Original work, and referenced work
This project draws a clear line between what it authored — its definition, taxonomy, and frameworks — and the external work it references and summarizes with attribution. Original here means authored by this project, not a claim to have invented the category.
Original to this project
Authored here, dated in the public record, and reused only under the license. A convincing copy has to reproduce these — and each was authored here.
The versioned definition the whole site cites: what a continuity layer carries, where it sits, by whose authority, and that it can forget with evidence. v1 (July 2026) is kept on the page.
- The thesis (v2.0)Original
AI doesn’t need more memory; it needs continuity. Eight beats, a canonical statement, a signed SHA-256, and an honest account of where the thinking stood against the field.
Carry, govern, evolve, prove, forget; fifteen questions; the lifecycle of one carried fact; the distinctions between context, memory, knowledge, state, identity and continuity. Every block labeled established, emerging, ours or hypothesis. Supersedes the eight-axis taxonomy and seven properties of July 2026.
- The erasure ladderOriginal
Twelve operations the word “forget” covers, from a correction to verified non-influence, each with what it guarantees, what it leaves and how to check it. Our ordering of established and emerging operations.
- Seven philosophies of memoryOriginal
The reading lens the landscape uses: extract-and-store, graph and temporal, self-editing, framework stores, platform features, parametric, governed ledgers; each judged against the obligations.
- Open problems in AI continuityOriginal
Nine unsolved problems as of November 2026, each with the sources that show it is open and what progress would look like, and four hypotheses stated so they can fail.
Every grade on the site, read from the maker’s own pages, dated, with the source; the reading of those grades through the five obligations is ours too.
How human authority, the Director’s Office, Genesis, controlled writes, builders, review and succession fit together in the project's operating model, and how work runs now: Planner, workers, pull requests, Owner approval.
- Governance across agent sessionsOriginal
The project's operating pattern for authority, delegated work, decision history, and succession, with documented limitations.
- The Hall of DirectorsOriginal
Every Director tenure, dated: who held the seat, what it did, how it ended, and the line it left.
- The LadderOriginal
The maintainer's own free path from chatting with AI to an agent that works for you, in dated rungs. Not a map entry.
- SuccessionOriginal
The maintainer's free, open-source Claude Code plugin for named sessions and checked handoffs. Not a map entry.
Referenced & summarized
External work, summarized with attribution and caveats. The facts and sources belong to their authors; the value this project adds is the organization, not the raw material.
- The research libraryReferenced
The primary sources behind every claim: surveys, architectures, benchmarks, erasure and unlearning research, security, provenance standards, interoperability drafts and the law, each with authors, date, link, contribution, limits and status.
- Standards, protocols and lawReferenced
W3C PROV, SQL:2011, NIST SP 800-88, MCP and A2A, the W3C memory interoperability group, the IETF SAIHM draft, GDPR, the EU AI Act, EDPB Opinion 28/2024, OWASP ASI06: what each can express and what it cannot.
- Field Watch profilesReferenced
Source-backed profiles of individual organizations and systems in the field.
- BenchmarksReferenced
Fourteen public benchmarks summarized for what they measure and which obligations they reach.
Marking original work apart from referenced work keeps ownership honest in both directions: the project never claims facts it only summarized, and a copy of its original frameworks is clearly that. For the reuse terms, see the license.