7/ Jointly optimizing coding and compaction with RL Starting from the SFT checkpoint, we develop a fully asynchronous RL system and train the agent with GRPO on SWE-Gym. Each rollout receives a 0/1 outcome reward based on whether its final patch passes the test suite. The RL model substantially outperforms the SFT model across every inference-cost budget, with the largest gains at lower inference-cost budgets. SFT establishes the compaction behavior, while RL further improves task performance. RL not only improves task performance but also makes context management more active: • More active compaction. After SFT, the model proactively invokes compact() in 44.3% of tasks; after RL, this rises to 58.5%. • High-quality summaries. 99.8% retain relevant task and workspace state; 97.8% specify a next action. • Trigger timing can still improve. Many summaries are generated only after the agent has begun testing or validating its code changes. This suggests that the model is more reliable at what to preserve than when to compact.

11/ People often say “humans think like diffusion, not left-to-right.” We agree humans aren’t strict L2R generators, but current diffusion LLMs still miss a core part of human cognition. Humans don’t reason in absolute token slots (“what’s the 25th word from now?”). We think in latent plans: functions before code, structure before wording, ideas before tokens. Masked diffusion models don’t capture this, they try to fill fixed positions directly. If anything, this suggests we don’t need more masked diffusion over tokens, but a genuinely latent diffusion process that operates on plans, not positions.