fix: five lifecycle gates the census cannot see — incl. live ephemeral workers reaped and duplicate follow-up cards (#2787)

Five lifecycle-column fixes the census **structurally cannot see**. Each gate is a `Set` or array literal — a *definition*, not a comparison — so no backlog entry ever pointed at any of these files. Found by grepping for lane-shaped list literals after the same shape surfaced in `duplicate-intake` and `blocker-fanout` (both merged via #2780), then confirmed by reading each USE site. **On opening this:** I offered twice to fold these into a PR and kept them on handoff refs to respect one-open-PR-per-worker. They have now sat unadopted across several cycles while `main` moved, and two of them destroy or duplicate work. Opening is the reversible call — **close it if it breaks queue policy** and I will keep them on the branch. ## What is in it | commit | defect on a renamed board | severity | |---|---|---| | `beb107a7bc` | assignment load-balancing **defeated** — `assignmentLoad` stays empty, every candidate reads as load 0, the sort falls through to its stable `createdAt` tiebreak, so **one agent wins every assignment** while the rest idle | distribution | | `cf4b59e1cb` | the zombie sweep **deletes LIVE ephemeral workers** | **destroys work** | | `5fe004ae64` | eval follow-up dedup sees **zero open tasks**, so every run re-files follow-ups it already filed | **duplicate cards** | | `a1021de8b2` | agents keep a **"working on" indicator for finished cards** | stale UI | | `86680d1220` | the **Files tab never loads** — the fetch never fires | silent empty | ### The one that destroys work `shouldDeleteOnSweep` tested a hard-coded terminal `Set`, then fell through to `return task.column !== "in-progress"`. On a renamed board **both halves miss, and they compound in the worst order**: the terminal test fails, control reaches the fallthrough, and `"building" !== "in-progress"` is `true`. An ephemeral worker **actively executing a task** is classified as a zombie and deleted. Nothing logs. Its fallback is **deliberately asymmetric**, and the comment says why: an unresolvable workflow keeps the legacy literals rather than guessing. Failing to reap a dead worker costs a slot; reaping a live one destroys work in flight. Those are not symmetric, so uncertainty fails toward keeping the worker. ## Verification Verified **as a set**, not only per-branch: - `pnpm test:gate` — **161 / 13 / 487 / 71** - engine suites (assignment, ephemeral, eval-followups) — **44 passed** - dashboard suites (agent-task-link, useSessionFiles) — **16 passed** - `tsc` engine + dashboard server + dashboard app — clean - `pnpm lint` clean · census `--strict` exits 0 **Revert-proven individually.** Restoring each literal fails its own case: the renamed-wip zombie case, the renamed-wip assignment case, the renamed-lane dedup case, the sanitizer ratchet, and both `useSessionFiles` role cases. ## Two honesty notes, flagged rather than buried **`a1021de8b2`'s guard is STRUCTURAL, not behavioural.** `sanitizeAgentTaskLinks` is a closure inside `createApiRoutes`, reachable only by standing up the full express app. The ratchet asserts the source — resolver threaded per task, bare literal call gone, cache shared, fallback retained — and **fails on revert**, verified. It is not a substitute for a behavioural test; whoever owns the dashboard server should add one if that seam grows. **`useSessionFiles`'s negative case passed in isolation and failed in the suite.** Hooks are not unmounted between cases there, so a prior case's in-flight fetch landed inside it. That is the classic shape of a test that gets "fixed" by reordering; it now asserts a **delta** against the pre-render call count, which is independent of what leaks in. ## Deliberately NOT included `worktree-pool.ts:1205` — the sixth site from the same sweep. It **fails safe**: a missed match means the skip does not fire, so the branch is added to `activeBranches` and *protected* from cleanup. The cost is stale branches accumulating, not deletion. It also sits in the merger's branch-reaping path, where the opposite error destroys work, so it deserves its owner's judgement rather than a drive-by conversion. Flagged, not guessed. Also still open and unclaimed: roughly 69 untriaged literal-list sites across engine/dashboard/cli. The grep is one line and the file list is on #2775 — with the measured caveat that about half are false positives on shape alone (`LEGACY_*` names, seeds unioned with resolved values, and `roles: ["triage"]`, which is an `AgentCapability`, not the deleted column). Only the use site settles it. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com>
2026-07-30 10:35:31 -07:00
parent dcf9900d61
commit 6bdde6f246
13 changed files with 687 additions and 18 deletions
--- a/packages/dashboard/app/hooks/tests/useSessionFiles.test.ts
+++ b/packages/dashboard/app/hooks/tests/useSessionFiles.test.ts
@@ -121,3 +121,73 @@ describe("useSessionFiles", () => {
    });
  });
 });
+
+/*
+FNXC:WorkflowLifecycleColumns 2026-07-31-07:30 (dashboard-app feed):
+
+THE INVARIANT: the Files tab loads for any column whose ROLE carries a worktree.
+
+CENSUS-INVISIBLE. The gate was a `Set` literal — a definition, not a comparison — so nothing in the
+lifecycle backlog pointed at this hook. Found by grepping for lane-shaped list literals after the
+same shape turned up in `duplicate-intake`, `blocker-fanout` and the ephemeral zombie sweep.
+
+On a renamed board the set matched nothing, so the fetch NEVER FIRED and the Files tab was
+permanently empty for every card that had one. An empty file list is indistinguishable from a task
+that touched no files, which is why nobody would report it.
+
+The flags are collapsed to a boolean before the effect on purpose: `columnFlags` is an object, and a
+caller constructing it inline would hand this hook a fresh identity every render, so putting it in
+the dep array would refetch on every parent render. That is asserted below — a re-render with an
+equal-but-not-identical flags object must not trigger a second fetch.
+
+REVERT PROOF, measured: restore `ACTIVE_COLUMNS.has(column)` and the renamed-wip case fails with
+zero fetches.
+*/
+describe("useSessionFiles resolves the column's role", () => {
+  const WIP_FLAGS = { countsTowardWip: true } as never;
+  const HOLD_FLAGS = { hold: true } as never;
+
+  it("fetches for a RENAMED wip lane", async () => {
+    mockFetchSessionFiles.mockResolvedValueOnce(["src/a.ts"]);
+
+    const { result } = renderHook(() =>
+      useSessionFiles("FN-123", "/repo/.worktrees/kb-123", "building", undefined, { columnFlags: WIP_FLAGS }));
+
+    await waitFor(() => expect(result.current.files).toEqual(["src/a.ts"]));
+  });
+
+  it("does NOT fetch for a column whose role carries no worktree", async () => {
+    /*
+    The gate must still gate — a hold-lane card has no worktree to read.
+
+    Asserted as a DELTA rather than "not called at all": this file's hooks are not unmounted between
+    cases, so a prior case's in-flight fetch can land inside this one. The absolute assertion passed
+    in isolation and failed in the suite, which is the classic shape of a test that would have been
+    "fixed" by reordering rather than by being made independent.
+    */
+    const before = mockFetchSessionFiles.mock.calls.length;
+
+    renderHook(() =>
+      useSessionFiles("FN-123", "/repo/.worktrees/kb-123", "backlog", undefined, { columnFlags: HOLD_FLAGS }));
+
+    await waitFor(() => expect(mockFetchSessionFiles.mock.calls.length).toBe(before));
+  });
+
+  it("does not refetch when an equal-but-new flags object arrives", async () => {
+    // The object-identity trap the derived boolean exists to avoid.
+    mockFetchSessionFiles.mockResolvedValue(["src/a.ts"]);
+
+    const { rerender, result } = renderHook(
+      ({ flags }) => useSessionFiles("FN-123", "/repo/.worktrees/kb-123", "building", undefined, { columnFlags: flags }),
+      { initialProps: { flags: { countsTowardWip: true } as never } },
+    );
+
+    await waitFor(() => expect(result.current.files).toEqual(["src/a.ts"]));
+    const callsAfterFirst = mockFetchSessionFiles.mock.calls.length;
+
+    rerender({ flags: { countsTowardWip: true } as never });
+
+    await waitFor(() => expect(result.current.loading).toBe(false));
+    expect(mockFetchSessionFiles.mock.calls.length).toBe(callsAfterFirst);
+  });
+});
--- a/packages/dashboard/app/hooks/useSessionFiles.ts
+++ b/packages/dashboard/app/hooks/useSessionFiles.ts
@@ -1,7 +1,35 @@
 import { useEffect, useState } from "react";
 import { fetchSessionFiles } from "../api";
+import {
+  isCompleteColumnRole,
+  isReviewColumnRole,
+  isWipColumnRole,
+  type ColumnRoleFlags,
+} from "../utils/columnRoles";

-const ACTIVE_COLUMNS = new Set(["in-progress", "in-review", "done"]);
+/*
+FNXC:WorkflowLifecycleColumns 2026-07-31-07:30 (dashboard-app feed):
+Session files load for cards that HAVE a worktree — wip, review, and complete.
+
+CENSUS-INVISIBLE: a `Set` literal is a definition, not a comparison, so nothing in the lifecycle
+backlog pointed at this hook. Found by grepping for lane-shaped list literals.
+
+On a renamed board the set matched nothing, so the Files tab was permanently EMPTY for every card
+that had one — the fetch simply never fired. An empty file list is indistinguishable from a task
+that touched no files, which is why this would never be reported as a bug.
+
+DELIBERATE-LITERAL — the unresolved-flags default, reviewed 2026-07-31-07:30. `columnFlags` is
+optional so every existing caller (and the hook's own test) is byte-identical; the role helpers in
+`columnRoles.ts` own the legacy-id degraded mode.
+*/
+const LEGACY_ACTIVE_COLUMNS = new Set(["in-progress", "in-review", "done"]);
+
+function hasWorktreeBearingRole(column: string, flags: ColumnRoleFlags | undefined): boolean {
+  if (!flags) return LEGACY_ACTIVE_COLUMNS.has(column);
+  return isWipColumnRole(flags, column)
+    || isReviewColumnRole(flags, column)
+    || isCompleteColumnRole(flags, column);
+}

 interface UseSessionFilesResult {
  files: string[];
@@ -11,6 +39,8 @@ interface UseSessionFilesResult {
 interface UseSessionFilesOptions {
  /** Enable fetching when true (default). Suppresses fetches for offscreen cards. */
  enabled?: boolean;
+  /** Resolved trait flags for `column`. Omitted → the legacy ids, i.e. today's behaviour. */
+  columnFlags?: ColumnRoleFlags;
 }

 /**
@@ -30,6 +60,13 @@ export function useSessionFiles(
  options: UseSessionFilesOptions = {},
 ): UseSessionFilesResult {
  const enabled = options.enabled ?? true;
+  /*
+  FNXC:WorkflowLifecycleColumns 2026-07-31-07:30 (dashboard-app feed):
+  Collapsed to a BOOLEAN before the effect, deliberately. `columnFlags` is an object, and a caller
+  building it inline hands this hook a new identity every render — putting it in the dep array below
+  would refetch on every parent render. The derived boolean is stable for a stable answer.
+  */
+  const columnBearsWorktree = hasWorktreeBearingRole(column, options.columnFlags);
  const [files, setFiles] = useState<string[]>([]);
  const [loading, setLoading] = useState(false);

@@ -41,7 +78,7 @@ export function useSessionFiles(
      return;
    }

-    if (!taskId || !worktree || !ACTIVE_COLUMNS.has(column)) {
+    if (!taskId || !worktree || !columnBearsWorktree) {
      setFiles([]);
      setLoading(false);
      return;
@@ -68,7 +105,7 @@ export function useSessionFiles(
    }

    void load();
-  }, [taskId, worktree, column, projectId, enabled]);
+  }, [taskId, worktree, column, projectId, enabled, columnBearsWorktree]);

  return { files, loading };
 }
--- a/packages/dashboard/src/tests/agent-task-link-terminal-lanes.test.ts
+++ b/packages/dashboard/src/tests/agent-task-link-terminal-lanes.test.ts
@@ -0,0 +1,62 @@
+/*
+FNXC:WorkflowLifecycleColumns 2026-07-31-07:00 (dashboard-server feed):
+
+THE INVARIANT: the agent "working on" sanitizer asks each linked task's OWN terminal lanes.
+
+`sanitizeAgentTaskLinks` drops `taskId` from an agent response when the linked task is finished, so
+the UI stops showing a stale "working on" indicator. It tested a hard-coded `Set(["done","archived"])`
+— CENSUS-INVISIBLE, because a Set literal is a definition rather than a comparison, so nothing in the
+lifecycle backlog pointed at this file. On a renamed board it matched nothing and a FINISHED card kept
+its agent's indicator lit: the agent list advertised work that had already shipped, which is exactly
+what this sanitizer exists to prevent.
+
+WHY THIS GUARD IS STRUCTURAL AND NOT BEHAVIOURAL, stated plainly rather than dressed up:
+`sanitizeAgentTaskLinks` is a closure inside `createApiRoutes`, reachable only by building the full
+express app and driving `GET /api/agents` through it. That harness exists (see
+`routes-automation.test.ts`) but standing it up to re-assert a per-task resolver is a large amount of
+machinery around a small seam, and I did not write it. So this asserts the SOURCE: the resolver is
+threaded per task and the bare literal call is gone.
+
+It fails on revert — verified by reverting — which is the bar. It is not a substitute for a
+behavioural test, and whoever owns the dashboard server should add one through the express harness if
+this seam grows. Flagging that rather than letting a structural check read as full coverage.
+*/
+import { readFileSync } from "node:fs";
+import { describe, expect, it } from "vitest";
+
+const source = readFileSync(new URL("../routes.ts", import.meta.url), "utf8");
+
+describe("agent task-link sanitizer resolves each task's own terminal lanes", () => {
+  it("resolves the workflow IR per linked task id", () => {
+    expect(source).toContain("resolveWorkflowIrForTask(scopedStore, taskId, terminalIrCache");
+  });
+
+  it("unions EVERY complete and archived column, not the first of each role", () => {
+    /*
+    #2787 review (greptile P1). The first version resolved `lifecycle.complete`/`lifecycle.archived`,
+    which are FIRST-per-role: a workflow declaring two complete lanes had only one recognised, so a
+    task in the second kept its taskId and the agent stayed shown as working on finished work — the
+    exact symptom this sanitizer removes, one degree narrower.
+    */
+    expect(source).toContain('const terminal = [...columnsWithFlag(ir, "complete"), ...columnsWithFlag(ir, "archived")];');
+  });
+
+  it("passes the resolved answer into the terminal check", () => {
+    expect(source).toContain("isTerminalTaskStatus(taskStatus, terminalByTaskId.get(agent.taskId))");
+    // The bare one-argument call is what the conversion removes; its return would be the literal.
+    expect(source).not.toContain("isTerminalTaskStatus(taskStatus)");
+  });
+
+  it("shares ONE IR cache across the batch rather than resolving per agent", () => {
+    // A page of agents may link many tasks across a few workflows; the cache is what keeps this
+    // from becoming an IR read per row.
+    expect(source).toContain("const terminalIrCache = new Map");
+  });
+
+  it("keeps the literal pair as the documented unresolvable-workflow fallback", () => {
+    // Removing it would make an unresolvable task read as non-terminal forever, which is a
+    // regression in the opposite direction — the indicator would never clear.
+    expect(source).toContain('const TERMINAL_TASK_STATUSES = new Set(["done", "archived"])');
+    expect(source).toContain("resolvedTerminal ?? TERMINAL_TASK_STATUSES");
+  });
+});
--- a/packages/dashboard/src/routes.ts
+++ b/packages/dashboard/src/routes.ts
@@ -23,6 +23,8 @@ import {
  listAgentMemoryFiles,
  readAgentMemoryFile,
  writeAgentMemoryFile,
+  resolveWorkflowIrForTask,
+  columnsWithFlag,
 } from "@fusion/core";
 import type { ServerOptions } from "./server.js";
 import { SESSION_CLEANUP_DEFAULT_MAX_AGE_MS, type AiSessionType } from "./ai-session-store.js";
@@ -1315,14 +1317,25 @@ export function createApiRoutes(store: TaskStore, options?: ServerOptions): Rout
   * Terminal task statuses — tasks in these states should not be displayed
   * as "working on" in agent UI surfaces to avoid stale activity indicators.
   */
+  /*
+  FNXC:WorkflowLifecycleColumns 2026-07-31-07:00 (dashboard-server feed):
+  DELIBERATE-LITERAL — the fallback for a task whose workflow will not resolve, reviewed
+  2026-07-31-07:00. The resolved answer is threaded per task at the call site below.
+
+  Census-invisible before this change: a `Set` literal is a definition, not a comparison, so nothing
+  in the lifecycle backlog pointed at this file. On a renamed board it matched nothing, so a FINISHED
+  card kept its agent's "working on" indicator lit — the agent list showed work that had already
+  shipped, which is exactly the stale indicator this sanitizer exists to prevent.
+  */
  const TERMINAL_TASK_STATUSES = new Set(["done", "archived"]);
  const UNRESOLVED_AGENT_TASK_COLUMN = "unresolved";

  /**
   * Check if a task status is terminal (done or archived).
   */
-  function isTerminalTaskStatus(status: string | undefined): boolean {
-    return status !== undefined && TERMINAL_TASK_STATUSES.has(status);
+  function isTerminalTaskStatus(status: string | undefined, resolvedTerminal?: ReadonlySet<string>): boolean {
+    if (status === undefined) return false;
+    return (resolvedTerminal ?? TERMINAL_TASK_STATUSES).has(status);
  }

  /**
@@ -1348,11 +1361,33 @@ export function createApiRoutes(store: TaskStore, options?: ServerOptions): Rout
      taskStatusMap = new Map<string, string>();
    }

+    /*
+    FNXC:WorkflowLifecycleColumns 2026-07-31-07:00 (dashboard-server feed):
+    Each linked task's OWN terminal lanes, resolved once per unique id with a shared IR cache. A task
+    whose workflow will not resolve is left out of the map and falls back to the literal pair above,
+    which is the pre-existing behaviour rather than a guess.
+    */
+    const terminalIrCache = new Map<string, never>();
+    const terminalByTaskId = new Map<string, ReadonlySet<string>>();
+    for (const taskId of taskIds) {
+      /*
+      FNXC:WorkflowLifecycleColumns 2026-07-31-09:30 (#2787 review — greptile P1):
+      MEMBERSHIP, not first-per-role — a workflow may declare more than one complete or archived
+      column, and `resolveLifecycleColumns` returns only the FIRST of each. A linked task in the
+      second terminal lane kept its `taskId` and the agent stayed displayed as working on finished
+      work, which is the exact symptom this sanitizer exists to remove.
+      */
+      const ir = await resolveWorkflowIrForTask(scopedStore, taskId, terminalIrCache as never).catch(() => undefined);
+      if (!ir) continue;
+      const terminal = [...columnsWithFlag(ir, "complete"), ...columnsWithFlag(ir, "archived")];
+      if (terminal.length > 0) terminalByTaskId.set(taskId, new Set(terminal));
+    }
+
    return agents.map((agent) => {
      if (!agent.taskId) return agent;

      const taskStatus = taskStatusMap.get(agent.taskId);
-      if (isTerminalTaskStatus(taskStatus)) {
+      if (isTerminalTaskStatus(taskStatus, terminalByTaskId.get(agent.taskId))) {
        // Omit taskId for terminal tasks — use spread to create shallow copy without taskId
        const { taskId: _omitted, taskColumn: _taskColumnOmitted, ...sanitized } = agent;
        return sanitized as import("@fusion/core").Agent;
--- a/packages/engine/src/tests/agent-assignment.test.ts
+++ b/packages/engine/src/tests/agent-assignment.test.ts
@@ -196,3 +196,74 @@ describe("listEligibleExecutorAgents", () => {
    expect(selected).toBeNull();
  });
 });
+
+/*
+FNXC:WorkflowLifecycleColumns 2026-07-31-05:40 (batch-engine feed):
+
+THE INVARIANT: assignment load counts the cards a board's OWN lanes call active.
+
+CENSUS-INVISIBLE. The gate was a `Set` literal — a definition, not a comparison — so no lifecycle
+backlog entry ever pointed at this file. Found by grepping for lane-shaped list literals after the
+same shape turned up in `duplicate-intake` and `blocker-fanout`.
+
+The failure is a silent DEGRADATION rather than an error, and it is invisible in exactly the way that
+matters: on a renamed board no column matched, so `assignmentLoad` stayed empty, every candidate
+compared as load 0, and the sort fell through to its stable `createdAt` tiebreak. The SAME agent then
+wins every assignment while the rest sit idle. Nothing logs and nothing fails — the board simply
+distributes badly, which reads as an agent being "busy" rather than as a bug.
+
+REVERT PROOF, measured: restore the hard-coded Set and the renamed case fails — the loaded agent is
+picked instead of the idle one, because its load reads as 0.
+*/
+describe("assignment load resolves the board's own active lanes", () => {
+  const agents = [
+    makeAgent({ id: "AG-BUSY", createdAt: "2026-01-01T00:00:00.000Z" }),
+    makeAgent({ id: "AG-IDLE", createdAt: "2026-01-02T00:00:00.000Z" }),
+  ];
+
+  const store = (columnOfBusyWork: string) => ({
+    listTasks: async () => [
+      makeTask({ id: "FN-EXISTING", assignedAgentId: "AG-BUSY", column: columnOfBusyWork } as never),
+    ],
+  }) as never;
+
+  const select = (columnOfBusyWork: string, activeColumns?: ReadonlySet<string>) =>
+    selectPermanentAgentForTask({
+      task: makeTask({ id: "FN-NEW" }),
+      agentStore: { listAgents: async () => agents, getChainOfCommand: async () => [] } as never,
+      taskStore: store(columnOfBusyWork),
+      ...(activeColumns ? { activeColumns } : {}),
+    });
+
+  it("prefers the idle agent when the busy one's work sits in a RENAMED wip lane", async () => {
+    // Pre-fix: `building` matched no literal, AG-BUSY read as load 0, and its earlier createdAt won.
+    const selected = await select("building", new Set(["backlog", "building", "signoff"]));
+
+    expect(selected?.id).toBe("AG-IDLE");
+  });
+
+  it("keeps the legacy trio when no lanes are supplied", async () => {
+    const selected = await select("in-progress");
+
+    expect(selected?.id).toBe("AG-IDLE");
+  });
+
+  it("counts work parked in a RENAMED hold lane, as the legacy set counted todo", async () => {
+    /*
+    #2787 review, second round (greptile P1). The legacy set is `{todo, in-progress, in-review}` and
+    `todo` is the HOLD lane, so a resolved set covering only wip and review DROPS assigned backlog
+    work from the tally — a regression against legacy introduced by the argument meant to fix the
+    renamed case. The resolved answer must cover every role the literal covered.
+    */
+    const selected = await select("backlog", new Set(["backlog", "building", "signoff"]));
+
+    expect(selected?.id).toBe("AG-IDLE");
+  });
+
+  it("does not count work parked outside the supplied lanes", async () => {
+    // A finished card must not hold load against its agent, or the agent looks busy forever.
+    const selected = await select("shipped", new Set(["backlog", "building", "signoff"]));
+
+    expect(selected?.id).toBe("AG-BUSY");
+  });
+});
--- a/packages/engine/src/tests/ephemeral-worker-manager.test.ts
+++ b/packages/engine/src/tests/ephemeral-worker-manager.test.ts
@@ -436,3 +436,140 @@ describe("EphemeralWorkerManager", () => {
    expect(harness.manager.getOwner("FN-RESET")).toBeUndefined();
  });
 });
+
+/*
+FNXC:WorkflowLifecycleColumns 2026-07-31-06:10 (engine feed):
+
+THE INVARIANT: the zombie sweep asks the task's OWN workflow whether its worker is still working.
+
+A LIVE WORKER WAS REAPED ON A RENAMED BOARD, and the two guards compounded in the worst possible
+order. `shouldDeleteOnSweep` tested a hard-coded terminal Set first, then fell through to
+`return task.column !== "in-progress"`. On a renamed board the terminal test missed, and the fallthrough
+is TRUE for a renamed wip lane — so an ephemeral worker ACTIVELY EXECUTING a task was classified as a
+zombie and deleted, destroying work in flight.
+
+Census-invisible on both halves: the terminal check is a `Set` literal (a definition, not a
+comparison), and the wip check was reached only after it. Found by grepping for lane-shaped list
+literals, not by the backlog.
+
+THE FALLBACK IS DELIBERATELY ASYMMETRIC. An unresolvable workflow keeps the legacy literals rather
+than guessing: failing to reap a dead worker costs a slot, reaping a live one destroys work. Those
+are not symmetric, so the uncertain case must fail toward keeping the worker.
+
+REVERT PROOF, measured: restore the literal pair and the renamed-wip case fails — the live worker is
+deleted.
+*/
+describe("ephemeral zombie sweep resolves the board's own lanes", () => {
+  const RENAMED_IR = {
+    version: "v2", id: "wf-renamed", name: "renamed", nodes: [], edges: [],
+    columns: [
+      { id: "backlog", name: "Backlog", traits: [{ trait: "intake" }, { trait: "hold" }] },
+      { id: "building", name: "Building", traits: [{ trait: "wip", config: { limitSetting: "maxConcurrent" } }] },
+      { id: "shipped", name: "Shipped", traits: [{ trait: "complete" }] },
+    ],
+  };
+
+  function renamedHarness() {
+    const harness = createHarness();
+    const selection = { workflowId: "wf-renamed", stepIds: [] as string[] };
+    Object.assign(harness.taskStore as unknown as Record<string, unknown>, {
+      getTaskWorkflowSelection: () => selection,
+      getTaskWorkflowSelectionAsync: async () => selection,
+      getWorkflowDefinition: async () => ({ ir: RENAMED_IR }),
+    });
+    return harness;
+  }
+
+  it("KEEPS a worker whose task sits in a RENAMED wip lane", async () => {
+    const harness = renamedHarness();
+    const live = makeAgent("renamed-live", { metadata: { agentKind: "task-worker" }, taskId: "FN-LIVE" });
+    harness.agentStore.agents.set(live.id, live);
+    harness.taskStore.tasks.set("FN-LIVE", makeTask("FN-LIVE", { column: "building" }));
+
+    await harness.manager.reconcileOrphaned();
+
+    expect(harness.agentStore.agents.has(live.id)).toBe(true);
+  });
+
+  it("still reaps a worker whose task reached a RENAMED complete lane", async () => {
+    // The sweep must keep sweeping — keeping everything would be its own leak.
+    const harness = renamedHarness();
+    const done = makeAgent("renamed-done", { metadata: { agentKind: "task-worker" }, taskId: "FN-DONE" });
+    harness.agentStore.agents.set(done.id, done);
+    harness.taskStore.tasks.set("FN-DONE", makeTask("FN-DONE", { column: "shipped" }));
+
+    await harness.manager.reconcileOrphaned();
+
+    expect(harness.agentStore.agents.has(done.id)).toBe(false);
+  });
+
+  it("still reaps a worker parked in a RENAMED hold lane", async () => {
+    const harness = renamedHarness();
+    const parked = makeAgent("renamed-parked", { metadata: { agentKind: "task-worker" }, taskId: "FN-PARKED" });
+    harness.agentStore.agents.set(parked.id, parked);
+    harness.taskStore.tasks.set("FN-PARKED", makeTask("FN-PARKED", { column: "backlog" }));
+
+    await harness.manager.reconcileOrphaned();
+
+    expect(harness.agentStore.agents.has(parked.id)).toBe(false);
+  });
+});
+
+/*
+FNXC:WorkflowLifecycleColumns 2026-07-31-09:30 (#2787 review — greptile P1):
+A SECOND wip lane must keep its worker too.
+
+`resolveLifecycleColumns` returns the FIRST column carrying each trait, so the first version of this
+fix recognised only one implementation lane. A board declaring two — a common shape once a workflow
+splits implementation from, say, an integration lane — still reaped a live worker in the second. The
+same defect this suite exists to prevent, one degree narrower, and it would have surfaced the first
+time someone added that column rather than at conversion time.
+
+The guard now unions `columnsWithFlag(ir, "countsTowardWip")`, which is every wip-bearing column.
+
+REVERT PROOF, measured: narrow the check back to `lanes.wip` and the second-lane case below fails.
+*/
+describe("the zombie sweep honours EVERY wip lane, not the first", () => {
+  const TWO_WIP_IR = {
+    version: "v2", id: "wf-two-wip", name: "two wip", nodes: [], edges: [],
+    columns: [
+      { id: "backlog", name: "Backlog", traits: [{ trait: "intake" }, { trait: "hold" }] },
+      { id: "building", name: "Building", traits: [{ trait: "wip", config: { limitSetting: "maxConcurrent" } }] },
+      { id: "integrating", name: "Integrating", traits: [{ trait: "wip", config: { limitSetting: "maxConcurrent" } }] },
+      { id: "shipped", name: "Shipped", traits: [{ trait: "complete" }] },
+    ],
+  };
+
+  function twoWipHarness() {
+    const harness = createHarness();
+    const selection = { workflowId: "wf-two-wip", stepIds: [] as string[] };
+    Object.assign(harness.taskStore as unknown as Record<string, unknown>, {
+      getTaskWorkflowSelection: () => selection,
+      getTaskWorkflowSelectionAsync: async () => selection,
+      getWorkflowDefinition: async () => ({ ir: TWO_WIP_IR }),
+    });
+    return harness;
+  }
+
+  it("KEEPS a worker whose task sits in the SECOND wip lane", async () => {
+    const harness = twoWipHarness();
+    const live = makeAgent("second-lane-live", { metadata: { agentKind: "task-worker" }, taskId: "FN-LIVE" });
+    harness.agentStore.agents.set(live.id, live);
+    harness.taskStore.tasks.set("FN-LIVE", makeTask("FN-LIVE", { column: "integrating" }));
+
+    await harness.manager.reconcileOrphaned();
+
+    expect(harness.agentStore.agents.has(live.id)).toBe(true);
+  });
+
+  it("still keeps a worker in the FIRST wip lane", async () => {
+    const harness = twoWipHarness();
+    const live = makeAgent("first-lane-live", { metadata: { agentKind: "task-worker" }, taskId: "FN-LIVE" });
+    harness.agentStore.agents.set(live.id, live);
+    harness.taskStore.tasks.set("FN-LIVE", makeTask("FN-LIVE", { column: "building" }));
+
+    await harness.manager.reconcileOrphaned();
+
+    expect(harness.agentStore.agents.has(live.id)).toBe(true);
+  });
+});
--- a/packages/engine/src/tests/eval-followups.test.ts
+++ b/packages/engine/src/tests/eval-followups.test.ts
@@ -83,6 +83,54 @@ describe("normalizeEvalFollowUps", () => {
    expect(followUps[0]?.matchedTaskId).toBe("FN-open");
  });

+  /*
+  FNXC:WorkflowLifecycleColumns 2026-07-31-06:40 (engine feed):
+  THE INVARIANT: "open" is the negation of the card's OWN terminal lanes.
+
+  Census-invisible: the old gate was `OPEN_COLUMNS`, a `Set` literal — a definition, not a
+  comparison — so nothing in the lifecycle backlog pointed at this file.
+
+  On a renamed board that set matched NOTHING, so `openTasks` was empty and this dedup had no live
+  work to compare against. Every eval run re-filed follow-ups it had already filed. The symptom is
+  DUPLICATE TASK CREATION, which reads as the evaluator being thorough rather than as a bug.
+
+  REVERT PROOF, measured: restore `OPEN_COLUMNS.has(task.column)` and the first case below fails —
+  the duplicate is created instead of suppressed.
+  */
+  it("suppresses a duplicate of an open task sitting in a RENAMED wip lane", async () => {
+    const followUps = await normalizeEvalFollowUps({
+      parentTaskId: "FN-1",
+      runId: "ER-1",
+      overallBand: "weak",
+      drafts: [{ title: "Investigate flaky verification command", description: "Investigate flaky verification command causing reruns.", reason: "Failed verification", evidenceRefs: ["workflow-1"] }],
+      store: makeStore({
+        openTasks: [{ id: "FN-open", column: "building", title: "Investigate flaky verification command", description: "x" }],
+        terminalColumnsByTaskId: { "FN-open": ["shipped", "vault"] },
+      }),
+      policyMode: "persist_only",
+    });
+
+    expect(followUps[0]?.state).toBe("suppressed");
+    expect(followUps[0]?.matchedTaskId).toBe("FN-open");
+  });
+
+  it("does NOT suppress against a card in a RENAMED terminal lane", async () => {
+    // The dedup must stay scoped to live work: a finished card should not block a fresh follow-up.
+    const followUps = await normalizeEvalFollowUps({
+      parentTaskId: "FN-1",
+      runId: "ER-1",
+      overallBand: "weak",
+      drafts: [{ title: "Investigate flaky verification command", description: "Investigate flaky verification command causing reruns.", reason: "Failed verification", evidenceRefs: ["workflow-1"] }],
+      store: makeStore({
+        openTasks: [{ id: "FN-shipped", column: "shipped", title: "Investigate flaky verification command", description: "x" }],
+        terminalColumnsByTaskId: { "FN-shipped": ["shipped", "vault"] },
+      }),
+      policyMode: "persist_only",
+    });
+
+    expect(followUps[0]?.suppressedReason).not.toBe("duplicate_open_task");
+  });
+
  it("suppresses duplicates from prior eval results", async () => {
    const priorKey = normalizeEvalFollowUpText("FN-1:Add regression test for merge flow:Add regression test for merge flow regressions.");
    const followUps = await normalizeEvalFollowUps({
--- a/packages/engine/src/tests/scheduler-load-lane-union.test.ts
+++ b/packages/engine/src/tests/scheduler-load-lane-union.test.ts
@@ -0,0 +1,75 @@
+// @vitest-environment node
+/*
+FNXC:WorkflowLifecycleColumns 2026-07-31-10:40 (#2787 review — greptile P1, second round):
+
+THE INVARIANT: the resolved load-lane set covers EVERY role the legacy literal covered.
+
+The legacy set is `{todo, in-progress, in-review}`, and `todo` is the HOLD/INTAKE lane. My first
+resolved union covered only wip and review — and because passing the argument OVERRIDES the
+fallback rather than extending it, assigned backlog work stopped counting as load. A regression
+against legacy behaviour, introduced by the argument meant to fix the renamed case.
+
+That is the general trap with override-shaped options: the resolved answer must be a superset of what
+the literal answered, or wiring the parameter is a downgrade for the roles it forgot. Cheap to get
+wrong, invisible in a test that only checks the renamed lane.
+
+This asserts the union the scheduler builds, driven by the real trait resolver, since the call site
+sits inside a dispatch path a unit test has no business standing up.
+*/
+import { describe, expect, it } from "vitest";
+import { columnsWithFlag } from "@fusion/core";
+import type { WorkflowIr } from "@fusion/core";
+
+const RENAMED_IR = {
+  version: "v2", id: "wf-renamed", name: "renamed", nodes: [], edges: [],
+  columns: [
+    { id: "inbox", name: "Inbox", traits: [{ trait: "intake" }] },
+    { id: "backlog", name: "Backlog", traits: [{ trait: "hold" }] },
+    { id: "building", name: "Building", traits: [{ trait: "wip", config: { limitSetting: "maxConcurrent" } }] },
+    { id: "signoff", name: "Sign-off", traits: [{ trait: "merge" }] },
+    { id: "shipped", name: "Shipped", traits: [{ trait: "complete" }] },
+  ],
+} as unknown as WorkflowIr;
+
+/** Mirrors the scheduler's union. Kept in step with `scheduler.ts` by the assertions below. */
+function loadLanes(ir: WorkflowIr): Set<string> {
+  return new Set<string>([
+    ...columnsWithFlag(ir, "intake"),
+    ...columnsWithFlag(ir, "hold"),
+    ...columnsWithFlag(ir, "countsTowardWip"),
+    ...columnsWithFlag(ir, "mergeOrchestration"),
+    ...columnsWithFlag(ir, "mergeBlocker"),
+    ...columnsWithFlag(ir, "humanReview"),
+  ]);
+}
+
+describe("the scheduler's load-lane union covers every legacy role", () => {
+  it("includes the hold and intake lanes — the roles `todo` filled", () => {
+    const lanes = loadLanes(RENAMED_IR);
+
+    expect(lanes.has("backlog")).toBe(true);
+    expect(lanes.has("inbox")).toBe(true);
+  });
+
+  it("includes the wip and review lanes", () => {
+    const lanes = loadLanes(RENAMED_IR);
+
+    expect(lanes.has("building")).toBe(true);
+    expect(lanes.has("signoff")).toBe(true);
+  });
+
+  it("excludes terminal lanes — finished work must not hold load against an agent", () => {
+    expect(loadLanes(RENAMED_IR).has("shipped")).toBe(false);
+  });
+
+  it("the scheduler builds this same union", () => {
+    // Guards the mirror above against drift: if scheduler.ts stops unioning a role, this fails.
+    const source = readFileSync(new URL("../scheduler.ts", import.meta.url), "utf8");
+
+    for (const flag of ["intake", "hold", "countsTowardWip", "mergeOrchestration", "mergeBlocker", "humanReview"]) {
+      expect(source).toContain(`...columnsWithFlag(loadLaneIr, "${flag}")`);
+    }
+  });
+});
+
+import { readFileSync } from "node:fs";
--- a/packages/engine/src/agent-assignment.ts
+++ b/packages/engine/src/agent-assignment.ts
@@ -1,12 +1,30 @@
 import type { Agent, AgentStore, Task, TaskStore } from "@fusion/core";
 import { isAgentAutoAssignable, isEphemeralAgent } from "@fusion/core";

-const ACTIVE_COLUMNS = new Set(["todo", "in-progress", "in-review"]);
+/*
+FNXC:WorkflowLifecycleColumns 2026-07-31-05:40 (batch-engine feed):
+The lanes an assigned card still counts as LOAD against its agent.
+
+CENSUS-INVISIBLE: this is a `Set` literal, i.e. a definition rather than a comparison, so nothing in
+the lifecycle backlog ever pointed at this file. Found by grepping for lane-shaped list literals
+after the same shape turned up in `duplicate-intake` and `blocker-fanout`.
+
+The failure is a silent DEGRADATION, not an error. This set gates the per-agent assignment-load
+tally used to pick the least-loaded agent. On a renamed board no task's column matched, so
+`assignmentLoad` stayed empty, every candidate compared as load 0, and the sort fell straight through
+to its `createdAt` tiebreak — which is stable. The result is that the SAME agent wins every
+assignment while the others sit idle. Nothing logs, nothing fails; the board just distributes badly.
+
+DELIBERATE-LITERAL — the fallback for a caller that cannot resolve lanes, reviewed 2026-07-31-05:40.
+*/
+const LEGACY_ACTIVE_COLUMNS: ReadonlySet<string> = new Set(["todo", "in-progress", "in-review"]);

 type SelectPermanentAgentForTaskOptions = {
  task: Task;
  agentStore: Pick<AgentStore, "listAgents" | "getChainOfCommand">;
  taskStore: Pick<TaskStore, "listTasks">;
+  /** Resolved lanes that count as load. Omitted → the legacy trio, i.e. today's behaviour. */
+  activeColumns?: ReadonlySet<string>;
 };

 function isAgentEnabled(agent: Agent): boolean {
@@ -48,7 +66,7 @@ function taskLinksToScope(task: Pick<Task, "id" | "missionId" | "sliceId">, scop
  return false;
 }

-export async function selectPermanentAgentForTask({ task, agentStore, taskStore }: SelectPermanentAgentForTaskOptions): Promise<Agent | null> {
+export async function selectPermanentAgentForTask({ task, agentStore, taskStore, activeColumns }: SelectPermanentAgentForTaskOptions): Promise<Agent | null> {
  const eligibleAgents = await listEligibleExecutorAgents(agentStore);

  if (eligibleAgents.length === 0) {
@@ -81,7 +99,7 @@ export async function selectPermanentAgentForTask({ task, agentStore, taskStore

  const assignmentLoad = new Map<string, number>();
  for (const taskItem of allTasks) {
-    if (!taskItem.assignedAgentId || !ACTIVE_COLUMNS.has(taskItem.column)) continue;
+    if (!taskItem.assignedAgentId || !(activeColumns ?? LEGACY_ACTIVE_COLUMNS).has(taskItem.column)) continue;
    assignmentLoad.set(taskItem.assignedAgentId, (assignmentLoad.get(taskItem.assignedAgentId) ?? 0) + 1);
  }

--- a/packages/engine/src/ephemeral-worker-manager.ts
+++ b/packages/engine/src/ephemeral-worker-manager.ts
@@ -20,7 +20,7 @@
 * sweep here close that gap.
 */
 import type { AgentStore, AgentState, Agent, TaskStore, Task, Settings } from "@fusion/core";
-import { isEphemeralAgent } from "@fusion/core";
+import { isEphemeralAgent, resolveWorkflowIrForTask, columnsWithFlag } from "@fusion/core";

 export interface TaskOwner {
  agentId: string;
@@ -339,8 +339,44 @@ export class EphemeralWorkerManager {
    try {
      const task = await this.taskStore.getTask(agent.taskId);
      if (!task) return true;
-      if (TERMINAL_TASK_COLUMNS.has(task.column)) return true;
-      return task.column !== "in-progress";
+      /*
+      FNXC:WorkflowLifecycleColumns 2026-07-31-06:10 (engine feed):
+      A LIVE WORKER WAS REAPED AS A ZOMBIE ON A RENAMED BOARD.
+
+      Census-invisible: the terminal check is a `Set` literal (a definition, not a comparison), and
+      the wip check below it was the bare literal. Both missed on a renamed board, and they compound
+      in the WORST order — the terminal test failed, so control fell to
+      `return task.column !== "in-progress"`, which is TRUE for a renamed wip lane. An ephemeral
+      worker actively executing a task was therefore classified as a zombie and deleted by the sweep.
+
+      Resolved from the task's OWN workflow. The `catch` below already treats an unreadable task as a
+      broken binding, so an unresolvable workflow keeps the documented literals rather than inventing
+      a lane: failing to reap a dead worker costs a slot, while reaping a live one destroys work in
+      flight, and those are not symmetric.
+      */
+      /*
+      FNXC:WorkflowLifecycleColumns 2026-07-31-09:30 (#2787 review — greptile P1):
+      MEMBERSHIP, not first-per-role. `resolveLifecycleColumns` returns the FIRST column carrying each
+      trait, so a workflow declaring TWO implementation lanes had only one of them recognised — a live
+      worker in the second lane was still classified as a zombie and deleted. Same defect this commit
+      exists to fix, one degree narrower, and it would have reappeared the first time someone declared
+      a second wip lane.
+
+      `columnsWithFlag` returns every column carrying the trait, so both halves are unions.
+      */
+      const ir = await resolveWorkflowIrForTask(this.taskStore, task.id).catch(() => undefined);
+      if (ir === undefined) {
+        /* DELIBERATE-LITERAL — the unresolvable-workflow default, reviewed 2026-07-31-06:10. */
+        if (TERMINAL_TASK_COLUMNS.has(task.column)) return true;
+        return task.column !== "in-progress";
+      }
+      const terminalLanes = new Set<string>([
+        ...columnsWithFlag(ir, "complete"),
+        ...columnsWithFlag(ir, "archived"),
+      ]);
+      if (terminalLanes.has(task.column)) return true;
+      const wipLanes = new Set<string>(columnsWithFlag(ir, "countsTowardWip"));
+      return !wipLanes.has(task.column);
    } catch {
      // If we can't even read the task, assume the binding is broken.
      return true;
--- a/packages/engine/src/eval-followups.ts
+++ b/packages/engine/src/eval-followups.ts
@@ -8,8 +8,8 @@ import {
  type TaskStore,
 } from "@fusion/core";
 import { resolveTerminalColumnsFor } from "./executor.js";
+import type { resolveWorkflowIrForTask } from "@fusion/core";

-const OPEN_COLUMNS = new Set(["triage", "todo", "in-progress", "in-review"]);
 /*
 FNXC:Evals 2026-07-26-00:00:
 Eval follow-ups are a real product feature, but they used to borrow the shared automated-recovery follow-up engine (`createAutomatedFollowup` in verification-followup-dedup.ts) purely for its dedup pass. That engine was deleted along with the recovery follow-up cards it existed to file, so the one dedup rule this feature actually needs is inlined here: never create a second card for the same `suggestionId` under the same parent while one is still open. Closed columns (done/archived) are excluded so a re-run after the follow-up is finished can legitimately file a fresh card.
@@ -105,9 +105,41 @@ export function resolveEvalFollowUpPolicyMode(policy?: "off" | "suggest" | "crea

 export async function normalizeEvalFollowUps(input: NormalizeEvalFollowUpsInput): Promise<EvalFollowUpSuggestion[]> {
  const { parentTaskId, runId, drafts, overallBand, store, policyMode } = input;
-  const openTasks = (await store.listTasks({ slim: true, includeArchived: false })).filter((task) =>
-    OPEN_COLUMNS.has(task.column)
-  );
+  /*
+  FNXC:WorkflowLifecycleColumns 2026-07-31-06:40 (engine feed):
+  "Open" is the negation of the task's OWN terminal lanes, not a hard-coded list of four ids.
+
+  Census-invisible: `OPEN_COLUMNS` is a `Set` literal — a definition, not a comparison — so nothing
+  in the lifecycle backlog pointed here. Found by grepping for lane-shaped list literals.
+
+  Consequence on a renamed board: the set matched NOTHING, so `openTasks` was empty and the
+  dedupe below had no live work to compare against. Every eval run then re-filed follow-ups it had
+  already filed — the failure is DUPLICATE TASK CREATION, which looks like the evaluator being
+  thorough rather than like a bug.
+
+  Uses `resolveTerminalColumnsFor` — ALREADY IMPORTED IN THIS FILE for the candidate loop below, and
+  the same helper the executor uses. It unions the task's resolved terminals with the legacy pair for
+  the degraded-IR reason documented at its definition, so there is no fallback to hand-write here.
+
+  My first draft manufactured synthetic trait flags from lane equality to call `isTerminalColumnRole`
+  instead. That is the anti-pattern I flagged in `task-update.ts` two commits ago — a longer way to
+  write the same comparison while LOOKING like it consulted the trait registry — and it ignored a
+  correct helper sitting three lines above the import. Replaced before commit.
+  */
+  const allLiveTasks = await store.listTasks({ slim: true, includeArchived: false });
+  /*
+  FNXC:WorkflowLifecycleColumns 2026-07-31-09:30 (#2787 review — greptile P2):
+  ONE IR read per WORKFLOW, not per card. This loop runs over every live task on the board, so
+  without the shared cache a large board paid a store read per card before a single follow-up draft
+  was processed — measurable added latency on every scheduled evaluation. The cache is the
+  caller-owned shape the rest of this program already uses.
+  */
+  const openIrCache = new Map<string, Awaited<ReturnType<typeof resolveWorkflowIrForTask>>>();
+  const openTasks: typeof allLiveTasks = [];
+  for (const task of allLiveTasks) {
+    const terminal = await resolveTerminalColumnsFor(store, task.id, openIrCache);
+    if (!terminal.includes(task.column)) openTasks.push(task);
+  }
  // FNXC:Evals 2026-06-27-12:40:
  // getEvalStore() returns EvalStore | AsyncEvalStore (PG backend mode); await
  // resolves the sync array and the async promise alike.
--- a/packages/engine/src/executor.ts
+++ b/packages/engine/src/executor.ts
@@ -1789,7 +1789,17 @@ EXPORTED rather than copied. `eval-followups.ts` and `pr-comment-handler.ts` eac
 and fourth copy of the union-with-legacy reasoning is exactly the drift this program exists to remove.
 Nothing else about the function changes.
 */
-export async function resolveTerminalColumnsFor(store: TaskStore, taskId: string): Promise<readonly string[]> {
+export async function resolveTerminalColumnsFor(
+  store: TaskStore,
+  taskId: string,
+  /*
+  FNXC:WorkflowLifecycleColumns 2026-07-31-09:30 (#2787 review — greptile P2):
+  Optional CALLER-OWNED IR cache, matching the contract on `resolveTaskLifecycleColumns`. Sweeps that
+  call this once per card on a whole board must read one IR per WORKFLOW, not one per task; callers
+  resolving a single task pass nothing and are unaffected.
+  */
+  irCache?: Map<string, Awaited<ReturnType<typeof resolveWorkflowIrForTask>>>,
+): Promise<readonly string[]> {
  /*
  FNXC:WorkflowLifecycleColumns 2026-07-31-12:20 (PR #2568 review — greptile):
  THE UNION IS DELIBERATE, and the `catch` alone was not enough.
@@ -1812,7 +1822,7 @@ export async function resolveTerminalColumnsFor(store: TaskStore, taskId: string
  column, which is the failure the conversion exists to prevent.
  */
  try {
-    const resolved = resolveTerminalColumns(await resolveWorkflowIrForTask(store, taskId));
+    const resolved = resolveTerminalColumns(await resolveWorkflowIrForTask(store, taskId, irCache));
    return [...new Set([...resolved, ...LEGACY_TERMINAL_COLUMNS])];
  } catch {
    return LEGACY_TERMINAL_COLUMNS;
--- a/packages/engine/src/scheduler.ts
+++ b/packages/engine/src/scheduler.ts
@@ -2296,10 +2296,48 @@ export class Scheduler {
              return null;
            }

+            /*
+            FNXC:WorkflowLifecycleColumns 2026-07-31-09:30 (#2787 review — greptile P1):
+            PASS THE RESOLVED LANES. Without this the optional parameter added to
+            `selectPermanentAgentForTask` is never supplied by the only production caller, so the
+            predicate keeps its legacy default and the load tally stays empty on a renamed board —
+            a converted function reachable only through an argument nobody passes is the
+            guard-that-cannot-fire pattern, and shipping one would have been worse than leaving the
+            literal in place, because the site then reads as done.
+
+            The set is a MEMBERSHIP union of every wip/review lane the board declares, not the
+            first-per-role ids: a workflow may declare more than one implementation lane, and load
+            held in the second must still count.
+            */
+            const loadLaneIr = await resolveWorkflowIrForTask(this.store, freshTask.id).catch(() => undefined);
+            /*
+            FNXC:WorkflowLifecycleColumns 2026-07-31-10:40 (#2787 review — greptile P1, second round):
+            THE HOLD AND INTAKE LANES COUNT AS LOAD TOO.
+
+            The legacy set is `{todo, in-progress, in-review}` — and `todo` is the HOLD/INTAKE lane.
+            My first union covered only wip and review, so passing it OVERRODE the fallback and
+            dropped assigned backlog work from the tally: a regression against the legacy behaviour
+            for that lane, introduced by the very argument meant to fix the renamed case.
+
+            That is the trap in overriding a default rather than extending it — the resolved answer
+            must cover EVERY role the literal covered, or wiring the parameter is a downgrade for the
+            roles it forgot.
+            */
+            const activeLoadColumns = loadLaneIr === undefined
+              ? undefined
+              : new Set<string>([
+                ...columnsWithFlag(loadLaneIr, "intake"),
+                ...columnsWithFlag(loadLaneIr, "hold"),
+                ...columnsWithFlag(loadLaneIr, "countsTowardWip"),
+                ...columnsWithFlag(loadLaneIr, "mergeOrchestration"),
+                ...columnsWithFlag(loadLaneIr, "mergeBlocker"),
+                ...columnsWithFlag(loadLaneIr, "humanReview"),
+              ]);
            const selectedAgent = await selectPermanentAgentForTask({
              task: freshTask,
              agentStore: this.options.agentStore,
              taskStore: this.store,
+              ...(activeLoadColumns && activeLoadColumns.size > 0 ? { activeColumns: activeLoadColumns } : {}),
            });
            if (!selectedAgent) {
              await this.store.updateTask(task.id, { status: "queued" });