Returning rejection notices within the same turn eliminated visible false-completion replies without changing strict episode success relative to next-turn notices.

model: Qwen3.6-35B · notice: next-turn

model: Qwen3.6-35B · notice: same-turn

model: Qwen3-235B · notice: next-turn

model: Qwen3-235B · notice: same-turn

model: Qwen3.6-35B · notice: next-turn · denominator: 86

model: Qwen3.6-35B · notice: same-turn · denominator: 89

model: Qwen3-235B · notice: next-turn · denominator: 27

model: Qwen3-235B · notice: same-turn · denominator: 17