# Story Fix-and-Re-Evaluate Loop Implementation Plan

> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.

**Goal:** Close the recursive-story-decomposition design spec's "fix-and-re-evaluate loop has to become real" gap. Today, when `finalizeArbitration` sets a story to `NEEDS_FIX`, nothing further happens automatically — a human must notice and manually intervene. This plan makes `StoryOrchestrator` automatically spawn a new builder-role fix attempt, re-point the story at it, and let the existing Builder → Evaluators → Arbitration chain run again from scratch against the new attempt — with no new pipeline, just the existing one re-entered against a new root.

**Architecture:** A story at status `NEEDS_FIX` is handled by a new `ensureFixAttempt` stage, checked first thing in `processStory` before the existing Builder/Evaluators/Arbitration flow. It spawns one new top-level `builder`-role task (`DependsOn: [oldRootID]`, purely for structural discoverability/audit trail — the rejected root is already `COMPLETED` so this dependency is immediately satisfied), whose instructions carry the original story spec/acceptance criteria plus the arbitration's structured rejection reasoning (read via `AgentChannel.ReportVerdict`'s `KindVerdictReported` event, the same mechanism `finalizeArbitration` already reads). It then re-points `story.RootTaskID` at the new task and resets `story.Status` to `IN_PROGRESS`. On the *next* tick, `processStory` finds the new root via the identical unmodified path every other root already takes — `ensureEvaluators`/`ensureArbitration`/`finalizeArbitration` all reused completely unchanged. Idempotency is structural (mirrors every other stage in this file): `ensureFixAttempt` looks for an existing `builder`-role dependent of the rejected root before spawning a new one, safe to call on every tick and safe across a process restart. A hardcoded `maxFixAttempts = 3` safety cap (a plain, deterministic counter — not real cost/escalation tiering, which the design spec's Non-Goals explicitly deferred) stops automatic spawning once a chain of that many consecutive fix attempts has been tried, leaving the story at `NEEDS_FIX` for a human exactly as it behaves today.

**Tech Stack:** Go, existing `internal/scheduler`/`internal/story`/`internal/task`/`internal/event` packages — no new packages, no schema changes (`story.Status` already lists `IN_PROGRESS` as a valid value; `story.RootTaskID` is already an unvalidated plain string field).

## Global Constraints

- The mechanism to re-point a story at a new attempt is **re-pointing `story.RootTaskID`** (not spawning the fix task as a dependent of the original root while leaving `RootTaskID` fixed forever) — this was an explicit design decision, made after comparing both approaches: re-pointing keeps every existing root-task-id-keyed structural idempotency check (`ensureEvaluators`, `ensureArbitration`) working unmodified against the new attempt, with no "find the latest attempt in the chain" logic needed anywhere.
- `maxFixAttempts = 3` is a **hardcoded safety net, not a designed cost/escalation policy** — the design spec's Non-Goals section explicitly deferred real depth/cost bounds to future work tied to cheap-model tiering. This plan's cap exists only to prevent a literal unbounded spend loop; it is not meant to be a tuned value.
- Do not touch `internal/api` (chatbot/REST MCP surface), the `builder` role's system prompt, or anything about per-node arbitrated review at depths below a story's root. Those remain separately deferred pieces of the recursive-story-decomposition design.

---

## Task 1: Automatic fix-attempt spawning on `NEEDS_FIX`

**Files:**
- Modify: `internal/scheduler/story_orchestrator.go` (`processStory`, `finalizeArbitration`, `arbitrationVerdict`; new `ensureFixAttempt`, `rejectionReasoning`, `fixAttemptDepth`, `maxFixAttempts`)
- Test: `internal/scheduler/story_orchestrator_test.go` (append)

**Interfaces:**
- Consumes: `event.KindVerdictReported` (existing), `StoryStore.ListEvents`/`ListDependents`/`CreateTask`/`UpdateTaskState`/`UpdateStory` (all existing, already used elsewhere in this file), `spawnRoleTask`/`findEvaluators`/`findArbitration` (existing helpers in this same file).
- Produces: nothing new for later work — this closes the last "documented simplification" this plan set out to close.

- [ ] **Step 1: Write the failing tests**

Append to `internal/scheduler/story_orchestrator_test.go`, directly after `TestStoryOrchestrator_ArbitrationApproves_SetsReviewReady` (the last test before `TestStoryOrchestrator_DoesNothing_WhenBuilderNotComplete`):

```go
// seedNeedsFixStory drives a story through the full Builder -> Evaluators ->
// Arbitration chain to a rejected verdict (mirrors
// TestStoryOrchestrator_ArbitrationRejects_SetsStoryNeedsFix's own setup),
// leaving it at status NEEDS_FIX with its original (rejected) root task
// still in the store -- the starting point ensureFixAttempt operates on.
// Returns the store, the story (as currently persisted), the rejected root
// task, the evaluators, and the arbitration task.
func seedNeedsFixStory(t *testing.T) (*fakeStoryStore, *story.Story, *task.Task, []*task.Task, *task.Task) {
	t.Helper()
	store, st, evaluators := seedStoryWithEvaluators(t, task.StateCompleted)
	oldRoot, err := store.GetTask(st.RootTaskID)
	if err != nil {
		t.Fatalf("get root: %v", err)
	}

	pool := &fakePool{}
	orch := &StoryOrchestrator{Store: store, Pool: pool}
	orch.Tick(context.Background()) // spawns arbitration

	arbitrations := store.dependentsWithRole(evaluators[0].ID, "planner")
	if len(arbitrations) != 1 {
		t.Fatalf("expected 1 arbitration task, got %d", len(arbitrations))
	}
	arb := arbitrations[0]

	payload, _ := json.Marshal(struct {
		Approved  bool   `json:"approved"`
		Reasoning string `json:"reasoning"`
	}{Approved: false, Reasoning: "breaks the running-log panel styling"})
	if err := store.CreateEvent(&event.Event{
		TaskID:  arb.ID,
		Kind:    event.KindVerdictReported,
		Actor:   event.ActorAgent,
		Payload: payload,
	}); err != nil {
		t.Fatalf("seed verdict event: %v", err)
	}
	store.setTaskState(arb.ID, task.StateCompleted)
	store.setTaskSummary(arb.ID, "found a real problem")

	orch.Tick(context.Background()) // finalizeArbitration runs, sets NEEDS_FIX

	stories, _ := store.ListStories(storage.StoryFilter{})
	var got *story.Story
	for _, s := range stories {
		if s.ID == st.ID {
			got = s
		}
	}
	if got == nil || got.Status != "NEEDS_FIX" {
		t.Fatalf("setup failed: story = %+v, want status NEEDS_FIX", got)
	}
	return store, got, oldRoot, evaluators, arb
}

// TestStoryOrchestrator_NeedsFix_SpawnsFixAttemptAndRepointsRoot proves the
// core mechanism: a story at NEEDS_FIX gets a new builder-role fix-attempt
// task spawned (depending on the rejected root), and the story is re-pointed
// at it (RootTaskID updated, Status reset to IN_PROGRESS) so the very next
// tick re-enters the normal Builder->Evaluators->Arbitration flow against
// the new attempt.
func TestStoryOrchestrator_NeedsFix_SpawnsFixAttemptAndRepointsRoot(t *testing.T) {
	store, st, oldRoot, _, _ := seedNeedsFixStory(t)

	pool := &fakePool{}
	orch := &StoryOrchestrator{Store: store, Pool: pool}
	orch.Tick(context.Background())

	fixAttempts := store.dependentsWithRole(oldRoot.ID, "builder")
	if len(fixAttempts) != 1 {
		t.Fatalf("expected exactly 1 fix-attempt task depending on the rejected root, got %d", len(fixAttempts))
	}
	fix := fixAttempts[0]
	if fix.State != task.StateQueued {
		t.Errorf("fix attempt State = %v, want QUEUED", fix.State)
	}
	if fix.ParentTaskID != "" {
		t.Errorf("fix attempt ParentTaskID = %q, want empty (top-level, DAG sibling not delegated subtask)", fix.ParentTaskID)
	}

	stories, _ := store.ListStories(storage.StoryFilter{})
	var got *story.Story
	for _, s := range stories {
		if s.ID == st.ID {
			got = s
		}
	}
	if got == nil {
		t.Fatal("story not found")
	}
	if got.RootTaskID != fix.ID {
		t.Errorf("story RootTaskID = %q, want the new fix-attempt task %q", got.RootTaskID, fix.ID)
	}
	if got.Status != "IN_PROGRESS" {
		t.Errorf("story Status = %q, want IN_PROGRESS", got.Status)
	}
}

// TestStoryOrchestrator_NeedsFix_FixAttemptInstructionsIncludeRejectionReasoning
// proves the fix attempt's instructions carry the arbitration's structured
// rejection reasoning (read via the KindVerdictReported event), not just a
// generic "try again" -- the whole point of automating this loop is that the
// next attempt has something concrete to act on.
func TestStoryOrchestrator_NeedsFix_FixAttemptInstructionsIncludeRejectionReasoning(t *testing.T) {
	store, _, oldRoot, _, _ := seedNeedsFixStory(t)

	pool := &fakePool{}
	orch := &StoryOrchestrator{Store: store, Pool: pool}
	orch.Tick(context.Background())

	fixAttempts := store.dependentsWithRole(oldRoot.ID, "builder")
	if len(fixAttempts) != 1 {
		t.Fatalf("expected exactly 1 fix-attempt task, got %d", len(fixAttempts))
	}
	instructions := fixAttempts[0].Agent.Instructions
	if !strings.Contains(instructions, "breaks the running-log panel styling") {
		t.Errorf("fix-attempt instructions do not include the arbitration's rejection reasoning:\n%s", instructions)
	}
	if !strings.Contains(instructions, "Test Story") {
		t.Errorf("fix-attempt instructions do not include the story name:\n%s", instructions)
	}
}

// TestStoryOrchestrator_NeedsFix_IdempotentAcrossTicks proves repeated ticks
// don't spawn more than one fix-attempt task -- after the first tick the
// story's status flips to IN_PROGRESS, so the NEEDS_FIX branch simply isn't
// re-entered on subsequent ticks (the normal Builder-not-complete-yet path
// takes over instead, since the fix attempt is freshly QUEUED).
func TestStoryOrchestrator_NeedsFix_IdempotentAcrossTicks(t *testing.T) {
	store, _, oldRoot, _, _ := seedNeedsFixStory(t)

	pool := &fakePool{}
	orch := &StoryOrchestrator{Store: store, Pool: pool}
	orch.Tick(context.Background())
	orch.Tick(context.Background())
	orch.Tick(context.Background())

	fixAttempts := store.dependentsWithRole(oldRoot.ID, "builder")
	if len(fixAttempts) != 1 {
		t.Fatalf("expected exactly 1 fix-attempt task after repeated ticks, got %d", len(fixAttempts))
	}
}

// TestStoryOrchestrator_EnsureFixAttempt_FindsExistingSpawnedAttempt proves
// ensureFixAttempt's own structural idempotency: simulating a restart
// between "fix task spawned" and "story RootTaskID/Status updated to point
// at it" -- calling ensureFixAttempt again must find the already-spawned
// builder-role dependent rather than spawning a second one, and still
// complete the re-pointing.
func TestStoryOrchestrator_EnsureFixAttempt_FindsExistingSpawnedAttempt(t *testing.T) {
	store, st, oldRoot, _, _ := seedNeedsFixStory(t)

	pool := &fakePool{}
	orch := &StoryOrchestrator{Store: store, Pool: pool}

	preexisting, err := orch.spawnRoleTask(context.Background(), "Fix attempt: pre-existing", "builder", []string{oldRoot.ID}, oldRoot, "fix it")
	if err != nil {
		t.Fatalf("seed pre-existing fix attempt: %v", err)
	}

	orch.ensureFixAttempt(context.Background(), st)

	fixAttempts := store.dependentsWithRole(oldRoot.ID, "builder")
	if len(fixAttempts) != 1 {
		t.Fatalf("expected exactly 1 fix-attempt task (the pre-existing one, not a new one), got %d", len(fixAttempts))
	}
	if fixAttempts[0].ID != preexisting.ID {
		t.Errorf("expected ensureFixAttempt to reuse the pre-existing task %s, got a different one %s", preexisting.ID, fixAttempts[0].ID)
	}
	if st.RootTaskID != preexisting.ID {
		t.Errorf("story RootTaskID = %q, want the pre-existing fix attempt %q", st.RootTaskID, preexisting.ID)
	}
	if st.Status != "IN_PROGRESS" {
		t.Errorf("story Status = %q, want IN_PROGRESS", st.Status)
	}
}

// TestStoryOrchestrator_NeedsFix_CapsAtMaxFixAttempts proves the safety net:
// once a chain of maxFixAttempts consecutive fix attempts already exists,
// ensureFixAttempt stops spawning new ones and leaves the story at
// NEEDS_FIX, exactly like today's fully-manual behavior.
func TestStoryOrchestrator_NeedsFix_CapsAtMaxFixAttempts(t *testing.T) {
	store, st, oldRoot, _, _ := seedNeedsFixStory(t)

	pool := &fakePool{}
	orch := &StoryOrchestrator{Store: store, Pool: pool}

	// Build a chain of maxFixAttempts prior fix-attempt tasks, each
	// depending on the one before it, ending at oldRoot -- simulating a
	// story that has already exhausted its automatic fix attempts.
	prev := oldRoot
	for i := 0; i < maxFixAttempts; i++ {
		nt, err := orch.spawnRoleTask(context.Background(), fmt.Sprintf("Fix attempt %d", i), "builder", []string{prev.ID}, oldRoot, "fix it")
		if err != nil {
			t.Fatalf("seed fix-attempt chain: %v", err)
		}
		prev = nt
	}
	st.RootTaskID = prev.ID
	if err := store.UpdateStory(st); err != nil {
		t.Fatalf("persist chained root: %v", err)
	}

	orch.Tick(context.Background())

	fixAttempts := store.dependentsWithRole(prev.ID, "builder")
	if len(fixAttempts) != 0 {
		t.Fatalf("expected no new fix-attempt task once maxFixAttempts is reached, got %d", len(fixAttempts))
	}

	stories, _ := store.ListStories(storage.StoryFilter{})
	var got *story.Story
	for _, s := range stories {
		if s.ID == st.ID {
			got = s
		}
	}
	if got.Status != "NEEDS_FIX" {
		t.Errorf("story Status = %q, want NEEDS_FIX (cap reached, no further automation)", got.Status)
	}
}
```

`internal/scheduler/story_orchestrator_test.go` already imports `"strings"` (used elsewhere in this file) — no import changes are needed for this task.

- [ ] **Step 2: Run tests to verify they fail**

Run: `go test ./internal/scheduler/ -run 'TestStoryOrchestrator_(NeedsFix|EnsureFixAttempt)' -v`
Expected: compile failure — `ensureFixAttempt`, `fixAttemptDepth`, and `maxFixAttempts` don't exist yet. Paste the actual output.

- [ ] **Step 3: Extend `arbitrationVerdict` to also return the rejection reasoning**

In `internal/scheduler/story_orchestrator.go`, find `arbitrationVerdict`:

```go
// arbitrationVerdict reads the arbitration task's own event stream for a
// KindVerdictReported event (AgentChannel.ReportVerdict) and returns its
// approved value. hasVerdict is false if no such event was ever recorded —
// callers must treat that as "no structured verdict was reported", not as
// an implicit rejection.
func (o *StoryOrchestrator) arbitrationVerdict(st *story.Story, arbitration *task.Task) (approved bool, hasVerdict bool) {
	events, err := o.Store.ListEvents(arbitration.ID, 0)
	if err != nil {
		o.logf("story orchestrator: list arbitration task events", "storyID", st.ID, "taskID", arbitration.ID, "error", err)
		return false, false
	}
	for _, e := range events {
		if e.Kind != event.KindVerdictReported {
			continue
		}
		var payload struct {
			Approved bool `json:"approved"`
		}
		if json.Unmarshal(e.Payload, &payload) == nil {
			approved = payload.Approved
			hasVerdict = true
		}
	}
	return approved, hasVerdict
}
```

Replace it with:

```go
// arbitrationVerdict reads the arbitration task's own event stream for a
// KindVerdictReported event (AgentChannel.ReportVerdict) and returns its
// approved value and reasoning text. hasVerdict is false if no such event was
// ever recorded — callers must treat that as "no structured verdict was
// reported", not as an implicit rejection. reasoning feeds ensureFixAttempt's
// automatic fix-attempt instructions; finalizeArbitration itself only needs
// approved/hasVerdict and discards reasoning.
func (o *StoryOrchestrator) arbitrationVerdict(st *story.Story, arbitration *task.Task) (approved bool, reasoning string, hasVerdict bool) {
	events, err := o.Store.ListEvents(arbitration.ID, 0)
	if err != nil {
		o.logf("story orchestrator: list arbitration task events", "storyID", st.ID, "taskID", arbitration.ID, "error", err)
		return false, "", false
	}
	for _, e := range events {
		if e.Kind != event.KindVerdictReported {
			continue
		}
		var payload struct {
			Approved  bool   `json:"approved"`
			Reasoning string `json:"reasoning"`
		}
		if json.Unmarshal(e.Payload, &payload) == nil {
			approved = payload.Approved
			reasoning = payload.Reasoning
			hasVerdict = true
		}
	}
	return approved, reasoning, hasVerdict
}
```

- [ ] **Step 4: Update `finalizeArbitration`'s call site**

In the same file, find this line inside `finalizeArbitration`:

```go
	approved, hasVerdict := o.arbitrationVerdict(st, arbitration)
```

Replace it with:

```go
	approved, _, hasVerdict := o.arbitrationVerdict(st, arbitration)
```

- [ ] **Step 5: Add the `NEEDS_FIX` branch to `processStory`**

In the same file, find the start of `processStory`:

```go
func (o *StoryOrchestrator) processStory(ctx context.Context, st *story.Story) {
	root, err := o.Store.GetTask(st.RootTaskID)
```

Replace it with:

```go
func (o *StoryOrchestrator) processStory(ctx context.Context, st *story.Story) {
	if st.Status == "NEEDS_FIX" {
		o.ensureFixAttempt(ctx, st)
		return
	}

	root, err := o.Store.GetTask(st.RootTaskID)
```

- [ ] **Step 6: Add `maxFixAttempts`, `ensureFixAttempt`, `rejectionReasoning`, and `fixAttemptDepth`**

In the same file, find the `optionalBool` function (it ends right before `processRetro`'s doc comment):

```go
// optionalBool returns a pointer to v if present is true, nil otherwise —
// used so KindArbitrationDecided's payload omits "approved" entirely
// (rather than encoding a misleading false) when no structured verdict was
// ever reported.
func optionalBool(v bool, present bool) *bool {
	if !present {
		return nil
	}
	return &v
}
```

Directly after it (still before `processRetro`'s doc comment), add:

```go
// maxFixAttempts caps automatic fix-attempt spawning (see ensureFixAttempt)
// to prevent an unbounded loop if arbitration keeps rejecting a story's
// work. This is a simple, hardcoded safety net, not real cost/escalation
// tiering — the design spec's Non-Goals explicitly deferred "no depth or
// cost bound in this design... left as a distinct future piece of work".
// Once hit, the story stays at NEEDS_FIX exactly like it does today, for a
// human to intervene manually via PUT /api/stories/{id}.
const maxFixAttempts = 3

// ensureFixAttempt handles a story sitting at NEEDS_FIX: it spawns one new
// top-level builder-role task depending on the rejected root (purely for
// structural discoverability/audit trail — the rejected root is already
// COMPLETED, so this dependency is immediately satisfied), whose
// instructions carry the original story spec/acceptance criteria plus the
// arbitration's rejection reasoning, then re-points st.RootTaskID at the new
// task and resets st.Status to IN_PROGRESS. The very next tick re-enters
// processStory's normal Builder->Evaluators->Arbitration flow against the
// new root, completely unchanged — no new pipeline, the existing one
// re-entered.
//
// Idempotency is structural, mirroring every other stage in this file: it
// looks for an existing builder-role dependent of the rejected root before
// spawning a new one, so calling this repeatedly (or after a restart between
// "task spawned" and "story updated") never spawns duplicates. The
// maxFixAttempts cap is only checked in the "spawn a new one" branch — a
// dependent that was already committed to being spawned still gets
// re-pointed to, regardless of the cap, since refusing to do so would leave
// a task dangling with nothing tracking it.
func (o *StoryOrchestrator) ensureFixAttempt(ctx context.Context, st *story.Story) {
	oldRoot, err := o.Store.GetTask(st.RootTaskID)
	if err != nil {
		o.logf("story orchestrator: fix attempt: get rejected root task", "storyID", st.ID, "rootTaskID", st.RootTaskID, "error", err)
		return
	}

	dependents, err := o.Store.ListDependents(oldRoot.ID)
	if err != nil {
		o.logf("story orchestrator: fix attempt: list root dependents", "storyID", st.ID, "error", err)
		return
	}
	var fix *task.Task
	for _, d := range dependents {
		if d.Agent.Role == "builder" {
			fix = d
			break
		}
	}

	if fix == nil {
		if depth := o.fixAttemptDepth(oldRoot); depth >= maxFixAttempts {
			o.logf("story orchestrator: fix attempt: max fix attempts reached, leaving story at NEEDS_FIX for a human", "storyID", st.ID, "depth", depth)
			return
		}
		reasoning := o.rejectionReasoning(st, oldRoot)
		instructions := fmt.Sprintf(
			"A previous attempt at story %q (task %s) was rejected by arbitration. Fix the issues and resubmit.\n\n"+
				"Story spec:\n%s\n\nAcceptance criteria:\n%s\n\nArbitration's rejection reasoning:\n%s",
			st.Name, oldRoot.ID, st.Spec, formatAcceptanceCriteria(st.AcceptanceCriteria), reasoning)
		nt, err := o.spawnRoleTask(ctx, "Fix attempt: "+st.Name, "builder", []string{oldRoot.ID}, oldRoot, instructions)
		if err != nil {
			o.logf("story orchestrator: fix attempt: spawn", "storyID", st.ID, "error", err)
			return
		}
		fix = nt
	}

	st.RootTaskID = fix.ID
	st.Status = "IN_PROGRESS"
	if err := o.Store.UpdateStory(st); err != nil {
		o.logf("story orchestrator: fix attempt: repoint story root", "storyID", st.ID, "error", err)
	}
}

// rejectionReasoning finds the rejected root's evaluator/arbitration chain
// (read-only, via findEvaluators/findArbitration — the same lookups
// processRetro uses) and returns the arbitration task's structured
// KindVerdictReported reasoning, falling back to its plain-text Summary if no
// structured verdict was reported, or a placeholder if neither is available.
func (o *StoryOrchestrator) rejectionReasoning(st *story.Story, oldRoot *task.Task) string {
	evaluators, ok := o.findEvaluators(oldRoot.ID)
	if !ok {
		return "(no arbitration reasoning found)"
	}
	arbitration, ok := o.findArbitration(evaluators)
	if !ok {
		return "(no arbitration reasoning found)"
	}
	_, reasoning, hasVerdict := o.arbitrationVerdict(st, arbitration)
	if hasVerdict && reasoning != "" {
		return reasoning
	}
	if arbitration.Summary != "" {
		return arbitration.Summary
	}
	return "(no arbitration reasoning found)"
}

// fixAttemptDepth walks backward through builder-role tasks connected by a
// single DependsOn edge — the exact shape ensureFixAttempt creates — counting
// how many fix attempts precede root. Stops as soon as it finds a task with
// anything other than exactly one DependsOn entry (the original, non-fix-
// attempt root, which was created by whatever process first submitted the
// story). Bounded by maxFixAttempts+1 iterations since callers only care
// whether the cap is hit, not the exact depth beyond that.
func (o *StoryOrchestrator) fixAttemptDepth(root *task.Task) int {
	depth := 0
	current := root
	for depth <= maxFixAttempts {
		if len(current.DependsOn) != 1 {
			break
		}
		prev, err := o.Store.GetTask(current.DependsOn[0])
		if err != nil {
			break
		}
		depth++
		current = prev
	}
	return depth
}
```

- [ ] **Step 7: Run tests to verify they pass**

Run: `go test ./internal/scheduler/ -run 'TestStoryOrchestrator_(NeedsFix|EnsureFixAttempt|Arbitration)' -v`
Expected: PASS, all of them — the new tests, plus the pre-existing `TestStoryOrchestrator_ArbitrationRejects_SetsStoryNeedsFix`, `TestStoryOrchestrator_ArbitrationApproves_SetsReviewReady`, and `TestStoryOrchestrator_ArbitrationCompletes_EmitsDecisionAndReviewReady` (unmodified, must still pass — they only exercise `arbitrationVerdict` indirectly through `finalizeArbitration`, which still compiles and behaves the same after Step 4's one-line signature-adjustment). Paste the actual output.

- [ ] **Step 8: Run the full package suite, then the full repo suite**

Run: `go test ./internal/scheduler/...`
Expected: PASS for everything.

Then run: `go test ./...` (the entire repo). This must also pass — a prior plan in this repo found that task-scoped tests passing is not sufficient on its own; grep the repo for any other caller of `arbitrationVerdict` before assuming none exists (`grep -rn "arbitrationVerdict" internal/`) — it should only be called from within `story_orchestrator.go` itself (by `finalizeArbitration` and the new `rejectionReasoning`), so Step 4's signature change should not require touching any other file. Paste the actual full-repo test output.

Note: if you see a transient one-off failure in `internal/api` under the race detector mentioning "sql: database is closed" during an unrelated webhook/websocket test's teardown, that is a known pre-existing flake unrelated to this change — simply rerun `go test ./...` once to confirm it doesn't reproduce, and proceed if it passes clean on rerun. If it fails a second time in a way connected to your actual changes, stop and call ask_user.

- [ ] **Step 9: Commit and push to `main` directly, matching this repo's existing workflow**

```bash
git add internal/scheduler/story_orchestrator.go internal/scheduler/story_orchestrator_test.go
git commit -m "feat(scheduler): automatically spawn and re-point a fix attempt when a story is rejected"
```

## Mandatory verification disclosure

When you call report_summary, paste the actual terminal output of every test command above — literal pass/fail counts, not a claim of success, including the full-repo `go test ./...` run. If anything here is ambiguous or conflicts with what you find in the actual repo, call ask_user and describe the specific conflict — don't guess or silently decide.

---

## Final Verification

- [ ] Run `go build ./...` — passes.
- [ ] Run `go test ./...` — passes, full repo.
- [ ] Run `grep -rn "ensureFixAttempt\|fixAttemptDepth\|rejectionReasoning\|maxFixAttempts" internal/` to visually confirm the File Map was actually touched (should all be in `internal/scheduler/story_orchestrator.go` and its test file only).