Session State Continuity & Anchor Replay Family (FAM-SSCAR) — Public release v1.1.0

By epistemic continuity we mean what long-running AI work must carry about its own knowledge, not its content: which claims are settled and which are still open, which are provisional, what has been designated as work to continue, and where each claim came from. Agent memory — conversation history, provider-side memory, handover documents — carries content across a session boundary; it does not carry this structure, and on the other side every sentence reads with the same weight. Epistemic continuity is that structure crossing the boundary, from one session to the next and from one model to another, and admitted on the far side only by a decision. Making that decision is the work of an agentic AI governance architecture: every claim keeps its provenance, the categories stay distinct, the same inputs always yield the same outcome, the next session fails closed when continuity cannot be established, and the decision leaves an audit trail from which a reviewer who was not present can reconstruct what crossed. The Session State Continuity & Anchor Replay Family specifies that decision — five agents and an integrator, each built to be verified — and governs agent memory rather than storing it. This record is the Family Pack: an execution-neutral normative specification from which a team can build the whole family — the five component agents (Verified Rehydration, Freshness Portability, Boundary Defense, Truth Arbitration and Audit Replay) and the family integrator that composes them — distributed free of charge. It does not prescribe a programming language, framework, or runtime, and it leaves storage to the execution layer. The package also carries the conformance tests, the connector tests, the verification runbook verify.md and an observational, non-normative NNT companion. License. The package is governed by the SSCAR family License, included as SSCAR_Family_License.md. In its words: "Where you obtained the package otherwise than at the point of distribution, you accept this license by accepting it at semantichor.com, in the same two confirmations, before any use of the package." and "A copy of the package held without acceptance confers nothing." The acceptance page is https://semantichor.com/families/sscar/license/. Integrity. SSCAR_Family_Pack.zip, 210 files, SHA-256 df876868946c5166f7279b3b0226d2bd01b8d4a3b2651075cea70692ebf51994. manifest_ref (the SHA-256 of SSCAR_Family_Manifest.sha256, bound in the License): 33ae0304298fe8891231455741dc512404cef9406e0f1c2ed5499945a3b497f5. The family supplementary derivations are archived as a separate record. Use of AI tools. The author constructed the canonical files of the six units of the family package, including the conformance tests and the connector annexes, reviewed them manually, and designed and calibrated the NNT harness. Large language models were used openly, as tools to accelerate research and production, with manual review: they assisted the drafting of the package's documents and, by design, built the five reconstruction baselines in the NNT companion, which are study artifacts, not reference implementations. The package's canonical files were constructed in one AI-provider environment, and the NNT campaign was run in a different AI-provider environment. No tool is an author; the author is accountable for the content. What matters is not which tools touched the documents but whether the package can be verified: its integrity is anchored by the Manifest, and its conformance suite and connector tests are runnable. Contact: [email protected]