans", nil, "_INBOX.x.4", func(ctx context.Context, _ json.RawMessage) (any, error) {
+			AnswersWithin(ctx, time.Hour)
+			<-release
+			return "late", nil
+		}, a.respond, nil)
+		close(finished)
+	}()
+	select {
+	case <-a.sent:
+	case <-time.After(2 * time.Second):
+		t.Fatal("a handler asking for an hour held its caller past LongestAnswer")
+	}
+	if got, _ := a.only()["result"].(map[string]any); got["running"] != true {
+		t.Fatalf("past LongestAnswer the caller was answered %v, not that it is running", a.only())
+	}
+	close(release)
+	<-finished
+}

testdata/store-tests

@@ -7,12 +7,14 @@
 ./cmd/mesh-controller TestAStopNamesWhatStaysOnTheMachinesAndWhatStillCarriesIt
 ./cmd/mesh-controller TestAStopWithoutAMergeLeavesItOutOfABatch
 ./cmd/mesh-controller TestSavingPlansAndMovingMergesIsOneAct
+./cmd/mesh-controller TestAPlanFollowedFromTheStoreAnswersWhenItMoves
 ./internal/inventory TestTheLastSendSaysWhatTheMachineAnswered
 ./internal/inventory TestAReportIsTheLastSendsAnswerOnlyWhenItIsAboutThatSend
 ./cmd/mesh-controller TestNodeShowAndStatusReadTheLastSendFromTheStore
 ./internal/inventory TestAnIssuedValueIsSealedToEveryHolderOnceAndNotAgain
 ./internal/inventory TestAnIssuedValueIsSealedAgainAfterAKeyChangeOrAnAcceptance
 ./internal/inventory TestAConditionsLooksAreKeptBesideTheBucket
+./internal/inventory TestAPlanIdHoldingAQuoteIsReadAsAPlainId
 ./cmd/mesh-controller TestTheConditionsVerbReadsTheLooksTheBucketNoLongerHolds
 ./internal/inventory TestAWalksStartIsKeptAndARefusedSaveLeavesItUnsaid
 ./cmd/mesh-controller TestAWalkLetGoAndStoppedByVerbsIsSaidStartedAndEnded
@@ -29,3 +31,6 @@
 ./cmd/mesh-controller TestADirectoryOwnersKeyIsRefusedThroughAVerb
 ./cmd/mesh-controller TestTheSendMakesNoValueForAnOutsideSecretNobodyGave
 ./cmd/mesh-controller TestAQuestionComposesWithoutAnOutsideSecretNobodyGave
+./cmd/mesh-controller TestReplay201
+./cmd/mesh-controller TestARollbackWhoseSendWasRefusedIsCompletedByTheNextSend
+./cmd/mesh-controller TestAGateJudgesTheOtherMovesBesideAKeptCore