feat(rate): single-rate reader + run_reader lane + answer-choice reuse (R3d)

generate/rate_comprehension/reader.py: read_rate_problem comprehends explicit single-rate prose into a RateProblem via structural recognizers — rate value (<N> <plural> per <singular>), duration (for/in <N> <unit>), standalone quantity (outside those spans), query (how many <unit> by unit-match, or speed in <X> per <Y> -> rate). The compound-unit-consistency check is the wrong=0 gate: a duration whose unit isn't the rate denominator (miles/hour for 30 minutes) REFUSES rate_unit_mismatch, never converts.

Refuses combined_rates (>=2 rate clauses), temporal_state (clock markers), missing_rate/time/quantity (underdetermined). run_reader lane + 'reader' CLI. R3 reader: setup_correct 8 / setup_wrong 0 / refused_correct 4 -> answers 6 correct / 0 wrong / 6 refused (the 6/0/6 target). Reuses R2 verify_answer_choice end-to-end. Off-serving. 6 reader tests + constructed unit-mismatch guard.

evals/rate_oracle/__main__.py

@@ -1,19 +1,28 @@
-"""CLI: validate the R3 rate gold.
+"""CLI: validate the R3 rate gold, or grade the R3 reader against it.
 
-    python -m evals.rate_oracle   # validate rate_gold.jsonl; exit 0 iff invalid == 0
+    python -m evals.rate_oracle           # validate rate_gold.jsonl; exit 0 iff invalid == 0
+    python -m evals.rate_oracle reader    # grade the reader; exit 0 iff setup_wrong == 0
+                                          #   and reason_mismatch == 0
 """
 
 from __future__ import annotations
 
 import json
+import sys
 
-from evals.rate_oracle.runner import run
+from evals.rate_oracle.runner import run, run_reader
 
 
 def main() -> int:
-    report = run()
+    lane = sys.argv[1] if len(sys.argv) > 1 else ""
+    if lane == "reader":
+        report = run_reader()
+        ok = report["setup_wrong"] == 0 and report["reason_mismatch"] == 0
+    else:
+        report = run()
+        ok = report["invalid"] == 0
     print(json.dumps(report, indent=2, default=str))
-    return 0 if report["invalid"] == 0 else 1
+    return 0 if ok else 1
 
 
 if __name__ == "__main__":

evals/rate_oracle/runner.py

@@ -129,4 +129,62 @@ def run() -> dict[str, Any]:
     }
 
 
-__all__ = ["EXPECTATIONS", "READER_REASONS", "SOLVER_REASONS", "gold_to_problem", "run", "validate_fixture"]
+def run_reader() -> dict[str, Any]:
+    """Grade the R3 rate reader against the gold (R3d).
+
+    Well-formed fixtures (``solved`` / ``solver_refuses``) must read to a setup whose signature
+    equals the gold's (``setup_correct``); a refusal is a miss; a mismatch is ``setup_wrong``.
+    ``reader_refuses`` fixtures must refuse with the gold's ``reader_reason`` (``refused_correct``);
+    a refusal with the wrong reason is ``reason_mismatch``; producing a setup is ``setup_wrong``.
+    Exit-0 criterion: ``setup_wrong == 0 and reason_mismatch == 0``.
+    """
+    from generate.meaning_graph.reader import Refusal
+    from generate.rate_comprehension.reader import read_rate_problem
+
+    fixtures = _load_rate_gold()
+    setup_correct = setup_wrong = setup_refused = refused_correct = reason_mismatch = 0
+    details: list[dict[str, Any]] = []
+    for fx in fixtures:
+        out = read_rate_problem(fx["text"])
+        fid = fx.get("id")
+        if fx["expect"] in ("solved", "solver_refuses"):
+            if isinstance(out, Refusal):
+                setup_refused += 1
+                details.append({"id": fid, "outcome": "setup_refused", "reason": out.reason})
+            elif rate_setup_signature(out) == rate_setup_signature(gold_to_problem(fx)):
+                setup_correct += 1
+                details.append({"id": fid, "outcome": "setup_correct"})
+            else:
+                setup_wrong += 1
+                details.append({"id": fid, "outcome": "setup_WRONG"})
+        else:  # reader_refuses
+            if isinstance(out, Refusal) and out.reason == fx["reader_reason"]:
+                refused_correct += 1
+                details.append({"id": fid, "outcome": "refused_correct", "reason": out.reason})
+            elif isinstance(out, Refusal):
+                reason_mismatch += 1
+                details.append({"id": fid, "outcome": "reason_mismatch", "got": out.reason, "want": fx["reader_reason"]})
+            else:
+                setup_wrong += 1
+                details.append({"id": fid, "outcome": "setup_WRONG_over_read"})
+    return {
+        "lane": "rate_oracle_reader",
+        "total": len(fixtures),
+        "setup_correct": setup_correct,
+        "setup_wrong": setup_wrong,
+        "setup_refused": setup_refused,
+        "refused_correct": refused_correct,
+        "reason_mismatch": reason_mismatch,
+        "details": details,
+    }
+
+
+__all__ = [
+    "EXPECTATIONS",
+    "READER_REASONS",
+    "SOLVER_REASONS",
+    "gold_to_problem",
+    "run",
+    "run_reader",
+    "validate_fixture",
+]

generate/rate_comprehension/__init__.py

@@ -13,6 +13,7 @@ conversion, no multi-equation systems. Those are later R3 slices.
 from __future__ import annotations
 
 from generate.rate_comprehension.model import RateProblem, RateRole
+from generate.rate_comprehension.reader import read_rate_problem
 from generate.rate_comprehension.solver import answer_unit, solve_rate
 from generate.rate_comprehension.units import (
     BaseUnit,
@@ -32,6 +33,7 @@ __all__ = [
     "answer_unit",
     "rate_from_quantity_over_time",
     "rate_times_time",
+    "read_rate_problem",
     "solve_rate",
     "time_from_quantity_over_rate",
 ]

generate/rate_comprehension/reader.py

@@ -0,0 +1,124 @@
+"""Single-rate prose reader (R3d): explicit rate prose -> RateProblem.
+
+Reads ONLY explicit single-rate problems and refuses everything else — combined/multi-rate,
+clock-time/temporal, unit-mismatched durations, and underdetermined (missing-piece) prose. The
+compound-unit consistency check (R3a) is the wrong=0 gate: a duration whose unit is not the rate's
+denominator (``60 miles per hour for 30 minutes``) REFUSES rather than silently converting.
+
+Recognized clauses (structural, not fixed strings):
+  - rate value     : ``<N> <plural> per <singular>``        -> rate value + RateUnit(num, denom)
+  - duration/time  : ``(for|in) <N> <unit>``                -> time value + unit
+  - quantity       : a standalone ``<N> <unit>`` (outside the rate/duration spans)
+  - query          : ``how many <unit>`` (role by unit match) | ``speed in <X> per <Y>`` (-> rate)
+
+Refusals: ``combined_rates`` (≥2 rate clauses), ``temporal_state`` (clock markers),
+``rate_unit_mismatch`` (duration/quantity unit ≠ rate unit), ``missing_rate`` / ``missing_time`` /
+``missing_quantity`` (underdetermined). Off-serving; deterministic.
+"""
+
+from __future__ import annotations
+
+import re
+
+from generate.meaning_graph.reader import Refusal
+from generate.rate_comprehension.model import RateProblem
+from generate.rate_comprehension.units import RateUnit
+
+_RATE_VALUE = re.compile(r"\b(\d+)\s+([a-z]+)\s+per\s+([a-z]+)\b")
+_RATE_QUERY = re.compile(r"\bspeed in ([a-z]+) per ([a-z]+)\b")
+_DURATION = re.compile(r"\b(?:for|in)\s+(\d+)\s+([a-z]+)\b")
+_HOW_MANY = re.compile(r"\bhow many ([a-z]+)\b")
+_DIGIT_NOUN = re.compile(r"\b(\d+)\s+([a-z]+)\b")
+_TEMPORAL = re.compile(r"\b\d+\s*(?:am|pm)\b|\barrived\b|\bleft at\b|\bo'?clock\b")
+
+
+def _singular(noun: str) -> str:
+    if noun.endswith("es") and noun[:-2].endswith(("x", "s", "z", "ch", "sh")):
+        return noun[:-2]
+    if noun.endswith("s") and len(noun) > 1:
+        return noun[:-1]
+    return noun
+
+
+def read_rate_problem(text: str) -> RateProblem | Refusal:
+    """Comprehend explicit single-rate prose into a typed RateProblem, or refuse."""
+    if not text or not text.strip():
+        return Refusal("empty")
+    t = text.lower()
+
+    if _TEMPORAL.search(t):
+        return Refusal("temporal_state", "elapsed clock time is not an explicit duration")
+    rate_clauses = _RATE_VALUE.findall(t)
+    if len(rate_clauses) >= 2:
+        return Refusal("combined_rates", "more than one rate is multi-rate (R3.2)")
+
+    # rate value clause (≤1)
+    rate_value: int | None = None
+    num_unit = denom_unit = None
+    rate_match = _RATE_VALUE.search(t)
+    if rate_match:
+        rate_value = int(rate_match.group(1))
+        num_unit, denom_unit = _singular(rate_match.group(2)), _singular(rate_match.group(3))
+
+    # duration / time clause
+    dur = _DURATION.search(t)
+    time_value = int(dur.group(1)) if dur else None
+    time_unit = _singular(dur.group(2)) if dur else None
+
+    # standalone quantity: a digit-noun outside the rate-clause and duration spans
+    spans = [m.span() for m in (rate_match, dur) if m is not None]
+    quantity_value = quantity_unit = None
+    for m in _DIGIT_NOUN.finditer(t):
+        if any(s <= m.start() < e for s, e in spans):
+            continue
+        quantity_value, quantity_unit = int(m.group(1)), _singular(m.group(2))
+        break
+
+    # query role + rate unit
+    rate_query = _RATE_QUERY.search(t)
+    how_many = _HOW_MANY.search(t)
+    if rate_query:
+        query = "rate"
+        rate_unit = RateUnit(_singular(rate_query.group(1)), _singular(rate_query.group(2)))
+    elif how_many:
+        asked = _singular(how_many.group(1))
+        if rate_value is None:
+            return Refusal("missing_rate", "no rate given and rate is not the question")
+        rate_unit = RateUnit(num_unit, denom_unit)
+        if asked == num_unit:
+            query = "quantity"
+        elif asked == denom_unit:
+            query = "time"
+        else:
+            return Refusal("query_target_unrecognized", asked)
+    else:
+        return Refusal("no_query")
+
+    # compound-unit consistency (the wrong=0 gate)
+    if time_value is not None and time_unit != rate_unit.denominator:
+        return Refusal("rate_unit_mismatch", f"duration {time_unit!r} ≠ rate denominator {rate_unit.denominator!r}")
+    if quantity_value is not None and quantity_unit != rate_unit.numerator:
+        return Refusal("rate_unit_mismatch", f"quantity {quantity_unit!r} ≠ rate numerator {rate_unit.numerator!r}")
+
+    # assemble by query (refusing an underdetermined setup)
+    if query == "quantity":
+        if rate_value is None:
+            return Refusal("missing_rate")
+        if time_value is None:
+            return Refusal("missing_time")
+        return RateProblem(rate_unit, rate_value, time_value, None, "quantity")
+    if query == "rate":
+        if quantity_value is None:
+            return Refusal("missing_quantity")
+        if time_value is None:
+            return Refusal("missing_time")
+        return RateProblem(rate_unit, None, time_value, quantity_value, "rate")
+    # query == "time"
+    if rate_value is None:
+        return Refusal("missing_rate")
+    if quantity_value is None:
+        return Refusal("missing_quantity")
+    return RateProblem(rate_unit, rate_value, None, quantity_value, "time")
+
+
+__all__ = ["read_rate_problem"]

tests/test_rate_reader.py

@@ -0,0 +1,65 @@
+"""Tests for the R3 single-rate reader (R3d).
+
+Pins wrong=0: every well-formed fixture reads to exactly the gold setup; every reader_refuses
+fixture refuses with its gold reason; and — end to end — every solved fixture reads → solves →
+ties to its labeled answer (reusing R2's answer-choice verifier). Plus the unit-mismatch gate.
+"""
+
+from __future__ import annotations
+
+from evals.rate_oracle.runner import _load_rate_gold, gold_to_problem, run_reader
+from evals.rate_oracle.signature import rate_setup_signature
+from generate.answer_choices.verify import ChoiceVerdict, verify_answer_choice
+from generate.meaning_graph.reader import Refusal
+from generate.rate_comprehension.reader import read_rate_problem
+from generate.rate_comprehension.solver import solve_rate
+
+
+def _by(expect: str) -> list[dict]:
+    return [f for f in _load_rate_gold() if f["expect"] == expect]
+
+
+def test_reader_lane_is_wrong_zero_and_complete() -> None:
+    r = run_reader()
+    assert r["setup_wrong"] == 0 and r["reason_mismatch"] == 0
+    assert r["setup_refused"] == 0
+    assert r["setup_correct"] == 8  # 6 solved + 2 solver_refuses (both have valid setups)
+    assert r["refused_correct"] == 4
+
+
+def test_reads_every_well_formed_fixture_to_gold_signature() -> None:
+    for fx in _by("solved") + _by("solver_refuses"):
+        out = read_rate_problem(fx["text"])
+        assert not isinstance(out, Refusal), f"{fx['id']}: {getattr(out, 'reason', '')}"
+        assert rate_setup_signature(out) == rate_setup_signature(gold_to_problem(fx)), fx["id"]
+
+
+def test_refuses_every_reader_refuse_fixture_with_reason() -> None:
+    for fx in _by("reader_refuses"):
+        out = read_rate_problem(fx["text"])
+        assert isinstance(out, Refusal) and out.reason == fx["reader_reason"], fx["id"]
+
+
+def test_read_solve_verify_end_to_end_for_solved() -> None:
+    for fx in _by("solved"):
+        problem = read_rate_problem(fx["text"])
+        assert not isinstance(problem, Refusal), fx["id"]
+        value = solve_rate(problem)
+        assert value == fx["gold"], fx["id"]
+        verdict = verify_answer_choice(value, fx["options"], fx["answer"])
+        assert isinstance(verdict, ChoiceVerdict) and verdict.status == "consistent"
+        assert verdict.computed_label == fx["answer"]
+
+
+def test_read_then_solver_refuses_for_solver_refuse_fixtures() -> None:
+    for fx in _by("solver_refuses"):
+        problem = read_rate_problem(fx["text"])
+        assert not isinstance(problem, Refusal), fx["id"]
+        out = solve_rate(problem)
+        assert isinstance(out, Refusal) and out.reason == fx["solver_reason"], fx["id"]
+
+
+def test_constructed_unit_mismatch_refuses() -> None:
+    # rate per hour, duration in days — does not compose -> refuse (never converts).
+    out = read_rate_problem("A car travels 70 miles per hour for 2 days. How many miles does it travel?")
+    assert isinstance(out, Refusal) and out.reason == "rate_unit_mismatch"