A phantom-gains audit found a single greedy decode manufacturing improvement on an untrained model.

Three rounds of LoRA self-training on Qwen3-8B against a frozen control turned up seven systematic measurement failures, including a ledger showing capability changes on a model that was never trained, largely an artifact of inference batching. arXiv After a per-problem exact test under FDR control, external distillation genuinely improved rarely-solved problems while self-training did not, and self-training corrupted previously-solved problems above the baseline failure rate. Two independent audits of self-improvement in one week, both negative.