From 88d8950d3e808e4e7d36e5940095f5af9606711a Mon Sep 17 00:00:00 2001 From: Shay Date: Sat, 6 Jun 2026 18:28:49 -0700 Subject: [PATCH] test(comprehension): expose R1 answer lane --- evals/setup_oracle/__init__.py | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/evals/setup_oracle/__init__.py b/evals/setup_oracle/__init__.py index 282550aa..4ea652d4 100644 --- a/evals/setup_oracle/__init__.py +++ b/evals/setup_oracle/__init__.py @@ -9,11 +9,11 @@ closes that gap: it compares the reader's comprehended STRUCTURE (the relations binding-graph reader). A wrong setup is a first-class failure even when its answer is right. `setup_wrong` is the load-bearing, wrong=0-critical count: a reading that misrepresents -the problem. v1 grades structure (facts + equations + question target/state/form); unit -modelling stays covered by the admissibility tests (a documented signature extension). +the problem. The R1 answer lane (PR-6b) runs only after setup is correct; unsupported +fixtures remain refusals and setup-wrong fixtures are never answer-scored. """ -from evals.setup_oracle.runner import run, run_r1 +from evals.setup_oracle.runner import run, run_r1, run_r1_answers from evals.setup_oracle.signature import ( gold_unknown_signature, reader_symbol_units, @@ -29,5 +29,6 @@ __all__ = [ "relation_signature", "run", "run_r1", + "run_r1_answers", "symbol_unit_signature", ]