ARTFEED — Contemporary Art Intelligence

TraceCoder: Explainable Code Generation with Position-Key Snippet Versioning

other · 2026-07-30

A new code generation concept, TraceCoder, addresses the black-box nature of LLM-based coding agents by introducing three complementary mechanisms: a relational snippet-history schema that records benchmark references, repair rounds, failure texts, and LLM explanations for full provenance; a browser-based visualization tool rendering history as heat-mapped, hover-annotated source code; and a competitive fractional position-key indexing scheme with tree-node delimiters for stable, lexicographically-ordered identifiers. TraceCoder was evaluated on 30 algorithmic programming tasks spanning string processing, enabling fine-grained tracking without disrupting surrounding lines.

Key facts

  • TraceCoder addresses black-box outputs of LLM-based coding agents.
  • It uses a relational snippet-history schema for provenance queries.
  • A browser-based visualization tool renders history as heat-mapped, hover-annotated source code.
  • A competitive fractional position-key indexing scheme assigns stable identifiers.
  • The scheme uses tree-node delimiters for fine-grained tracking.
  • Evaluation was on 30 algorithmic programming tasks.
  • Tasks span string processing.
  • The approach enables auditing and explainability.

Entities

Sources