diff --git a/.changeset/capacity-pool-sentinel.md b/.changeset/capacity-pool-sentinel.md
new file mode 100644
index 0000000000..7c062e5c5c
--- /dev/null
+++ b/.changeset/capacity-pool-sentinel.md
@@ -0,0 +1,7 @@
+---
+"@runfusion/fusion": patch
+---
+
+summary: Fix the column capacity check so a task with no workflow selection is counted against the limit.
+category: fix
+dev: `moves.ts` asked `countActiveInCapacitySlotAsync` for pool `"builtin:coding"` while the counter buckets selection-less rows under `DEFAULT_WORKFLOW_POOL_ID`, so the count was always 0 and a finite limit could never bind. Both sides now derive the pool through the shared `resolveCapacityPoolId`. NOTE: no operator-visible change yet — the capacity block is still gated on `experimentalFeatures.workflowColumns`, which nothing in production sets (Phase A3 R2). If that gate is removed, this becomes user-visible and the changeset should be re-categorised.
diff --git a/package.json b/package.json
index 2ca66b86d2..b6986dab4d 100644
--- a/package.json
+++ b/package.json
@@ -14,14 +14,14 @@
   "type": "module",
   "packageManager": "pnpm@10.33.0",
   "scripts": {
-    "pretest": "node scripts/check-no-nohup.mjs && node scripts/check-no-cwd-relative-dashboard-test-reads.mjs && node scripts/check-no-kill-4040.mjs && node scripts/check-no-getdatabase.mjs && node scripts/check-no-node-only-core-imports-in-dashboard.mjs && node scripts/check-pi-versions-pinned.mjs && node scripts/check-no-test-timeout-appeasement.mjs && node scripts/check-changeset-format.mjs && node scripts/check-routes-modular.mjs",
-    "pretest:full": "node scripts/check-no-nohup.mjs && node scripts/check-no-cwd-relative-dashboard-test-reads.mjs && node scripts/check-no-kill-4040.mjs && node scripts/check-no-getdatabase.mjs && node scripts/check-no-node-only-core-imports-in-dashboard.mjs && node scripts/check-pi-versions-pinned.mjs && node scripts/check-no-test-timeout-appeasement.mjs && node scripts/check-changeset-format.mjs && node scripts/check-routes-modular.mjs",
+    "pretest": "node scripts/check-no-nohup.mjs && node scripts/check-no-cwd-relative-dashboard-test-reads.mjs && node scripts/check-no-kill-4040.mjs && node scripts/check-no-getdatabase.mjs && node scripts/check-capacity-pool-id.mjs && node scripts/check-no-node-only-core-imports-in-dashboard.mjs && node scripts/check-pi-versions-pinned.mjs && node scripts/check-no-test-timeout-appeasement.mjs && node scripts/check-changeset-format.mjs && node scripts/check-routes-modular.mjs",
+    "pretest:full": "node scripts/check-no-nohup.mjs && node scripts/check-no-cwd-relative-dashboard-test-reads.mjs && node scripts/check-no-kill-4040.mjs && node scripts/check-no-getdatabase.mjs && node scripts/check-capacity-pool-id.mjs && node scripts/check-no-node-only-core-imports-in-dashboard.mjs && node scripts/check-pi-versions-pinned.mjs && node scripts/check-no-test-timeout-appeasement.mjs && node scripts/check-changeset-format.mjs && node scripts/check-routes-modular.mjs",
     "check:line-count": "node scripts/check-file-line-count.mjs",
     "check:routes-modular": "node scripts/check-routes-modular.mjs",
     "check:changesets": "node scripts/check-changeset-format.mjs",
     "check:quarantine-ledger": "node scripts/check-quarantine-ledger.mjs",
     "check:mock-completeness": "node scripts/check-mock-completeness.mjs",
-    "test:gate": "node scripts/check-no-nohup.mjs && node scripts/check-no-cwd-relative-dashboard-test-reads.mjs && node scripts/check-no-kill-4040.mjs && node scripts/check-no-getdatabase.mjs && node scripts/check-no-node-only-core-imports-in-dashboard.mjs && node scripts/check-pi-versions-pinned.mjs && node scripts/check-no-test-timeout-appeasement.mjs && node scripts/check-changeset-format.mjs && node scripts/check-mock-completeness.mjs && sh -c 'pnpm --filter @fusion/engine test:core & engine_pid=$!; pnpm --filter @fusion/core test:pg-gate & pg_pid=$!; status=0; wait $engine_pid || status=1; wait $pg_pid || status=1; exit $status' && pnpm --filter @runfusion/fusion test:ci-shape",
+    "test:gate": "node scripts/check-no-nohup.mjs && node scripts/check-no-cwd-relative-dashboard-test-reads.mjs && node scripts/check-no-kill-4040.mjs && node scripts/check-no-getdatabase.mjs && node scripts/check-capacity-pool-id.mjs && node scripts/check-no-node-only-core-imports-in-dashboard.mjs && node scripts/check-pi-versions-pinned.mjs && node scripts/check-no-test-timeout-appeasement.mjs && node scripts/check-changeset-format.mjs && node scripts/check-mock-completeness.mjs && sh -c 'pnpm --filter @fusion/engine test:core & engine_pid=$!; pnpm --filter @fusion/core test:pg-gate & pg_pid=$!; status=0; wait $engine_pid || status=1; wait $pg_pid || status=1; exit $status' && pnpm --filter @runfusion/fusion test:ci-shape",
     "smoke:boot": "node scripts/boot-smoke.mjs",
     "local": "node scripts/start-local.mjs",
     "dev": "node scripts/dev-with-memory.mjs",
diff --git a/packages/core/src/__tests__/postgres/move-path-equivalence.pg.test.ts b/packages/core/src/__tests__/postgres/move-path-equivalence.pg.test.ts
index 426c798ec1..1090d4804e 100644
--- a/packages/core/src/__tests__/postgres/move-path-equivalence.pg.test.ts
+++ b/packages/core/src/__tests__/postgres/move-path-equivalence.pg.test.ts
@@ -370,34 +370,43 @@ pgTest("move-path equivalence — the flag gates MORE than side effects (Phase A
     expect(hooksErr!.message).toContain("Unknown column for this workflow");
   });
 
-  it("UNPROVEN: in-transaction column capacity did NOT reject on EITHER path in this fixture", async () => {
+  it("DIVERGENCE: in-transaction capacity rejects on the HOOKS path only — the inline path cannot run it", async () => {
     /*
-    HONEST NEGATIVE RESULT — recorded rather than dropped, because the code gate
-    is real and the practical impact is not what reading it suggests.
+    FNXC:WorkflowCapacity 2026-07-28-19:40 (pool-id sentinel fix):
+    WAS `UNPROVEN: … did NOT reject on EITHER path`. That test recorded an honest
+    negative result and left the cause open: "something further in
+    (`resolveColumnCapacity`'s limit resolution, or what
+    `countActiveInCapacitySlotAsync` counts as an occupant) keeps the check from
+    firing even when the flag is on. This suite does not establish which." It
+    also predicted its own obsolescence: "if a future change makes this reject,
+    that is the capacity gate coming alive."
 
-    STRUCTURALLY, `moves.ts` wraps the whole in-transaction capacity block in
-    `if (useWorkflow && workflowIr && fromColumn !== toColumn)`, so it cannot run
-    on the LIVE inline path at all. The obvious inference is "converging onto the
-    hooks path turns store-level capacity rejection ON for every project" — a
-    serious blast radius, since `capacity-exhausted` is a code the graph column
-    boundary parks on and the promote route surfaces to operators.
+    THE ANSWER, established by the sentinel fix: neither of those guesses. The
+    counter and the limit resolution were both fine. `moves.ts` asked the counter
+    for occupants of pool `"builtin:coding"` while the counter buckets
+    selection-less rows under `DEFAULT_WORKFLOW_POOL_ID`, so the count came back
+    0 for a pool nothing is ever placed in. Both sides now derive the pool
+    through `resolveCapacityPoolId`, and the hooks path rejects.
 
-    EMPIRICALLY that inference does not hold up: with `maxConcurrent: 1` and a
-    wip column already occupied, the second move was ACCEPTED on both paths. So
-    something further in (`resolveColumnCapacity`'s limit resolution, or what
-    `countActiveInCapacitySlotAsync` counts as an occupant — a task with no
-    session/agent may not count) keeps the check from firing even when the flag
-    is on. This suite does not establish which.
-
-    Consequence for the convergence decision: the capacity blast radius is
-    UNQUANTIFIED, not absent. It needs its own investigation before either path
-    is made authoritative — this test pins today's observed behavior so that
-    investigation starts from a fact instead of the code reading.
+    The blast-radius question this test was holding open is therefore ANSWERED for
+    the hooks path and STILL OPEN for the inline one: the inline path remains
+    structurally unable to run the block (`if (useWorkflow && …)`), so converging
+    the paths still turns store-level capacity rejection on for every project.
+    That convergence stays an operator decision — see the R2 note in
+    workflow-capacity-invariant.pg.test.ts.
     */
     const store = h.store();
     await store.updateSettings({ maxConcurrent: 1 });
 
+    /* Each phase starts from an EMPTY wip column. Before the gate bound, the two phases could share
+       one fixture because nothing ever counted occupants; now they cannot — the inline phase leaves
+       two cards in wip, and the hooks phase's own HOLDER move would trip the cap before the
+       contended move under test ever runs (observed: "column at capacity (2/1)"). Evacuating is
+       what keeps this a test of the contender's move rather than of fixture residue. */
     async function fillThenMoveSecond(): Promise<Error | null> {
+      for (const stale of await store.listTasks({ includeArchived: false })) {
+        if (stale.column === "in-progress") await store.deleteTask(stale.id);
+      }
       const first = await store.createTask({ description: "capacity holder" });
       await store.moveTask(first.id, "todo");
       await store.moveTask(first.id, "in-progress");
@@ -407,11 +416,14 @@ pgTest("move-path equivalence — the flag gates MORE than side effects (Phase A
     }
 
     await setPath("inline");
+    // Unchanged: the block is unreachable on this path regardless of the pool id.
     expect(await fillThenMoveSecond()).toBeNull();
 
     await setPath("hooks");
-    // If a future change makes this reject, that is the capacity gate coming
-    // alive — and this failure is the signal to re-open the blast-radius question.
-    expect(await fillThenMoveSecond()).toBeNull();
+    const hooksErr = await fillThenMoveSecond();
+    expect(hooksErr).toBeInstanceOf(Error);
+    expect((hooksErr as unknown as { rejection?: { code?: string } }).rejection?.code).toBe(
+      "capacity-exhausted",
+    );
   });
 });
diff --git a/packages/core/src/__tests__/postgres/workflow-capacity-invariant.pg.test.ts b/packages/core/src/__tests__/postgres/workflow-capacity-invariant.pg.test.ts
index d7a9a3e44b..7a035b209e 100644
--- a/packages/core/src/__tests__/postgres/workflow-capacity-invariant.pg.test.ts
+++ b/packages/core/src/__tests__/postgres/workflow-capacity-invariant.pg.test.ts
@@ -111,9 +111,25 @@ pgTest("in-transaction column capacity — ground truth (Phase A3)", () => {
     return { error, secondColumn: (await store.getTask(contender.id))?.column };
   }
 
-  it("DEFECT (R2): on the LIVE inline path the in-txn capacity check cannot run at all", async () => {
-    // Not a surprise, but it is the reason the defect is latent rather than
-    // active today: the whole block is inside `if (useWorkflow && …)`.
+  it("DEFECT (R2, STILL LIVE): on the production inline path the in-txn capacity check cannot run at all", async () => {
+    /*
+    FNXC:WorkflowCapacity 2026-07-28-19:05:
+    R1 IS FIXED; R2 IS NOT, and this test is the standing evidence. The whole
+    capacity block sits inside `if (useWorkflow && …)`, and `useWorkflow` reads
+    `experimentalFeatures.workflowColumns === true`, which NOTHING in production
+    sets (it is absent from DEFAULT_GLOBAL_SETTINGS and has no writer outside
+    tests). So on the path real projects take, the gate still does not run at all
+    and cards still enter wip past the cap.
+
+    Concretely, measured on this suite's fixture with maxConcurrent 1:
+        flag OFF, no selection  -> ADMITTED   (this test)
+        flag OFF, selection     -> ADMITTED
+        flag ON,  no selection  -> REFUSED    (was ADMITTED before the R1 fix)
+        flag ON,  selection     -> REFUSED
+    Making the gate bind for real projects means removing the `useWorkflow`
+    condition from the capacity block — a separate, larger blast radius than the
+    sentinel fix, and an operator decision rather than a drive-by.
+    */
     const store = h.store();
     await store.updateSettings({ maxConcurrent: 1 });
     await setPath("inline");
@@ -124,12 +140,14 @@ pgTest("in-transaction column capacity — ground truth (Phase A3)", () => {
     expect(secondColumn).toBe("in-progress"); // limit of 1, two occupants
   });
 
-  it("DEFECT (R1): flag-ON, a NO-SELECTION task still slips the limit — the pool sentinels disagree", async () => {
+  it("FIXED (R1): flag-ON, a NO-SELECTION task is now REFUSED at the limit", async () => {
     /*
-    THE ACTUAL BUG. `moves.ts:319` asks the counter for pool "builtin:coding";
-    `countActiveInCapacitySlotAsyncImpl` buckets no-selection rows under
-    "__default-workflow__". No occupant is ever counted, so the limit cannot
-    bind. Flip this expectation to a rejection when the sentinel is fixed.
+    FNXC:WorkflowCapacity 2026-07-28-19:05:
+    Was `DEFECT (R1)`, asserting the wrong-but-current outcome. The pool-id
+    sentinel is fixed: `moves.ts` and the counter now BOTH derive the pool through
+    `resolveCapacityPoolId`, so a selection-less row is bucketed and asked about
+    under the same key and the limit binds. Flipping this expectation is the
+    acceptance test the original ratchet named.
     */
     const store = h.store();
     await store.updateSettings({ maxConcurrent: 1 });
@@ -137,8 +155,12 @@ pgTest("in-transaction column capacity — ground truth (Phase A3)", () => {
 
     const { error, secondColumn } = await fillWipThenAdmitSecond();
 
-    expect(error).toBeNull();
-    expect(secondColumn).toBe("in-progress");
+    expect(error).toBeInstanceOf(Error);
+    expect((error as unknown as { rejection?: { code?: string } }).rejection?.code).toBe(
+      "capacity-exhausted",
+    );
+    expect(secondColumn).toBe("todo"); // refused, stays put
+
   });
 
   it("DISCRIMINATOR: with an EXPLICIT builtin:coding selection the sentinels agree and the limit BINDS", async () => {
@@ -181,8 +203,8 @@ pgTest("in-transaction column capacity — ground truth (Phase A3)", () => {
   invariant nobody can prove is an invariant nobody has; this is the proof
   obligation, written down.
   */
-  it.fails(
-    "INVARIANT (currently BROKEN): a move into a full capacity column is refused, even with no workflow selection",
+  it(
+    "INVARIANT (HOLDS on the flag-ON path): a move into a full capacity column is refused, even with no workflow selection",
     async () => {
       const store = h.store();
       await store.updateSettings({ maxConcurrent: 1 });
diff --git a/packages/core/src/index.gate.ts b/packages/core/src/index.gate.ts
index fa12580c2a..e226a70954 100644
--- a/packages/core/src/index.gate.ts
+++ b/packages/core/src/index.gate.ts
@@ -416,7 +416,7 @@ export {
 } from "./plugin-gate-verdict.js";
 export type { PluginGateVerdict, ColumnPluginGate } from "./plugin-gate-verdict.js";
 // ── U6: workflow capacity (WIP) resolution shared by store + sweep ───────────
-export { resolveColumnCapacity, DEFAULT_WORKFLOW_POOL_ID } from "./workflow-capacity.js";
+export { resolveColumnCapacity, DEFAULT_WORKFLOW_POOL_ID, resolveCapacityPoolId } from "./workflow-capacity.js";
 export type { ColumnCapacity } from "./workflow-capacity.js";
 // ── U5: workflow lifecycle reconciliation (switch / edit / delete) ───────────
 export {
diff --git a/packages/core/src/index.ts b/packages/core/src/index.ts
index fe56e7cd21..8abfe3f004 100644
--- a/packages/core/src/index.ts
+++ b/packages/core/src/index.ts
@@ -447,7 +447,7 @@ export {
 } from "./plugin-gate-verdict.js";
 export type { PluginGateVerdict, ColumnPluginGate } from "./plugin-gate-verdict.js";
 // ── U6: workflow capacity (WIP) resolution shared by store + sweep ───────────
-export { resolveColumnCapacity, resolveWipBudgetColumns, DEFAULT_WORKFLOW_POOL_ID } from "./workflow-capacity.js";
+export { resolveColumnCapacity, resolveWipBudgetColumns, DEFAULT_WORKFLOW_POOL_ID, resolveCapacityPoolId } from "./workflow-capacity.js";
 export { createWorkflowEventBus, getWorkflowEventBus, emitWorkflowLifecycleEvent, resetWorkflowEventBusForTesting } from "./workflow-events.js";
 export type { WorkflowEventBus, WorkflowEventSubscriber, WorkflowEventSubscription } from "./workflow-events.js";
 export { findWorkflowEventShapeViolations, isIdsOnlyWorkflowEvent, MAX_ID_VALUE_LENGTH } from "./types/workflow-events.js";
diff --git a/packages/core/src/task-store/moves.ts b/packages/core/src/task-store/moves.ts
index 650d2cf6e1..a61773a2f0 100644
--- a/packages/core/src/task-store/moves.ts
+++ b/packages/core/src/task-store/moves.ts
@@ -19,7 +19,7 @@ import {isBuiltinWorkflowId, getBuiltinWorkflow, resolveDefaultWorkflowIr, DEFAU
 import {parseWorkflowIr} from "../workflow-ir.js";
 import {findWorkflowColumn, resolveColumnPluginGates} from "../plugin-gate-verdict.js";
 import {getTraitRegistry, resolveColumnFlags} from "../trait-registry.js";
-import {resolveColumnCapacity, resolveWipBudgetColumns} from "../workflow-capacity.js";
+import {resolveColumnCapacity, resolveWipBudgetColumns, resolveCapacityPoolId} from "../workflow-capacity.js";
 import {
   type TransitionColumnFacts,
   evaluateCapacityRejection,
@@ -315,9 +315,19 @@ export async function moveTaskInternalImpl(store: TaskStore, id: string, toColum
     // capacity check is not a guard (U6 fills the enforcement; U4 leaves a
     // pass-through slot). An explicit option value wins; otherwise derive it.
     const bypassGuards = store.resolveWorkflowBypassGuards(moveSource, options);
-    const effectiveWorkflowIdForMove = useWorkflow
-      ? (await store.getTaskWorkflowSelectionAsync(id))?.workflowId ?? "builtin:coding"
-      : "builtin:coding";
+    /*
+    FNXC:WorkflowCapacity 2026-07-28-19:05 (pool-id sentinel fix):
+    ONE selection read, TWO derived ids, because they are two different things
+    that were previously conflated into one variable — and the conflation is the
+    defect. The capacity POOL key must match how the counter buckets rows
+    (`resolveCapacityPoolId`, shared); the WORKFLOW id is telemetry and must stay
+    a real workflow id, never the bucketing sentinel.
+    */
+    const workflowSelectionForMove = useWorkflow
+      ? await store.getTaskWorkflowSelectionAsync(id)
+      : undefined;
+    const effectiveWorkflowIdForMove = workflowSelectionForMove?.workflowId ?? DEFAULT_WORKFLOW_ID;
+    const capacityPoolIdForMove = resolveCapacityPoolId(workflowSelectionForMove?.workflowId);
     const workflowIr: WorkflowIr | undefined = useWorkflow
       ? await resolveTaskWorkflowIrForMove(store, id)
       : undefined;
@@ -932,7 +942,7 @@ export async function moveTaskInternalImpl(store: TaskStore, id: string, toColum
               store.countActiveInCapacitySlotAsync({
                 tx,
                 targetColumn: budgetColumn,
-                workflowId: effectiveWorkflowIdForMove,
+                workflowId: capacityPoolIdForMove,
                 countPending,
                 excludeTaskId: id,
               }),
diff --git a/packages/core/src/task-store/project-store-ops.ts b/packages/core/src/task-store/project-store-ops.ts
index aac0e95e78..7b31b60441 100644
--- a/packages/core/src/task-store/project-store-ops.ts
+++ b/packages/core/src/task-store/project-store-ops.ts
@@ -9,6 +9,7 @@
  * instance as its first parameter and performs byte-identical work.
  */
 import {TaskStore, storeLog, WORKFLOW_COMPILED_STEP_TEMPLATE_PREFIX, WORKFLOW_MOVE_POLICY_TIMEOUT_MS} from "../store.js";
+import { resolveCapacityPoolId } from "../workflow-capacity.js";
 import {TransitionRejectionError} from "./errors.js";
 import * as schema from "../postgres/schema/index.js";
 import {and, eq, isNull, ne, or, sql} from "drizzle-orm";
@@ -764,7 +765,7 @@ export function countActiveInCapacitySlotSyncImpl(store: TaskStore, params: { ta
 
     let count = 0;
     for (const row of rows) {
-      const effectiveWorkflowId = row.wid ?? TaskStore.DEFAULT_WORKFLOW_POOL_ID;
+      const effectiveWorkflowId = resolveCapacityPoolId(row.wid);
       if (effectiveWorkflowId !== workflowId) continue;
 
       if (row.col === targetColumn) {
@@ -816,7 +817,7 @@ export async function countActiveInCapacitySlotAsyncImpl(store: TaskStore, param
 
     let count = 0;
     for (const row of rows) {
-      const effectiveWorkflowId = row.wid ?? TaskStore.DEFAULT_WORKFLOW_POOL_ID;
+      const effectiveWorkflowId = resolveCapacityPoolId(row.wid);
       if (effectiveWorkflowId !== workflowId) continue;
 
       if (row.col === targetColumn) {
diff --git a/packages/core/src/task-store/task-store-helpers.ts b/packages/core/src/task-store/task-store-helpers.ts
index 84c770fdfa..7824fe2ce9 100644
--- a/packages/core/src/task-store/task-store-helpers.ts
+++ b/packages/core/src/task-store/task-store-helpers.ts
@@ -11,6 +11,7 @@
  */
 
 import { TaskStore } from "../store.js";
+import { resolveCapacityPoolId } from "../workflow-capacity.js";
 import { isBuiltinWorkflowId } from "../builtin-workflows.js";
 import { InsightStore } from "../insight-store.js";
 import { ResearchStore } from "../research-store.js";
@@ -287,7 +288,7 @@ export function resolveTaskCustomFieldDefsSyncImpl(store: TaskStore, taskId: str
 
 export function resolveEffectiveWorkflowIdSyncImpl(store: TaskStore, taskId: string): string {
     const selection = store.getTaskWorkflowSelection(taskId);
-    return selection?.workflowId ?? TaskStore.DEFAULT_WORKFLOW_POOL_ID;
+    return resolveCapacityPoolId(selection?.workflowId);
 }
 
 export async function clearTaskWorkflowSelectionImpl(store: TaskStore, taskId: string): Promise<void> {
diff --git a/packages/core/src/workflow-capacity.ts b/packages/core/src/workflow-capacity.ts
index 4b2549c933..bd022e1b15 100644
--- a/packages/core/src/workflow-capacity.ts
+++ b/packages/core/src/workflow-capacity.ts
@@ -34,6 +34,33 @@ const DEFAULT_WIP_COLUMN_ID = "in-progress";
  *  is not a real workflow row id (no `builtin:`/custom collision possible). */
 export const DEFAULT_WORKFLOW_POOL_ID = "__default-workflow__";
 
+/*
+FNXC:WorkflowCapacity 2026-07-28-19:05 (pool-id sentinel fix):
+THE one place the "no selection → which pool?" convention is expressed.
+
+It exists because the convention was previously restated at each end of the
+comparison, and the two restatements disagreed: the COUNTER bucketed
+selection-less rows under `DEFAULT_WORKFLOW_POOL_ID` while `moves.ts` asked the
+counter for pool `"builtin:coding"`. Nothing ever landed in the pool being asked
+about, so the count came back 0 and a finite limit could never bind — the
+in-transaction capacity gate was structurally dead for every default-workflow
+task (Phase A3, R1).
+
+A shared constant alone would NOT have prevented that: both sides had the
+constant available and one of them still wrote a literal. Both sides now call
+THIS function, so "what pool does a selection-less task belong to" has exactly
+one answer and no call site is in a position to disagree with it.
+
+NOT to be confused with `DEFAULT_WORKFLOW_ID` ("builtin:coding"). That is a real,
+resolvable workflow row id and is the correct fallback when the value is used to
+RESOLVE AN IR (as `scheduler.ts` does). This is a bucketing key that deliberately
+cannot collide with any workflow id. Using either one in the other's role is the
+bug this function exists to make unspellable.
+*/
+export function resolveCapacityPoolId(selectionWorkflowId: string | null | undefined): string {
+  return selectionWorkflowId ?? DEFAULT_WORKFLOW_POOL_ID;
+}
+
 /** Resolved capacity configuration for a single column. */
 export interface ColumnCapacity {
   /** True when the column carries a capacity (`wip`/`countsTowardWip`) trait. */
diff --git a/packages/engine/src/__tests__/capacity-pool-id-check.test.ts b/packages/engine/src/__tests__/capacity-pool-id-check.test.ts
new file mode 100644
index 0000000000..76833984b2
--- /dev/null
+++ b/packages/engine/src/__tests__/capacity-pool-id-check.test.ts
@@ -0,0 +1,177 @@
+/*
+FNXC:WorkflowCapacity 2026-07-28-22:45 (PR #2488 review — ratchet regression suite):
+
+THE ACCEPTANCE TEST FOR A GUARD IS NOT "IT PASSES ON MAIN".
+
+The first version of `check-capacity-pool-id` passed on main and passed on the
+REINTRODUCED ORIGINAL DEFECT, because it matched one spelling of the fallback
+(`?? DEFAULT_WORKFLOW_POOL_ID`) and the real bug used another
+(`?? "builtin:coding"`). Green on the bug it exists to prevent is worse than no
+guard: it stops anyone looking.
+
+So every form the guard must catch is pinned here as a case. The first fixture is
+the ACTUAL PRE-FIX `moves.ts` shape, reduced — if this suite ever goes green on
+that, the ratchet has silently narrowed again and this file is what says so.
+
+The negative cases matter just as much: `?? "builtin:coding"` is the legitimate
+default for a WORKFLOW id in about eight places (scheduler's IR-resolution key,
+task creation, analytics). A guard that fired on those would be suppressed until
+it rotted, so the rule is deliberately about what reaches a CAPACITY SINK, not
+about a spelling.
+*/
+import { describe, expect, it } from "vitest";
+
+import { findViolationsInSource } from "../../../../scripts/lib/capacity-pool-id-check.mjs";
+
+const rules = (src: string): string[] =>
+  (findViolationsInSource("packages/core/src/task-store/example.ts", src) as Array<{ rule: string }>).map(
+    (v) => v.rule,
+  );
+
+describe("check-capacity-pool-id ratchet — forms it MUST catch", () => {
+  it("catches the ORIGINAL DEFECT: a `?? \"builtin:coding\"` local reaching the counter", () => {
+    /* The reduced pre-fix moves.ts. The previous regex version passed on this. */
+    const src = `
+      async function moveTaskInternalImpl(store: any, id: string) {
+        const effectiveWorkflowIdForMove = useWorkflow
+          ? (await store.getTaskWorkflowSelectionAsync(id))?.workflowId ?? "builtin:coding"
+          : "builtin:coding";
+        await store.countActiveInCapacitySlotAsync({
+          tx, targetColumn: budgetColumn, workflowId: effectiveWorkflowIdForMove, countPending,
+        });
+      }
+    `;
+    expect(rules(src)).toContain("unresolved-pool-into-capacity-sink");
+  });
+
+  it("catches a bare literal passed inline to the counter", () => {
+    const src = `
+      async function f(store: any) {
+        await store.countActiveInCapacitySlotAsync({ workflowId: "builtin:coding" });
+      }
+    `;
+    expect(rules(src)).toContain("unresolved-pool-into-capacity-sink");
+  });
+
+  it("catches a MULTILINE fallback — the AST sees one node regardless of formatting", () => {
+    const src = `
+      const poolId =
+        selection
+          ?.workflowId
+        ??
+          DEFAULT_WORKFLOW_POOL_ID;
+    `;
+    expect(rules(src)).toContain("sentinel-fallback");
+  });
+
+  it("catches a DEEPLY QUALIFIED sentinel reference", () => {
+    const src = `const poolId = selection?.workflowId ?? Foo.Bar.Baz.DEFAULT_WORKFLOW_POOL_ID;`;
+    expect(rules(src)).toContain("sentinel-fallback");
+  });
+
+  it("catches the sentinel's RAW STRING VALUE, not just its constant name", () => {
+    const src = `const poolId = selection?.workflowId ?? "__default-workflow__";`;
+    expect(rules(src)).toContain("sentinel-fallback");
+  });
+
+  it("catches an underived POSITIONAL argument to countCapacitySlot", () => {
+    const src = `
+      const workflowId = byTask.get(task.id) ?? "builtin:coding";
+      const n = countCapacitySlot(allTasks, byTask, budgetColumns, workflowId, countPending);
+    `;
+    expect(rules(src)).toContain("unresolved-pool-into-capacity-sink");
+  });
+
+  it("catches a sink fed by a local that is NOT resolver-derived", () => {
+    const src = `
+      const poolId = someOtherThing();
+      await store.countActiveInCapacitySlotAsync({ workflowId: poolId });
+    `;
+    expect(rules(src)).toContain("unresolved-pool-into-capacity-sink");
+  });
+});
+
+describe("check-capacity-pool-id ratchet — forms it must NOT flag", () => {
+  it("accepts a direct resolver call at the sink", () => {
+    const src = `
+      await store.countActiveInCapacitySlotAsync({
+        workflowId: resolveCapacityPoolId(selection?.workflowId),
+      });
+    `;
+    expect(rules(src)).toEqual([]);
+  });
+
+  it("accepts a local derived from the resolver, declared before OR after the sink", () => {
+    const before = `
+      const poolId = resolveCapacityPoolId(selection?.workflowId);
+      await store.countActiveInCapacitySlotAsync({ workflowId: poolId });
+    `;
+    expect(rules(before)).toEqual([]);
+  });
+
+  it("accepts `?? \"builtin:coding\"` when it is a WORKFLOW id and never reaches a capacity sink", () => {
+    /* scheduler.ts's IR-resolution key. Flagging this would make the guard noise
+       that gets suppressed — the rule is about the sink, not the spelling. */
+    const src = `
+      const workflowIdByTaskId = new Map<string, string>();
+      workflowIdByTaskId.set(task.id, selection?.workflowId ?? "builtin:coding");
+      const ir = await resolveWorkflowIrById(store, workflowId, cache);
+    `;
+    expect(rules(src)).toEqual([]);
+  });
+
+  it("accepts naming the sentinel constant where nothing is derived from a selection", () => {
+    /* scheduler's capacity DIAGNOSTIC label — no selection input, nothing to drift. */
+    const src = `const perColumnGates = [{ workflowId: DEFAULT_WORKFLOW_POOL_ID, columnId: "in-progress" }];`;
+    expect(rules(src)).toEqual([]);
+  });
+});
+
+describe("check-capacity-pool-id ratchet — it fails CLOSED", () => {
+  /*
+  FNXC:WorkflowCapacity 2026-07-28-23:30 (PR #2488 review):
+  These were ONE test titled "unparseable" that actually simulated a read()
+  throw and asserted "unreadable" — it never reached the parse path at all. A
+  test that misreports its own subject is this PR's entire failure mode in
+  miniature, so they are split and each now exercises the path it names.
+
+  Writing the second one surfaced a real defect: `ts.createSourceFile` is
+  error-TOLERANT and does not throw on malformed syntax, so the try/catch it was
+  meant to cover was unreachable and the "unparseable" rule could never fire.
+  Detection now reads `parseDiagnostics`.
+  */
+  it("reports an UNREADABLE file rather than skipping it", async () => {
+    const { findViolations } = await import("../../../../scripts/lib/capacity-pool-id-check.mjs");
+    const out = (findViolations([
+      {
+        file: "packages/core/src/broken.ts",
+        read: () => {
+          throw new Error("EACCES");
+        },
+      },
+    ]) as Array<{ rule: string }>).map((v) => v.rule);
+    // "could not inspect" must never render as "inspected and clean".
+    expect(out).toContain("unreadable");
+  });
+
+  it("reports an UNPARSEABLE file — real malformed syntax, reaching the parse check", () => {
+    /* Genuinely malformed TypeScript. A partial AST can silently lack the `??`
+       nodes and sink calls the rules look for, so "parsed badly" must not read
+       as "inspected and clean". */
+    const malformed = `
+      const x = { unclosed: (((  ;
+      function )( {
+    `;
+    const out = (findViolationsInSource("packages/core/src/malformed.ts", malformed) as Array<{ rule: string }>).map(
+      (v) => v.rule,
+    );
+    expect(out).toContain("unparseable");
+  });
+
+  it("does NOT report well-formed source as unparseable", () => {
+    /* The negative half: a diagnostics-based check that fired on valid syntax
+       would be noise, and noise gets suppressed. */
+    const fine = `const poolId = resolveCapacityPoolId(selection?.workflowId);`;
+    expect(rules(fine)).toEqual([]);
+  });
+});
diff --git a/packages/engine/src/__tests__/workflow-lifecycle-live-e2e.pg.test.ts b/packages/engine/src/__tests__/workflow-lifecycle-live-e2e.pg.test.ts
index a2129ceaf3..228eb95495 100644
--- a/packages/engine/src/__tests__/workflow-lifecycle-live-e2e.pg.test.ts
+++ b/packages/engine/src/__tests__/workflow-lifecycle-live-e2e.pg.test.ts
@@ -686,6 +686,59 @@ pgDescribe("live lifecycle E2E: real graph + real PostgreSQL store", () => {
       expect(r.acted).toBe(false);
     });
   });
+  /*
+  FNXC:WorkflowCapacity 2026-07-28-19:20 (pool-id sentinel fix — E2E acceptance):
+
+  The gate must BIND, and it must bind IN BOTH DIRECTIONS. A test that only asserts "held at cap"
+  passes trivially if the mover simply never admits anything — including if a future change breaks
+  admission outright — so each case is run twice against the SAME fixture with only the cap
+  changed: at limit 1 the card is HELD, at limit 2 the identical card is ADMITTED.
+
+  NO WORKFLOW SELECTION, deliberately. That is the whole defect: `moves.ts` asked the counter for
+  pool `"builtin:coding"` while the counter buckets selection-less rows under
+  `DEFAULT_WORKFLOW_POOL_ID`, so nothing was ever counted. A card WITH a selection makes both
+  sentinels agree and the gate already bound before the fix — seeding one here would have produced a
+  green test that never touched the bug. (Verified: with the fix reverted, this row fails.)
+
+  FLAG-ON, and labelled as such. The capacity block sits inside `if (useWorkflow && …)` and
+  `useWorkflow` reads a settings key nothing in production sets (Phase A3 R2, still live). So this
+  row proves the SENTINEL is fixed on the only path where the gate can execute at all; it does NOT
+  prove production behavior changed. Read the R2 note in workflow-capacity-invariant.pg.test.ts
+  before treating this as "capacity is enforced".
+  */
+  describe.each([
+    { limit: 1, expected: "held" as const },
+    { limit: 2, expected: "admitted" as const },
+  ])("in-transaction capacity gate (maxConcurrent=$limit → $expected)", ({ limit, expected }) => {
+    it(`a selection-less card at the wip boundary is ${expected}`, async () => {
+      const store = h.store();
+      await store.updateSettings({ maxConcurrent: limit } as never);
+      // The capacity block's enclosing flag is GLOBAL-scoped; a project-scoped write is silently
+      // dropped and the test then measures the unflagged path while claiming otherwise.
+      await store.updateGlobalSettings({ experimentalFeatures: { workflowColumns: true } } as never);
+
+      const holder = await store.createTask({ description: "capacity holder" });
+      await store.moveTask(holder.id, "todo");
+      await store.moveTask(holder.id, "in-progress");
+
+      const contender = await store.createTask({ description: "capacity contender" });
+      await store.moveTask(contender.id, "todo");
+      const error = await store
+        .moveTask(contender.id, "in-progress")
+        .then(() => null, (e: unknown) => e as Error);
+
+      store.taskCache.delete(contender.id);
+      const persisted = (await store.getTask(contender.id)).column;
+
+      if (expected === "held") {
+        expect((error as unknown as { rejection?: { code?: string } })?.rejection?.code).toBe("capacity-exhausted");
+        expect(persisted).toBe("todo"); // observed state: refused, stays put
+      } else {
+        expect(error).toBeNull();
+        expect(persisted).toBe("in-progress"); // the same fixture admits once the cap allows it
+      }
+    });
+  });
 });
 
 /*
diff --git a/packages/engine/src/hold-release.ts b/packages/engine/src/hold-release.ts
index f000e89593..876e4fb49a 100644
--- a/packages/engine/src/hold-release.ts
+++ b/packages/engine/src/hold-release.ts
@@ -43,7 +43,7 @@ import {
   resolveColumnAdjacency,
   PLAN_REVIEW_GROUP_ID,
   ACTIVE_WORKFLOW_WORK_ITEM_STATES,
-  DEFAULT_WORKFLOW_POOL_ID,
+  resolveCapacityPoolId,
   TransitionRejectionError,
   resolveWorkflowIrForTask,
   isUnplannedSeedPrompt,
@@ -113,9 +113,11 @@ export interface HoldReleaseResult {
 
 async function effectiveWorkflowId(store: TaskStore, taskId: string): Promise<string> {
   try {
-    return (await store.getTaskWorkflowSelectionAsync(taskId))?.workflowId ?? DEFAULT_WORKFLOW_POOL_ID;
+    return resolveCapacityPoolId((await store.getTaskWorkflowSelectionAsync(taskId))?.workflowId);
   } catch {
-    return DEFAULT_WORKFLOW_POOL_ID;
+    /* An unreadable selection is indistinguishable from no selection for pooling
+       purposes, so it takes the same bucket rather than a second convention. */
+    return resolveCapacityPoolId(undefined);
   }
 }
 
@@ -439,7 +441,7 @@ function countCapacitySlot(
 ): number {
   let count = 0;
   for (const t of allTasks) {
-    if ((effectiveWorkflowIdByTask.get(t.id) ?? DEFAULT_WORKFLOW_POOL_ID) !== workflowId) continue;
+    if (resolveCapacityPoolId(effectiveWorkflowIdByTask.get(t.id)) !== workflowId) continue;
     if (budgetColumns.has(t.column)) {
       count += 1;
       continue;
@@ -573,7 +575,7 @@ export async function runHoldReleaseSweep(
       }
       const capacity = resolveColumnCapacity(ir, target, settings);
       if (capacity.hasCapacity && Number.isFinite(capacity.limit)) {
-        const workflowId = effectiveWorkflowIdByTask.get(task.id) ?? DEFAULT_WORKFLOW_POOL_ID;
+        const workflowId = resolveCapacityPoolId(effectiveWorkflowIdByTask.get(task.id));
         // U4/KTD-9: count occupants across every column sharing the target's
         // budget (a shared `limitSetting` pools multiple wip columns).
         const budgetColumns = new Set(resolveWipBudgetColumns(ir, target));
diff --git a/scripts/check-capacity-pool-id.mjs b/scripts/check-capacity-pool-id.mjs
new file mode 100644
index 0000000000..aadc03ebba
--- /dev/null
+++ b/scripts/check-capacity-pool-id.mjs
@@ -0,0 +1,68 @@
+#!/usr/bin/env node
+/*
+FNXC:WorkflowCapacity 2026-07-28-22:30 (PR #2488 review — ratchet rebuilt):
+CLI wrapper. The rules and their rationale live in
+scripts/lib/capacity-pool-id-check.mjs; the regression suite that pins each form
+this guard must catch lives in
+packages/engine/src/__tests__/capacity-pool-id-check.test.ts.
+
+Wired into `pretest`, `pretest:full`, and the blocking `test:gate`.
+*/
+import { readFileSync } from "node:fs";
+import { execSync } from "node:child_process";
+
+import { findViolations, RESOLVER } from "./lib/capacity-pool-id-check.mjs";
+
+let files;
+try {
+  files = execSync("git ls-files 'packages/*/src/**/*.ts' 'packages/*/src/*.ts'", {
+    encoding: "utf8",
+    maxBuffer: 64 * 1024 * 1024,
+  })
+    .split("\n")
+    .map((f) => f.trim())
+    .filter(Boolean)
+    // Test sources are excluded using the repo's own guideline shape — the
+    // `{test,spec}.{ts,tsx}` family — because the earlier `.test.ts`-only suffix
+    // would have scanned a `.spec.ts` sitting directly under `packages/<pkg>/src/`
+    // as production source and flagged its fixtures as real violations.
+    .filter((f) => !f.includes("__tests__") && !/\.(test|spec)\.tsx?$/.test(f));
+} catch (err) {
+  // FAIL CLOSED: if we cannot even enumerate the files, we have checked nothing.
+  console.error(`check-capacity-pool-id: could not list files — ${err?.message ?? err}`);
+  process.exit(1);
+}
+
+if (files.length === 0) {
+  console.error("check-capacity-pool-id: file list is EMPTY — refusing to report success on zero files.");
+  process.exit(1);
+}
+
+const violations = findViolations(files.map((file) => ({ file, read: () => readFileSync(file, "utf8") })));
+
+if (violations.length > 0) {
+  const byRule = {
+    "unresolved-pool-into-capacity-sink":
+      `a value reaching a capacity counter's \`workflowId\` must come from ${RESOLVER}().\n` +
+      "  Two enforcement surfaces that each derive the pool themselves WILL drift: that is how the\n" +
+      '  in-transaction gate silently stopped binding (moves.ts derived `?? "builtin:coding"` while the\n' +
+      "  counter bucketed under the sentinel, so nothing was ever counted).",
+    "sentinel-fallback":
+      "the pool sentinel must not be restated in a `??` fallback — call the resolver instead.",
+    unreadable: "a tracked source file could not be read, so it was NOT inspected.",
+    unparseable: "a tracked source file could not be parsed, so it was NOT inspected.",
+  };
+
+  console.error("\ncheck-capacity-pool-id: FAILED\n");
+  for (const rule of Object.keys(byRule)) {
+    const hits = violations.filter((v) => v.rule === rule);
+    if (hits.length === 0) continue;
+    console.error(`${rule}: ${byRule[rule]}\n`);
+    for (const v of hits) console.error(`  ${v.file}:${v.line}: ${v.text}`);
+    console.error("");
+  }
+  console.error(`Fix by deriving the pool through ${RESOLVER}(selection?.workflowId).\n`);
+  process.exit(1);
+}
+
+console.log(`check-capacity-pool-id: ok (${files.length} files inspected)`);
diff --git a/scripts/lib/capacity-pool-id-check.mjs b/scripts/lib/capacity-pool-id-check.mjs
new file mode 100644
index 0000000000..3b045ea35a
--- /dev/null
+++ b/scripts/lib/capacity-pool-id-check.mjs
@@ -0,0 +1,235 @@
+/*
+FNXC:WorkflowCapacity 2026-07-28-22:30 (PR #2488 review — ratchet rebuilt):
+
+WHY THIS IS AN AST CHECK AND NOT A REGEX.
+
+The first version of this guard matched `?? DEFAULT_WORKFLOW_POOL_ID` on one line.
+It therefore MISSED `?? "builtin:coding"` — the actual defect the whole change
+exists to fix — and passed green on the reintroduced bug (verified, not assumed).
+A guard that reports success without checking is worse than no guard, because it
+stops anyone looking. The claim being made is structural ("no capacity pool id is
+derived except through the resolver"), so it is checked structurally.
+
+TWO COMPLEMENTARY RULES. Neither alone is sufficient:
+
+  RULE 1 — THE SINK RULE (the one that catches the original defect).
+  Every argument that flows into a capacity counter's `workflowId` must be
+  `resolveCapacityPoolId(...)`, or a local whose initializer is that call. The
+  original defect passed `effectiveWorkflowIdForMove`, a local initialized from a
+  `??` literal — so this rule fires on it regardless of WHICH literal was used, on
+  one line or twenty. This is the rule that expresses the real invariant: the
+  banned thing is not a spelling, it is an underived value reaching the counter.
+
+  RULE 2 — THE SENTINEL RULE (cheap, catches restatements of the convention).
+  No `??` whose right-hand side is the pool sentinel — under ANY qualification
+  depth (`X.Y.Z.DEFAULT_WORKFLOW_POOL_ID`) or as its raw string value
+  ("__default-workflow__") — outside the module that owns the convention. Being
+  AST-based, a fallback split across lines is the same node and is caught.
+
+WHY `?? "builtin:coding"` IS NOT BANNED OUTRIGHT. That literal is the legitimate
+default for a WORKFLOW id in ~8 places (scheduler's IR-resolution key,
+task-creation, analytics). It is only a bug when it reaches a CAPACITY POOL, and
+that distinction is exactly what Rule 1 encodes. Banning the spelling everywhere
+would be cargo-culting the rule past the thing it protects, and would have to be
+suppressed so often it would rot.
+
+FAIL CLOSED. An unreadable or unparseable file is reported as a violation, never
+skipped — "could not inspect" must not render as "inspected and clean", which is
+the same failure shape as the regex that could not see the defect.
+*/
+import ts from "typescript";
+
+/** The module that owns the convention; the resolver's own `??` lives here. */
+export const CONVENTION_OWNER = "packages/core/src/workflow-capacity.ts";
+
+/** The canonical resolver every pool-id derivation must go through. */
+export const RESOLVER = "resolveCapacityPoolId";
+
+/** Capacity counters whose `workflowId` input is a pool id. */
+export const CAPACITY_SINKS = new Set([
+  "countActiveInCapacitySlotAsync",
+  "countActiveInCapacitySlotSync",
+  "countActiveInCapacitySlot",
+  "countCapacitySlot",
+]);
+
+/** `countCapacitySlot(allTasks, byTask, budgetColumns, workflowId, countPending)` */
+const POSITIONAL_SINKS = { countCapacitySlot: 3 };
+
+const SENTINEL_CONST = "DEFAULT_WORKFLOW_POOL_ID";
+const SENTINEL_VALUE = "__default-workflow__";
+
+/** Right-most name of a possibly-qualified reference, at any depth. */
+function tailName(node) {
+  let cur = node;
+  while (ts.isPropertyAccessExpression(cur)) cur = cur.name;
+  return ts.isIdentifier(cur) ? cur.text : undefined;
+}
+
+function isResolverCall(node) {
+  return (
+    node &&
+    ts.isCallExpression(node) &&
+    tailName(node.expression) === RESOLVER
+  );
+}
+
+/** True when `expr` is the resolver call, or a local initialized from one. */
+function isDerivedThroughResolver(expr, resolverLocals) {
+  if (!expr) return false;
+  if (isResolverCall(expr)) return true;
+  if (ts.isIdentifier(expr)) return resolverLocals.has(expr.text);
+  // `cond ? resolver(a) : resolver(b)` is still derived through the resolver.
+  if (ts.isConditionalExpression(expr)) {
+    return (
+      isDerivedThroughResolver(expr.whenTrue, resolverLocals) &&
+      isDerivedThroughResolver(expr.whenFalse, resolverLocals)
+    );
+  }
+  if (ts.isAsExpression(expr) || ts.isParenthesizedExpression(expr)) {
+    return isDerivedThroughResolver(expr.expression, resolverLocals);
+  }
+  return false;
+}
+
+/**
+ * Analyse one source file.
+ * @returns {Array<{file:string,line:number,rule:string,text:string}>}
+ */
+export function findViolationsInSource(file, text) {
+  const violations = [];
+  /*
+  FNXC:WorkflowCapacity 2026-07-28-23:30 (PR #2488 review):
+  Parse failure is detected via `parseDiagnostics`, NOT via a try/catch.
+  `ts.createSourceFile` is error-TOLERANT: given `function )( {` it returns a
+  source file carrying diagnostics rather than throwing, so the catch this
+  replaced was unreachable and the "unparseable" rule could never fire. That made
+  the guard's own fail-closed claim overstated in exactly the way this PR is
+  about — a check advertising a capability it did not have. A file whose syntax
+  did not parse yields a partial AST, so its `??` nodes and sink calls may simply
+  be absent: reporting it clean would be reporting "not inspected" as "inspected".
+  The defensive catch is kept for a genuine internal error, but detection is the
+  diagnostics check.
+  */
+  let sf;
+  try {
+    sf = ts.createSourceFile(file, text, ts.ScriptTarget.Latest, true, ts.ScriptKind.TS);
+  } catch (err) {
+    return [{ file, line: 0, rule: "unparseable", text: `parser threw: ${String(err && err.message)}` }];
+  }
+  const parseErrors = sf.parseDiagnostics ?? [];
+  if (parseErrors.length > 0) {
+    const first = ts.flattenDiagnosticMessageText(parseErrors[0].messageText, " ");
+    return [
+      {
+        file,
+        line: 0,
+        rule: "unparseable",
+        text: `${parseErrors.length} syntax error(s), first: ${first} — a file that did not parse was NOT inspected`,
+      },
+    ];
+  }
+
+  const lineOf = (node) => sf.getLineAndCharacterOfPosition(node.getStart(sf)).line + 1;
+  const snippet = (node) => node.getText(sf).replace(/\s+/g, " ").slice(0, 140);
+
+  // Locals whose initializer is a resolver call — collected first so a sink that
+  // reads one is accepted regardless of declaration order within the file.
+  const resolverLocals = new Set();
+  const collect = (node) => {
+    if (ts.isVariableDeclaration(node) && ts.isIdentifier(node.name)) {
+      if (isDerivedThroughResolver(node.initializer, resolverLocals)) resolverLocals.add(node.name.text);
+    }
+    ts.forEachChild(node, collect);
+  };
+  collect(sf);
+  // Second pass: a local initialized from another resolver local.
+  collect(sf);
+
+  const visit = (node) => {
+    // ── RULE 2: sentinel restated in a `??` fallback ──────────────────────────
+    if (
+      ts.isBinaryExpression(node) &&
+      node.operatorToken.kind === ts.SyntaxKind.QuestionQuestionToken &&
+      file !== CONVENTION_OWNER
+    ) {
+      const rhs = node.right;
+      const isSentinelConst = tailName(rhs) === SENTINEL_CONST;
+      const isSentinelValue = ts.isStringLiteral(rhs) && rhs.text === SENTINEL_VALUE;
+      if (isSentinelConst || isSentinelValue) {
+        violations.push({
+          file,
+          line: lineOf(node),
+          rule: "sentinel-fallback",
+          text: snippet(node),
+        });
+      }
+    }
+
+    // ── RULE 1: something underived reaching a capacity counter ───────────────
+    if (ts.isCallExpression(node)) {
+      const callee = tailName(node.expression);
+      if (callee && CAPACITY_SINKS.has(callee)) {
+        // Object-literal form: `count...({ workflowId: <expr> })`
+        for (const arg of node.arguments) {
+          if (!ts.isObjectLiteralExpression(arg)) continue;
+          for (const prop of arg.properties) {
+            if (!ts.isPropertyAssignment(prop)) continue;
+            if (tailName(prop.name) !== "workflowId") continue;
+            if (!isDerivedThroughResolver(prop.initializer, resolverLocals)) {
+              violations.push({
+                file,
+                line: lineOf(prop),
+                rule: "unresolved-pool-into-capacity-sink",
+                text: snippet(prop),
+              });
+            }
+          }
+        }
+        // Positional form.
+        const idx = POSITIONAL_SINKS[callee];
+        if (idx !== undefined && node.arguments.length > idx) {
+          const arg = node.arguments[idx];
+          if (!isDerivedThroughResolver(arg, resolverLocals)) {
+            violations.push({
+              file,
+              line: lineOf(arg),
+              rule: "unresolved-pool-into-capacity-sink",
+              text: snippet(arg),
+            });
+          }
+        }
+      }
+    }
+
+    ts.forEachChild(node, visit);
+  };
+  visit(sf);
+
+  return violations;
+}
+
+/**
+ * @param {Array<{file:string, read:() => string}>} entries
+ * @returns {Array<{file:string,line:number,rule:string,text:string}>}
+ */
+export function findViolations(entries) {
+  const out = [];
+  for (const entry of entries) {
+    let text;
+    try {
+      text = entry.read();
+    } catch (err) {
+      // FAIL CLOSED: an uninspectable file is a violation, not a pass.
+      out.push({
+        file: entry.file,
+        line: 0,
+        rule: "unreadable",
+        text: `could not read (${String(err && err.message)}) — a file that cannot be inspected must not report as clean`,
+      });
+      continue;
+    }
+    out.push(...findViolationsInSource(entry.file, text));
+  }
+  return out;
+}