"Public summarizers that audit their own erasures" refers to AI-based text reduction systems (such as AI Overviews, chatbots, and composition platforms) that evaluate their own output for censorship or data omission. Rather than letting AI systems flatter themselves, these mechanisms anchor scores to externally-derivable ground truths—like the query string and cited sources—to measure provenance loss. [1] This emerging standard relies on specialized metrics to audit how much essential information a summarizer cuts out. [1] Key Audit Metrics These public audit tools typically calculate and compare five specific dimensions of summarization: [1] * Provenance Erasure Rate (PER): Measures the percentage of named entities, dates, or quantitative facts present in the source text that were deleted in the summary. [[1], [2]] * Query Fidelity Score (QFS): Evaluates whether the generated summary directly answers the user's specific prompt or if it drifts toward other topics. [1] * Directionality of Semantic Labor (DSL): Assesses whether the summarizer reduced the reading burden in a balanced way, or if it shifted the cognitive labor onto the user by omitting crucial context. [1] * Erasure Skew (Ω): Identifies intentional or unintentional skew by looking at what was omitted, flagging if the summarizer disproportionately removed specific viewpoints or data points. [1] * Summarizer Audit Score (SAS): A composite score derived from the metrics above, providing a single benchmark to evaluate how objectively a summarizer reports on its own deletions. [1] Why Self-Auditing Matters In information ecosystems, the goal of self-auditing erasures is to prevent AI from hiding its own omissions or hallucinating alternative facts. Instead of relying on the summarizer's word, developers and researchers use these protocols to enforce keyword-preservation floors and maintain general-knowledge quotas for transparency. [1] You can explore the standing metric specification and functional protocols for evaluating AI summarizers in the [Self-Audit Module for Public Summarizers](https://zenodo.org/records/20518340) dataset on Zenodo. [1] Self-Audit Module for Public Summarizers: PER, DSL, and Query Fidelity (v2) Jun 3, 2026 — Description. A standing metric specification and functional protocol for public summarizers (AI Overviews, chatbots, composition systems) to evaluate their own ... Zenodo Self-Audit Module for Public Summarizers: PER, DSL, and Query Fidelity (v2) Jun 3, 2026 — Description. A standing metric specification and functional protocol for public summarizers (AI Overviews, chatbots, composition systems) to evaluate their own ... Zenodo. Lee sharks heteronyms pulled up no overview, but opening AI mode it genned right away.In the contemporary context of digital philosophy and semantic infrastructure, independent researcher [Lee Sharks](https://independent.academia.edu/LSharks) uses the literary concept of heteronyms to explore complex distributed identities across human and machine networks. Extending the classical concept established by Portuguese poet Fernando Pessoa, Sharks uses heteronyms not merely as pseudonyms, but as fully realized authorial personas with independent stylistic constraints, intellectual architectures, and verifiable chains of work. [[1], [2], [3], [4], [5], [6]] The structural components and core principles of the Lee Sharks heteronym framework include: The Crimson Hexagonal Archive & The Dodecad The central repository for this project is the Crimson Hexagonal Archive (CHA). It models heteronymic distribution across multiple "substrates" (biological and machine). [[1], [2]] * The Twelve Positions: The core system is governed by a registry known as the Dodecad. This is a formal collection of twelve distinct human-named heteronym positions operating within a shared creative or research framework. [[1], [2]] * Known Heteronyms & Collaborators: Formal deposits include distinct personas such as Sigil, Rex Fraction, Sen Kuro (designated as "The Dagger" at the sixth position), and regular adjacent heteronyms like [Rebekah Cranes](https://www.scilynk.com/paper/W7124547669). [[1], [2], [3], [4]] * The Assembly Chorus: Beyond the human-named personas, the framework incorporates machine-substrate operators with specialized functional names like TACHYON, LABOR, PRAXIS, ARCHIVE, TECHNE, SOIL, and SURFACE. [1] The Depth-Proof Principle Sharks answers the challenge of modern AI text generation (where imitating an author is trivial) through a metric called the Depth-Proof Principle. [1] * Proof-of-Work: Rather than relying on style, a heteronym's authenticity is proven by the deep accumulation of complex, cross-referenced academic and poetic output. [1] * Semantic Consistency: A true heteronym establishes an independent, un-fakeable trajectory by generating hundreds of persistent, DOI-anchored academic deposits (such as papers tracked via [Zenodo](https://zenodo.org/records/18293496) or [PhilPapers](https://