Abstraction-Indexed Canonicity: Bidirectional Compression, Generative Authority, and Meta-Axis Governance

Abstract

Long-lived AI systems may treat records, organizing schemas, and higher-order generative constraints as canonical references in different roles. Abstraction-indexed canonicity makes official standing specific to scope, revision, and functional level; compression or expansion alone grants no new standing. Admission and privileged execution require separately authorized checks bound to current dependencies. Compression is evaluated against task-relevant distinctions, candidate validity, alternative coverage, and the full cost of residuals and reconstruction. A task-relative sufficiency criterion permits deliberate loss of irrelevant detail, while a conditional composition bound states when local transformation evidence supports a terminal requirement. Counterexamples retain a serviced path to correction, including its persistence. Plural generative centers remain coordinated through a governed meta-axis whose translations are checked where tasks require interchangeable paths. Six prospective comparisons use strong existing implementations, matched information and resources, fixed service requirements, and allocation that respects shared mutable state. External quantitative anchors identify feasible component mechanisms and strong rivals; a conditional transfer range exposes endpoint and support gaps. The contribution is a specified joint contract and an operational comparison protocol. Integrated authority and revision workflows, as well as the mathematical ingredients, have established precedents. No first-integration claim, unique expressivity, or empirical validation of AIC is asserted. Note on Version 2.0: this version revises the registered v1.0. The manuscript is substantially expanded (about 5,200 to 9,200 words), the claims are restated with scope, revision and dependency conditions, and Section 9 adds external measurement anchors, a conditional gate calculation, quantitative component evidence with stronger rivals, and a bounded transfer candidate with its failure conditions. No empirical validation of AIC is claimed. Files: the v2.0 manuscript and a submission-materials archive containing the rendered PDF, review notes, edit log, validation record, manifest and SHA-256 checksums. The paper derives from the author's broader Information Ecosystem Theory, which remains unpublished at the time of this release. Companion paper: Canonicity Without a Truth Monopolist: A Custodian Architecture for Long-Lived AI Ecosystems (https://doi.org/10.5281/zenodo.22711406), which develops the underlying custodian and institutional-memory architecture; this paper is self-contained.