@@ -0,0 +1,24 @@
+---
+'@objectstack/spec': minor
+---
+
+fix(spec)!: the filter save and validate doors refuse an `addDays` offset whose `{ $field }` is a dotted path, in the words SQL push-down already refuses it with on every read
+
+An `addDays` offset reads its days from a column of the same record: `{ $field: 'due_on', addDays: { $field: 'grace_days' } }`, or `{ $now: true, addDays: { $field: 'review_sla_days' } }`. A dotted offset such as `addDays: { $field: 'contract_type.review_sla_days' }` passed `FieldReferenceSchema`, `NowAnchorSchema`, `FilterConditionSchema`, the dataset `filter` and measure `filter`, and `os validate`, and then failed every SQL read with `INVALID_FILTER` / 400. The one executor that walks the path, the memory evaluator, reads a lookup's id there on a stored row and counts the offset as zero days. Now every authoring door refuses it, naming the path: `the addDays offset "contract_type.review_sla_days" is a dotted path, and SQL push-down compiles same-table column references only (no relation traversal, no alias-qualified columns).`
+
+This retires the earlier reading that the dotted offset validates and is answered in memory (the #14104 ruling-record correction 5550346132, and the spec pin that accepted `addDays: { $field: 'duty.grace_days' }`). Ruling A on #25808 bounds the offset to a same-record column.
+
+Clause-②: no (narrowing) — the save and validate doors refuse a dotted `addDays` `$field` that every SQL read already refused. Nothing is widened.
+
+Metadata-protocol: narrowed (a filter, a dataset filter or a dataset measure filter may no longer carry an `addDays` offset whose `$field` is a dotted path)
+
+<!-- adr-0087: not-required (no-migration-prescription) No key of any schema is added, removed or renamed. The dotted offset never compiled on SQL push-down and, on a stored row, read as zero days in memory, so there is no mechanical rewrite that keeps what the author meant: the related value has to be copied onto a numeric column of the row, which is an application change the author makes. -->
+
+**BREAKING** accept-set narrowing on the `addDays` offset (`@objectstack/spec`).
+
+- **Before.** `FieldReferenceSchema`, `NowAnchorSchema`, the ordering slots of `FieldOperatorsSchema`, `FilterConditionSchema` (data filters, dataset `filter`, dataset measure `filter`) and `os validate` accepted `addDays: { $field: 'a.b' }`. SQL reads answered `INVALID_FILTER` / 400; a dataset measure filter answered it with the reason withheld.
+- **Now.** Each of those doors refuses it. `parseFilterAST` and the engine's filter doors (`where`, `aggregations[i].filter`, `having`) refuse it with `INVALID_FILTER` / 400 and the path named, so a dataset measure filter that never met the save door now gets the reason too.
+- **Unchanged.** A same-record offset column, a literal offset, and the reference's own `$field` (a dotted base reference keeps its asymmetry: resolved in memory, refused by SQL push-down).
+- **Who sees it.** An author who read the offset through a relation. Every SQL deployment already refused that filter on read.
+
+**The remedy.** Copy the related value onto a numeric column of the record (on create and whenever the relation changes) and name that column: `addDays: { $field: 'review_sla_days' }`.
@@ -509,9 +509,11 @@ On SQL push-down the offset compiles only between two temporal columns of the sa
 class (`date` with `date`, `datetime` with `datetime`) against a numeric offset column,
 on every dialect the `$field` compiler covers (SQLite, PostgreSQL, MySQL); the
 memory evaluator matches, and a fractional offset *value* is truncated toward zero on
-both. The same-table rule applies to the offset column too: `addDays: { $field:
-'duty.grace_days' }` (a relation path) is resolved by the memory evaluator and refused
-by SQL push-down with `INVALID_FILTER`, exactly as a dotted `$field` is.
+both. The same-table rule applies to the offset column too, and the offset is held to it
+at save time: `addDays: { $field: 'duty.grace_days' }` (a relation path) is refused with
+`INVALID_FILTER` by `os validate`, by the filter and dataset save doors and by every read,
+in the SQL push-down's own words. Only the referenced `$field` itself keeps the older
+asymmetry — resolved by the memory evaluator, refused by SQL push-down.
 
 **A dataset's `include` does not widen this.** A dataset compiles a `LEFT JOIN` for every
 relation it `include`s (ADR-0071, up to three hops), and on the native-SQL path those joins
@@ -521,7 +523,8 @@ declines it so that the driver, which alone holds the declared field set, the co
 and the tenant column the cross-field rules check, compiles it (#7598) — and the driver
 compiles single-table statements, with no join in scope. So
 `completed_on <= due_on + duty.grace_days` is refused with `INVALID_FILTER` even when `duty`
-is in the dataset's `include`; measured and pinned in
+is in the dataset's `include` — at save time now, and by the driver for a dataset that
+never met the save door; measured and pinned in
 `packages/services/service-analytics/src/__tests__/include-relation-cross-field-boundary.test.ts`
 (#15103). The spelling that answers on both paths is the same-table one — every column the
 reference names, offset included, is a column of the object the filter runs over.
@@ -583,9 +586,11 @@ strategy declines a query carrying an anchor so that the driver compiles it, as
 
 **The SLA held on a related record.** The shape that compiles is the same-table one: the SLA
 (or its negative) propagated onto the row, or the dataset's base object carrying the column.
-A dotted offset (`addDays: { $field: 'queue.sla_days' }`) validates, is resolved by the
-memory evaluator, and is refused by SQL push-down with `INVALID_FILTER`, exactly as a dotted
-`$field` is — a dataset's `include` does not widen it (see above).
+A dotted offset (`addDays: { $field: 'queue.sla_days' }`) is refused with `INVALID_FILTER`
+when the filter, dataset or stack is saved or validated, in the SQL push-down's words — a
+dataset's `include` does not widen it (see above). The memory evaluator would walk the path,
+but on a stored row the relation holds the related record's id, so the offset would read as
+zero days: copy the related value onto a numeric column of the row instead.
 
 ### Date, Datetime, and Time Filters
 
 
@@ -240,7 +240,7 @@ A `{ $field }` reference to the numeric column holding the day offset
 
 | Property | Type | Required | Description |
 | :--- | :--- | :--- | :--- |
-| **$field** | `string` | ✅ | Numeric column whose value is the number of days to add |
+| **$field** | `string` | ✅ | Numeric column of the same record whose value is the number of days to add (no dotted path) |
 
 
 ---
@@ -389,7 +389,7 @@ A `{ $field }` reference to the numeric column holding the day offset
 
 | Property | Type | Required | Description |
 | :--- | :--- | :--- | :--- |
-| **$field** | `string` | ✅ | Numeric column whose value is the number of days to add |
+| **$field** | `string` | ✅ | Numeric column of the same record whose value is the number of days to add (no dotted path) |
 
 
 ---
 
@@ -339,15 +339,20 @@ describe('[#14104] the dataset, valued end to end on the engine path', () => {
     expect(Number(byTitle.get('f')?.late ?? 0)).toBe(0);
   });
 
-  it('the dotted offset spelling of the ruling validates, routes, and is refused by SQL push-down by name', async () => {
-    // Validates at the schema door (the memory evaluator walks the path).
+  it('[#26601] the dotted offset spelling of the ruling is refused at the schema door, and past it by SQL push-down', async () => {
+    // Refused at the schema door: the offset column is a column of the same
+    // record (ruling A on #25808), and the memory evaluator, which walks the
+    // path, reads a lookup's id there on a stored row — zero days, silently.
     const dotted = { completed_on: { $lte: { $field: 'due_on', addDays: { $field: 'duty.grace_days' } } } };
-    expect(DatasetSchema.safeParse({
+    const saved = DatasetSchema.safeParse({
       ...TASK_HEALTH,
       measures: [{ name: 'done_on_time', aggregate: 'count', filter: dotted }],
-    }).success).toBe(true);
-    // …and on a SQL deployment it is a loud INVALID_FILTER, never a wrong count:
-    // the 2026-08-06 same-table ruling on #5222, inherited by the offset.
+    });
+    expect(saved.success).toBe(false);
+    expect(saved.error?.issues[0]?.message).toContain('"duty.grace_days" is a dotted path');
+    // …and a dataset that never met that door is a loud INVALID_FILTER on a SQL
+    // deployment, never a wrong count: the 2026-08-06 same-table ruling on
+    // #5222, inherited by the offset.
     let err: WireBearingError | null = null;
     try {
       await service.queryDataset(
 
@@ -106,8 +106,16 @@ const RELATION_BARE: FilterCondition = {
   completed_on: { $lte: { $field: 'duty.grace_days' } },
 };
 
-/** The dataset: `duty` IS in `include`, and a dimension already reads through it. */
-const DATASET = DatasetSchema.parse({
+/**
+ * The dataset: `duty` IS in `include`, and a dimension already reads through it.
+ *
+ * [#26601] The offset arm is a DOTTED `addDays` offset, which the dataset write
+ * door now refuses (ruling A on #25808 bounds the offset to a same-record
+ * column). So the dataset is parsed with the same-table filter in that slot
+ * and the dotted one put back PAST the save door — the read-time half of the
+ * boundary is still this file's to pin, for a row that never met that door.
+ */
+const PARSED = DatasetSchema.parse({
   name: 'task_health_by_duty',
   label: 'Task health by duty',
   object: TASK,
@@ -119,10 +127,14 @@ const DATASET = DatasetSchema.parse({
   measures: [
     { name: 'total', aggregate: 'count' },
     { name: 'done_on_time', aggregate: 'count', filter: SAME_TABLE_ON_TIME },
-    { name: 'done_on_time_by_duty', aggregate: 'count', filter: RELATION_OFFSET_ON_TIME },
+    { name: 'done_on_time_by_duty', aggregate: 'count', filter: SAME_TABLE_ON_TIME },
     { name: 'done_by_duty_bare', aggregate: 'count', filter: RELATION_BARE },
   ],
 });
+const DATASET = {
+  ...PARSED,
+  measures: PARSED.measures.map((m) => (m.name === 'done_on_time_by_duty' ? { ...m, filter: RELATION_OFFSET_ON_TIME } : m)),
+} as typeof PARSED;
 
 interface WireBearingError extends Error {
   code?: string;
@@ -212,6 +224,17 @@ describe('[#15103] a cross-field comparand on an `include`d relation — the bou
     });
   });
 
+  describe('[#26601] the save door refuses the dotted OFFSET; the dotted bare reference still validates', () => {
+    it('DatasetSchema refuses the offset arm at its slot, in the SQL read\'s words, and keeps the bare arm', () => {
+      const offset = DatasetSchema.safeParse({ ...PARSED, measures: [{ name: 'by_duty', aggregate: 'count', filter: RELATION_OFFSET_ON_TIME }] });
+      expect(offset.success).toBe(false);
+      expect(offset.error?.issues[0]?.path).toEqual(['measures', 0, 'filter', 'completed_on', '$lte']);
+      expect(offset.error?.issues[0]?.message).toContain('"duty.grace_days" is a dotted path');
+      expect(offset.error?.issues[0]?.message).toContain('same-table column references only');
+      expect(DatasetSchema.safeParse({ ...PARSED, measures: [{ name: 'by_duty', aggregate: 'count', filter: RELATION_BARE }] }).success).toBe(true);
+    });
+  });
+
   describe('the `$field` COMPARAND never reaches that chain — refused even with `duty` in `include`', () => {
     for (const [arm, measure, ref] of [
       ['offset', 'done_on_time_by_duty', 'duty.grace_days'],
 
@@ -332,12 +332,14 @@ describe('[#25808] the dataset, valued end to end on the engine path', () => {
     expect(Number(byTitle.get('e')?.past_grace ?? 0)).toBe(0);
   });
 
-  it('the dotted offset spelling validates at the schema door and is refused by SQL push-down by name', async () => {
+  it('[#26601] the dotted offset spelling is refused at the schema door, and past it by the read, by name', async () => {
     const dotted = { completed_on: { $lte: { $now: true, addDays: { $field: 'duty.sla_days' } } } };
-    expect(DatasetSchema.safeParse({
+    const saved = DatasetSchema.safeParse({
       ...TASK_HEALTH,
       measures: [{ name: 'within_grace', aggregate: 'count', filter: dotted }],
-    }).success).toBe(true);
+    });
+    expect(saved.success).toBe(false);
+    expect(saved.error?.issues[0]?.message).toContain('"duty.sla_days" is a dotted path');
     const err = await (async () => {
       try {
         await service.queryDataset(
 
@@ -350,3 +350,95 @@ describe('[#26413] the now-anchor is closed at the door, as NowAnchorSchema decl
     }
   });
 });
+
+describe('[#26413 review] the subDays clause answers subDays alone, so a one-key refusal stays inside the bound', () => {
+  const NOT_APPLIED_TAIL = 'an unapplied filter would have returned the UNFILTERED result set.';
+
+  it('names the subDays rule when subDays is among the keys, and only then', () => {
+    const subDays = refusalOf(() => normalizeFilterComparandTypes({ opened_at: { $lt: { $now: true, addDays: -30, subDays: 7 } } }));
+    expect(subDays?.message).toContain('there is no subDays: a negative addDays subtracts');
+    const note = refusalOf(() => normalizeFilterComparandTypes({ opened_at: { $lt: { $now: true, addDays: -30, note: 'x' } } }));
+    expect(note?.code).toBe('INVALID_FILTER');
+    expect(note?.message).toContain('undeclared key "note" — the anchor takes $now and addDays only.');
+    expect(note?.message).not.toContain('subDays');
+  });
+
+  it('a one-key refusal under the engine prefix, under $and, keeps its last sentence under 500 characters', () => {
+    // The contract review's measured case was 496 characters with the clause
+    // printed for `note`; a longer field and a path under `$and` pushed it over.
+    const err = refusalOf(() => normalizeFilterComparandTypes(
+      { $and: [{ status: 'open' }, { review_opened_at: { $lte: { $now: true, addDays: -30, note: 'x' } } }] },
+      "find('support_ticket')",
+    ));
+    expect(err?.code).toBe('INVALID_FILTER');
+    expect(err!.message.length).toBeLessThanOrEqual(500);
+    expect(err!.message.endsWith(NOT_APPLIED_TAIL)).toBe(true);
+  });
+});
+
+describe('[#26601] a dotted addDays offset is refused at the door, on a reference and on an anchor', () => {
+  const REF = 'contract_type.review_sla_days';
+  /** `driver-sql`'s reason for the same refusal on every SQL read, verbatim. */
+  const SQL_REASON = `the addDays offset "${REF}" is a dotted path, and SQL push-down compiles same-table `
+    + 'column references only (no relation traversal, no alias-qualified columns).';
+  const COMPARANDS: Array<[string, Record<string, unknown>]> = [
+    ['the { $field, addDays } pair', { $field: 'review_due_at', addDays: { $field: REF } }],
+    ['the now-anchor', { $now: true, addDays: { $field: REF } }],
+    ['the now-anchor in its resolved form', { $now: '2026-03-03T12:00:00.000Z', addDays: { $field: REF } }],
+  ];
+
+  it('refuses it at every scalar comparison, in the SQL read\'s own words, with the envelope', () => {
+    for (const [label, comparand] of COMPARANDS) {
+      for (const op of ['$eq', '$ne', '$gt', '$gte', '$lt', '$lte']) {
+        const err = refusalOf(() => normalizeFilterComparandTypes({ review_started_at: { [op]: comparand } }));
+        expect(err?.code, `${label} ${op}`).toBe(StandardErrorCode.enum.INVALID_FILTER);
+        expect(err?.status, `${label} ${op}`).toBe(400);
+        expect(err?.message, `${label} ${op}`).toContain(`at where.review_started_at.${op} is refused: ${SQL_REASON}`);
+      }
+    }
+  });
+
+  it('refuses it under $and / $or / $not and through parseFilterAST\'s two forms', () => {
+    for (const [label, comparand] of COMPARANDS) {
+      for (const filter of [
+        { $and: [{ status: 'open' }, { review_started_at: { $lt: comparand } }] },
+        { $or: [{ review_started_at: { $gte: comparand } }] },
+        { $not: { review_started_at: { $lt: comparand } } },
+      ]) {
+        expect(refusalOf(() => normalizeFilterComparandTypes(filter))?.code, `${label} ${JSON.stringify(filter)}`).toBe('INVALID_FILTER');
+      }
+      expect(refusalOf(() => parseFilterAST({ review_started_at: { $lt: comparand } }))?.code, label).toBe('INVALID_FILTER');
+      expect(refusalOf(() => parseFilterAST(['review_started_at', '<', comparand]))?.code, label).toBe('INVALID_FILTER');
+    }
+  });
+
+  it('the refusal at the measure-filter position keeps its last sentence under 500 characters', () => {
+    const err = refusalOf(() => normalizeFilterComparandTypes(
+      { $and: [{ status: 'open' }, { contract_review_started_at: { $lte: { $now: true, addDays: { $field: 'contract_type_of_the_agreement.review_sla_days_offset' } } } }] },
+      "aggregate('clm_contract_review')",
+      'aggregations[0].filter',
+    ));
+    expect(err?.code).toBe('INVALID_FILTER');
+    expect(err!.message.length).toBeLessThanOrEqual(500);
+    expect(err!.message.endsWith('an unapplied filter would have returned the UNFILTERED result set.')).toBe(true);
+  });
+
+  it('CONTROL: a same-record offset column, a literal offset and a dotted BASE reference are kept by reference', () => {
+    for (const comparand of [
+      { $field: 'review_due_at', addDays: { $field: 'review_sla_days' } },
+      { $now: true, addDays: { $field: 'review_sla_days' } },
+      { $now: true, addDays: -3 },
+      // The reference's own `$field` is not this rule's: memory walks it, SQL refuses it.
+      { $field: 'contract_type.review_due_at' },
+      { $field: 'contract_type.review_due_at', addDays: 2 },
+    ]) {
+      const filter = { review_started_at: { $lt: comparand } };
+      expect(normalizeFilterComparandTypes(filter), JSON.stringify(comparand)).toBe(filter);
+    }
+  });
+
+  it('CONTROL: the bare implicit form keeps its own fate — an operator spec named $field, not judged here', () => {
+    const filter = { review_started_at: { $field: 'review_due_at', addDays: { $field: REF } } };
+    expect(normalizeFilterComparandTypes(filter)).toBe(filter);
+  });
+});