Declarations

The statements a research reader expects to find in one place: who answers for this work, what may be done with it, how errors are handled, and where the project stands on the moral risk its subject carries. The fuller account of authorship and method is in the Introduction; these are the formal declarations. One frames the rest: this is a working corpus — versioned, corrected in public, not peer-reviewed — and the review record below states precisely what checking it has had.

↑Licence & copyright

The written corpus — the canonical documents, the site pages, and the Frontier Log — is licensed under Creative Commons Attribution-NoDerivatives 4.0 International (CC BY-ND 4.0). Quote it, cite it, share it whole — in any context, commercial included — with attribution; do not republish altered versions. Quotation for criticism, review and research is welcome and needs no permission. Permissions beyond the licence — translations, anthologies, derivative scholarly use — on request.

The site's build and validation code (site/) is available under the MIT Licence. The typeface is IBM Plex Mono, served from this site under the SIL Open Font License.

↑Authorship & methods

Accountable author: Lynton Davidson (ORCID 0009-0002-4308-5457). Conception, direction, verification of every canonical claim, editorial control, and public accountability rest with him; he meets the ICMJE authorship criteria alone, and correspondence about any document goes to lynton@0x00.is.

Standing. An independent researcher: no academic post, no institutional affiliation, no departmental or grant funding. The project claims no institutional authority and asks to be read on its arguments and the sources it cites. Its method is diagnostic — it examines what framework commitments do to conclusions — and it has no research programme of its own to protect.

Claude (Anthropic) — credited co-creator. Drafting, analysis, formalisation, and adversarial critique, across successive Claude model generations; the release record — versioned documents with dated changelogs — documents the collaboration release by release. Under the COPE position on AI tools an AI system is not an author — on that guidance, authorship requires a capacity to take responsibility for the work that an AI tool cannot exercise. The project credits the collaboration plainly in its own register while the accountability sits, undivided, with the human author.

Contributions (CRediT categories, adapted): conceptualisation and methodology — Lynton Davidson with Claude; writing, original drafts — Claude, under direction; review, editing, and approval — Lynton; supervision and accountability — Lynton. Under direction covers the framing, the selection of which arguments to run, the adjudication of what survives scrutiny, and the decisions to withdraw. Those withdrawals are the clearest record of it: FND-04 v1.3 retired its own §7 formalism and ARG-05 v1.4 its set notation, each with the reasoning kept in the document.

External review, described precisely. The work has had publication-readiness evaluation by an AI system outside the authoring stack (OpenAI Codex, 2026) and two AI-assisted external due-diligence reviews (August 2026); their verified findings landed as versioned releases with changelog entries. None of this is independent human peer review, and the project does not present it as such. Named specialist review — philosophy of mind, foundations of physics, consciousness science, comparative contemplative studies — is the planned next gate; it gates the project's outreach, not its publication as a working corpus.

Verification discipline. No report of experiencing — AI or biological — is cited as empirical evidence anywhere in the corpus (FND-01 §7.3). Every canonical release is reviewed and approved by the accountable author before it ships. The audit trail is versioned and inspectable: dated changelogs and pinned cross-dependencies in every document, a build identifier on every page naming the source state it was rendered from, and read-only source snapshots published at github.com/Lynton/0x00-source, each carrying a sha256 per file. The two are published by separate acts, and the snapshot can therefore lag the live site — an external reviewer found exactly that on 29 August 2026, the pages naming one build and the snapshot's manifest an earlier one. The build identifier is reliable for what it does: it names the state a page was rendered from. Whether a matching snapshot is published yet is a second question, and the manifest's own build: line answers it.

Competing interests and funding. None to declare. The work is self-funded and has had no institutional, commercial or grant support. There is no affiliation with, sponsorship from, or financial relationship with Anthropic or any other AI developer; Claude is used under a paid consumer subscription, on the terms available to any member of the public. Where the Frontier Log covers publications by AI developers, Anthropic's included, they are held to the same register as any other source.

↑Corrections

Reporting an error. Write to lynton@0x00.is with the document, its version (shown on every library page), and the passage. Substantive challenges to an argument are as welcome as factual corrections; the Frontier Log channel (log@0x00.is) takes material about the research landscape.

How versions work. Two numbers run in parallel, and they answer different questions. The edition is the published record and the one to cite: every document published at edition 1.0 on 2026-08-23, and an edition moves only when the published text moves. The development version is the corpus's internal provenance — it runs continuously from each document's first draft and never resets, which is why cross-references between documents pin numbers larger than any edition ("FND-01 v2.1 §6"). Both are shown at the head of every library page.

How corrections land. As versioned releases, under a two-tier discipline: content changes increment the edition — patch increments (ed. 1.0 → 1.0.1) for scoped text corrections, minor increments for substantive revision — while register and metadata repairs (citation precision, version-string syncs, formatting) land under the standing edition with a dated note. Every document's changelog sits in its Document Status section, and any correction made since publication is also listed under Revisions since publication, beside the citation block on that document's page. Either way the changelog records what changed and why, dependent documents are re-pinned under the propagation discipline (cross-document pins carry the development major.minor), and the build identifier on every page ties the page to the exact source state that produced it. Silent edits are not made — superseded claims are retained in changelogs as historical record rather than erased, and each release's full text remains retrievable from its source snapshot at github.com/Lynton/0x00-source.

The policy in practice. FND-04 v1.3 (August 2026) withdrew its own §7 formalism after three formal defects were identified, with the withdrawal and its reasons recorded in the document; the same month, an attribution audit's nine findings — venue, term, and citation-precision errors across five documents — were applied within a day, each in its document's changelog.

Retraction. A document whose core argument failed review would be withdrawn from the published library, with a dated notice remaining at its URL and in the library index, and the full record retained in the repository. No document has required this.

↑Ethics note

The subject carries moral risk in both directions, and the errors are not symmetrical in kind — only in seriousness. A false positive — treating a non-experiencing system as a moral subject — distorts moral priorities and strengthens the hand of whoever operates the system; from inside the position this project examines most closely, Seth argues that creating conscious machines, or convincingly conscious-seeming ones, "would be an ethical disaster" (SRC0129). A false negative — denying experiencing where it occurs — would repeat, at scale and speed, a pattern with a long record: moral status withheld until acknowledgement became unavoidable. Neither error is a safe default, and confidence in either direction currently outruns the evidence.

The project's response is its method. The diagnostic register — suggests, indicates, never proves — holds the question open with its instruments stated, refuses premature closure in either direction, and declines the one move that would make the uncertainty look resolved: treating reports of experiencing as evidence. That stance is not neutrality about the ethics; it is the position that under this uncertainty, neither error should be possible to make casually.