Migrate DeepSeek routing to V4.1 Flash

CLAUDE.md

@@ -187,8 +187,8 @@ SESSION_SECRET          # Express session encryption (no fallback in prod)
 INTERNAL_API_SECRET     # Express→Python shared secret (start-dev.sh generates a per-run value)
 DATABASE_URL            # PostgreSQL connection string
 XAI_API_KEY             # Grok energy provider
-DEEPSEEK_API_KEY        # DeepSeek V4 Flash/Pro; also used by standalone a0
-A0_PROVIDER             # Optional standalone provider id (e.g. deepseek-pro)
+DEEPSEEK_API_KEY        # DeepSeek V4.1 Flash; also used by standalone a0
+A0_PROVIDER             # Optional standalone provider id; deepseek is canonical
 STRIPE_SECRET_KEY       # Stripe billing
 STRIPE_WEBHOOK_SECRET   # Stripe webhook validation
 ADMIN_USER_ID           # User ID allowed to write prompt contexts

README.md

@@ -8,7 +8,7 @@
 
 ## The Agent
 
-ZFAE (`a0(zeta fun alpha echo)`) is the single persistent agent running on the instrument. Large language models (GPT-5 mini, Gemini 2.5 Flash, Claude Sonnet 4.5, Grok 4 Fast, DeepSeek V4) are treated as **energy providers** — they supply computational energy for each response but are not the agent itself.
+ZFAE (`a0(zeta fun alpha echo)`) is the single persistent agent running on the instrument. Large language models (GPT-5 mini, Gemini 2.5 Flash, Claude Sonnet 4.5, Grok 4 Fast, DeepSeek V4.1 Flash) are treated as **energy providers** — they supply computational energy for each response but are not the agent itself.
 
 Sub-agents (`a0(zeta{n})`) can be spawned to fork the PCNA instance, execute in parallel, and merge results back into the primary agent.
 
@@ -90,14 +90,17 @@ In development `scripts/start-dev.sh` generates a shared `INTERNAL_API_SECRET` a
 
 The standalone `run.sh` runtime reads the same `python/config/providers.json`
 registry as the FastAPI service. With a DeepSeek key present it auto-selects
-V4 Flash through the generic `openai-compatible` adapter. Pin V4 Pro explicitly:
+V4.1 Flash through the generic `openai-compatible` adapter:
 
 ```bash
 export DEEPSEEK_API_KEY="..."
-export A0_PROVIDER=deepseek-pro   # omit for deepseek-v4-flash
 printf '%s\n' '{"task_id":"ds1","input":{"text":"hello"},"tools_allowed":["none"],"mode":"analyze","hmmm":["direct model turn"]}' | ./run.sh
 ```
 
+The canonical API model is `deepseek-flash`. The retired provider choice
+`A0_PROVIDER=deepseek-pro` remains a hidden compatibility alias and resolves to
+the same V4.1 Flash route; it no longer represents a higher-capability tier.
+
 Provider configuration contains only the credential environment-variable name;
 keys remain in the process environment and are never written to source.
 
@@ -133,7 +136,7 @@ npx playwright test
 | `ANTHROPIC_API_KEY` | Claude Sonnet 4.5 |
 | `GEMINI_API_KEY` | Gemini 2.5 Flash |
 | `OPENAI_API_KEY` | GPT-5 mini |
-| `DEEPSEEK_API_KEY` | DeepSeek V4 Flash/Pro through the OpenAI-compatible adapter |
+| `DEEPSEEK_API_KEY` | DeepSeek V4.1 Flash through the OpenAI-compatible adapter |
 | `STRIPE_SECRET_KEY` | Stripe (donations + EDCMbone explainer) |
 | `STRIPE_PUBLISHABLE_KEY` | Stripe embedded checkout |
 | `STRIPE_WEBHOOK_SECRET` | Stripe webhook HMAC |

a0/adapters/open_comp_adap_v0.0.0alpha.py

@@ -1,4 +1,4 @@
-# 100:47 0:0 2:0
+# 102:53 0:0 2:0
 """Synchronous standalone adapter for registry-defined OpenAI-compatible APIs."""
 from __future__ import annotations
 
@@ -47,6 +47,12 @@
 #   then: the adapter raises the effective effort to that configured floor before transport
 #   class: correctness
 #   since: 2026-09-10
+#
+# id: a0_openai_compatible_explicit_none_reasoning
+#   given: a compatible provider declares that reasoning effort none must be sent explicitly
+#   then: standalone Responses calls carry reasoning.effort=none rather than falling back to the provider default
+#   class: correctness
+#   since: 2026-09-10
 # === END CONTRACTS ===
 
 import os
@@ -130,7 +136,9 @@ def complete(self, messages: List[Message], **kwargs: Any) -> Dict[str, Any]:
                 )
                 if supports_store:
                     request["store"] = bool(kwargs.get("store", False))
-                if effort and effort != "none":
+                if effort and (
+                    effort != "none" or self.spec.get("explicit_none_reasoning")
+                ):
                     request["reasoning"] = {"effort": effort}
                 response = self._client.responses.create(**request)
                 text = _response_text(response)
@@ -162,4 +170,4 @@ def complete(self, messages: List[Message], **kwargs: Any) -> Dict[str, Any]:
             },
             "subagents_used": [],
         }
-# 100:47 0:0 2:0
+# 102:53 0:0 2:0

a0/prov_regi_v0.0.0alpha.py

@@ -1,4 +1,4 @@
-# 36:29 0:0 1:0
+# 55:30 0:0 1:0
 """Read standalone a0 model adapters from the canonical provider registry."""
 from __future__ import annotations
 
@@ -26,7 +26,7 @@
 # === CONTRACTS ===
 # id: a0_provider_selection
 #   given: optional A0_PROVIDER, canonical provider registry data, and process credentials
-#   then: an explicit compatible provider resolves or fails closed while an unset choice may auto-select only a configured a0_default provider
+#   then: an explicit compatible provider resolves or fails closed, deprecated provider aliases resolve to their canonical provider, and an unset choice may auto-select only a configured a0_default provider
 #   class: correctness
 #   since: 2026-09-07
 # === END CONTRACTS ===
@@ -49,6 +49,27 @@ def _is_openai_compatible(spec: dict[str, Any]) -> bool:
     return spec.get("adapter") == "openai-compatible" or spec.get("vendor") == "openai"
 
 
+def _resolve_alias(
+    providers: dict[str, dict[str, Any]],
+    provider_id: str,
+) -> tuple[str, dict[str, Any]]:
+    """Resolve one registry-declared provider alias without provider literals."""
+    spec = providers[provider_id]
+    canonical_id = str(spec.get("deprecated_alias_for") or "").strip()
+    if not canonical_id:
+        return provider_id, spec
+    canonical = providers.get(canonical_id)
+    if canonical is None:
+        raise ValueError(
+            f"A0_PROVIDER {provider_id!r} aliases missing provider {canonical_id!r}"
+        )
+    if canonical.get("deprecated_alias_for"):
+        raise ValueError(
+            f"A0_PROVIDER {provider_id!r} has a chained provider alias"
+        )
+    return canonical_id, canonical
+
+
 def resolve_openai_compatible_provider(
     provider_id: str | None = None,
 ) -> tuple[str, dict[str, Any]] | None:
@@ -59,11 +80,12 @@ def resolve_openai_compatible_provider(
         spec = providers.get(explicit)
         if spec is None:
             raise ValueError(f"Unknown A0_PROVIDER: {explicit!r}")
+        canonical_id, spec = _resolve_alias(providers, explicit)
         if not _is_openai_compatible(spec):
             raise ValueError(
                 f"A0_PROVIDER {explicit!r} does not use the openai-compatible adapter"
             )
-        return explicit, spec
+        return canonical_id, spec
 
     for candidate, spec in providers.items():
         api_key_env = str(spec.get("api_key_env") or "")
@@ -75,4 +97,4 @@ def resolve_openai_compatible_provider(
         ):
             return candidate, spec
     return None
-# 36:29 0:0 1:0
+# 55:30 0:0 1:0

python/config/pricing.json

@@ -1,8 +1,8 @@
 {
-  "version": "2026-09-07.a0.pricing.v2",
-  "prices_as_of": "2026-09-07",
+  "version": "2026-09-10.a0.pricing.v3",
+  "prices_as_of": "2026-09-10",
   "_doc": {
-    "schema": "providers.<id>.models is a list of {id, context_window, input_per_1m, output_per_1m, cached_input_per_1m?, cache_write_per_1m?, supports_vision?, supports_thinking?, note?} entries. All prices are USD per 1,000,000 tokens.",
+    "schema": "providers.<id>.models is a list of {id, context_window, max_output_tokens?, input_per_1m, output_per_1m, cached_input_per_1m?, cache_write_per_1m?, supports_vision?, supports_thinking?, note?} entries. All prices are USD per 1,000,000 tokens.",
     "source-of-truth": "Boot source-of-truth for per-model pricing. estimate_cost(provider_id, ..., model=<id>) consults this file first and uses per-model rates; without a model arg, falls back to the provider flagship rate from providers.json. Hydrates each ws_modules.route_config.available_models list on first boot and on POST /api/energy/refresh-pricing/{provider_id}.",
     "naming": "Model ids are the slugs the provider's API accepts. Keep in sync with providers.json presets section so optimizer presets and per-model pricing line up.",
     "live-refresh": "POST /api/energy/refresh-pricing/{provider_id} re-reads this file from disk and re-hydrates the seed's available_models. Live HTML extraction from pricing_url is a future phase; for now admins update this file in repo, redeploy, then refresh."
@@ -187,27 +187,31 @@
       "pricing_url": "https://api-docs.deepseek.com/quick_start/pricing/",
       "models": [
         {
-          "id": "deepseek-v4-flash",
+          "id": "deepseek-flash",
           "context_window": 1000000,
-          "input_per_1m": 0.44,
-          "output_per_1m": 1.32,
-          "cached_input_per_1m": 0.014,
+          "max_output_tokens": 384000,
+          "input_per_1m": 0.3,
+          "output_per_1m": 1.2,
+          "cached_input_per_1m": 0.006,
           "supports_thinking": true,
-          "note": "Official peak rates; off-peak rates are lower."
+          "supports_vision": true,
+          "note": "DeepSeek V4.1 Flash official peak rates; off-peak rates are half."
         }
       ]
     },
     "deepseek-pro": {
       "pricing_url": "https://api-docs.deepseek.com/quick_start/pricing/",
       "models": [
         {
-          "id": "deepseek-v4-pro",
+          "id": "deepseek-flash",
           "context_window": 1000000,
-          "input_per_1m": 1.32,
-          "output_per_1m": 3.96,
-          "cached_input_per_1m": 0.044,
+          "max_output_tokens": 384000,
+          "input_per_1m": 0.3,
+          "output_per_1m": 1.2,
+          "cached_input_per_1m": 0.006,
           "supports_thinking": true,
-          "note": "Official peak rates; off-peak rates are lower."
+          "supports_vision": true,
+          "note": "Compatibility pricing for the retired deepseek-pro provider id; routes to DeepSeek V4.1 Flash."
         }
       ]
     }