01 / A CORRECTED MISUNDERSTANDINGRejected
One variable changed the argument.
Other questions remained.
Adaptive Stacking · Learning What to Remember for Non-Markovian Reinforcement Learning
The discussion distinguished the capacity of a memory stack from the length of history an environment requires. The area chair identified that confusion in a critique of the theory. The paper was still rejected, with concerns about evaluation scope and comparisons remaining.
THE OBSERVER’S QUESTION
When a central misunderstanding is corrected, how should the remaining objections be weighed?
Follow the change in the discussion
The archived meta-review acknowledges the distinction between a small memory buffer and a much longer relevant history. It also identifies gaps in complex visual evaluation and comparisons with specialized memory architectures.
This lets us separate the correction of a particular argument from an assessment of the whole paper. The published analysis uses the available discussion and manuscript material; no independent experiment reproduction is claimed.
02 / A DIFFERENT VIEW OF CONTRIBUTIONAccepted
A low-score acceptance.
A disagreement about value.
PLAGUE · Plug-and-play Framework for Lifelong Adaptive Generation of Multi-turn Jailbreaks
The archive shows ratings of 2, 2, 4 and 2, alongside a poster acceptance. Reviewers questioned novelty and evaluation. The area chair valued the engineering contribution and cited additional comparisons and clarifications in the response.
THE OBSERVER’S QUESTION
How should a useful engineering contribution be judged when reviewers disagree about what counts as novelty?
Look beyond the numerical contrast
The meta-review describes additional budget-matched comparisons and ablations. It also expresses disagreement with some novelty objections and with the expectation that an attack paper must propose defenses.
A prediction that reviewers might raise their ratings is distinct from a recorded rating change. The archived mean is a dated snapshot, and the acceptance needs to be read alongside the arguments and process context.