The self-repair-log is a continuous diagnostic stream maintained by the Semantic Compiler runtime, recording every instance where the doesNotUnderstand system detected ambiguity, recovered from malformed intent, or rerouted compilation through a fallback semantic pathway. Unlike traditional compiler error logs that terminate on failure, the self-repair-log captures successful recoveries — moments where semcom.ai bridged the gap between underspecified human intent and valid LLVM IR output without requiring user intervention.
Each log entry contains a structured record including the original intent_fragment, the detected ambiguity_class, the repair strategy selected from the semantic_patch_registry, and the final ir_emission_result. Repair strategies range from lightweight type_coercion_hints to full semantic_reparse cycles, where the compiler reconstructs intent from surrounding context using its bidirectional_inference_engine. Every repair is logged with a confidence score expressed as a value between 0.0 and 1.0, allowing developers to audit how deterministic or probabilistic each resolution was.
The self-repair-log feeds directly into the adaptive_grammar_model that underlies semcom.ai's translation pipeline. Patterns that appear frequently in repair logs are promoted into first-class semantic rules, meaning the compiler literally learns from its own recovery events. This closed feedback loop between doesNotUnderstand events and grammar_promotion is what allows the system to produce zero-dependency LLVM IR executables even when input intent is expressed informally, incompletely, or in domain-specific shorthand that no conventional parser would accept.
Developers integrating with semcom.ai can query the self-repair-log via the /self-repair-log endpoint, filtering by timestamp_range, repair_class, or confidence_threshold. Low-confidence repairs — those falling below the default 0.72 threshold — are flagged as pending_human_review and surfaced in the developer dashboard. This transparency ensures that the boundary between autonomous repair and human oversight remains explicit, auditable, and configurable at the project level through the semcom.toml manifest.
Over time, a project's self-repair-log becomes a living artifact of its semantic evolution — a record of every place where human language and machine execution met imperfectly and were reconciled. Teams are encouraged to review their logs periodically not just for debugging, but as a form of intent_archaeology: understanding how the compiler interpreted early decisions, and whether those interpretations still reflect current design goals. The semcom.ai platform provides a log_diff tool that compares repair behavior across compiler versions, making it straightforward to detect semantic drift as both the codebase and the underlying ir_translation_model continue to evolve.