feat(learn): the classifier program and its response cache (decision-engine 6.4)

learn/program.py: the declared signature (code, signals, roles, family in;
verdict, family, confidence, rationale with cited lines out), the
baseline instruction, rendering as [system: instruction + demonstrations]
-- byte-identical for every candidate, so the provider's prefix cache can
hit -- then [user: retrieved examples with their labels, the candidate
without one]. Signals pass validate_x first (no EXCLUDED_FIELDS, missing
instruments in words). Answers are one JSON object; one that does not
parse or cites no line inside the excerpt is asked once more at
temperature 0 and otherwise becomes unsure/parse_failed, never
not_vulnerable. Classify(k): sample 1 at 0.0, the rest at 0.7; score s =
mean p_j; majority verdict, ties unsure; a majority unsure cannot BLOCK.
Caller caches each response as responses/<sha>.json keyed by (model,
Model parameters digest, messages, temperature, sample, json_object) and
replays at $0; spend is priced from the recorded usage once per distinct
request, so a replay reports the same usd. Program.decide asserts the
evaluation boundary on the rendered ids before any call.

MeteredClient.complete passes temperature/logprobs through (only when
set); DeepSeekChatClient sends temperature beside thinking=disabled.

Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_01EmM1j46qgKMh3zqD1QMwEo

src/openultrasast/model/audit.py

@@ -73,9 +73,9 @@ SPEC_OWNED = {
         "learned-decision-engine Req 8.1: model role classification per file chunk (model_sinks.py moved in); "
         "its Run template comes with the harvest (task 5)"
     ),
-    "learn.retrieve": (
-        "learned-decision-engine Req 3.2/3.4: retrieval from the local memory and the evaluation boundary; "
-        "the program (task 6.4) and `ousast learn compile` (6.5) are its callers"
+    "learn.program": (
+        "learned-decision-engine Req 3.1/3.3: the AI classifier program (signature, Retrieve -> Classify(k), response "
+        "cache); `ousast learn compile` (task 6.5) and the decide task (task 8) are its callers"
     ),
     "plane.workspaces": "ai-service-plane task 4: Workspace manifests per case pin of a population; `ousast plane workspaces`",
     "plane.runner": "ai-service-plane Req 4.1-4.4: the ax-task-runner entrypoint (PID 1 of the task image); delivers the output directory",

src/openultrasast/model/endpoint.py

@@ -106,13 +106,26 @@ class DeepSeekChatClient:
         timeout_seconds: int = 60,
         json_object: bool = False,
         logprobs: bool = False,
+        temperature: float | None = None,
     ) -> ChatResponse:
         response = self._call(
-            model=model, messages=messages, tools=tools, timeout_seconds=timeout_seconds, json_object=json_object, logprobs=logprobs
+            model=model,
+            messages=messages,
+            tools=tools,
+            timeout_seconds=timeout_seconds,
+            json_object=json_object,
+            logprobs=logprobs,
+            temperature=temperature,
         )
         if json_object and not (response.content or "").strip():
             response = self._call(
-                model=model, messages=messages, tools=tools, timeout_seconds=timeout_seconds, json_object=json_object, logprobs=logprobs
+                model=model,
+                messages=messages,
+                tools=tools,
+                timeout_seconds=timeout_seconds,
+                json_object=json_object,
+                logprobs=logprobs,
+                temperature=temperature,
             )  # documented behaviour: JSON mode occasionally returns empty content. One retry, never a loop.
         return response
 
@@ -131,10 +144,13 @@ class DeepSeekChatClient:
         timeout_seconds: int,
         json_object: bool,
         logprobs: bool,
+        temperature: float | None = None,
     ) -> ChatResponse:
         extra: dict[str, object] = {}
         if self._disable_thinking:
             extra["thinking"] = {"type": "disabled"}  # otherwise temperature is silently ignored
+        if temperature is not None:
+            extra["temperature"] = temperature  # overrides the raw call's default 0
         if json_object:
             extra["response_format"] = {"type": "json_object"}
         if logprobs: