Report bounded schedule dispatch summaries for Waterline refresh

docs/server-reference.md

@@ -1258,10 +1258,38 @@ with at most 64 Unicode scalars / 256 UTF-8 bytes. The nullable UTC timestamp
 preserves six fractional digits, for example `2026-11-05T12:00:01.123456Z`, and
 matches describe's `info.last_skipped_at`. Invalid or oversized stored values
 return HTTP 503 `storage_unavailable`. A later successful fire preserves an
-earlier skip, so this pair is not the latest overall dispatch outcome. Full
-last-dispatch/buffer summaries and published capability remain pending Server
-#1014. An empty list cannot prove pair support; consumers retain describe
-fallback and their current polling policy.
+earlier skip, so this pair is not the latest overall dispatch outcome. The
+dispatch summary has its own authority below. An empty list cannot prove
+support for either contract.
+
+Native list rows and describe responses also carry a top-level
+`dispatch_summary` v1. Older producers omit it. The complete object contains
+`version`, `state`, `last_outcome` and `buffered_actions_count`:
+
+| State | Last outcome | Current buffer count |
+| --- | --- | --- |
+| `known` | Complete recorded dispatch | Integer from 0 through 9007199254740991 |
+| `never` | `null`, established only by new creation | `0` |
+| `unavailable` | `null`, including legacy or invalid stored authority | `null` |
+
+A known outcome has exactly `id`, `outcome`, `recorded_at`, `reason`,
+`workflow_instance_id` and `workflow_run_id`. Outcomes are `started`, `buffered`,
+`skipped` or `failed`. Started outcomes name both new uppercase ULIDs and have
+a null reason. Other outcomes have null workflow identities. Skipped and failed
+outcomes keep the complete reason, including empty strings and whitespace;
+buffered outcomes have a null reason. Times are UTC with six fractional digits.
+The complete summary is bounded to 4096 serialized UTF-8 bytes. An outcome that
+cannot fit the complete contract replaces earlier authority with unavailable
+without changing the underlying scheduling operation or truncating its reason.
+
+A valid unavailable summary overrides older legacy outcome evidence. Absent,
+null, partial or unsupported summaries provide no v1 authority. A later describe
+owns the whole summary, including absence or null; do not merge it with a list
+snapshot. Count-only buffer changes retain the outcome ID and time, so a buffered
+outcome can coexist with a current count of zero. Neither timestamp ranking nor
+the latest-workflow pointer determines the latest dispatch. Current workflow
+execution status remains separate. Native schema upgrades leave existing rows
+unknown and do not reconstruct summaries from legacy arrays.
 
 Schedule listing is namespace-scoped and never returns deleted schedules. The
 optional `status` (`active` or `paused`) and `workflow_type` filters are exact

resources/platform-protocol-specs/control-plane-api.openapi.yaml

@@ -1217,10 +1217,74 @@ components:
               spec: { type: [object, "null"], additionalProperties: true }
               action: { type: [object, "null"], additionalProperties: true }
               search_attributes: { type: [object, "null"], additionalProperties: true }
+              dispatch_summary: { $ref: "#/components/schemas/ScheduleDispatchSummary" }
               created_at: { type: string, format: date-time }
               updated_at: { type: string, format: date-time }
         schedule_count: { type: integer, minimum: 0, maximum: 200 }
         next_page_token: { type: [string, "null"] }
+    ScheduleDispatchSummary:
+      description: >-
+        Native bounded dispatch authority, also at the top level of describe.
+        Older producers omit this member. Partial, null or unsupported objects
+        supply no v1 authority. A valid unavailable object overrides legacy
+        outcome evidence. A later describe owns its entire summary.
+      type: object
+      additionalProperties: false
+      required: [version, state, last_outcome, buffered_actions_count]
+      properties:
+        version: { type: integer, const: 1 }
+        state: { type: string, enum: [known, never, unavailable] }
+        last_outcome:
+          oneOf:
+            - { $ref: "#/components/schemas/ScheduleDispatchOutcome" }
+            - { type: "null" }
+        buffered_actions_count: { type: [integer, "null"], minimum: 0, maximum: 9007199254740991 }
+      oneOf:
+        - properties:
+            state: { const: known }
+            last_outcome: { $ref: "#/components/schemas/ScheduleDispatchOutcome" }
+            buffered_actions_count: { type: integer }
+        - properties:
+            state: { const: never }
+            last_outcome: { type: "null" }
+            buffered_actions_count: { const: 0 }
+        - properties:
+            state: { const: unavailable }
+            last_outcome: { type: "null" }
+            buffered_actions_count: { type: "null" }
+    ScheduleDispatchOutcome:
+      description: >-
+        Last committed dispatch, independently of current execution status.
+        Reason strings retain all text within 64 Unicode scalars and 256 UTF-8
+        bytes. Oversized or incomplete outcomes produce unavailable, without
+        changing the underlying scheduling operation. The complete summary is
+        at most 4096 serialized UTF-8 bytes.
+      type: object
+      additionalProperties: false
+      required: [id, outcome, recorded_at, reason, workflow_instance_id, workflow_run_id]
+      properties:
+        id: { type: string, pattern: "^[0-7][0-9A-HJKMNP-TV-Z]{25}$" }
+        outcome: { type: string, enum: [started, buffered, skipped, failed] }
+        recorded_at: { type: string, format: date-time, pattern: '^\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}\.\d{6}Z$' }
+        reason: { type: [string, "null"], maxLength: 64 }
+        workflow_instance_id: { type: [string, "null"], pattern: "^[0-7][0-9A-HJKMNP-TV-Z]{25}$" }
+        workflow_run_id: { type: [string, "null"], pattern: "^[0-7][0-9A-HJKMNP-TV-Z]{25}$" }
+      oneOf:
+        - properties:
+            outcome: { const: started }
+            reason: { type: "null" }
+            workflow_instance_id: { type: string }
+            workflow_run_id: { type: string }
+        - properties:
+            outcome: { const: buffered }
+            reason: { type: "null" }
+            workflow_instance_id: { type: "null" }
+            workflow_run_id: { type: "null" }
+        - properties:
+            outcome: { enum: [skipped, failed] }
+            reason: { type: string }
+            workflow_instance_id: { type: "null" }
+            workflow_run_id: { type: "null" }
     ScheduleListError:
       type: object
       additionalProperties: true

rust/README.md

@@ -420,6 +420,8 @@ table; commands admitted earlier have no digest. Stop the native nodes, back up
 the database and run `schema-upgrade` before starting this binary on an
 earlier schema; rolling back means restoring the backup.
 
+Schedule dispatch summaries add a nullable scalar on `workflow_schedules`: SQLite schema 14, PostgreSQL and MariaDB 15, MySQL 16. Existing rows remain unknown, without an array scan or backfill. New dispatches and buffer mutations maintain the summary atomically. This step also requires stopped nodes, a backup and `schema-upgrade`; restore the backup to roll back. The wire states and bounds are documented in [Schedules](../docs/server-reference.md#schedules).
+
 Implemented paths cover health/schema readiness, limited capability discovery,
 workflow start/describe/paginated history, worker registration/heartbeat/removal,
 workflow/activity polling, workflow lease renewal and both completion paths.