Add Scout lifecycle and filter maintenance APIs

docs/plans/2026-07-12-0915-framework-coroutine-state-lifecycle-audit-ledger.md

@@ -981,7 +981,7 @@ Append package entries in checklist order. Keep each entry compact but complete
 | `database-03` | Defect and parity defect | Major | High | Lazy refresh hooks run in the wrong coroutine/lifetime, runtime opt-out is ignored, and current assertion shapes are missing | Use the actual runtime coroutine flag, install hooks on every owned connection, make refresh state exception-safe, and port current multiple-row, iterable, and connection-aware assertions |
 | `foundation-16` | Package metadata and documentation defect | Major | High | Foundation's split package omits direct dependencies and provenance while several new public APIs and runtime limitations are undiscoverable | Declare only direct split dependencies, add focused metadata coverage and provenance, and update seven task-first user guides plus the Kernel contract note |
 | `foundation-17` | Integration-test harness defect | Major | High | Foundation's Meilisearch waiter discards failed task results and cleanup returns before its asynchronous index deletions finish, allowing consumers to report success after failed work or race the next test | Check every awaited result, propagate failures, wait for exact cleanup tasks, and keep Foundation as the single owner |
-| `foundation-18` | Integration-test harness defect | Major | High | Foundation's Algolia cleanup returns before its asynchronous index deletions finish, allowing the next test to reuse an index while deletion remains queued | Wait for each exact index-deletion task before setup or teardown continues |
+| `foundation-18` | Integration-test harness defect | Major | High | Foundation's Algolia cleanup returns before its asynchronous index deletions finish, allowing the next test to reuse an index while deletion remains queued | Wait for each exact index-deletion task and require a published terminal result before setup or teardown continues |
 
 - **Approved owner gates:** The owner approved the process-local/testing-only array maintenance driver, nullable redirect and current Laravel API additions, the truthful `Http\Kernel` contract expansion required by middleware configuration, the `0600` default for newly decrypted plaintext, and correcting verified upstream defects rather than preserving parity bugs. The Kernel contract change is documented for custom implementations. The existing worker maintenance wrapper remains a periodic same-process snapshot and does not make the array driver cross-process.
 - **Important rejected concerns:** Do not add shutdown-callback registries, PHPUnit meta-fixtures, a generic finalizer, request-scoped Application/Vite clones, Carbon managers, locks, watchers, retry loops, PID incarnation tracking, publication/delete transaction services, arbitrary-stream copy transactions, SQL parsers, TTY emulation, or broad PHPStan impurity annotations. The supported paths are covered by existing lifecycle, coroutine-context, Filesystem replacement, Process, and typed-reflection primitives. Retain `Filesystem::delete()` cache clearing because the full caller set and PHP stat cache require it; use targeted cache invalidation only where raw command/native postconditions demand it.
@@ -1563,7 +1563,7 @@ Append package entries in checklist order. Keep each entry compact but complete
 | `scout-19`, `scout-20`, `scout-36` | Correct provenance, concise actionable differences, current public docs/CLI language, and the identified Scout test typing. |
 | `scout-38`, `database-22` | Instantiate Scout search attributes and Database `CollectedBy` so positional and named arguments share the constructor-owned contract and malformed declarations fail fast. |
 | `foundation-17` | Make Foundation's Meilisearch test waiter reject failed awaited tasks and propagate timeouts, and make cleanup wait for its exact deletion tasks; delete Scout's duplicate wait path, use a service-valid custom-key fixture, and document the identifier restriction. |
-| `foundation-18` | Make Foundation's Algolia cleanup wait for each exact index-deletion task before setup or teardown continues. |
+| `foundation-18` | Make Foundation's Algolia cleanup wait for each exact index-deletion task and require a published terminal result before setup or teardown continues. |
 
 - **Worker and operation ownership:** Engines, SDK clients, and reusable transports remain worker-shared. Mutable builders, paginators, import failure state, observer suppression, and detached Typesense wrappers remain operation/coroutine-local. No per-request client, static reset, second cleanup registry, lock, or unbounded retained state is introduced.
 - **Approved Laravel-facing gates:** The owner approved top-level `scout.after_commit` / `SCOUT_AFTER_COMMIT`; three observer-required `SearchableInterface` methods; model-selected/container-substituted builders and paginators; Database raw pagination honoring its advertised model capability; Typesense rejecting unknown comparison operators; and four temporary wrapper objects per remote Typesense operation to eliminate unbounded per-name retention. Scout job-option properties remain intentionally untyped because current Laravel supports subclasses that redeclare them.

docs/plans/2026-08-02-1006-scout-current-parity-queue-and-search-lifecycles.md

@@ -133,7 +133,7 @@ Apart from the approved `database-21` behavior correction, no accepted item adds
 | `database-23` | Descending cursor type defect | Major | `forPageBeforeId()` accepts string keys so descending chunk and lazy traversal work beyond the first page. |
 | `scout-40` | Integer range defect and upstream defect | Major | Descending queue-import ranges use overflow-safe remaining-distance arithmetic; the ascending branch uses the same exact formulation instead of relying on mixed numeric tie-breaking. |
 | `foundation-17` | Integration-test harness defect | Major | Foundation's Meilisearch waiter reports failed awaited tasks and timeouts, while cleanup waits for its exact deletion tasks; Scout uses that single owner, a service-valid custom-key fixture, and documents Meilisearch's identifier alphabet. |
-| `foundation-18` | Integration-test harness defect | Major | Foundation's Algolia cleanup waits for each exact index-deletion task before setup or teardown continues. |
+| `foundation-18` | Integration-test harness defect | Major | Foundation's Algolia cleanup waits for each exact index-deletion task and requires a published terminal result before setup or teardown continues. |
 
 ## Implementation design
 
@@ -321,9 +321,9 @@ The two existing `DatabaseQueryBuilderTest` methods retain their current signatu
 
 ### 9. External search integration task truthfulness (`foundation-17`, `foundation-18`)
 
-Use Foundation's `InteractsWithMeilisearch::waitForMeilisearchTasks()` as the single wait owner. Preserve its current selection of all pending tasks, inspect each terminal result returned by `waitForTask()`, throw its error when the status is not `succeeded`, and let timeout or transport exceptions propagate. Meilisearch cleanup waits for the exact `taskUid` values returned by its deletions; Algolia cleanup waits for each exact index name and `taskID`. Setup failures propagate, while teardown remains exception-safe. Delete the duplicate Support implementation; Scout keeps its longer timeout through parent delegation.
+Use Foundation's `InteractsWithMeilisearch::waitForMeilisearchTasks()` as the single wait owner. Select pending tasks whose index UID begins with the current test worker's prefix, inspect each terminal result returned by `waitForTask()`, throw its error when the status is not `succeeded`, and let timeout or transport exceptions propagate. Meilisearch cleanup waits for the exact `taskUid` values returned by its deletions; Algolia cleanup waits for each exact index name and `taskID` and requires a published terminal result because its SDK can swallow a polling failure and return null. Setup failures propagate, while teardown remains exception-safe. Delete the duplicate Support implementation; Scout keeps its longer timeout through parent delegation.
 
-The shared custom-key fixture uses a Meilisearch-valid hyphenated key while remaining distinct from the database key. Keep the positive remote precondition before queued removal. Add deterministic harness coverage for successful, failed, and timed-out waits rather than racing the real service, and state the Meilisearch identifier alphabet once in the custom-key guide. Do not poll in production, change `Engine::update(): void`, scan already-terminal global failures, filter by test prefix, or add a task watermark.
+The shared custom-key fixture uses a Meilisearch-valid hyphenated key while remaining distinct from the database key. Keep the positive remote precondition before queued removal. Add deterministic harness coverage for successful, failed, and timed-out waits rather than racing the real service, and state the Meilisearch identifier alphabet once in the custom-key guide. Do not poll in production, change `Engine::update(): void`, scan already-terminal global failures, or add a task watermark.
 
 ## Test plan