Implement 0080 PromptGroup arity enforcement

CHANGELOG.md

@@ -9,6 +9,7 @@ The format follows [Keep a Changelog](https://keepachangelog.com/en/1.1.0/). The
 ### Added
 
 - **PromptManager service-wide default cache TTL** (proposal 0086, prompt-management §6, spec v0.79.0). `PromptManager` construction accepts an optional `default_cache_ttl_seconds`, applied to any `fetch` or `get` that omits a per-call `cache_ttl_seconds`. Resolution follows a precedence chain: an explicit per-call value (including `0` force-fresh) wins; otherwise the manager default applies; otherwise nothing is forwarded and the backend's own caching governs. An omitted or explicit-`None` per-call value both select the default, so resolution does not depend on argument presence. A negative default is rejected at construction, and the per-call negative rejection is unchanged. `get` delegates to `fetch` and inherits the chain, and the bundled caching backends need no change (they already honor a resolved `cache_ttl_seconds`). Conformance fixture 036 is un-deferred.
+- **PromptGroup arity enforcement** (proposal 0080, prompt-management §10 / §11, spec v0.75.0). Constructing a `PromptGroup` with fewer than two members (an empty or single-member group) now raises a categorized `PromptGroupInvalid` at construction, before any render or LLM call. This replaces the previous bare `ValueError` (which pydantic folded into a `ValidationError` carrying no error category) and adds a `prompt_group_invalid` category to the prompt-management error set. `PromptGroupInvalid` is non-transient and is exported from `openarmature.prompts`. Conformance fixture 035 is un-deferred.
 
 ## [0.16.0] — 2026-07-18
 

conformance.toml

@@ -795,10 +795,12 @@ note = "Retrieval-provider OpenAI-compatible embeddings wire mapping (§8 / §8.
 
 # Spec v0.75.0 (proposal 0080).  PromptGroup arity enforcement
 # (prompt-management §10 / §11 -- construct-time raise plus the new
-# prompt_group_invalid error category).  Not-yet; prompt-management
-# fixture 035 defers with it.
+# prompt_group_invalid error category).  Implemented since 0.17.0;
+# prompt-management fixture 035 pins it.
 [proposals."0080"]
-status = "not-yet"
+status = "implemented"
+since = "0.17.0"
+note = "PromptGroup already rejected fewer than two members, but as a bare ValueError (pydantic folds it into a ValidationError, which has no category). Now raises a categorized PromptGroupInvalid (new category prompt_group_invalid, prompt-management §11) at construction: PromptGroupInvalid is a PromptError (not a ValueError), so pydantic propagates it unwrapped from the model validator and the caller sees .category. Non-transient, not in PROMPT_TRANSIENT_CATEGORIES; exported from openarmature.prompts. Fixture 035 (single-member + empty group, each asserting raises.category = prompt_group_invalid) is un-deferred. Harness: PromptManagementFixture.backends is now optional (035 is cases-only with per-case backends, one case needing none) and the cases-runner builds per-case backends when a case declares its own backends key (else shares the top-level, preserving 016) and runs a manager-less case's direct-target calls instead of skipping them (035's empty-group construct_prompt_group needs no manager)."
 
 # Spec v0.76.0 (proposal 0081).  Conformance-adapter value-matcher
 # vocabulary (§5.10) -- ratifies the fixture matcher tokens already in

docs/concepts/prompts.md

@@ -329,10 +329,11 @@ Canonical patterns the primitive covers:
 - **Map-reduce over chunks**: `[chunk_classify_1..N, synthesize]`.
 
 The N=2 case ("classifier + follow-up") is the simplest;
-larger groups work under the same primitive. The group rejects
-empty and single-member shapes; single-prompt tagging is
-already served by the per-prompt observability attributes
-below.
+larger groups work under the same primitive. Constructing a
+group with fewer than two members raises `PromptGroupInvalid`
+at construction time, before any render or call; single-prompt
+tagging is already served by the per-prompt observability
+attributes below.
 
 ## Observability propagation
 

src/openarmature/prompts/__init__.py

@@ -9,11 +9,13 @@
     with_active_prompt_group,
 )
 from .errors import (
+    PROMPT_GROUP_INVALID,
     PROMPT_NOT_FOUND,
     PROMPT_RENDER_ERROR,
     PROMPT_STORE_UNAVAILABLE,
     PROMPT_TRANSIENT_CATEGORIES,
     PromptError,
+    PromptGroupInvalid,
     PromptNotFound,
     PromptRenderError,
     PromptStoreUnavailable,
@@ -38,6 +40,7 @@
 )
 
 __all__ = [
+    "PROMPT_GROUP_INVALID",
     "PROMPT_NOT_FOUND",
     "PROMPT_RENDER_ERROR",
     "PROMPT_STORE_UNAVAILABLE",
@@ -57,6 +60,7 @@
     "PromptBackend",
     "PromptError",
     "PromptGroup",
+    "PromptGroupInvalid",
     "PromptManager",
     "PromptNotFound",
     "PromptRenderError",

src/openarmature/prompts/errors.py

@@ -7,6 +7,7 @@
 PROMPT_NOT_FOUND = "prompt_not_found"
 PROMPT_RENDER_ERROR = "prompt_render_error"
 PROMPT_STORE_UNAVAILABLE = "prompt_store_unavailable"
+PROMPT_GROUP_INVALID = "prompt_group_invalid"
 
 # Mirrors openarmature.llm.errors.TRANSIENT_CATEGORIES. Retry-middleware
 # classifiers MAY import this to identify transient prompt-management
@@ -123,3 +124,17 @@ def __init__(
         self.label = label
         self.backends_tried = backends_tried
         self.causes = causes
+
+
+class PromptGroupInvalid(PromptError):
+    """Raised when a ``PromptGroup`` construction violates a
+    group-validity rule. Currently raised when ``members`` contains
+    fewer than two elements (an empty or single-member group). Raised
+    at construction time, before any render or LLM call, so an invalid
+    group never reaches observability emission.
+
+    Non-transient: a caller-contract violation. Constructing again with
+    the same members will not succeed without changing them.
+    """
+
+    category = PROMPT_GROUP_INVALID

src/openarmature/prompts/group.py

@@ -4,6 +4,7 @@
 
 from pydantic import BaseModel, ConfigDict, model_validator
 
+from .errors import PromptGroupInvalid
 from .prompt import PromptResult
 
 
@@ -31,6 +32,10 @@ class PromptGroup(BaseModel):
 
     @model_validator(mode="after")
     def _check_min_two_members(self) -> PromptGroup:
+        # PromptGroupInvalid is not a ValueError, so pydantic propagates it
+        # unwrapped from the validator (only ValueError / AssertionError get
+        # folded into a ValidationError) -- the categorized error reaches the
+        # caller with its ``category``.
         if len(self.members) < 2:
-            raise ValueError("prompt group: members MUST contain at least two PromptResult instances")
+            raise PromptGroupInvalid("prompt group: members MUST contain at least two PromptResult instances")
         return self

tests/conformance/harness/prompt_management.py

@@ -235,7 +235,10 @@ class FixtureExpectedTopLevel(_PermissiveModel):
 
 
 class PromptManagementFixture(_StrictModel):
-    backends: list[FixtureBackendSpec]
+    # Optional: fixture 035 is cases-only with per-case backends (and an
+    # empty-group case that needs no backend at all), so the top level
+    # declares none (proposal 0080).
+    backends: list[FixtureBackendSpec] = []
     # Fixture 016 uses a top-level ``cases:`` list to split into
     # independent sub-cases that share the backends declaration but
     # each have their own manager + calls. The runner walks the list

tests/conformance/test_prompt_management.py

@@ -620,12 +620,6 @@ def _message_to_dict_for_compare(message: Message) -> dict[str, Any]:
     "032-cross-variable-substring-stability": (
         "Proposal 0047 wire-byte stability (expected_shared_prefix directive); queued for v0.13.0"
     ),
-    # ----- v0.16.0 spec-pin bump (v0.70.1 -> v0.84.0) -------------------
-    # Proposal 0080 (PromptGroup arity enforcement, spec v0.75.0) -- fixture
-    # 035 uses a cases-only shape (no backends) the PM fixture model doesn't
-    # accept, and asserts the construct-time prompt_group_invalid raise that
-    # python does not yet implement. Defers until a later v0.16.0 PR.
-    "035-prompt-group-arity-rejection": ("Proposal 0080 PromptGroup arity enforcement; not implemented"),
 }
 
 
@@ -671,8 +665,11 @@ async def test_prompt_management_fixture(fixture_path: Path) -> None:
             if call.capture_as is not None and raised is None:
                 captures[call.capture_as] = result
 
-    # Cases-form fixtures (016) split into independent sub-cases that
-    # share the backends but use their own per-case manager + calls.
+    # Cases-form fixtures split into independent sub-cases sharing the
+    # captures dict. Fixture 016 shares the top-level backends across cases;
+    # fixture 035 (proposal 0080) declares backends per case (one case needs
+    # none), so a case that brings its own ``backends`` key gets fresh
+    # backends and any other shares the top-level set.
     cases = raw.get("cases")
     if cases:
         for case in cases:
@@ -684,18 +681,32 @@ async def test_prompt_management_fixture(fixture_path: Path) -> None:
                 **{k: v for k, v in case.items() if k not in {"name", "description"}},
             }
             case_fixture = PromptManagementFixture.model_validate(case_payload)
+            # A case that declares its own ``backends`` REPLACES the top-level
+            # set (not a union) -- the payload merge already gives case keys
+            # override semantics. No current fixture needs a case to add to the
+            # shared backends; if one ever does, merge here instead.
+            case_backends = (
+                {spec.name: MockPromptBackend(spec) for spec in case_fixture.backends}
+                if "backends" in case
+                else backends
+            )
             case_manager_pairs = [
                 (case_fixture.manager, case_fixture.calls),
                 (case_fixture.secondary_manager, case_fixture.secondary_calls),
                 (case_fixture.tertiary_manager, case_fixture.tertiary_calls),
             ]
             for manager_spec, manager_calls in case_manager_pairs:
-                if manager_spec is None:
-                    continue
-                manager = _build_manager(manager_spec, backends, resolvers_map)
+                # Build a manager only when one is declared; a manager-less
+                # case still runs its direct-target calls (035's empty-group
+                # construct_prompt_group needs no manager).
+                manager = (
+                    _build_manager(manager_spec, case_backends, resolvers_map)
+                    if manager_spec is not None
+                    else None
+                )
                 for call in manager_calls:
-                    result, raised = await _run_call(call, backends, manager, captures)
-                    _assert_per_call(call, result, raised, backends)
+                    result, raised = await _run_call(call, case_backends, manager, captures)
+                    _assert_per_call(call, result, raised, case_backends)
                     if call.capture_as is not None and raised is None:
                         captures[call.capture_as] = result
             if case_fixture.expected is not None:

tests/unit/test_prompts.py

@@ -20,6 +20,7 @@
 
 from openarmature.llm.messages import Message, UserMessage
 from openarmature.prompts import (
+    PROMPT_GROUP_INVALID,
     PROMPT_NOT_FOUND,
     PROMPT_RENDER_ERROR,
     PROMPT_STORE_UNAVAILABLE,
@@ -28,6 +29,7 @@
     Prompt,
     PromptError,
     PromptGroup,
+    PromptGroupInvalid,
     PromptManager,
     PromptNotFound,
     PromptRenderError,
@@ -51,9 +53,11 @@ def test_error_categories_match_spec() -> None:
     assert PromptNotFound.category == "prompt_not_found"
     assert PromptRenderError.category == "prompt_render_error"
     assert PromptStoreUnavailable.category == "prompt_store_unavailable"
+    assert PromptGroupInvalid.category == "prompt_group_invalid"
     assert PROMPT_NOT_FOUND == "prompt_not_found"
     assert PROMPT_RENDER_ERROR == "prompt_render_error"
     assert PROMPT_STORE_UNAVAILABLE == "prompt_store_unavailable"
+    assert PROMPT_GROUP_INVALID == "prompt_group_invalid"
 
 
 def test_transient_categories_contains_only_store_unavailable() -> None:
@@ -167,8 +171,13 @@ def test_prompt_result_rejects_empty_messages() -> None:
 
 
 def test_prompt_group_rejects_zero_members() -> None:
-    with pytest.raises(ValueError, match="at least two"):
+    # Categorized prompt_group_invalid (proposal 0080), not a bare ValueError:
+    # PromptGroupInvalid is not a ValueError, so pydantic propagates it
+    # unwrapped from the model validator rather than folding it into a
+    # ValidationError.
+    with pytest.raises(PromptGroupInvalid, match="at least two") as exc_info:
         PromptGroup(group_name="g", members=[])
+    assert exc_info.value.category == "prompt_group_invalid"
 
 
 def test_prompt_group_rejects_one_member() -> None:
@@ -184,8 +193,9 @@ def test_prompt_group_rejects_one_member() -> None:
         fetched_at=prompt.fetched_at,
         rendered_at=datetime.now(UTC),
     )
-    with pytest.raises(ValueError, match="at least two"):
+    with pytest.raises(PromptGroupInvalid, match="at least two") as exc_info:
         PromptGroup(group_name="g", members=[pr])
+    assert exc_info.value.category == "prompt_group_invalid"
 
 
 def test_prompt_group_accepts_two_or_more_members() -> None: