diff --git a/src/dashboards/executive.dashboard.ts b/src/dashboards/executive.dashboard.ts
index 8895f83..fe0d4d0 100644
--- a/src/dashboards/executive.dashboard.ts
+++ b/src/dashboards/executive.dashboard.ts
@@ -1,8 +1,16 @@
 import type { Dashboard } from '@objectstack/spec/ui';
 
 /**
- * 管理层 — DESIGN.md §09's second dashboard: 生效合同额 · 90 天内到期 ·
- * 高风险合同 · 审批瓶颈（各台阶平均停留）· 按方向的合同额趋势.
+ * 管理层 — DESIGN.md §09's second dashboard.
+ *
+ * §09's brick list is deliberately NOT copied here, for the reason PR #74 gave
+ * when it took the sibling copy off `legal.dashboard.ts`: a copy drifts, a
+ * pointer cannot. This one had already drifted. It advertised one brick this
+ * file has never built — 审批瓶颈（各台阶平均停留）— and §09 stopped asking for
+ * that brick on 2026-09-09 (维护者裁定, PR #40 `1127e52`), replacing it with the
+ * per-rung routing volume the four `route_*` tiles below deliver. §09 as merged
+ * (PR #70, `30fd863`) is the source; every widget below carries its own comment
+ * for its own shape.
  *
  * Every widget binds `contract_metrics`, so the two global filters below reach
  * a column that exists on every one of them. See `legal.dashboard.ts` for the
@@ -136,10 +144,19 @@ export const ExecutiveDashboard: Dashboard = {
       colorVariant: 'danger',
       layout: { x: 9, y: 0, w: 3, h: 2 },
     },
-    // ─── §09's 审批瓶颈, as the four rungs of DESIGN.md §04's ladder ───────
+    // ─── The four rungs of DESIGN.md §04's ladder, read as ROUTING VOLUME ───
+    //
+    // These four tiles are the brick §09 puts at this position, and they build
+    // it in full: one tile per rung, reading the four `route_*` flags F2 stamps
+    // on the contract. §09 says of this position that it carries traffic and
+    // not dwell — the board and the design agree here, and §09 (PR #70,
+    // `30fd863`) is where that wording lives rather than a copy of it here.
     //
-    // §09 asks for 各台阶平均停留 — average DWELL per approval step. Two
-    // independent facts rule the dwell out, both measured rather than assumed:
+    // The average DWELL per rung that §09 listed BEFORE 2026-09-09 is not
+    // deliverable, and §09 no longer asks for it: the maintainer ruling in
+    // PR #40 (`1127e52`) placed it outside the V1.0 surface, and §09 now gives
+    // these two measurements — both taken on this board — as the reasons it
+    // declines:
     //
     //   1. There is no dwell to read. `sys_approval_request` holds 0 rows on a
     //      stock `pnpm demo` — the seed stamps `approval_status` directly and
@@ -147,10 +164,19 @@ export const ExecutiveDashboard: Dashboard = {
     //   2. Even with rows, a duration is not expressible in the semantic layer:
     //      see the measurement in `cycle-time.dataset.ts`.
     //
+    // ⚠️ Neither reads as a TODO. Fact 1 is independent of the platform and
+    // survives any upgrade; and the in-flight platform work
+    // (`objectstack-ai/objectstack#16737`) makes the wrong path ERROR rather
+    // than adding date arithmetic, so the dwell does not arrive when it closes
+    // either. ⛔ Do not re-promise the metric on the strength of that issue, and
+    // ⛔ do not stamp it from an application-side daily job — §09 forbids that
+    // route in as many words, as replicating a platform rule in the app.
+    //
     // What IS persisted is the routing: F2 stamps four booleans on the contract,
     // one per rung. These four tiles read them — how much traffic each rung
-    // carries. Each title names its rung and the group title says "Routing
-    // Load", never "bottleneck", because volume is not dwell.
+    // carries. Each tile is titled `Routes: <rung>`; no title or description on
+    // this board — here or in either translation bundle — says "bottleneck" or
+    // "dwell", because volume is neither.
     //
     // FOUR TILES, not one four-measure chart, and the shape was forced by a
     // measurement. The rungs are four COLUMNS, not four values of one column, so
diff --git a/src/dashboards/legal.dashboard.ts b/src/dashboards/legal.dashboard.ts
index b3a3850..7ee7356 100644
--- a/src/dashboards/legal.dashboard.ts
+++ b/src/dashboards/legal.dashboard.ts
@@ -140,15 +140,29 @@ export const LegalDashboard: Dashboard = {
     },
     {
       /**
-       * §09's 超 SLA tile, with the one honest change its threshold had to
-       * take. The SLA is PER TYPE (`clm_contract_type.review_sla_days`: 2, 3, 5
-       * or 10 days across the nine seeded types), and a breach is a row-wise
-       * comparison of `review_started_at` against another object's column.
-       * Analytics cannot express that: there is no formula filter (§12 gap
-       * #10), and a cross-object filter is refused outright on the strategy
-       * every bucketed query lands on. The alternative — a persisted
-       * `review_due_at` stamped by the daily job — is the §12 gap-#7 shape and
-       * belongs to card 09, not here.
+       * The review-ageing tile, at a FIXED 30-day threshold written into its
+       * own title — which is the brick §09 prescribes here, not a degraded
+       * stand-in for one. §09 places the per-type「超 SLA」 outside the V1.0
+       * delivery surface (维护者裁定 2026-09-09, PR #40 `1127e52`; §09 as
+       * merged, PR #70 `30fd863`) and asks in its place for exactly this: a
+       * fixed threshold, self-declared in the brick's title.
+       *
+       * The measurement behind that ruling was taken here, which is why it is
+       * recorded here. The SLA is PER TYPE (`clm_contract_type.review_sla_days`:
+       * 2, 3, 5 or 10 days across the nine seeded types), and a breach is a
+       * row-wise comparison of `review_started_at` against another object's
+       * column. Analytics cannot express that: there is no formula filter
+       * (§12 gap #10), and a cross-object filter is refused outright on the
+       * strategy every bucketed query lands on.
+       *
+       * ⛔ The remaining route — a persisted `review_due_at` stamped by a daily
+       * job, the §12 gap-#7 shape — is closed for this metric, and not merely
+       * unbuilt: §09 forbids filling the per-type breach from an
+       * application-side job because that replicates a platform rule inside the
+       * application, and decision #31 ruled the same way on the notification
+       * side (1C + 2B, 2026-09-09 — see `legal-review-sla.flow.ts`, whose F3
+       * reminder fires on this same fixed 30 days so the tile and the reminder
+       * cannot tell two stories).
        *
        * So the threshold is FIXED, stated in the title, and above every seeded
        * SLA: a review older than 30 days is late under any of the nine. The
@@ -179,20 +193,29 @@ export const LegalDashboard: Dashboard = {
     // ─── Row 2: throughput, period over period ───────────────────────────
     {
       /**
-       * §09 asks for 平均周转（本月 vs 上月） — an AVERAGE TURNAROUND compared
-       * with the previous period. The comparison half is delivered exactly as
-       * the card requires, with the platform primitive (`compareTo`, the one
-       * shape `DatasetSelection.compareTo` implements) rather than a hardcoded
-       * delta. The average-duration half is NOT delivered, and this tile
-       * measures THROUGHPUT instead: how many contracts legal got to `approved`
-       * in the window.
+       * §09's 审批吞吐 brick: how many contracts legal got to `approved` in the
+       * window, against the previous period — and the period comparison uses
+       * the platform primitive §09 names (`compareTo`, the one shape
+       * `DatasetSelection.compareTo` implements) rather than a hardcoded delta.
+       * This tile IS that brick; it is not a substitute for a missing one.
        *
-       * Why: a duration cannot be computed in the semantic layer on this
+       * §09 did ask for 平均周转 — an AVERAGE TURNAROUND — until 2026-09-09,
+       * when the maintainer ruling in PR #40 (`1127e52`) placed 各段时长 outside
+       * the V1.0 delivery surface and rewrote this brick to throughput (§09 as
+       * merged: PR #70, `30fd863`). The measurement behind that ruling was
+       * taken here: a duration cannot be computed in the semantic layer on this
        * platform version — `Field.datetime` persists ISO text, so `AVG()` over
        * a stage stamp answers the average YEAR (2025.9166…), and the dataset
        * layer has no expression in which to subtract two dates. The full
-       * measurement is in `cycle-time.dataset.ts`. Substituting a number that
-       * is honest and saying so beats rendering a plausible one that is not.
+       * measurement is in `cycle-time.dataset.ts`. A number that is honest and
+       * says so beats a plausible one that is not, which is the reasoning §09
+       * itself now carries.
+       *
+       * ⚠️ This tile does not become an average turnaround when
+       * `objectstack-ai/objectstack#16737` closes. That in-flight fix makes the
+       * wrong path ERROR instead of returning a plausible fake number; it does
+       * not add date arithmetic, so the duration stays uncomputable after it
+       * lands. ⛔ Do not read the tracker as a queue this metric is waiting in.
        *
        * The window is stated HERE, on the widget, and it is bounded on both
        * ends on purpose: `DatasetWidget` lowers only a `{ $gte, $lte }` pair
diff --git a/src/datasets/cycle-time.dataset.ts b/src/datasets/cycle-time.dataset.ts
index 55a5600..dbb67bb 100644
--- a/src/datasets/cycle-time.dataset.ts
+++ b/src/datasets/cycle-time.dataset.ts
@@ -4,8 +4,15 @@ import { defineDataset } from '@objectstack/spec/ui';
  * How far contracts get, and when — DESIGN.md §09's second dataset, by
  * contract type and legal owner.
  *
- * ## ⚠️ §09 asks for 各段时长 (per-stage DURATIONS). This dataset does not carry
- * one, because on this platform version none can be computed honestly.
+ * ## ⚠️ No duration measure — and §09 does not ask for one
+ *
+ * §09's bullet for this dataset asks for stage-reached counts and rules out any
+ * duration measure, which is exactly what is declared below. 各段时长 (per-stage
+ * DURATIONS) was placed outside the V1.0 delivery surface by the maintainer
+ * ruling of 2026-09-09 (PR #40, `1127e52`; §09 as merged, PR #70, `30fd863`),
+ * and the measurements in this comment are the ones §09 now cites as its
+ * reason. They were taken here, on this dataset — which is why they are
+ * recorded here, and not because this file falls short of the design.
  *
  * Measured on `@objectstack/spec` 17.3.0 + `@objectstack/service-analytics`
  * 17.3.0 + better-sqlite3 13.0.3, against the 820-row demo:
@@ -27,10 +34,19 @@ import { defineDataset } from '@objectstack/spec/ui';
  * form is the one that renders a clean, plausible number. Shipping it would be
  * exactly the "seeded number that looks computed" AGENTS.md forbids.
  *
- * A duration needs a PERSISTED numeric — the same shape DESIGN.md §12 already
- * prescribes for the other date-arithmetic gap (#7: "到期、逾期由日任务盖戳字段"),
- * i.e. a daily job stamping `clm_contract`. That field does not exist and
- * creating it is a schema change, not an analytics one. Raised on the PR.
+ * ⚠️ None of this changes when `objectstack-ai/objectstack#16737` closes. The
+ * in-flight platform fix makes the wrong path ERROR instead of returning that
+ * plausible number; it does not add date arithmetic, so a duration remains
+ * uncomputable after it lands. The unlock criterion decision #31 records is
+ * "this repo upgrades to a version where the wrong path actually errors" — the
+ * end of a fake number, not the arrival of the metric.
+ *
+ * A duration would need a PERSISTED numeric — the shape DESIGN.md §12 gives the
+ * other date-arithmetic gap (#7), a daily job stamping the column, as the
+ * obligation and payment sweeps do for 到期 / 逾期. ⛔ That route is closed for
+ * this one, and not merely unbuilt: §09 forbids filling 各段时长 or the per-type
+ * 「超 SLA」 from an application-side daily job, because it replicates a platform
+ * rule inside the application.
  *
  * ## What it DOES carry
  *