整合官方 Sub2API v0.2.16 并保留本站契约

.downstream/upstream-base

@@ -1 +1 @@
-v0.2.15
+v0.2.16

.downstream/upstream-review-v0.2.16.md

@@ -0,0 +1,46 @@
+# Official v0.2.16 integration
+
+- Official tag: `v0.2.16`, commit `01842576b3d5c1e2545b5a99ad555d36e28ed306`.
+- Downstream starting commit: `157836bdc12424a71f1d0bd30a37fcca38842fc9`.
+- The 17 initial conflicts cover gateway request/error handling, pricing, model
+  discovery, dependency injection, effort metadata, and account UI/tests. Ent and
+  Wire are regenerated from the combined schemas and provider sets. Existing
+  runtime dependency pins and Go 1.27.2 remain.
+- Haiku 5.5 registration, validation, bridge, sampling and tier pricing follow
+  official behavior. Its downstream effort entry is adaptive with medium default
+  effort. Official API capacity evidence is 1M/128K, verified on 2026-10-13 from
+  the [model overview](https://platform.claude.com/docs/en/models/haiku-5-5/overview).
+- Codex descriptors retain zero generated capacity defaults, including Sol's
+  upstream 500k default. Account, group and client capacities still use the shared
+  source-bound resolver. Sol subscription model-card pricing exemption uses the
+  actual account and upstream model, without changing API-key/custom-interval
+  pricing or downstream DeepSeek time-pricing application.
+- Structured-input Chat fallback retains the official complete-body and safe
+  retry guards, plus downstream prompt-echo restoration, raw account classifiers,
+  request cancellation and session identity. Native Responses preserves downstream
+  request-terminal metadata (`NextAccountStop`, no account-health penalty), which
+  the inherited no-retry case verifies directly.
+- Gemini explicit caching, Vertex authentication/model discovery, model-level
+  cooldown scopes, group translations and UI fixes follow official behavior.
+  Scheduler projections retain both downstream capabilities and the new cache
+  authentication/TTL fields; catalog readers share bounded response handling and
+  retain downstream Anthropic/Gemini pagination.
+- Ordinary OpenAI HTTP scope tests use the production failover marker. Explicit
+  model-scoped 401 rules are also evaluated before the unmarked count/probe auth
+  fast path; missing/account-scoped rules retain the existing downstream behavior,
+  and credential-level/OAuth failures keep their authentication handling. The
+  unmarked API-key path still records its scheduling-neutral last upstream error.
+- Existing Codex identity and strict STATE borrowing, Pelican jobs, prompt binding,
+  full administrator visibility, account/model mapping, capacity evidence and
+  flat-console hooks remain. The model picker keeps its explicit confirmation and
+  stale-response guard while showing the upstream sync error message.
+- Official migrations `242_group_i18n`, `243_gemini_cached_contents`,
+  `244_gemini_cached_contents_channel_usage` and
+  `245_gemini_cached_contents_mutation_lease` are unchanged. Full filenames are
+  migration identities; matching numeric prefixes do not replace historical SQL.
+  Source VERSION is corrected from the official tag's stale 0.2.15 to 0.2.16.
+
+Local validation uses synthetic upstreams, UI fixtures and disposable PostgreSQL
+18.1. Normal required PR CI remains authoritative. Source validation, immutable
+release and US production availability are recorded separately in the operator
+workspace delivery evidence. No real model generation is part of this integration.

DOWNSTREAM.md

@@ -5,7 +5,7 @@ This fork maintains the `codexrip` patch set over official Sub2API releases.
 ## Source baseline
 
 - Integrated official baseline: the release recorded in `.downstream/upstream-base`.
-- Retained compatibility decisions: [upstream review](.downstream/upstream-review-v0.2.15.md).
+- Retained compatibility decisions: [upstream review](.downstream/upstream-review-v0.2.16.md).
 - The operator workspace `docs/sub2api.md` owns the production version and image digest. A source merge or Release does not establish deployment completion.
 
 ## Settings and account operations
@@ -186,9 +186,21 @@ Scheduled discovery may identify newer official releases. It cannot deploy
 production. Each integration records its selected official commit, conflicts and
 retained downstream contracts before a new immutable release is published.
 
-`.downstream/upstream-risk.json` records the v0.2.14-to-v0.2.15 integration against
-downstream commit `d469422e94b0655b22dd9d0a826decda61ffb91e`. The risk gate
-recomputes those exact inputs. `initial_conflict_files` lists the 40
-conflicts resolved during this merge; `merge_conflicts: []` records that none
-remain. A `review_required` candidate needs a real completed review and the
-`upstream-reviewed` label before promotion.
+`.downstream/upstream-risk.json` records the v0.2.15-to-v0.2.16 integration against
+downstream commit `157836bdc12424a71f1d0bd30a37fcca38842fc9`. The risk gate
+recomputes those exact inputs. `initial_conflict_files` records the 17 resolved
+conflicts; `merge_conflicts: []` records that none remain. The candidate requires
+a completed review and the `upstream-reviewed` label before promotion.
+
+Gemini explicit caches follow the official account/key binding, storage billing,
+mutation lease and recovery rules. Vertex API-key authentication and service-account
+model discovery, account/model temporary-unschedulable scopes and group display
+translations follow v0.2.16. Existing unconfigured accounts and rules retain their
+stored behavior. Official migrations 242 through 245 keep their original filenames
+and coexist with downstream migrations of the same numeric prefixes.
+
+Haiku 5.5 uses adaptive thinking and medium default effort; its API limits are
+resolved through the official evidence catalog. Local Codex descriptors still
+carry no capacity defaults, including the official Sol 500k generation default.
+Sol's subscription model-card long-context exemption follows the actual credential
+and upstream model; API-key and custom-interval pricing keep their own policies.

Makefile

@@ -5,6 +5,7 @@ FRONTEND_CRITICAL_VITEST := \
 	src/api/__tests__/client.spec.ts \
 	src/api/__tests__/tokenRefresh.spec.ts \
 	src/api/__tests__/keys.bulkUpdate.spec.ts \
+	src/composables/__tests__/useModelWhitelist.spec.ts \
 	src/components/account/__tests__/OpenAIReferralCell.spec.ts \
 	src/components/account/__tests__/OpenAIReferralCell.transport.spec.ts \
 	src/components/account/__tests__/OpenAIQuotaResetCell.spark_shadow.spec.ts \

backend/cmd/server/VERSION

@@ -1 +1 @@
-0.2.15
+0.2.16

backend/cmd/server/wire.go

@@ -117,6 +117,7 @@ func provideCleanup(
 	subscriptionExpiry *service.SubscriptionExpiryService,
 	usageCleanup *service.UsageCleanupService,
 	idempotencyCleanup *service.IdempotencyCleanupService,
+	geminiCachedContent *service.GeminiCachedContentService,
 	batchImageCleanup *service.BatchImageCleanupService,
 	batchImageWorker *service.BatchImageWorkerRuntime,
 	pricing *service.PricingService,
@@ -280,6 +281,12 @@ func provideCleanup(
 				}
 				return nil
 			}},
+			{"GeminiCachedContentService", func() error {
+				if geminiCachedContent != nil {
+					geminiCachedContent.Stop()
+				}
+				return nil
+			}},
 			{"BatchImageCleanupService", func() error {
 				if batchImageCleanup != nil {
 					batchImageCleanup.Stop()

backend/cmd/server/wire_gen_test.go

@@ -84,6 +84,7 @@ func minimalDependencyCleanup(gatewayCache service.GatewayCache, autoReset *serv
 		subscriptionExpirySvc,
 		&service.UsageCleanupService{},
 		idempotencyCleanupSvc,
+		service.NewGeminiCachedContentService(nil, nil, cfg),
 		&service.BatchImageCleanupService{},
 		nil, // batchImageWorker
 		pricingSvc,