Search for a command to run...
Current AI systems have no structural representation of their own knowledge boundaries. A system with no such representation fills gaps with inference presented as knowledge, producing outputs indistinguishable in surface confidence from well-grounded ones. This paper introduces Chronicle, an architectural framework that addresses this problem and a second one: AI systems cannot maintain coherent understanding across time, because no deployed architecture treats temporal structure as a first-class concern. Chronicle proposes a unified epistemic ledger containing two co-resident record types. Chronicle-K entries provide hierarchical temporal compression of accumulated knowledge through a four-layer architecture, achieving O(log N) query complexity and asymptotic storage stabilization. A system with ten years of operational history queries that history at Layer 3 resolution without scanning ten years of records. Chronicle-G entries provide a persistent, structured, compounding record of knowledge boundaries: every gap encountered during query processing is logged as a typed GapRecord, tracked through a defined lifecycle, and compressed through the same four-layer hierarchy as Chronicle-K. Gap baseline establishment proceeds via a four-process dual-method triangulation mechanism combining Bayesian accumulation and Kalman tracking, converging on gap boundary estimates from independent directions before committing them to topographic classification. Neither record type is sufficient without the other. A ledger with only Chronicle-K entries is an efficient memory system that still fills gaps with false confidence. A ledger with only Chronicle-G entries has no accumulated knowledge foundation to anchor gap history against. Together, they produce a bidirectional query resolution architecture in which Chronicle-G is consulted before synthesis, not after, and every response carries a four-layer structured output: high-confidence findings, moderate-confidence findings, identified gaps, and actionable research directions. At sufficient operational scale, Chronicle-G produces something not previously available in any architecture known to the author: a continuously refined, empirically grounded map of where AI systems consistently fail to answer the questions people are actually asking. The framework is domain-general and regulatory-relevant, providing structural support for the traceability and human oversight requirements of EU AI Act Articles 13 and 14 and FDA guidance on AI-enabled medical devices as a byproduct of normal operation rather than a reporting layer added afterward.