feat(engine): per-intent LLM chain-head rotation (opt-in llm_rotate)

Implements the per-intent rotation shape from #85: a task created with
llm_rotate=true distributes worker intents round-robin over its LLM chain
heads, so lane A can hunt on one model while lane B hunts on another
(different lenses per intent, no extra tokens). Default stays the classic
ordered failover from the active head; planner/goals/mainagent keep the
active head always.

Mechanics: the engine pins the next chain head into the run context when a
worker claims an intent; the pin travels through the provider resolver
(hooks.current now takes ctx). A pin is only a dispatch preference - it must
belong to the task chain and never bypasses failover state (an exhausted
chain stays exhausted). Schema: tasks.llm_rotate boolean, default false,
idempotent ALTER; exposed on the create payload and the task DTO.

Closes #85 (the ensemble/P4 shape stays parked as before)

db/schema.sql

@@ -522,6 +522,7 @@ CREATE TABLE IF NOT EXISTS tasks (
     llm_profile_id BIGINT REFERENCES llm_profiles(id) ON DELETE SET NULL,
     active_llm_profile_id BIGINT REFERENCES llm_profiles(id) ON DELETE SET NULL,
     llm_chain_revision BIGINT NOT NULL DEFAULT 0,
+    llm_rotate     BOOLEAN NOT NULL DEFAULT false,
     company_id     BIGINT REFERENCES companies(id) ON DELETE SET NULL,
     parent_ref     TEXT,
     timeout_seconds INTEGER NOT NULL DEFAULT 0,
@@ -545,6 +546,7 @@ CREATE TRIGGER trg_tasks_upd BEFORE UPDATE ON tasks
 -- planner heartbeat trigger interval (seconds); backfills old databases. Default 300s (5min). See docs/planner-trigger-impl-plan.md
 ALTER TABLE tasks ADD COLUMN IF NOT EXISTS plan_heartbeat_seconds INTEGER NOT NULL DEFAULT 300;
 ALTER TABLE tasks ADD COLUMN IF NOT EXISTS plan_quiet_mode BOOLEAN NOT NULL DEFAULT true;
+ALTER TABLE tasks ADD COLUMN IF NOT EXISTS llm_rotate BOOLEAN NOT NULL DEFAULT false;
 -- Concurrency-limit hold state; backfills old databases. true=queued due to the concurrency limit, auto-starts when a slot frees up.
 ALTER TABLE tasks ADD COLUMN IF NOT EXISTS queued BOOLEAN NOT NULL DEFAULT false;
 -- Asset coverage feature switch; backfills old databases. true (default)=compute/show test coverage, auto-accumulate test scope,

db/tasks.go

@@ -33,6 +33,9 @@ type Task struct {
 	LLMChainRevision   int64      `json:"-"`
 	LLMFailoverState   string     `json:"llm_failover_state,omitempty"`
 	LLMFailoverReason  string     `json:"llm_failover_reason,omitempty"`
+	// LLMRotate: per-intent chain-head rotation switch (#85); false = classic
+	// ordered failover from the active head.
+	LLMRotate bool `json:"llm_rotate"`
 	SourceTaskIDs      []int64    `json:"source_task_ids,omitempty"`
 	CompanyIDs         []int64    `json:"company_ids,omitempty"`
 	ParentRef          string     `json:"parent_ref,omitempty"` // parent task id (recorded by orchestration spawn; empty=top level)
@@ -165,6 +168,11 @@ type TaskCreateOptions struct {
 	CoverageEnabled *bool
 	// InterceptRules are task-level asset interception rules, written into task_intercept_rules in the same transaction as the task at creation.
 	InterceptRules []TaskInterceptRuleInput
+	// LLMRotate enables per-intent round-robin rotation of the task's LLM chain
+	// heads (#85): each claimed worker intent runs on the next profile in the
+	// chain instead of the single active head. nil = off (default), preserving
+	// the ordered-failover semantics.
+	LLMRotate *bool
 }
 
 // CreateTaskWithOptions creates an exploration, task, direct source relations,
@@ -239,10 +247,14 @@ VALUES ($1, 'fact', $2, 0, 'origin', 'system')`, expID, string(originPayload));
 		PlanQuietMode:   quietMode,
 		CoverageEnabled: coverageEnabled,
 	}
+	rotate := false
+	if opts.LLMRotate != nil {
+		rotate = *opts.LLMRotate
+	}
 	if err := tx.QueryRow(`
-INSERT INTO tasks(name, category_id, description, goal, exploration_id, llm_profile_id, active_llm_profile_id, timeout_seconds, plan_heartbeat_seconds, plan_quiet_mode, coverage_enabled)
-VALUES ($1,$2,$3,$4,$5,$6,$6,$7,$8,$9,$10)
-RETURNING id, status, paused, created_at`, opts.Name, opts.CategoryID, description, goal, expID, active, opts.TimeoutSeconds, opts.PlanHeartbeatSeconds, quietMode, coverageEnabled).Scan(&t.ID, &t.Status, &t.Paused, &t.CreatedAt); err != nil {
+INSERT INTO tasks(name, category_id, description, goal, exploration_id, llm_profile_id, active_llm_profile_id, timeout_seconds, plan_heartbeat_seconds, plan_quiet_mode, coverage_enabled, llm_rotate)
+VALUES ($1,$2,$3,$4,$5,$6,$6,$7,$8,$9,$10,$11)
+RETURNING id, status, paused, created_at`, opts.Name, opts.CategoryID, description, goal, expID, active, opts.TimeoutSeconds, opts.PlanHeartbeatSeconds, quietMode, coverageEnabled, rotate).Scan(&t.ID, &t.Status, &t.Paused, &t.CreatedAt); err != nil {
 		return nil, err
 	}
 	if err := insertTaskRelations(tx, t.ID, opts.SourceTaskIDs); err != nil {
@@ -358,11 +370,11 @@ func insertTaskLLMProfiles(tx *sql.Tx, taskID int64, profileIDs []int64) error {
 
 const taskCols = `id, COALESCE(name,''), category_id,
 COALESCE((SELECT category.name FROM task_categories category WHERE category.id=tasks.category_id),''),
-description, goal, exploration_id, status, paused, queued, queued_at, COALESCE(queue_mode,''), llm_profile_id, active_llm_profile_id, COALESCE(parent_ref,''), pinned_at, created_at, completed_at, COALESCE(timeout_seconds,0), COALESCE(plan_heartbeat_seconds,300), COALESCE(plan_quiet_mode,true), COALESCE(coverage_enabled,true), first_run_at, deadline_at`
+description, goal, exploration_id, status, paused, queued, queued_at, COALESCE(queue_mode,''), llm_profile_id, active_llm_profile_id, COALESCE(parent_ref,''), pinned_at, created_at, completed_at, COALESCE(timeout_seconds,0), COALESCE(plan_heartbeat_seconds,300), COALESCE(plan_quiet_mode,true), COALESCE(coverage_enabled,true), COALESCE(llm_rotate,false), first_run_at, deadline_at`
 
 func scanTask(sc interface{ Scan(...any) error }) (*Task, error) {
 	var t Task
-	if err := sc.Scan(&t.ID, &t.Name, &t.CategoryID, &t.CategoryName, &t.Description, &t.Goal, &t.ExplorationID, &t.Status, &t.Paused, &t.Queued, &t.QueuedAt, &t.QueueMode, &t.LLMProfileID, &t.ActiveLLMProfileID, &t.ParentRef, &t.PinnedAt, &t.CreatedAt, &t.CompletedAt, &t.TimeoutSeconds, &t.PlanHeartbeatSeconds, &t.PlanQuietMode, &t.CoverageEnabled, &t.FirstRunAt, &t.DeadlineAt); err != nil {
+	if err := sc.Scan(&t.ID, &t.Name, &t.CategoryID, &t.CategoryName, &t.Description, &t.Goal, &t.ExplorationID, &t.Status, &t.Paused, &t.Queued, &t.QueuedAt, &t.QueueMode, &t.LLMProfileID, &t.ActiveLLMProfileID, &t.ParentRef, &t.PinnedAt, &t.CreatedAt, &t.CompletedAt, &t.TimeoutSeconds, &t.PlanHeartbeatSeconds, &t.PlanQuietMode, &t.CoverageEnabled, &t.LLMRotate, &t.FirstRunAt, &t.DeadlineAt); err != nil {
 		return nil, err
 	}
 	t.Pinned = t.PinnedAt != nil

server/dto.go

@@ -55,6 +55,7 @@ type TaskDTO struct {
 	LLMProfileID       *int64             `json:"llm_profile_id,omitempty"` // LLM profile used for this task; nil = default
 	LLMProfileIDs      []int64            `json:"llm_profile_ids"`
 	ActiveLLMProfileID *int64             `json:"active_llm_profile_id,omitempty"`
+	LLMRotate          bool               `json:"llm_rotate"`
 	LLMFailoverState   string             `json:"llm_failover_state"`
 	LLMFailoverReason  string             `json:"llm_failover_reason,omitempty"`
 	SourceTaskIDs      []string           `json:"source_task_ids"`
@@ -130,6 +131,7 @@ func taskDTO(t *Task, status string) TaskDTO {
 		LLMProfileID:       llmState.ProfileID,
 		LLMProfileIDs:      profileIDs,
 		ActiveLLMProfileID: llmState.ActiveID,
+		LLMRotate:          t.LLMRotate,
 		LLMFailoverState:   llmState.FailoverState,
 		LLMFailoverReason:  llmState.FailoverReason,
 		SourceTaskIDs:      sourceIDs,

server/engine.go

@@ -1017,6 +1017,14 @@ func (e *Engine) runWorkerStep(ctx context.Context, t *Task, name string, worker
 // must already hold one task-operation admission for the whole sequence so a delete
 // cannot observe quiescence between the LLM return and the final DB writes.
 func (e *Engine) runIntent(ctx context.Context, t *Task, name string, worker *agent.Worker, intent *db.Node, requestID, message string) bool {
+	// Per-intent LLM rotation (#85): each claimed worker intent runs on the
+	// next chain head. The pin travels through the run context into the
+	// provider resolver; non-worker roles and failover keep the active head.
+	if t.LLMRotate {
+		if head, ok := t.nextRotatedProfile(); ok {
+			ctx = withLLMProfilePin(ctx, head)
+		}
+	}
 	hasChatMessage := message != ""
 	e.stampFirstRun(t) // first real execution → stamp first_run_at + compute the deadline (only for tasks with a timeout)
 	e.touch(t.ID)

server/manager.go

@@ -51,6 +51,10 @@ type Task struct {
 	ActiveLLMProfileID *int64  `json:"active_llm_profile_id,omitempty"`
 	LLMChainRevision   int64   `json:"-"`
 	LLMFailoverState   string  `json:"llm_failover_state,omitempty"`
+	// LLMRotate (#85): per-intent round-robin over the task's LLM chain heads.
+	LLMRotate  bool `json:"llm_rotate,omitempty"`
+	rotateMu   sync.Mutex
+	rotateNext int
 	LLMFailoverReason  string  `json:"llm_failover_reason,omitempty"`
 	SourceTaskIDs      []int64 `json:"source_task_ids,omitempty"`
 	CompanyIDs         []int64 `json:"company_ids,omitempty"`
@@ -178,6 +182,22 @@ func (t *Task) llmStateSnapshot() taskLLMState {
 	}
 }
 
+// nextRotatedProfile returns the chain head for the NEXT claimed worker
+// intent under per-intent rotation (#85): strict round-robin over the task's
+// chain. The active head stays authoritative for every non-worker role and for
+// failover ordering; rotation only distributes worker lanes across heads.
+func (t *Task) nextRotatedProfile() (int64, bool) {
+	t.rotateMu.Lock()
+	defer t.rotateMu.Unlock()
+	ids := t.LLMProfileIDs
+	if len(ids) == 0 {
+		return 0, false
+	}
+	pick := ids[t.rotateNext%len(ids)]
+	t.rotateNext++
+	return pick, true
+}
+
 func (t *Task) setLLMState(profileID, activeID *int64, profileIDs []int64, revision int64, state, reason string) bool {
 	t.llmMu.Lock()
 	defer t.llmMu.Unlock()
@@ -840,7 +860,7 @@ func taskFromPG(pt *pgdb.Task, store *pgdb.ExplorationStore, ic *intercept.Inter
 		QueuedAt: unixNanoOrZero(pt.QueuedAt), QueueMode: pt.QueueMode,
 		CompletedAt: unixOrZero(pt.CompletedAt), Status: pt.Status, ParentRef: pt.ParentRef,
 		LLMProfileID:  pt.LLMProfileID,
-		LLMProfileIDs: append([]int64(nil), pt.LLMProfileIDs...), ActiveLLMProfileID: pt.ActiveLLMProfileID,
+		LLMProfileIDs: append([]int64(nil), pt.LLMProfileIDs...), ActiveLLMProfileID: pt.ActiveLLMProfileID, LLMRotate: pt.LLMRotate,
 		LLMChainRevision: pt.LLMChainRevision,
 		LLMFailoverState: pt.LLMFailoverState, LLMFailoverReason: pt.LLMFailoverReason,
 		SourceTaskIDs:  append([]int64(nil), pt.SourceTaskIDs...),

server/server.go

@@ -1506,7 +1506,8 @@ type createTaskReq struct {
 	LLMProfileIDs        []int64  `json:"llm_profile_ids,omitempty"`   // ordered task-level profile chain; the first entry takes effect initially
 	SourceTaskIDs        []string `json:"source_task_ids,omitempty"`   // direct, read-only inherited source tasks
 	CompanyIDs           []int64  `json:"company_ids,omitempty"`       // links the company scope and snapshots its current assets; does not copy assets or force intent generation
-	TimeoutSeconds       int      `json:"timeout_seconds"`             // task-level timeout (seconds); 0/omitted = no limit
+	TimeoutSeconds       int      `json:"timeout_seconds"`
+	LLMRotate            *bool    `json:"llm_rotate"`             // task-level timeout (seconds); 0/omitted = no limit
 	PlanHeartbeatSeconds int      `json:"plan_heartbeat_seconds"`      // planner heartbeat interval (seconds); 0/omitted = default 600 (10min); floor = default = 600, lower values are raised to 600
 	PlanQuietMode        *bool    `json:"plan_quiet_mode,omitempty"`   // planner quiet-mode switch (#86); omitted = default on; false = edge rounds always plan
 	SeedFirstIntent      *bool    `json:"seed_first_intent,omitempty"` // issue one seed intent at creation (content = description + goal) so a worker can start right away without waiting for the first planner round; omitted/null = default off, the standard plan-then-execute flow. Pass true explicitly to enable (for CTF tasks often solved in one work, this skips the pre-start planner round).
@@ -1580,6 +1581,7 @@ func (s *Server) createTask(w http.ResponseWriter, r *http.Request) {
 		TimeoutSeconds: req.TimeoutSeconds, PlanHeartbeatSeconds: req.PlanHeartbeatSeconds, PlanQuietMode: req.PlanQuietMode,
 		CoverageEnabled: req.CoverageEnabled,
 		InterceptRules:  interceptRules,
+		LLMRotate:       req.LLMRotate,
 	})
 	if err != nil {
 		if errors.Is(err, db.ErrTaskCategoryInvalid) || errors.Is(err, db.ErrTaskCategoryNotFound) {

server/task_llm.go

@@ -95,8 +95,25 @@ type taskLLMSelection struct {
 	retry agent.RetryConfig
 }
 
+// llmProfilePin carries a per-run chain-head override through the call stack
+// (#85 per-intent rotation). The engine sets it when launching a worker run;
+// resolution stays within the task's chain and never bypasses failover state.
+type llmProfilePinKey struct{}
+
+func withLLMProfilePin(ctx context.Context, profileID int64) context.Context {
+	return context.WithValue(ctx, llmProfilePinKey{}, profileID)
+}
+
+func llmProfilePinFrom(ctx context.Context) (int64, bool) {
+	if ctx == nil {
+		return 0, false
+	}
+	id, ok := ctx.Value(llmProfilePinKey{}).(int64)
+	return id, ok
+}
+
 type taskLLMStreamHooks struct {
-	current    func() (taskLLMSelection, error)
+	current    func(context.Context) (taskLLMSelection, error)
 	exhaust    func(taskLLMSelection, error) (db.TaskLLMTransition, error)
 	transition func(taskLLMSelection, db.TaskLLMTransition, error)
 }
@@ -108,6 +125,14 @@ type taskLLMStreamHooks struct {
 // to the task chain, and to the global configuration when the chain is empty.
 // The returned profile id is non-zero only on the task-chain path — streamTaskLLM uses it to decide whether a quota error should advance the task's failover state (binding/global paths leave the chain state untouched, preserving existing semantics).
 func (r *taskLLMRuntime) current() (taskLLMSelection, error) {
+	return r.currentFor(context.Background())
+}
+
+// currentFor resolves like current, honoring a per-run profile pin set by the
+// engine for per-intent rotation (#85). A pin is only a dispatch preference:
+// the chain must still be in a ready state (an exhausted chain with no active
+// head stays exhausted) and the pinned id must belong to the task's chain.
+func (r *taskLLMRuntime) currentFor(ctx context.Context) (taskLLMSelection, error) {
 	taskNum, err := parseTaskID(r.taskID)
 	if err != nil {
 		return taskLLMSelection{}, err
@@ -131,6 +156,14 @@ func (r *taskLLMRuntime) current() (taskLLMSelection, error) {
 			return sel, &taskLLMError{taskID: r.taskID, chainExhausted: true, cause: errors.New("all selected profiles are quota exhausted")}
 		}
 		sel.profileID = *pt.ActiveLLMProfileID
+		if pin, ok := llmProfilePinFrom(ctx); ok && pin != sel.profileID {
+			for _, id := range pt.LLMProfileIDs {
+				if id == pin {
+					sel.profileID = pin
+					break
+				}
+			}
+		}
 		prov, cfg, ok := r.s.providerForProfile(sel.profileID)
 		if !ok {
 			return sel, fmt.Errorf("LLM profile #%d is missing or invalid", sel.profileID)
@@ -201,7 +234,7 @@ func parseTaskID(id string) (int64, error) {
 // exhausted, and how to emit a failover transition.
 func (r *taskLLMRuntime) streamHooks() taskLLMStreamHooks {
 	return taskLLMStreamHooks{
-		current: r.current,
+		current: func(ctx context.Context) (taskLLMSelection, error) { return r.currentFor(ctx) },
 		exhaust: func(selection taskLLMSelection, cause error) (db.TaskLLMTransition, error) {
 			taskNum, _ := parseTaskID(r.taskID)
 			transition, err := r.s.m.pg.MarkTaskLLMProfileQuotaExhaustedAtRevision(taskNum, selection.profileID, selection.revision, cause.Error())
@@ -236,7 +269,7 @@ func (r *taskLLMRuntime) Complete(ctx context.Context, req llm.CompletionRequest
 
 func completeTaskLLM(ctx context.Context, taskID string, req llm.CompletionRequest, hooks taskLLMStreamHooks) (llm.Message, string, llm.Usage, error) {
 	for {
-		selection, err := hooks.current()
+		selection, err := hooks.current(ctx)
 		if err != nil {
 			return llm.Message{}, "", llm.Usage{}, err
 		}
@@ -287,7 +320,7 @@ func completeTaskLLM(ctx context.Context, taskID string, req llm.CompletionReque
 func streamTaskLLM(ctx context.Context, taskID string, req llm.CompletionRequest, hooks taskLLMStreamHooks) iter.Seq2[llm.StreamEvent, error] {
 	return func(yield func(llm.StreamEvent, error) bool) {
 		for {
-			selection, err := hooks.current()
+			selection, err := hooks.current(ctx)
 			if err != nil {
 				yield(llm.StreamEvent{}, err)
 				return

server/task_llm_rotation_test.go

@@ -0,0 +1,32 @@
+package server
+
+import (
+	"context"
+	"testing"
+)
+
+func TestLLMProfilePinRoundTrip(t *testing.T) {
+	ctx := context.Background()
+	if _, ok := llmProfilePinFrom(ctx); ok {
+		t.Fatal("empty context must carry no pin")
+	}
+	pinned := withLLMProfilePin(ctx, 42)
+	got, ok := llmProfilePinFrom(pinned)
+	if !ok || got != 42 {
+		t.Fatalf("pin roundtrip got=%d ok=%v", got, ok)
+	}
+}
+
+func TestNextRotatedProfileRoundRobin(t *testing.T) {
+	tk := &Task{LLMProfileIDs: []int64{3, 1, 2}}
+	want := []int64{3, 1, 2, 3, 1}
+	for i, w := range want {
+		got, ok := tk.nextRotatedProfile()
+		if !ok || got != w {
+			t.Fatalf("rotation %d: got=%d ok=%v want=%d", i, got, ok, w)
+		}
+	}
+	if _, ok := (&Task{}).nextRotatedProfile(); ok {
+		t.Fatal("empty chain must not rotate")
+	}
+}

server/task_llm_test.go

@@ -246,7 +246,7 @@ func TestTaskLLMStreamRetriesPreStreamQuotaOnNextProfile(t *testing.T) {
 	active := int64(11)
 	transitions := 0
 	hooks := taskLLMStreamHooks{
-		current: func() (taskLLMSelection, error) {
+		current: func(ctx context.Context) (taskLLMSelection, error) {
 			if active == 11 {
 				return taskLLMSelection{profileID: 11, provider: first}, nil
 			}
@@ -288,7 +288,7 @@ func TestTaskLLMStreamRetriesAfterNonContentStartEvent(t *testing.T) {
 	}}
 	active := int64(11)
 	hooks := taskLLMStreamHooks{
-		current: func() (taskLLMSelection, error) {
+		current: func(ctx context.Context) (taskLLMSelection, error) {
 			if active == 11 {
 				return taskLLMSelection{profileID: 11, provider: first}, nil
 			}
@@ -319,7 +319,7 @@ func TestTaskLLMStreamRetriesStalePreStreamFailureAgainstReplacementChain(t *tes
 	second := &scriptedLLMProvider{events: []llm.StreamEvent{{Type: llm.SETextDelta, Text: "replacement"}}}
 	current := first
 	hooks := taskLLMStreamHooks{
-		current: func() (taskLLMSelection, error) {
+		current: func(ctx context.Context) (taskLLMSelection, error) {
 			profileID := int64(11)
 			if current == second {
 				profileID = 22
@@ -353,7 +353,7 @@ func TestTaskLLMStreamDoesNotSwitchForOrdinaryErrors(t *testing.T) {
 			provider := &scriptedLLMProvider{err: errors.New(message)}
 			exhausted := 0
 			hooks := taskLLMStreamHooks{
-				current: func() (taskLLMSelection, error) {
+				current: func(ctx context.Context) (taskLLMSelection, error) {
 					return taskLLMSelection{profileID: 11, provider: provider}, nil
 				},
 				exhaust: func(taskLLMSelection, error) (db.TaskLLMTransition, error) {
@@ -382,7 +382,7 @@ func TestTaskLLMStreamAdvancesAfterMidStreamQuotaWithoutReplay(t *testing.T) {
 	second := &scriptedLLMProvider{events: []llm.StreamEvent{{Type: llm.SETextDelta, Text: "next-call"}}}
 	active := int64(11)
 	hooks := taskLLMStreamHooks{
-		current: func() (taskLLMSelection, error) {
+		current: func(ctx context.Context) (taskLLMSelection, error) {
 			if active == 11 {
 				return taskLLMSelection{profileID: 11, provider: first}, nil
 			}
@@ -418,7 +418,7 @@ func TestTaskLLMStreamMarkFailureAfterToolStartSkipsWorkerReplay(t *testing.T) {
 		err:    quota,
 	}
 	hooks := taskLLMStreamHooks{
-		current: func() (taskLLMSelection, error) {
+		current: func(ctx context.Context) (taskLLMSelection, error) {
 			return taskLLMSelection{profileID: 11, provider: provider}, nil
 		},
 		exhaust: func(taskLLMSelection, error) (db.TaskLLMTransition, error) {
@@ -467,7 +467,7 @@ func TestTaskLLMStreamRetriesTransientPreCommitOnSameProvider(t *testing.T) {
 	}
 	exhausted := 0
 	hooks := taskLLMStreamHooks{
-		current: func() (taskLLMSelection, error) {
+		current: func(ctx context.Context) (taskLLMSelection, error) {
 			return taskLLMSelection{profileID: 11, provider: provider}, nil
 		},
 		exhaust: func(taskLLMSelection, error) (db.TaskLLMTransition, error) {
@@ -499,7 +499,7 @@ func TestTaskLLMStreamSurfacesTransientAfterRetriesExhausted(t *testing.T) {
 		failErr:   errors.New("dial tcp: connection reset by peer"),
 	}
 	hooks := taskLLMStreamHooks{
-		current: func() (taskLLMSelection, error) {
+		current: func(ctx context.Context) (taskLLMSelection, error) {
 			return taskLLMSelection{profileID: 11, provider: provider}, nil
 		},
 		exhaust: func(taskLLMSelection, error) (db.TaskLLMTransition, error) {
@@ -524,7 +524,7 @@ func TestTaskLLMStreamDoesNotRetryDeterministicRejection(t *testing.T) {
 		failErr:   errors.New("anthropic: status 400: messages.1: invalid request"),
 	}
 	hooks := taskLLMStreamHooks{
-		current: func() (taskLLMSelection, error) {
+		current: func(ctx context.Context) (taskLLMSelection, error) {
 			return taskLLMSelection{profileID: 11, provider: provider}, nil
 		},
 		exhaust: func(taskLLMSelection, error) (db.TaskLLMTransition, error) {
@@ -620,7 +620,7 @@ func TestTaskLLMStreamStopsWhenChainExhausted(t *testing.T) {
 	provider := &scriptedLLMProvider{err: errors.New("billing_not_active")}
 	transitions := 0
 	hooks := taskLLMStreamHooks{
-		current: func() (taskLLMSelection, error) {
+		current: func(ctx context.Context) (taskLLMSelection, error) {
 			return taskLLMSelection{profileID: 11, provider: provider}, nil
 		},
 		exhaust: func(taskLLMSelection, error) (db.TaskLLMTransition, error) {