feat(openrouter): pin and sort chat providers; LoCoMo v21 on glm-5.3-flash

.env.example

@@ -69,6 +69,19 @@ REMEMBERSTACK_OPENROUTER_MAX_COMPLETION_TOKENS=32000
 # effort literals. Example: pin flash models to none while keeping high for a
 # reasoning model.
 # REMEMBERSTACK_OPENROUTER_REASONING_EFFORT_MAP={"z-ai/glm-4.7-flash":"none","openai/gpt-5.6-luna":"high"}
+
+# Restrict CHAT completions to named OpenRouter providers (allowlist of provider
+# slugs, not quantization tags). `only` bounds the pool, so failover moves
+# between these and can never leave them. Unset = ordinary marketplace routing.
+# REMEMBERSTACK_OPENROUTER_CHAT_PROVIDER_ONLY=z-ai,novita,deepinfra,gmicloud
+#
+# How OpenRouter orders those providers: price | throughput | latency.
+# Unset uses OpenRouter's default, which weights price heavily -- and the
+# cheapest endpoint is usually the most contended, so every call AND retry
+# lands on it and dead-letters together when it is overloaded. Note that
+# allow_fallbacks does NOT rescue this: a provider-returned 429 is surfaced
+# as "Provider returned error", not treated as the provider being unavailable.
+# REMEMBERSTACK_OPENROUTER_CHAT_PROVIDER_SORT=throughput
 # Debug only: retain raw schema-invalid completions under the private app-state
 # volume. Disabled by default because captured output may repeat source text.
 # REMEMBERSTACK_OPENROUTER_INVALID_COMPLETION_CAPTURE_DIR=/var/lib/rememberstack/openrouter-invalid-completions

benchmarks/locomo/__init__.py

@@ -1 +1 @@
-"""The pinned RS-LoCoMo-Full-v20 benchmark adapter."""
+"""The pinned RS-LoCoMo-Full-v21 benchmark adapter."""

benchmarks/locomo/cli.py

@@ -118,7 +118,7 @@ def _parser() -> argparse.ArgumentParser:
     parser = argparse.ArgumentParser(
         prog="python -m benchmarks.locomo",
         description=(
-            "RS-LoCoMo-Full-v20: prepare is local; ingest/answer/judge require "
+            "RS-LoCoMo-Full-v21: prepare is local; ingest/answer/judge require "
             "explicit execution acknowledgements"
         ),
     )

benchmarks/locomo/dataset.py

@@ -140,10 +140,10 @@ def load_manifest(tier: str) -> QuestionManifest:
         )
     if manifest.dataset_commit != DATASET_COMMIT:
         raise DatasetValidationError(
-            "manifest dataset commit is not RS-LoCoMo-Full-v20"
+            "manifest dataset commit is not RS-LoCoMo-Full-v21"
         )
     if manifest.dataset_sha256 != DATASET_SHA256:
-        raise DatasetValidationError("manifest dataset hash is not RS-LoCoMo-Full-v20")
+        raise DatasetValidationError("manifest dataset hash is not RS-LoCoMo-Full-v21")
     actual = item_ids_hash(item_ids=manifest.item_ids)
     if actual != manifest.item_ids_sha256:
         raise DatasetValidationError(

benchmarks/locomo/model.py

@@ -1,4 +1,4 @@
-"""Typed values for the full-system RS-LoCoMo-Full-v20 protocol."""
+"""Typed values for the full-system RS-LoCoMo-Full-v21 protocol."""
 
 from __future__ import annotations
 
@@ -28,11 +28,11 @@
 Category = Literal[1, 2, 3, 4, 5]
 RetainedCategory = Literal[1, 2, 3, 4]
 Tier = Literal["smoke", "development", "publication"]
-ProtocolKey = Literal["full-v20"]
-ProtocolName = Literal["RS-LoCoMo-Full-v20"]
+ProtocolKey = Literal["full-v21"]
+ProtocolName = Literal["RS-LoCoMo-Full-v21"]
 SourceTimezoneBasis = Literal["assumed_utc"]
-AnswerAgentModel = Literal["openai/gpt-5.6-luna"]
-JudgeModel = Literal["openai/gpt-5.6-luna"]
+AnswerAgentModel = Literal["z-ai/glm-5.3-flash"]
+JudgeModel = Literal["z-ai/glm-5.3-flash"]
 FailureKind = Literal[
     "readiness", "tool", "reader", "judge", "accounting", "invalid_response", "missing"
 ]
@@ -120,7 +120,7 @@ class QuestionManifest(FrozenModel):
 class RunConfiguration(FrozenModel):
     """Immutable identity of one prepared benchmark run."""
 
-    protocol_name: ProtocolName = "RS-LoCoMo-Full-v20"
+    protocol_name: ProtocolName = "RS-LoCoMo-Full-v21"
     adapter_version: NonEmpty
     prepared_at: datetime
     repository_revision: NonEmpty
@@ -139,10 +139,10 @@ class RunConfiguration(FrozenModel):
     api_timeout_seconds: float = Field(default=60.0, gt=0)
     knowledge_mode: Literal["not_composed"] = "not_composed"
     document_binding_generation: Literal["document-t0-v1"] = "document-t0-v1"
-    answer_agent_model: AnswerAgentModel = "openai/gpt-5.6-luna"
+    answer_agent_model: AnswerAgentModel = "z-ai/glm-5.3-flash"
     answer_agent_reasoning_effort: Literal["none"] = "none"
     answer_word_cap: int | None = Field(default=None, ge=1)
-    judge_model: JudgeModel = "openai/gpt-5.6-luna"
+    judge_model: JudgeModel = "z-ai/glm-5.3-flash"
     judge_reasoning_effort: Literal["none"] = "none"
     answer_agent_temperature: float = Field(default=0.0, ge=0, le=2)
     judge_temperature: float = Field(default=0.0, ge=0, le=2)
@@ -393,7 +393,7 @@ class SessionDiagnosticSummary(FrozenModel):
 class RunSummary(FrozenModel):
     """Publication-ready local aggregate with no hidden denominator."""
 
-    protocol_name: ProtocolName = "RS-LoCoMo-Full-v20"
+    protocol_name: ProtocolName = "RS-LoCoMo-Full-v21"
     protocol_fingerprint: NonEmpty
     tier: Tier
     questions: int = Field(ge=1)

benchmarks/locomo/protocol.py

@@ -29,9 +29,11 @@
 from rememberstack.model import Envelope
 from rememberstack.model import ToolDescriptor
 
-PROTOCOL_NAME: Final = "RS-LoCoMo-Full-v20"
-DEFAULT_PROTOCOL_KEY: Final = "full-v20"
-ADAPTER_VERSION: Final = "locomo-full-adapter-2026.09-complete-values-v20"
+PROTOCOL_NAME: Final = "RS-LoCoMo-Full-v21"
+DEFAULT_PROTOCOL_KEY: Final = "full-v21"
+ADAPTER_VERSION: Final = (
+    "locomo-full-adapter-2026.09-complete-values-glm53-pinned-providers-v21"
+)
 MAX_TOOL_CALLS: Final = 8
 MAX_AGENT_CALLS: Final = 9
 ANSWER_READER_RETRY_BUDGET: Final = 2
@@ -87,30 +89,32 @@
 EXPECTED_INGEST_MODEL_BINDINGS: Final[Mapping[str, str]] = MappingProxyType(
     {
         "chunk_embedding": "qwen/qwen3-embedding-8b",
-        "claim_extraction": "openai/gpt-5.6-luna",
-        "context_prefix": "openai/gpt-5.6-luna",
+        "claim_extraction": "z-ai/glm-5.3-flash",
+        "context_prefix": "z-ai/glm-5.3-flash",
         "entity_observation_embedding": "qwen/qwen3-embedding-8b",
-        "fact_label": "openai/gpt-5.6-luna",
-        "observation_frontier": "openai/gpt-5.6-luna",
-        "observation_small": "openai/gpt-5.6-luna",
+        "fact_label": "z-ai/glm-5.3-flash",
+        "observation_frontier": "z-ai/glm-5.3-flash",
+        "observation_small": "z-ai/glm-5.3-flash",
+        "openrouter_chat_provider_only": "z-ai,novita,deepinfra,gmicloud",
+        "openrouter_chat_provider_sort": "throughput",
         "openrouter_embedding_provider": "nebius",
         "openrouter_embedding_provider_order": "unset",
         "openrouter_max_completion_tokens": "32000",
         "openrouter_reasoning_effort": "auto",
-        "openrouter_reasoning_effort_map": '{"openai/gpt-5.6-luna": "high"}',
+        "openrouter_reasoning_effort_map": '{"z-ai/glm-5.3-flash": "high"}',
         "p1_embedding": "qwen/qwen3-embedding-8b",
-        "relation_normalization": "openai/gpt-5.6-luna",
-        "section_role": "openai/gpt-5.6-luna",
-        "section_summary": "openai/gpt-5.6-luna",
-        "skeleton_check": "openai/gpt-5.6-luna",
-        "structure_fallback": "openai/gpt-5.6-luna",
-        "supersession_frontier": "openai/gpt-5.6-luna",
-        "supersession_small": "openai/gpt-5.6-luna",
+        "relation_normalization": "z-ai/glm-5.3-flash",
+        "section_role": "z-ai/glm-5.3-flash",
+        "section_summary": "z-ai/glm-5.3-flash",
+        "skeleton_check": "z-ai/glm-5.3-flash",
+        "structure_fallback": "z-ai/glm-5.3-flash",
+        "supersession_frontier": "z-ai/glm-5.3-flash",
+        "supersession_small": "z-ai/glm-5.3-flash",
     }
 )
-ANSWER_AGENT_MODEL: Final = "openai/gpt-5.6-luna"
+ANSWER_AGENT_MODEL: Final = "z-ai/glm-5.3-flash"
 ANSWER_AGENT_REASONING_EFFORT: Final = "none"
-JUDGE_MODEL: Final = "openai/gpt-5.6-luna"
+JUDGE_MODEL: Final = "z-ai/glm-5.3-flash"
 JUDGE_REASONING_EFFORT: Final = "none"
 TEMPERATURE: Final = 0.0
 
@@ -211,8 +215,8 @@ class LoCoMoProtocol:
     answer_word_cap: int | None = None
 
 
-_FULL_V20 = LoCoMoProtocol(
-    key="full-v20",
+_FULL_V21 = LoCoMoProtocol(
+    key="full-v21",
     name=PROTOCOL_NAME,
     answer_agent_model=ANSWER_AGENT_MODEL,
     judge_model=JUDGE_MODEL,
@@ -234,7 +238,7 @@ class LoCoMoProtocol:
 )
 
 PROTOCOL_REGISTRY: Final[Mapping[ProtocolKey, LoCoMoProtocol]] = MappingProxyType(
-    {_FULL_V20.key: _FULL_V20}
+    {_FULL_V21.key: _FULL_V21}
 )
 
 

benchmarks/locomo/runner.py

@@ -411,7 +411,7 @@ def ingest_sample(
         _require_current_ingest_bindings(model_bindings=build.model_bindings)
         if build.document_binding_generation != EXPECTED_DOCUMENT_BINDING_GENERATION:
             raise ExecutionGuardError(
-                "deployment document binding generation differs from RS-LoCoMo-Full-v20"
+                "deployment document binding generation differs from RS-LoCoMo-Full-v21"
             )
         _require_current_query_surface(context=context, client=client)
         _require_exact_live_ingests(
@@ -570,7 +570,7 @@ def answer_sample(
     ):
         raise ExecutionGuardError(
             "the deployment did not report the exact completed"
-            " RS-LoCoMo-Full-v20 pipeline, live graph, and fresh P3 projection"
+            " RS-LoCoMo-Full-v21 pipeline, live graph, and fresh P3 projection"
         )
     _require_serving_revision(context=context, readiness=readiness)
     prior_readiness = context.state.readiness.get(sample_id)
@@ -1177,7 +1177,7 @@ def _validate_run(
     """Recompute immutable run identity before any local or remote stage."""
     selected_protocol = protocol_for_name(configuration.protocol_name)
     if configuration.dataset_sha256 != DATASET_SHA256:
-        raise BenchmarkRunError("run dataset hash is not RS-LoCoMo-Full-v20")
+        raise BenchmarkRunError("run dataset hash is not RS-LoCoMo-Full-v21")
     if item_ids_hash(item_ids=manifest.item_ids) != manifest.item_ids_sha256:
         raise BenchmarkRunError("run manifest item hash changed")
     if manifest_bytes_hash(manifest=manifest) != configuration.manifest_sha256:
@@ -1187,7 +1187,7 @@ def _validate_run(
     if manifest.tier != configuration.tier:
         raise BenchmarkRunError("run manifest tier changed")
     if configuration.dataset_commit != DATASET_COMMIT:
-        raise BenchmarkRunError("run dataset commit is not RS-LoCoMo-Full-v20")
+        raise BenchmarkRunError("run dataset commit is not RS-LoCoMo-Full-v21")
     if configuration.adapter_version != ADAPTER_VERSION:
         raise BenchmarkRunError("run adapter version differs from current code")
     if _models_hash(values=documents) != configuration.documents_sha256:
@@ -1445,7 +1445,7 @@ def _require_current_ingest_bindings(*, model_bindings: dict[str, str]) -> None:
             if model_bindings.get(name) != expected.get(name)
         )
         raise ExecutionGuardError(
-            "deployment ingest model bindings differ from RS-LoCoMo-Full-v20: "
+            "deployment ingest model bindings differ from RS-LoCoMo-Full-v21: "
             + ", ".join(mismatches)
         )
 

benchmarks/locomo/sharding/run_shard.sh

@@ -25,7 +25,7 @@ run_dir=$2
 dataset_path=$3
 python_bin=${LOCOMO_PYTHON:-.venv/bin/python}
 tier=${LOCOMO_TIER:-publication}
-protocol=${LOCOMO_PROTOCOL:-full-v20}
+protocol=${LOCOMO_PROTOCOL:-full-v21}
 mount_root=${LOCOMO_MOUNT_ROOT:-$run_dir/.mounts}
 max_documents=${LOCOMO_MAX_DOCUMENTS:-100}
 max_questions=${LOCOMO_MAX_QUESTIONS:-1540}
@@ -50,28 +50,30 @@ export GOOGLE_APPLICATION_CREDENTIALS=${LOCOMO_GCP_CREDENTIALS_FILE:-/etc/rememb
 export GOOGLE_API_CERTIFICATE_CONFIG=${LOCOMO_GCP_CERTIFICATE_CONFIG_FILE:-/etc/rememberstack/locomo-gcs/certificate-config.json}
 export GOOGLE_API_USE_CLIENT_CERTIFICATE=true
 
-# RS-LoCoMo-Full-v20's non-secret ingest identity. Override ambient self-host
+# RS-LoCoMo-Full-v21's non-secret ingest identity. Override ambient self-host
 # defaults so every shard runs the exact Luna/Qwen pipeline the protocol checks.
-export REMEMBERSTACK_STRUCTURER_MODEL=openai/gpt-5.6-luna
-export REMEMBERSTACK_SKELETON_CHECK_MODEL=openai/gpt-5.6-luna
-export REMEMBERSTACK_ROLE_MODEL=openai/gpt-5.6-luna
-export REMEMBERSTACK_SUMMARY_MODEL=openai/gpt-5.6-luna
+export REMEMBERSTACK_STRUCTURER_MODEL=z-ai/glm-5.3-flash
+export REMEMBERSTACK_SKELETON_CHECK_MODEL=z-ai/glm-5.3-flash
+export REMEMBERSTACK_ROLE_MODEL=z-ai/glm-5.3-flash
+export REMEMBERSTACK_SUMMARY_MODEL=z-ai/glm-5.3-flash
 export REMEMBERSTACK_E1_EMBEDDING_MODEL=qwen/qwen3-embedding-8b
-export REMEMBERSTACK_E1_PREFIX_MODEL=openai/gpt-5.6-luna
-export REMEMBERSTACK_E2_EXTRACT_MODEL=openai/gpt-5.6-luna
-export REMEMBERSTACK_E3_NORMALIZE_MODEL=openai/gpt-5.6-luna
+export REMEMBERSTACK_E1_PREFIX_MODEL=z-ai/glm-5.3-flash
+export REMEMBERSTACK_E2_EXTRACT_MODEL=z-ai/glm-5.3-flash
+export REMEMBERSTACK_E3_NORMALIZE_MODEL=z-ai/glm-5.3-flash
 export REMEMBERSTACK_OBS_EMBEDDING_MODEL=qwen/qwen3-embedding-8b
-export REMEMBERSTACK_OBS_SMALL_MODEL=openai/gpt-5.6-luna
-export REMEMBERSTACK_OBS_FRONTIER_MODEL=openai/gpt-5.6-luna
-export REMEMBERSTACK_ADJUDICATOR_SMALL_MODEL=openai/gpt-5.6-luna
-export REMEMBERSTACK_ADJUDICATOR_FRONTIER_MODEL=openai/gpt-5.6-luna
+export REMEMBERSTACK_OBS_SMALL_MODEL=z-ai/glm-5.3-flash
+export REMEMBERSTACK_OBS_FRONTIER_MODEL=z-ai/glm-5.3-flash
+export REMEMBERSTACK_ADJUDICATOR_SMALL_MODEL=z-ai/glm-5.3-flash
+export REMEMBERSTACK_ADJUDICATOR_FRONTIER_MODEL=z-ai/glm-5.3-flash
 export REMEMBERSTACK_P1_EMBEDDING_MODEL=qwen/qwen3-embedding-8b
-export REMEMBERSTACK_P1_LABEL_MODEL=openai/gpt-5.6-luna
+export REMEMBERSTACK_P1_LABEL_MODEL=z-ai/glm-5.3-flash
 export REMEMBERSTACK_OPENROUTER_EMBEDDING_PROVIDER=nebius
 unset REMEMBERSTACK_OPENROUTER_EMBEDDING_PROVIDER_ORDER
 export REMEMBERSTACK_OPENROUTER_MAX_COMPLETION_TOKENS=32000
 unset REMEMBERSTACK_OPENROUTER_REASONING_EFFORT
-export REMEMBERSTACK_OPENROUTER_REASONING_EFFORT_MAP='{"openai/gpt-5.6-luna":"high"}'
+export REMEMBERSTACK_OPENROUTER_REASONING_EFFORT_MAP='{"z-ai/glm-5.3-flash":"high"}'
+export REMEMBERSTACK_OPENROUTER_CHAT_PROVIDER_ONLY=z-ai,novita,deepinfra,gmicloud
+export REMEMBERSTACK_OPENROUTER_CHAT_PROVIDER_SORT=throughput
 export REMEMBERSTACK_OPENROUTER_INVALID_COMPLETION_CAPTURE_DIR=/var/lib/rememberstack/invalid-completions
 
 [[ -x "$python_bin" ]] || die "Python is not executable: $python_bin"
@@ -139,6 +141,8 @@ attest_worker_environment() {
     REMEMBERSTACK_ADJUDICATOR_FRONTIER_MODEL
     REMEMBERSTACK_P1_EMBEDDING_MODEL
     REMEMBERSTACK_P1_LABEL_MODEL
+    REMEMBERSTACK_OPENROUTER_CHAT_PROVIDER_ONLY
+    REMEMBERSTACK_OPENROUTER_CHAT_PROVIDER_SORT
     REMEMBERSTACK_OPENROUTER_EMBEDDING_PROVIDER
     REMEMBERSTACK_OPENROUTER_EMBEDDING_PROVIDER_ORDER
     REMEMBERSTACK_OPENROUTER_MAX_COMPLETION_TOKENS

benchmarks/locomo/sharding/store_backup.py

@@ -48,6 +48,8 @@
     "fact_label": "REMEMBERSTACK_P1_LABEL_MODEL",
     "observation_frontier": "REMEMBERSTACK_OBS_FRONTIER_MODEL",
     "observation_small": "REMEMBERSTACK_OBS_SMALL_MODEL",
+    "openrouter_chat_provider_only": "REMEMBERSTACK_OPENROUTER_CHAT_PROVIDER_ONLY",
+    "openrouter_chat_provider_sort": "REMEMBERSTACK_OPENROUTER_CHAT_PROVIDER_SORT",
     "openrouter_embedding_provider": "REMEMBERSTACK_OPENROUTER_EMBEDDING_PROVIDER",
     "openrouter_embedding_provider_order": (
         "REMEMBERSTACK_OPENROUTER_EMBEDDING_PROVIDER_ORDER"
@@ -69,6 +71,8 @@
     "supersession_small": "REMEMBERSTACK_ADJUDICATOR_SMALL_MODEL",
 }
 UNSET_MODEL_BINDINGS = {
+    "openrouter_chat_provider_only": "unset",
+    "openrouter_chat_provider_sort": "default",
     "openrouter_embedding_provider": "auto",
     "openrouter_embedding_provider_order": "unset",
     "openrouter_max_completion_tokens": "unset",

compose.yaml

@@ -21,6 +21,8 @@ x-app: &app
     REMEMBERSTACK_OPENROUTER_MAX_COMPLETION_TOKENS: ${REMEMBERSTACK_OPENROUTER_MAX_COMPLETION_TOKENS:-}
     REMEMBERSTACK_OPENROUTER_REASONING_EFFORT: ${REMEMBERSTACK_OPENROUTER_REASONING_EFFORT:-}
     REMEMBERSTACK_OPENROUTER_REASONING_EFFORT_MAP: ${REMEMBERSTACK_OPENROUTER_REASONING_EFFORT_MAP:-}
+    REMEMBERSTACK_OPENROUTER_CHAT_PROVIDER_ONLY: ${REMEMBERSTACK_OPENROUTER_CHAT_PROVIDER_ONLY:-}
+    REMEMBERSTACK_OPENROUTER_CHAT_PROVIDER_SORT: ${REMEMBERSTACK_OPENROUTER_CHAT_PROVIDER_SORT:-}
     REMEMBERSTACK_OPENROUTER_INVALID_COMPLETION_CAPTURE_DIR: ${REMEMBERSTACK_OPENROUTER_INVALID_COMPLETION_CAPTURE_DIR:-}
     REMEMBERSTACK_SELFHOST_DEPLOYMENT_ID: ${REMEMBERSTACK_SELFHOST_DEPLOYMENT_ID}
     REMEMBERSTACK_SELFHOST_DEPLOYMENT_SLUG: ${REMEMBERSTACK_SELFHOST_DEPLOYMENT_SLUG}