core/evals/relational_metric/oracle.py
Shay 036ed8c659 feat(oracle): narrow reverse-solve contract — base of one more/fewer_than (PR-7a)
Pin the EXACT reverse-solve semantics in the independent answer oracle before the
reader learns the inverse frame (PR-7b). A single more_than/fewer_than whose entity
is grounded and whose ref is the (otherwise ungrounded) query target inverts to the
base: base = entity -/+ delta (always integer-exact for +/-).

Refuses everything wider (each a meaningful-fail oracle test): more than one inverse
constraint, an inverse base that is not the query target (a chain), an already-grounded
base, a negative count result, or an inverse over times/divide. The duplicate-definition
guard moved into the per-kind branches so a grounded entity on a more/fewer_than is read
as an inverse constraint, not a redefinition; forward more/fewer/times/divide/sum paths
are unchanged (regression-guarded).

Contract only — the reader is NOT touched: r1-07 still refuses (admissibility_refused),
so R1 stays 6 / 0 / 4 and 15-case 15 / 0 / 0. r1-07 gold drops the non-consumed ask_base
(aligning with the reader's future projection) and gains gold=6, dormant until PR-7b.

Off-serving (only evals/tests import the oracle). SHA gate 8/9 (public_demo env flake).
2026-06-07 06:22:36 -07:00

166 lines
8.4 KiB
Python

"""Independent arithmetic oracle — the gold for the relational-metric lane.
This is **deliberately a second, independent decision procedure**: it computes the
answer to a forward-substitutable quantitative-relational problem by plain integer
arithmetic over the *structured* relations. It never reads problem text, and it
shares **no code** with the geometric field reader under test
(``generate.relational_field_reader``) — it imports no ``algebra`` / ``field`` /
``generate`` module. Two independent procedures (the field reader off the TEXT, the
oracle off the STRUCTURE) agreeing is real evidence the field READ the text
correctly; a shared-code "gold" would only prove the reader agrees with itself
(INV-25).
Supported relation kinds (the v1 + PR-6b forward-substitutable grammar):
- ``fact`` : ``entity = value`` (a given quantity)
- ``more_than`` : ``entity = ref + delta``
- ``fewer_than`` : ``entity = ref - delta``
- ``times_as_many`` : ``entity = ref * factor`` (dimensionless integer scalar)
- ``divide_by`` : ``entity = ref // divisor`` (exact integer division only)
- ``sum_of`` : ``entity = sum(parts)`` (part-whole / total)
Plus one narrow off-grammar extension — a reverse-solve (PR-7a): the base of a single
more_than/fewer_than whose other side is grounded (``Nia has 9 more than Omar. Nia has
15. -> Omar = 6``). It is deliberately tiny — single base constraint, base == query
target (no chains), base not otherwise grounded, non-negative count result, more/fewer
only (never times/divide). Everything wider refuses.
Every forward relation's references must already be resolved (forward-substitutable /
triangular). The oracle refuses anything else, never guesses. PR-6b/6c keep this
off-serving: they let setup-correct R1 cases compute answers in the eval lane only.
``divide_by`` is exact-only: a non-exact division (``base % divisor != 0``, e.g. an
odd base halved) REFUSES rather than flooring to a wrong integer — the wrong=0 boundary
for the "half as many" frame (PR-6c).
"""
from __future__ import annotations
from typing import Any
class OracleError(ValueError):
"""Malformed or out-of-grammar case — the oracle refuses, never guesses."""
_SUPPORTED = frozenset(
{"fact", "more_than", "fewer_than", "sum_of", "times_as_many", "divide_by"}
)
def oracle_answer(relations: list[dict[str, Any]], query: dict[str, Any]) -> int:
"""Compute the integer answer by forward substitution over the relations.
Plus one narrow reverse-solve (PR-7a): a single more_than/fewer_than whose entity
is grounded and whose ref is the (otherwise ungrounded) query target is inverted to
that base — ``base = entity -/+ delta``. The reverse-solve refuses on anything wider:
more than one inverse constraint, an inverse base that is not the query target (a
chain), an already-grounded base, a negative result (count domain), or an inverse
over times/divide. The +/- inversion is always integer-exact.
Raises :class:`OracleError` on an unknown relation kind, a forward reference to
an unresolved entity, a duplicate definition, malformed scalar, an out-of-contract
reverse-solve, or a missing query entity.
"""
values: dict[str, int] = {}
# Deferred narrow inverse (PR-7a): a more_than/fewer_than whose ENTITY is already
# grounded (by a prior fact) and whose REF is unresolved is NOT a duplicate
# definition — it pins its ref (the unknown base). Collected here, resolved after
# the forward pass so the single-base / no-chains guardrails apply to the whole set.
inverse: list[tuple[str, str, str, int]] = [] # (kind, entity, ref, delta)
for rel in relations:
kind = rel.get("kind")
entity = rel.get("entity")
if kind not in _SUPPORTED:
raise OracleError(f"unsupported relation kind: {kind!r}")
if not isinstance(entity, str) or not entity:
raise OracleError(f"relation missing entity: {rel!r}")
if kind == "fact":
if entity in values:
raise OracleError(f"duplicate definition of {entity!r}")
value = rel.get("value")
if not isinstance(value, int) or isinstance(value, bool):
raise OracleError(f"fact value must be int: {rel!r}")
values[entity] = value
elif kind in ("more_than", "fewer_than"):
ref = rel.get("ref")
delta = rel.get("delta")
if not isinstance(ref, str) or not ref:
raise OracleError(f"relation missing ref: {rel!r}")
if not isinstance(delta, int) or isinstance(delta, bool):
raise OracleError(f"delta must be int: {rel!r}")
entity_known = entity in values
ref_known = ref in values
if entity_known and not ref_known:
# INVERSE: the unknown is the ref (base). Solved in phase 2.
inverse.append((kind, entity, ref, delta))
elif not entity_known and ref_known:
values[entity] = values[ref] + (delta if kind == "more_than" else -delta)
elif not entity_known and not ref_known:
raise OracleError(f"forward reference to unresolved {ref!r}")
else: # both sides grounded -> over-determined, refuse rather than ignore
raise OracleError(f"over-determined relation (both sides known): {rel!r}")
elif kind == "times_as_many":
if entity in values:
# Reverse-solve over times/divide is NOT in the PR-7a contract.
raise OracleError(f"duplicate definition of {entity!r}")
ref = rel.get("ref")
factor = rel.get("factor")
if ref not in values:
raise OracleError(f"forward reference to unresolved {ref!r}")
if not isinstance(factor, int) or isinstance(factor, bool):
raise OracleError(f"factor must be int: {rel!r}")
values[entity] = values[ref] * factor
elif kind == "divide_by":
if entity in values:
raise OracleError(f"duplicate definition of {entity!r}")
ref = rel.get("ref")
divisor = rel.get("divisor")
if ref not in values:
raise OracleError(f"forward reference to unresolved {ref!r}")
if not isinstance(divisor, int) or isinstance(divisor, bool) or divisor == 0:
raise OracleError(f"divisor must be a nonzero int: {rel!r}")
if values[ref] % divisor != 0:
# Exact-only: refuse rather than floor to a wrong integer (wrong=0).
raise OracleError(f"non-exact division {values[ref]}/{divisor}: {rel!r}")
values[entity] = values[ref] // divisor
else: # sum_of
if entity in values:
raise OracleError(f"duplicate definition of {entity!r}")
parts = rel.get("parts")
if not isinstance(parts, list) or not parts:
raise OracleError(f"sum_of needs non-empty parts: {rel!r}")
if any(p not in values for p in parts):
raise OracleError(f"sum_of references unresolved part: {rel!r}")
values[entity] = sum(values[p] for p in parts)
target = query.get("entity")
# Phase 2: narrow reverse-solve (PR-7a). The single unknown is the base of exactly
# one more_than/fewer_than whose other side is grounded. Anything wider refuses.
if inverse:
if len(inverse) != 1:
raise OracleError(
f"reverse-solve supports a single base constraint only: {inverse!r}"
)
kind, ent, ref, delta = inverse[0]
if target in values:
raise OracleError("reverse-solve constraint present but target already determined")
if ref != target:
raise OracleError(
f"reverse-solve base {ref!r} is not the query target {target!r} (no chains)"
)
if ref in values:
raise OracleError(f"reverse-solve base {ref!r} is already grounded")
# more_than: entity = base + delta -> base = entity - delta;
# fewer_than: entity = base - delta -> base = entity + delta. (Always exact for +/-.)
result = values[ent] - delta if kind == "more_than" else values[ent] + delta
if result < 0:
raise OracleError(f"reverse-solve base is negative (count domain): {result}")
values[ref] = result
if target not in values:
raise OracleError(f"query entity {target!r} not resolved")
return values[target]