Read the explanation
The bundled synthetic baseline contains four claims and four events. Two claims have tester-owned events with nonempty artifact references. One tester event lacks a reference, and one purchase event belongs to the community. Two of the four critical questions therefore remain evidence gaps. Bars share one hundred twenty pixels per claim. A reference is supplied text, not independently authenticated evidence or proof of a real transaction. The synthetic repaired dataset supplies a tester purchase reference and a thesis reference. Verified claims rise from two to four, with the same one hundred twenty pixels per claim. Critical gaps fall from two to zero, changing the local verdict to firsthand basis complete. This demonstrates the classification rule only. It does not register accounts, purchase tokens, authenticate artifacts or establish product quality, investment suitability or legality. Claim CSV requires four named columns; event CSV requires six. These bars share eighty pixels per column. Duplicate identifiers, invalid actors and invalid timestamps block reconciliation. Among matching events, the earliest chronological event at the strongest available category is selected. A failed compile leaves the earlier valid result available; export also records current input text, so users should recompile successfully before treating exported input provenance as matching its matrix.