Form, Not Content? A Preregistered, Placebo-Controlled Evaluation of Learned Error-Conditioned Self-Repair Through Prompts and Weights in Frozen Small Code Models
Frozen small code LLMs are deployed locally, yet the information guiding a retry after a failed attempt is still measured without placebo controls in the self-repair literature. We treat a failed program as a conjecture and an execution counterexample as an oracle-relative refutation, and introduce PoPE (Popperian Placebo-controlled Evaluation): a methodology for measuring whether evidence that falsifies LLM-generated code can be used operationally by that same model. In PoPE, error content is paired with channel-specific placebos that keep the predeclared scaffold while ablating task-relevant
Lineage graph
Paper → model → repo connections mined from source citations (Tier-1 exact match).
Why these links exist
Every edge carries a method, confidence, and the source snippet that justified it — so bad links are debuggable.
- PossiblePossibly related (embedding) · 54%Retrace-1.5B →
- PossiblePossibly related (embedding) · 46%FailproofAI/failproofai →
- LinkedLinked via arxiv author · 85%Mehmet Iscan →
“Form, Not Content? A Preregistered, Placebo-Controlled Evaluation of Learned Error-Conditioned Self-Repair Through Promp”
- PossiblePossibly related (embedding) · 52%ndcorder/outputguard →
- FuzzySimilar title/name (fuzzy) · 87%f/prompts.chat →
“Fuzzy title match (0.94): “Form, Not Content? A Preregistered, Placebo-Controlled Evalu” ≈ “f/prompts.chat””
- FuzzySimilar title/name (fuzzy) · 59%mangiucugna/json_repair →
“Fuzzy title match (0.73): “Form, Not Content? A Preregistered, Placebo-Controlled Evalu” ≈ “mangiucugna/json_repair””
- PossiblePossibly related (embedding) · 47%Why Self-Correction Loops Can Degrade Reliability in LLM Pipelines (85% Down to 62%) →