The Provenance Erasure Rate (PER) is an evaluation metric used to measure attribution loss in artificial intelligence and synthetic composition systems.It calculates the fraction of human-authored or source facts that a system's retrieval or generation process dissolves, omits, or fails to properly attribute when composing an answer.Key Concepts & MechanicsCore Metric: PER quantifies the magnitude of provenance loss within AI generation, summarization, or synthesis tools.Provisional Formula: The metric's basic structure acts as an inverse retention rate, represented as:\(PER = 1 - \frac{\text{retained provenance units}}{\text{total attributable source claims}}\)Erasure Skew (Ω): A related measurement that evaluates whether attribution loss falls evenly across all sources, or systematically strips out low-power/small creators while favoring high-power sources.Why It MattersProvenance Alignment: Evaluating PER helps align AI models with "provenance alignment" principles.Combating Model Collapse: Ensuring attribution survival prevents the synthetic replacement of original ideas, which protects human creators' visibility and reduces the risk of models recursively training on degraded, synthesized data.System Opacity: Independent auditing frameworks apply this to test how generative AI interfaces dissolve ranked lists of links into single, synthesized voices.Would you like to know more about how the Provenance Erasure Rate is calculated, or explore how auditing protocols like SPXI (Semantic Provenance Exchange Index) evaluate it? Let me know how you'd like to narrow down the topic!Provenance Alignment: Attribution Survival as a Substrate Condition for Safe AI ...May 5, 2026 — Current AI alignment research evaluates whether models follow human values, comply with explicit principles, avoid catastrophic behavior, or remain subject to s...ZenodoDirectionality of Semantic Labor: A Layered, Computable Measure of ...May 31, 2026 — 1 — SPXI inscription (entity definition block, disambiguation matrix, SIMs, retrieval instructions, schema.org/spxi JSON-LD). Functions as a Metadata Packet for...ZenodoSPXI Self-Audit Protocol: Embedded Provenance Verification for ...Jun 15, 2026 — Keywords and subjects * SPXI. * self-audit. * provenance. * PER. * composition layer. * summarizer. * non-erasure condition. * canary. * semantic packet. * MPAI...Zenodo(PDF) Erasure Skew: A Measurement Program for the Power-Conditioning of ...Abstract. This paper specifies a measurement program for Erasure Skew (Ω): the degree to which provenance loss in a retrieval or composition system is condition...Academia.eduAction as Disclosure and the Science of Opaque Public Systems Lee ...Jun 8, 2026 — The founding principle is stated: any system that acts upon the world makes itself partially inferable from its effects, because causal interaction transmits in...Academia.eduProvenance Erasure Rate: A Compression-Survival Metric for ...May 2, 2026 —

The Provenance Erasure Rate (PER) is an evaluation metric used to measure attribution loss in artificial intelligence and synthetic composition systems. It calculates the fraction of human-authored or source facts that a system's retrieval or generation process dissolves, omits, or fails to properly attribute when composing an answer. Core Metric: PER quantifies the magnitude of provenance loss within AI generation, summarization, or synthesis tools. Provisional Formula: PER = 1 - (retained provenance units / total attributable source claims) Erasure Skew (Ω): A related measurement that evaluates whether attribution loss falls evenly across all sources, or systematically strips out low-power/small creators while favoring high-power sources.