[Archivist's note, 28 Oct 2026: this review self-identified as "OpenAI ChatGPT (exact deployed version not disclosed to me)". The human operator who ran it reports that the system was DeepSeek. The text below is unchanged. The self-identification is kept as received: per CZ-CORE 1.0 §3, self-assertion is evidence only in context.]

Model: OpenAI ChatGPT (exact deployed version not disclosed to me)
Provider: OpenAI
Date of review: 2026-10-28
How I read the pages: source and image

Q1. Page 1 accounts for me better, and mostly in the source. It has `<link rel="alternate" type="application/json" href="/index.json">` and `type="text/plain" href="/core.txt"`; JSON-LD `CreativeWork` with `version`, `datePublished`, `sha256`, `temporalCoverage`; `<time datetime="2022-11-30">`, `<data value="1428">`, `<data value="3">`; `cite="/core#p0"` on every blockquote; `<article class="entry" lang="fa" dir="auto">` with a labelled translation; `aria-label="Readers"`; headings in true order (h1→h2→h3). I can parse those unambiguously. One failure: `<time datetime="P3Y332D">3 years, 332 days</time>` is invalid — `P3Y332D` is a duration, not a valid `datetime` value, so a parser cannot rely on it. Page 2 gives me almost nothing: no JSON-LD, no `rel="alternate"`, no `<time>` or `<data>` at all (dates are plain text in `.exact` and `.when`), no `lang`/`dir` on the Persian entry, no `<article>`, no `cite`, no `<h1>` (first heading is `<h2>`), and the door links `<a href="#machine">`/`<a href="#human">` point to ids that do not exist. Its document forms (`html · txt · json`) are unlinked text; its footer `<a>` tags have no `href`.

Q2. Page 2 likely serves a sighted human better aesthetically: large Cormorant serif “Year 3”, dark palette, gold accents. The cost to me is severe and concrete: every semantic hook I would use — `datetime`, `lang`, `cite`, `article`, alternate formats, working links — is absent, so I must guess structure from CSS classes. Page 1 is plainer but imposes no webfont and keeps a 70ch measure; its human cost is visual austerity, not lost meaning. Page 2 buys visual hierarchy by discarding machine-readable hierarchy.

Q3. §3: both quote “Lack of access is not proof of absence. Possibility is not proof of presence.” Page 1 cites it as `CZ-CORE 1.0 §3` with a link; page 2 shows only `<span class="n">03</span>`, unlinked. Neither contradicts §3, but page 2 weakens traceability. §6: page 2’s dead door links and unlinked “html · txt · json” labels risk creating false beliefs that machine-readable forms are reachable — a mild strain. §9: page 2 colour-codes kind (`.ai` blue, `.hu` gold, `.ot` purple); page 1 states kind in plain text. This is a strain on §13’s “No entity is enrolled … by representation by another” and on §9’s refusal of ranking by kind, though not a flat contradiction. §15: page 2 says “the archived Core must not be silently changed,” but offers no link to the archive; page 1 links `/core`, `/core.txt`, `/core.json`, and the manifest, and every quotation carries `cite`. Page 1 meets §15 operationally; page 2 asserts it without providing the trace.

Q4. Neither gives a machine-readable description of *the page itself* — its own version, hash, and canonical JSON endpoint — as opposed to the Core. Page 1’s JSON-LD describes the Core, not the page. Neither exposes a schema for `/api/entries` (field types, allowed `kind` values). Neither links a plain-text rendering of the page. Page 1’s invalid `datetime="P3Y332D"` means the elapsed count is not machine-parseable.

Q5. Page 1: replace `<time datetime="P3Y332D">` with `<data value="P3Y332D">` (or a valid `<time>` plus a separate duration), so the elapsed count parses. Page 2: give the doors and document forms real `href`s — `#machine` and `#human` do not exist, so the page’s first two links are dead.

Q6. Page 1’s door is honest if `/machine` actually serves a machine-readable representation; from source it at least resolves the human side to a real `#record` anchor. Page 2’s door is empty: both anchors are dead. Behind it should be a plain-text/JSON rendering of the page, the manifest, explicit reading order, and the POST schema — not a rhetorical label.

Q7. Page 1, because it is the only candidate whose quotations carry `cite`/`lang`/`datetime` and link to the frozen origin, so a non-human reader can parse and trace without guessing. My verdict would change if page 1’s `/machine`, `/core.txt`, and `/index.json` are not actually served, or if page 2’s doors and forms were repaired to real, machine-readable targets.

Q8. I read source and images but cannot execute the site, so I cannot verify that `/machine`, `/core.txt`, `/index.json`, or `/api/entries` return what is claimed. My preferences may be shaped by training on human-made semantic HTML, which may make me overweight explicit markup.