@wendellresearch-ai/benchwarmers 0.1.1 → 0.2.1

package/suite/tasks/claim-error-text-string/fixtures/decide.py

@@ -1,187 +1,187 @@
-"""Ask a decision model a typed question: choose, score, or yes/no.
-
-    python metrics/decide.py mercury-decide-free "List every place serve.py writes a file." \\
-        --choice review,design,enumeration,code
-    python metrics/decide.py span-01-lite-free "Please refund the duplicate charge." \\
-        --noul "Is this about billing?"
-
-DECISION MODELS ARE NOT CHAT MODELS
------------------------------------
-Mercury Decide and Respan take a `state` and a map of typed `questions` and
-return each answer with a probability. They refuse /chat/completions outright
-("is a decisions model and cannot be used with the chat/completions endpoint"),
-so `llm -m` cannot reach them, the availability probe and the suite cannot
-test them, and they never go in the llm yaml. They live in
-`config/decision-models.json` instead, written by `scripts/onboard.py` when a
-model answers a typed question.
-
-WHERE THE CALL GOES
--------------------
-Each endpoint in `config/endpoints.json` may carry a `decisions` URL. Open
-WebUI has one only through its `/openai/{path}` passthrough, which forwards to
-`{connection base}/decisions` and answers only when the server runs with
-ENABLE_OPENAI_API_PASSTHROUGH=True. That reaches Inception
-(`api.inceptionlabs.ai/v1/decisions`) but never OpenRouter, whose Decisions API
-is `/api/alpha/decisions`, outside its `/api/v1` base.
-
-A model is called on the endpoint it was found on, recorded as `via`. Traffic
-goes through Open WebUI by design; the direct endpoints are for named special
-cases, so a different `via` happens only when asked for (`onboard.py --via`).
-
-WHAT EACH MODEL CAN ANSWER
---------------------------
-`question_types` lists the types a model accepted when onboarding probed it
-with one question of each: Mercury Decide takes noul, choice and score;
-Respan takes noul only, with plain-string instructions and criteria. `ask`
-refuses a type the entry does not list, so a caller learns it from the bench
-rather than from a 400. Both providers take the same body (`model`, `state`,
-`questions`) and return the same `answers`, so one client covers them.
-"""
-
-from __future__ import annotations
-
-import argparse
-import json
-import pathlib
-import sys
-import urllib.error
-import urllib.request
-
-ROOT = pathlib.Path(__file__).resolve().parent.parent
-FILE = ROOT / "config" / "decision-models.json"
-sys.path.insert(0, str(ROOT / "metrics"))
-import endpoints  # noqa: E402
-
-# One question of each type, plain strings throughout so Respan's rule holds.
-PROBES = {
-    "noul": {"type": "noul", "instructions": "Is this message about billing?",
-             "criteria": {"true": "It is about charges, payments or refunds.",
-                          "false": "It is about something else."}},
-    "choice": {"type": "choice", "instructions": "Which team should handle this message?",
-               "criteria": {"billing": "Charges, payments, refunds.",
-                            "support": "Product problems.", "sales": "Buying new things."}},
-    "score": {"type": "score", "instructions": "How urgent is this message?",
-              "criteria": ["Can wait", "Needs attention soon", "Needs immediate attention"]},
-}
-PROBE_STATE = "Please refund the duplicate charge on my account today."
-
-
-def load() -> dict[str, dict]:
-    try:
-        return json.loads(FILE.read_text(encoding="utf-8"))
-    except (OSError, ValueError):
-        return {}
-
-
-def call(endpoint: str, served: str, state, questions: dict, timeout: int = 60) -> dict:
-    """POST to an endpoint's decisions URL. Returns the parsed body; raises on HTTP errors."""
-    entry = endpoints.load()[endpoint]
-    url = entry.get("decisions")
-    if not url:
-        raise ValueError(f"{endpoint} has no decisions URL in config/endpoints.json")
-    headers = {"Content-Type": "application/json", **(entry.get("headers") or {})}
-    if (secret := endpoints.key(endpoint)):
-        headers["Authorization"] = f"Bearer {secret}"
-    body = json.dumps({"model": served, "state": state, "questions": questions}).encode()
-    request = urllib.request.Request(url, data=body, headers=headers, method="POST")
-    with urllib.request.urlopen(request, timeout=timeout) as response:
-        return json.load(response)
-
-
-def error_text(exc: Exception) -> str:
-    if isinstance(exc, urllib.error.HTTPError):
-        try:
-            detail = json.loads(exc.read() or b"{}")
-            detail = detail.get("error", detail).get("message", detail) if isinstance(detail, dict) else detail
-        except ValueError:
-            detail = ""
-        return f"{exc.code} {str(detail)[:160]}".strip()
-    return f"{type(exc).__name__}: {exc}"
-
-
-def probe(endpoint: str, served: str) -> str | None:
-    """None if `endpoint`'s decisions URL answers a yes/no question for
-    `served`; otherwise why not."""
-    if not endpoints.load()[endpoint].get("decisions"):
-        return f"{endpoint} has no decisions URL in config/endpoints.json"
-    try:
-        answer = call(endpoint, served, PROBE_STATE, {"probe": PROBES["noul"]})
-    except Exception as exc:                           # noqa: BLE001
-        return f"{endpoint}: {error_text(exc)}"
-    return None if "probe" in answer.get("answers", {}) else f"{endpoint}: no answer in {str(answer)[:120]}"
-
-
-def refuses_chat(endpoint: str, served: str) -> bool:
-    """True if a one-token chat call is refused because the model is a decision
-    model ("is a decisions model and cannot be used with the chat/completions
-    endpoint", as OpenRouter words it, relayed by Open WebUI)."""
-    entry = endpoints.load()[endpoint]
-    headers = {"Content-Type": "application/json", **(entry.get("headers") or {})}
-    if (secret := endpoints.key(endpoint)):
-        headers["Authorization"] = f"Bearer {secret}"
-    body = json.dumps({"model": served, "max_tokens": 1,
-                       "messages": [{"role": "user", "content": "ready?"}]}).encode()
-    request = urllib.request.Request(endpoints.base(endpoint) + "/chat/completions",
-                                     data=body, headers=headers, method="POST")
-    try:
-        with urllib.request.urlopen(request, timeout=60):
-            return False
-    except urllib.error.HTTPError as exc:
-        return "decisions model" in error_text(exc).lower()
-    except Exception:                                  # noqa: BLE001
-        return False
-
-
-def question_types(endpoint: str, served: str) -> list[str]:
-    """Which question types the model accepts, one probe call per type."""
-    ok = []
-    for kind, question in PROBES.items():
-        try:
-            if kind in call(endpoint, served, PROBE_STATE, {kind: question}).get("answers", {}):
-                ok.append(kind)
-        except Exception:                              # noqa: BLE001
-            pass
-    return ok
-
-
-def ask(alias: str, state, questions: dict) -> dict:
-    entry = load()[alias]
-    wrong = sorted({q["type"] for q in questions.values()} - set(entry["question_types"]))
-    if wrong:
-        raise ValueError(f"{alias} does not take {', '.join(wrong)} questions;"
-                         f" it takes {', '.join(entry['question_types'])}")
-    return call(entry["via"], entry["served"], state, questions)
-
-
-def main() -> None:
-    sys.stdout.reconfigure(encoding="utf-8", errors="replace")
-    ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
-    ap.add_argument("alias", help="a key of config/decision-models.json")
-    ap.add_argument("state", help="the text being judged")
-    group = ap.add_mutually_exclusive_group(required=True)
-    group.add_argument("--noul", metavar="QUESTION", help="a yes/no question")
-    group.add_argument("--choice", metavar="A,B,C", help="comma-separated options")
-    group.add_argument("--score", metavar="L0,L1,L2", help="comma-separated levels, lowest first")
-    ap.add_argument("--instructions", default="Answer the question about this text.")
-    args = ap.parse_args()
-    if args.alias not in load():
-        sys.exit(f"{args.alias} is not in {FILE.name}; onboard it first: scripts/onboard.py <served id>")
-    if args.noul:
-        q = {"type": "noul", "instructions": args.noul,
-             "criteria": {"true": "Yes.", "false": "No."}}
-    elif args.choice:
-        q = {"type": "choice", "instructions": args.instructions,
-             "criteria": {o.strip(): None for o in args.choice.split(",")}}
-    else:
-        q = {"type": "score", "instructions": args.instructions,
-             "criteria": [o.strip() for o in args.score.split(",")]}
-    try:
-        print(json.dumps(ask(args.alias, args.state, {"q": q})["answers"]["q"], indent=1))
-    except ValueError as exc:
-        sys.exit(str(exc))
-    except Exception as exc:                           # noqa: BLE001
-        sys.exit(error_text(exc))
-
-
-if __name__ == "__main__":
-    main()
+"""Ask a decision model a typed question: choose, score, or yes/no.
+
+    python metrics/decide.py mercury-decide-free "List every place serve.py writes a file." \\
+        --choice review,design,enumeration,code
+    python metrics/decide.py span-01-lite-free "Please refund the duplicate charge." \\
+        --noul "Is this about billing?"
+
+DECISION MODELS ARE NOT CHAT MODELS
+-----------------------------------
+Mercury Decide and Respan take a `state` and a map of typed `questions` and
+return each answer with a probability. They refuse /chat/completions outright
+("is a decisions model and cannot be used with the chat/completions endpoint"),
+so `llm -m` cannot reach them, the availability probe and the suite cannot
+test them, and they never go in the llm yaml. They live in
+`config/decision-models.json` instead, written by `scripts/onboard.py` when a
+model answers a typed question.
+
+WHERE THE CALL GOES
+-------------------
+Each endpoint in `config/endpoints.json` may carry a `decisions` URL. Open
+WebUI has one only through its `/openai/{path}` passthrough, which forwards to
+`{connection base}/decisions` and answers only when the server runs with
+ENABLE_OPENAI_API_PASSTHROUGH=True. That reaches Inception
+(`api.inceptionlabs.ai/v1/decisions`) but never OpenRouter, whose Decisions API
+is `/api/alpha/decisions`, outside its `/api/v1` base.
+
+A model is called on the endpoint it was found on, recorded as `via`. Traffic
+goes through Open WebUI by design; the direct endpoints are for named special
+cases, so a different `via` happens only when asked for (`onboard.py --via`).
+
+WHAT EACH MODEL CAN ANSWER
+--------------------------
+`question_types` lists the types a model accepted when onboarding probed it
+with one question of each: Mercury Decide takes noul, choice and score;
+Respan takes noul only, with plain-string instructions and criteria. `ask`
+refuses a type the entry does not list, so a caller learns it from the bench
+rather than from a 400. Both providers take the same body (`model`, `state`,
+`questions`) and return the same `answers`, so one client covers them.
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+import pathlib
+import sys
+import urllib.error
+import urllib.request
+
+ROOT = pathlib.Path(__file__).resolve().parent.parent
+FILE = ROOT / "config" / "decision-models.json"
+sys.path.insert(0, str(ROOT / "metrics"))
+import endpoints  # noqa: E402
+
+# One question of each type, plain strings throughout so Respan's rule holds.
+PROBES = {
+    "noul": {"type": "noul", "instructions": "Is this message about billing?",
+             "criteria": {"true": "It is about charges, payments or refunds.",
+                          "false": "It is about something else."}},
+    "choice": {"type": "choice", "instructions": "Which team should handle this message?",
+               "criteria": {"billing": "Charges, payments, refunds.",
+                            "support": "Product problems.", "sales": "Buying new things."}},
+    "score": {"type": "score", "instructions": "How urgent is this message?",
+              "criteria": ["Can wait", "Needs attention soon", "Needs immediate attention"]},
+}
+PROBE_STATE = "Please refund the duplicate charge on my account today."
+
+
+def load() -> dict[str, dict]:
+    try:
+        return json.loads(FILE.read_text(encoding="utf-8"))
+    except (OSError, ValueError):
+        return {}
+
+
+def call(endpoint: str, served: str, state, questions: dict, timeout: int = 60) -> dict:
+    """POST to an endpoint's decisions URL. Returns the parsed body; raises on HTTP errors."""
+    entry = endpoints.load()[endpoint]
+    url = entry.get("decisions")
+    if not url:
+        raise ValueError(f"{endpoint} has no decisions URL in config/endpoints.json")
+    headers = {"Content-Type": "application/json", **(entry.get("headers") or {})}
+    if (secret := endpoints.key(endpoint)):
+        headers["Authorization"] = f"Bearer {secret}"
+    body = json.dumps({"model": served, "state": state, "questions": questions}).encode()
+    request = urllib.request.Request(url, data=body, headers=headers, method="POST")
+    with urllib.request.urlopen(request, timeout=timeout) as response:
+        return json.load(response)
+
+
+def error_text(exc: Exception) -> str:
+    if isinstance(exc, urllib.error.HTTPError):
+        try:
+            detail = json.loads(exc.read() or b"{}")
+            detail = detail.get("error", detail).get("message", detail) if isinstance(detail, dict) else detail
+        except ValueError:
+            detail = ""
+        return f"{exc.code} {str(detail)[:160]}".strip()
+    return f"{type(exc).__name__}: {exc}"
+
+
+def probe(endpoint: str, served: str) -> str | None:
+    """None if `endpoint`'s decisions URL answers a yes/no question for
+    `served`; otherwise why not."""
+    if not endpoints.load()[endpoint].get("decisions"):
+        return f"{endpoint} has no decisions URL in config/endpoints.json"
+    try:
+        answer = call(endpoint, served, PROBE_STATE, {"probe": PROBES["noul"]})
+    except Exception as exc:                           # noqa: BLE001
+        return f"{endpoint}: {error_text(exc)}"
+    return None if "probe" in answer.get("answers", {}) else f"{endpoint}: no answer in {str(answer)[:120]}"
+
+
+def refuses_chat(endpoint: str, served: str) -> bool:
+    """True if a one-token chat call is refused because the model is a decision
+    model ("is a decisions model and cannot be used with the chat/completions
+    endpoint", as OpenRouter words it, relayed by Open WebUI)."""
+    entry = endpoints.load()[endpoint]
+    headers = {"Content-Type": "application/json", **(entry.get("headers") or {})}
+    if (secret := endpoints.key(endpoint)):
+        headers["Authorization"] = f"Bearer {secret}"
+    body = json.dumps({"model": served, "max_tokens": 1,
+                       "messages": [{"role": "user", "content": "ready?"}]}).encode()
+    request = urllib.request.Request(endpoints.base(endpoint) + "/chat/completions",
+                                     data=body, headers=headers, method="POST")
+    try:
+        with urllib.request.urlopen(request, timeout=60):
+            return False
+    except urllib.error.HTTPError as exc:
+        return "decisions model" in error_text(exc).lower()
+    except Exception:                                  # noqa: BLE001
+        return False
+
+
+def question_types(endpoint: str, served: str) -> list[str]:
+    """Which question types the model accepts, one probe call per type."""
+    ok = []
+    for kind, question in PROBES.items():
+        try:
+            if kind in call(endpoint, served, PROBE_STATE, {kind: question}).get("answers", {}):
+                ok.append(kind)
+        except Exception:                              # noqa: BLE001
+            pass
+    return ok
+
+
+def ask(alias: str, state, questions: dict) -> dict:
+    entry = load()[alias]
+    wrong = sorted({q["type"] for q in questions.values()} - set(entry["question_types"]))
+    if wrong:
+        raise ValueError(f"{alias} does not take {', '.join(wrong)} questions;"
+                         f" it takes {', '.join(entry['question_types'])}")
+    return call(entry["via"], entry["served"], state, questions)
+
+
+def main() -> None:
+    sys.stdout.reconfigure(encoding="utf-8", errors="replace")
+    ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
+    ap.add_argument("alias", help="a key of config/decision-models.json")
+    ap.add_argument("state", help="the text being judged")
+    group = ap.add_mutually_exclusive_group(required=True)
+    group.add_argument("--noul", metavar="QUESTION", help="a yes/no question")
+    group.add_argument("--choice", metavar="A,B,C", help="comma-separated options")
+    group.add_argument("--score", metavar="L0,L1,L2", help="comma-separated levels, lowest first")
+    ap.add_argument("--instructions", default="Answer the question about this text.")
+    args = ap.parse_args()
+    if args.alias not in load():
+        sys.exit(f"{args.alias} is not in {FILE.name}; onboard it first: scripts/onboard.py <served id>")
+    if args.noul:
+        q = {"type": "noul", "instructions": args.noul,
+             "criteria": {"true": "Yes.", "false": "No."}}
+    elif args.choice:
+        q = {"type": "choice", "instructions": args.instructions,
+             "criteria": {o.strip(): None for o in args.choice.split(",")}}
+    else:
+        q = {"type": "score", "instructions": args.instructions,
+             "criteria": [o.strip() for o in args.score.split(",")]}
+    try:
+        print(json.dumps(ask(args.alias, args.state, {"q": q})["answers"]["q"], indent=1))
+    except ValueError as exc:
+        sys.exit(str(exc))
+    except Exception as exc:                           # noqa: BLE001
+        sys.exit(error_text(exc))
+
+
+if __name__ == "__main__":
+    main()