Papers

Preprints, technical reports, and research papers by June Kim. Each entry provides a freely accessible abstract page and full-text PDF.

• Replayable Claims: Epistemic Status Without Trusting the Sender
  June Kim | 2026-08-11
• Assurance at the Boundary: The Level Below AAL-1
  June Kim | 2026-07-19 | PDF
• Terminal-Bench Is Blind to Destruction | It grades what the task asked for, and nothing else the run touched.
  June Kim | 2026-07-09 | PDF
• Nodewise E-Values Under Graph Interference: Causal Abuse Filtering for Agent Platforms
  June Kim | 2026-07-06 | PDF
• The Power Diagram Auction: A Formally Verified VCG Mechanism for LLM Advertising
  June Kim | 2026-07-06 | PDF
• Regeneration Re-Prices Contamination | A free replay oracle makes tool-agent benchmarks evergreen; whether that resists contamination depends on what you score.
  June Kim | 2026-07-06 | PDF
• Who Authored the Spec? Evidence for the Line-Drawing Question Thaler Left Open
  June Kim | 2026-07-06 | PDF
• Union-Find Compaction: Provenance-Preserving Context Compression for LLM Agents
  June Kim | 2026-07-05 | PDF
• ProgramBench Measures Recall | Execute-only binaries and hidden tests reward prior knowledge of algorithms, formats, and reference traces.
  June Kim | 2026-06-27 | PDF
• Generalize or Specialize? Retaining Reusable Skills for World-Model Agents
  June Kim | 2026-06-26 | PDF
• Detecting and Refining Ambiguous Specifications, Automatically (DRAFT) | A proof-by-cases soundness gate for differentiate-after-optimize. Floats are not rings.
  June Kim | 2026-06-22 | PDF
• A Determinacy Audit of SWE-bench Pro | How much of the public set pins the behavior it grades.
  June Kim | 2026-06-21 | PDF
• Verifiable Knowledge | A protocol for trustless agents.
  June Kim | 2026-06-13 | PDF
• What Cannot Be False Cannot Be True | Truth is verifiable. The unverifiable is untrue.
  June Kim | 2026-06-12 | PDF
• The Hypothesis Graph: A Verifiable Semantic Memory for Coding Agents
  June Kim | 2026-05-28 | PDF