Add bounded schedule list counters and current-run summary

docs/schedule-list-scalar-summary.md

@@ -0,0 +1,83 @@
+# Schedule list scalar summary draft
+
+This is the smaller source slice selected in Server #1014 comment 6644878850.
+It is not the full last-dispatch outcome contract or published capability. Waterline
+#222 keeps the existing describe fallback. No faster polling is justified by this
+partial projection, and describe remains unchanged.
+
+Each list item adds:
+
+| Field | Complete value | Unavailable or malformed storage |
+| --- | --- | --- |
+| `failures_count` | Nonnegative stored signed-64-bit-decodable integer | Decode/negative value fails the request |
+| `skipped_trigger_count` | Same counter contract | Same failure |
+| `latest_workflow_instance_id` | Stored identity or null, at most 191 Unicode scalars / 764 UTF-8 bytes | Empty/oversized/invalid UTF-8 fails; a pruned instance preserves this identity |
+| `current_run` | Nullable `{run_id,status,status_bucket,is_terminal}` from that instance's current run | Missing/pruned/wrong-namespace/wrong-instance target is null, meaning unknown |
+
+The instance and run joins require exact identities and exact schedule namespace
+on both sides. MySQL/MariaDB retain a candidate equality usable by each existing index, then
+apply an exact binary residual to avoid case/accent/trailing-space collation
+matches. The page namespace predicate uses the same candidate/residual pattern. A current run ID and the instance's complete current-run
+pointer have at most 26 Unicode scalars / 104 UTF-8 bytes. Status is a complete
+value at most 32 UTF-8 bytes and must be one of the existing statuses:
+`pending`, `running`, `waiting` map to running/nonterminal; `completed` maps to
+completed/terminal; `failed`, `cancelled`, `terminated` map to failed/terminal.
+An unknown or empty status is invalid storage, never a default healthy state.
+The dispatch count is not a run status and the current run is not a last-dispatch
+outcome. Missing current-run information does not imply completion or no action.
+
+For newly projected scalars, SQL transfers the **whole** value only within its
+byte bound and returns an independent invalid flag otherwise. Rust checks the
+complete decoded scalar count and enum before returning a response. No prefix
+can become a plausible identity or status. NUL and astral scalars remain exact in SQLite/MySQL. PostgreSQL text cannot
+store NUL, which is an unsupported storage-domain cell, not a decoder pass.
+Invalid UTF-8 setup limitations remain distinct from SQLite decoder refusal.
+Malformed/oversized scalar metadata refuses the page with the existing typed
+HTTP 503 `storage_unavailable` envelope; values are not included in diagnostics.
+Backend decode failures retain the existing storage-unavailable response.
+
+| SQL boundary | SQLite | PostgreSQL | MySQL / MariaDB |
+| --- | --- | --- | --- |
+| Byte size including NUL | `length(CAST(column AS BLOB))` | `OCTET_LENGTH(column)` | `OCTET_LENGTH(column)` |
+| Scalar projection | `CASE WHEN bytes<=bound THEN column ELSE NULL END` | Same | Same |
+| Invalid flag | Non-null and empty or over-bound, separately from SQL null | Same | Same |
+| Namespace / join identity | Exact text comparison | Exact text comparison | Candidate `left=right` plus `BINARY left=BINARY right` |
+
+The page selects explicit existing list columns and bounded scalar expressions.
+It excludes `recent_actions`, `buffered_actions`, `last_skip_reason` and
+`last_skipped_at` entirely. Recent/last-action, buffer and last-skip fields remain
+**deferred**, including their malformed-member, Unicode and timestamp semantics.
+No SQL date conversion, substring preview, JSON prefix or invented event ordering
+is introduced. The existing legacy action/input/spec/note/search-attribute fields
+keep their previous size surface; this does not claim a bounded whole response
+or constant database bytes/CPU.
+
+The one-to-one joins add no row-level calls. Existing request normalization,
+closed grammar, bound values, filter fingerprint, signed cursor, stale-anchor
+check, order, page size and plus-one lookahead stay intact. Cursor probes select
+only schedule ID/created-at and retain the same filters. A plain page uses one
+query; either cursor presence or registered-definition lookup adds one; both use
+three. Registered attributes still have the existing native unsupported-value
+behavior, outside this slice.
+
+Source vectors cover a 50-row page and its continuation with/without a system
+query (1/2/2/3 calls), malformed excluded arrays, independent counters, missing
+and wrong-namespace links, all seven statuses, pruned targets, exact NUL/astral
+identities and empty/scalar-count/byte-count failures. The existing MySQL/MariaDB and PostgreSQL disposable source suites invoke the
+same complete decoder/join controls, including maximum legal unsigned counters
+and exact HTTP 503/no-value responses on MySQL/MariaDB. The shared unsigned
+decoder remains unchanged with its original 409 behavior. MySQL/MariaDB EXPLAIN
+controls retain actual primary-key/access-type output for each referenced join
+after seeding 256 unrelated instances/runs. Those controls require successful
+normal backend qualification; source authorship alone is not plan evidence.
+The existing HTTP schedule-catalog corpus retains its frozen published-PHP
+fixture and expected responses. Native recordings require all four new fields
+on every list row. Before comparing the older shared response, the comparator
+checks counters and the latest identity against preceding backfill responses,
+the unclaimed current-run state, and the retained trigger-audit identity pair.
+Only those checked additions are removed from that comparison; raw observations
+and every existing field remain intact. Older published PHP recordings can
+retain their original list shape.
+Normal source checks and Server review must qualify the exact SQL/decoder on each backend before this
+can be called contract-compliant. No new runtime/image/qualification dispatch,
+schema/UDF, persisted summary, release or frozen-artifact admission follows.

docs/server-reference.md

@@ -1203,6 +1203,10 @@ heartbeat, complete, and fail responses include `run_closed_reason` and
 - `POST /api/schedules/{id}/trigger` — Trigger immediately
 - `POST /api/schedules/{id}/backfill` — Backfill missed runs
 
+The native development [scalar summary draft](schedule-list-scalar-summary.md) adds
+counters, latest-instance identity and a nullable current-run summary. Full last-
+dispatch/buffer/skip summaries and published capability remain pending Server #1014.
+
 Schedule listing is namespace-scoped and never returns deleted schedules. The
 optional `status` (`active` or `paused`) and `workflow_type` filters are exact
 matches. `query` accepts equality predicates joined by `AND` for

rust/src/runtime/mysql.rs

@@ -951,6 +951,16 @@ mod tests {
         }
     }
 
+    #[tokio::test]
+    #[ignore = "requires the existing disposable mysql source suite"]
+    async fn qualification_schedule_list_complete_summary_and_join_plans() {
+        let db=Database::new().await;
+        let store=super::super::store::Store::new(MySqlStorage::open(db.options.clone()).await.unwrap().into_pool());
+        store.qualify_schedule_list_summary().await;
+        store.close().await;
+        db.remove().await;
+    }
+
     #[tokio::test]
     #[ignore = "requires disposable MySQL/MariaDB; run qualification_ explicitly"]
     async fn qualification_schedule_creation_keeps_original_audit_and_unsigned_counters() {

rust/src/runtime/postgres.rs

@@ -534,6 +534,16 @@ mod tests {
     /// 0.9.0 it did (vendor/sqlx-postgres/DURABLE-PATCH.md): the next borrower's
     /// probe ran inside it, native's read snapshot `BEGIN` then failed with
     /// 25001, and every later statement on that connection with 25P02.
+    #[tokio::test]
+    #[ignore = "requires the existing disposable postgres source suite"]
+    async fn qualification_schedule_list_complete_summary_and_join_plans() {
+        let db=Database::new().await;
+        let store=super::super::store::Store::new(PostgresStorage::open(db.options.clone()).await.unwrap().into_pool());
+        store.qualify_schedule_list_summary().await;
+        store.close().await;
+        db.remove().await;
+    }
+
     #[tokio::test]
     #[ignore = "requires a disposable PostgreSQL; run the qualification_ tests explicitly"]
     async fn qualification_a_begin_cancelled_in_flight_leaves_no_transaction_behind() {