cookbook/environments/_21_math/TEST_LOG.md
Tested 2026-07-20 live against gpt-5.5, agno 2.7.4, using
.venvs/demo/bin/python with OPENAI_API_KEY loaded from .envrc.
Status: PASS
Description: Saturated multiplication beside a calibrated modular recurrence.
Result: 12 attempts in 29s. single-product saturated at 6/6, while
recurrence-8 landed at 5/6 (0.833), producing the intended visible transition from
an easy row to a true learning-zone row.
Status: PASS
Description: Product plus digit-sum, scaling, and remainder transform.
Result: 12 attempts in 56s. The shorter chain saturated at 6/6; the longer
16-digit chain split 4/6 (0.667). The explicit 0 < pass_rate < 1 selection returned
only long-chain.
Status: PASS
Description: Eight-, nine-, and ten-round nonlinear recurrence ladder.
Result: 18 attempts in 82s, all scored. rounds-8 was 3/6 (0.500),
rounds-9 was 3/6 (0.500), and rounds-10 was 4/6 (0.667). Every row occupied
the middle band without the 30s-plus incomplete behavior seen in the discarded
11- and 15-round calibration tasks.