feat(ADR-0170/W1): widen inject_from_match return type — no behavior change (#374)

First implementation PR of the ADR-0170 wave. Type-level widening only:
the recognizer-injector dispatch now returns
``tuple[InjectorEmission, ...]`` where
``InjectorEmission = CandidateInitial | CandidateOperation``.

The existing ``inject_discrete_count_statement`` continues to emit only
``CandidateInitial`` — the widening unlocks but does not exercise
operation emission. Subsequent W2-W5 PRs ship the per-injector emission
shapes:

- W2 — DCS-S1 acquisition verbs (CandidateOperation(add))
- W3 — A1 currency_amount (CandidateInitial reimplementation)
- W4 — A3 multiplicative_aggregation (CandidateInitial(product))
- W5 — A4 temporal_aggregation (deferred until apply_rate primitive)

## Changes

### `generate/recognizer_anchor_inject.py`
- New `InjectorEmission = Union[CandidateInitial, CandidateOperation]`
- `inject_from_match` return type widened to
  `tuple[InjectorEmission, ...]`
- `__all__` exports `InjectorEmission`
- Documentation comment names ADR-0170 §"Implementation outline"

### `generate/math_candidate_graph.py` (admissibility dispatch)
The per-statement admission loop now dispatches admissibility on the
concrete candidate type:

  if isinstance(c, CandidateInitial):
      if _initial_admissible(c): admitted.append(c)
  elif isinstance(c, CandidateOperation):
      if roundtrip_admissible(c): admitted.append(c)

No new admission semantics — each type is gated by the predicate it was
already gated by elsewhere in the codebase. The dispatch unifies the
injector path with the parser path.

### `tests/test_adr_0170_w1_injector_type_widening.py` (new)
- Pin: `InjectorEmission` union members are exactly the two candidate types
- Pin: `inject_from_match` return type is widened
- Pin: `inject_discrete_count_statement` still emits CandidateInitial (W1
  is type-level only)
- Hazard pin: case 0050 remains refused
- Hazard pin: unparseable-verb refusal path (#359) unchanged
- Anti-regression: canonical DCS narrow-form extraction still works

## Test plan

- tests/test_adr_0170_w1_injector_type_widening.py: 6 passed (new)
- tests/test_adr_0163_d2_discrete_count_injection.py: 21 passed
  (existing D.2 v1 injector regression)
- tests/test_brief_11b_audit_artifact.py + step2_lexicon +
  recognizer_skip_wrong_zero + brief_11_audit: 55 passed
- tests/test_candidate_graph_recognizer_wiring.py: 7 passed
- tests/test_candidate_domain_partition.py: 5 passed
- tests/test_adr_0131_G2_comparatives + G4 + G5 + S1_rate_events:
  130 passed
- Total: 225 passed
- evals/gsm8k_math/train_sample/v1: counts=correct=3 refused=47 wrong=0
  (unchanged; verified no behavioral regression)

## Hard invariants

- `wrong == 0` preserved (admissibility dispatch is type-aware but
  semantically identical to the parser path's gating)
- ADR-0166: no new eval lanes
- No teaching-store / pack mutation
- Five-layer wrong=0 safety net (ADR-0163.D.2) intact
- Reader path unchanged

generate/math_candidate_graph.py

@@ -727,9 +727,24 @@ def parse_and_solve(
                     )
                     injected = inject_from_match(recognizer_match, s)
                     if injected:
-                        admitted: list[SentenceChoice] = [
-                            c for c in injected if _initial_admissible(c)
-                        ]
+                        # ADR-0170 — dispatch admissibility on the
+                        # concrete candidate type. CandidateInitial uses
+                        # the existing _initial_admissible gate;
+                        # CandidateOperation uses the parser's
+                        # roundtrip_admissible gate (same predicate
+                        # operations from the regex path already pass
+                        # through). No new admission semantics — each
+                        # type is gated by the predicate it was always
+                        # gated by; the dispatch just unifies the
+                        # injector path with the parser path.
+                        admitted: list[SentenceChoice] = []
+                        for c in injected:
+                            if isinstance(c, CandidateInitial):
+                                if _initial_admissible(c):
+                                    admitted.append(c)
+                            elif isinstance(c, CandidateOperation):
+                                if roundtrip_admissible(c):
+                                    admitted.append(c)
                         if len(admitted) == len(injected) and admitted:
                             per_sentence_choices.append(
                                 _collapse_per_sentence_ties(admitted)

generate/recognizer_anchor_inject.py

@@ -42,13 +42,23 @@ section) is preserved across this module:
 
 from __future__ import annotations
 
-from typing import Mapping
+from typing import Mapping, Union
 
 from evals.refusal_taxonomy.shape_categories import ShapeCategory
-from generate.math_candidate_parser import CandidateInitial
+from generate.math_candidate_parser import CandidateInitial, CandidateOperation
 from generate.math_problem_graph import InitialPossession, MathGraphError, Quantity
 from generate.recognizer_match import RecognizerMatch
 
+# ADR-0170 — the widened injector emission type. Per-category injectors
+# may emit a tuple of ``CandidateInitial`` (existing) or
+# ``CandidateOperation`` (new, ADR-0170). The downstream
+# ``per_sentence_choices`` aggregator dispatches admissibility on the
+# concrete type (``_initial_admissible`` vs ``roundtrip_admissible``).
+# No new admission paths are introduced by the widening itself; new
+# emission shapes ship in subsequent per-injector PRs (ADR-0170 §"impl
+# outline" W2/W3/W4/W5).
+InjectorEmission = Union[CandidateInitial, CandidateOperation]
+
 
 # ---------------------------------------------------------------------------
 # Public surface
@@ -58,12 +68,15 @@ from generate.recognizer_match import RecognizerMatch
 def inject_from_match(
     match: RecognizerMatch,
     sentence: str,
-) -> tuple[CandidateInitial, ...]:
+) -> tuple[InjectorEmission, ...]:
     """Dispatch a recognizer match to its per-category injector.
 
     Returns an empty tuple when the category has no v1 injector or when
-    the v1 injector refused.  Skip-only behavior (the round-2 default)
-    is the empty-tuple result.
+    the v1 injector refused. Per ADR-0170, the return type is now
+    ``tuple[InjectorEmission, ...]`` (``CandidateInitial | CandidateOperation``)
+    so per-category injectors can emit operations as well as initials.
+    The v1 ``discrete_count_statement`` injector continues to emit only
+    ``CandidateInitial`` — the widening is type-level only in this PR.
     """
     injector = _INJECTORS.get(match.category)
     if injector is None:
@@ -260,6 +273,7 @@ _INJECTORS: Mapping[ShapeCategory, "type"] = {
 
 
 __all__ = [
+    "InjectorEmission",
     "inject_from_match",
     "inject_discrete_count_statement",
 ]

tests/test_adr_0170_w1_injector_type_widening.py

@@ -0,0 +1,159 @@
+"""ADR-0170 W1 — type widening pinning tests.
+
+These tests pin the no-behavior-change widening of
+``inject_from_match``'s return type. The contract becomes
+``tuple[CandidateInitial | CandidateOperation, ...]`` so per-category
+injectors can emit operations as well as initials. The existing
+``inject_discrete_count_statement`` still emits only ``CandidateInitial``;
+the widening is type-level only in this PR.
+
+References:
+- docs/decisions/ADR-0170-injector-contract-widening.md §"Implementation
+  outline" W1 (this PR)
+- docs/handoff/DCS-S1-FINDING.md — the investigation that surfaced the
+  contract gap
+- PR #369 (A2) — the schema-refusal that first observed the gap
+"""
+
+from __future__ import annotations
+
+from typing import get_type_hints, Union, get_args, get_origin
+
+import pytest
+
+from generate.math_candidate_parser import CandidateInitial, CandidateOperation
+from generate.recognizer_anchor_inject import (
+    InjectorEmission,
+    inject_from_match,
+    inject_discrete_count_statement,
+)
+
+
+# ---------------------------------------------------------------------------
+# Type-level contract
+# ---------------------------------------------------------------------------
+
+
+def test_injector_emission_union_includes_both_candidate_types():
+    """``InjectorEmission`` is the union of ``CandidateInitial`` and
+    ``CandidateOperation``. Future injector PRs can emit either."""
+    args = get_args(InjectorEmission)
+    assert CandidateInitial in args
+    assert CandidateOperation in args
+    # No third type smuggled in — the union is exactly two members.
+    assert len(args) == 2
+
+
+def test_inject_from_match_return_type_is_widened():
+    """The dispatcher returns a tuple of ``InjectorEmission`` (not just
+    ``CandidateInitial``). This pins the W1 widening; reverting to a
+    narrower return type without an explicit ADR amendment fails this
+    test."""
+    hints = get_type_hints(inject_from_match)
+    return_type = hints["return"]
+    # tuple[InjectorEmission, ...]
+    assert get_origin(return_type) is tuple
+    inner = get_args(return_type)
+    # tuple[X, ...] reports (X, Ellipsis)
+    assert inner[-1] is Ellipsis
+    inner_type = inner[0]
+    # The element type is either InjectorEmission directly or its
+    # Union[CandidateInitial, CandidateOperation] expansion.
+    if get_origin(inner_type) is Union:
+        members = set(get_args(inner_type))
+        assert {CandidateInitial, CandidateOperation}.issubset(members)
+    else:
+        # Annotated alias case — resolve once.
+        assert inner_type is InjectorEmission
+
+
+# ---------------------------------------------------------------------------
+# Behavioral pin — existing DCS injector unchanged
+# ---------------------------------------------------------------------------
+
+
+def test_discrete_count_injector_still_emits_only_candidate_initial():
+    """W1 is type-level only. The existing
+    ``inject_discrete_count_statement`` returns ``CandidateInitial`` —
+    not ``CandidateOperation`` — at runtime. This is the byte-identical
+    behavior guarantee for the W1 PR.
+
+    Mechanically: pre-W1 the function returned
+    ``tuple[CandidateInitial, ...]``. Post-W1 it still does. Subsequent
+    PRs (W2 DCS-S1 acquisition, W3 currency, W4 multiplicative) widen
+    the per-injector emission shapes; W1 ships only the dispatcher
+    contract."""
+    import inspect
+    sig = inspect.signature(inject_discrete_count_statement)
+    # With ``from __future__ import annotations`` the return annotation
+    # is stored as a string. The W1 pin is that the existing DCS
+    # injector's *narrower* return type is unchanged — only the
+    # dispatcher (``inject_from_match``) widens.
+    assert sig.return_annotation == "tuple[CandidateInitial, ...]"
+
+
+# ---------------------------------------------------------------------------
+# Behavioral pin — case 0050 hazard
+# ---------------------------------------------------------------------------
+
+
+def test_case_0050_remains_refused_post_widening():
+    """The widening must not weaken the wrong=0 invariant. Case 0050
+    refuses pre-W1 and must continue to refuse post-W1."""
+    from generate.math_candidate_graph import parse_and_solve
+    case_text = (
+        "Mark does a gig every other day for 2 weeks.  "
+        "For each gig, he plays 3 songs.  "
+        "2 of the songs are 5 minutes long and the last song is twice that long.  "
+        "How many minutes did he play?"
+    )
+    result = parse_and_solve(case_text)
+    assert not result.is_admitted, (
+        f"case 0050 admitted post-W1 — wrong=0 hazard re-introduced: "
+        f"answer={result.answer!r} graph={result.selected_graph!r}"
+    )
+
+
+def test_unparseable_verb_still_refuses_post_widening():
+    """The recognizer-no-injection refusal path (the #359 wrong=0 fix)
+    is unchanged by the W1 widening. Unparseable verbs still produce
+    explicit refusals, not silent admissions."""
+    from generate.math_candidate_graph import parse_and_solve
+    result = parse_and_solve(
+        "Sam has 5 apples. Sam contemplates 3 apples. "
+        "How many apples does Sam have?"
+    )
+    assert not result.is_admitted
+    assert result.refusal_reason is not None
+
+
+# ---------------------------------------------------------------------------
+# Anti-regression: existing DCS path still admits
+# ---------------------------------------------------------------------------
+
+
+def test_existing_dcs_admission_path_unchanged():
+    """A canonical narrow-form DCS sentence (proper noun + has + count
+    + observed counted_noun) still admits via the existing injector.
+    The widening must not regress the v1 admission path."""
+    from generate.recognizer_match import _try_extract_discrete_count_anchor, _padded_lower
+    from generate.math_candidate_graph import _load_ratified_registry_or_empty
+
+    reg = _load_ratified_registry_or_empty()
+    dcs_specs = [
+        r.canonical_pattern for r in reg
+        if r.shape_category.value == "discrete_count_statement"
+    ]
+    assert dcs_specs, "no ratified DCS recognizer on main"
+    spec = dcs_specs[0]
+
+    stmt = "Nicole has 400 Pokemon cards."
+    padded = _padded_lower(stmt)
+    anchor = _try_extract_discrete_count_anchor(stmt, padded, spec)
+    assert anchor is not None, (
+        "canonical DCS extraction regressed post-W1 — "
+        f"'Nicole has 400 Pokemon cards.' should extract"
+    )
+    assert anchor["subject_role"] == "Nicole"
+    assert anchor["count_token"] == "400"
+    assert anchor["counted_noun"] == "Pokemon cards"