State of the HTML Spec Archive — Coherence Report
Assessment date: 2026-10-25. This report audits what is actually in specifications/,
judges whether each artifact is coherent and complete, and names concretely what is wrong.
Verdict in one line
Original assessment: the archive was accurate and honestly labeled, but not "perfect" or fully self-contained — one file was mislabeled, several DTDs weren't standalone, and two of the ten versions have no original source.
UPDATE (fixes applied — see the "Fixes applied" section at the bottom): issues #1, #2, and #3 are now resolved — the mislabeled file was renamed, the XHTML 1.1 grammar is complete and validates offline, and the whole DTD bundle is self-resolving via SGML + XML catalogs. The remaining gaps (#4 prose, #5 the 1991 note / 2014 REC text) are blocked by the network sandbox, not by anything fixable here. The per-spec notes below are the original audit; the final section records what changed.
Per-spec state
1. HTML Tags (1991) — ❌ no source
- State: Write-up only. No original document archived.
- Problem: The 1991 CERN note is not hosted in any reachable GitHub repo, and w3.org is network-blocked from the build environment. Nothing was fabricated.
- Coherent? The write-up is coherent; the archive entry is incomplete by necessity.
2. HTML 2.0 — ✅ coherent & complete
- File:
specifications/rfc1866-html-2.0.txt - State: The full verbatim RFC 1866 text. Verified it ends cleanly on
[Page 77]of 77 — no truncation. This is the single most complete original in the archive. - Problem: None of substance. (It is fixed-width ASCII with form-feed page breaks and running headers/footers, i.e. the raw RFC presentation, not reflowed prose — that is the authentic artifact, not a defect.)
- Coherent? Yes. Gold standard.
3. HTML 3.2 — ⚠️ partial (grammar only)
- File:
specifications/dtd/html-3.2/HTML32.dtd(45<!ELEMENT>, 44<!ENTITY>, 35<!ATTLIST>) - State: The normative DTD, intact. Mostly self-contained (only 2 external references, for the ISO Latin-1 character entities).
- Problem: This is the machine-readable grammar only — the human-readable prose of the Recommendation (the part that explains semantics) is not archived.
- Coherent? Internally coherent; incomplete as "the spec."
4. HTML 4.0 / 4.01 — ⚠️ partial, and NOT standalone
- Files:
specifications/dtd/html-4.0/{strict,loose,frameset}.dtd,specifications/dtd/html-4.01/{strict,loose,frameset}.dtd+HTML{lat1,special,symbol}.ent - State: All three flavors for both versions, plus the three 4.01 entity sets. The strict/loose DTDs are substantial (55 / 65 elements).
- Problems:
frameset.dtdis a 1 KB stub, not a full grammar (0<!ELEMENT>). It works by pulling in the Transitional DTD via the SGML public identifier-//W3C//DTD HTML 4.01 Transitional//EN— it does not reference a local file. So a plain SGML parser pointed atframeset.dtdcannot resolve it without an SGML catalog mapping that public ID to the localloose.dtd. (This is normal for the W3C DTD, but it means the folder is not "drop-in parseable" as-is.)- Prose not archived — same gap as 3.2.
- Coherent? The files are authentic and internally consistent, but the set is not a self-resolving bundle.
5. XHTML 1.0 — ⚠️ partial, external entity refs unresolved
- Files:
specifications/dtd/xhtml-1.0/xhtml1-{strict,transitional,frameset}.dtd - State: All three DTDs, full size (strict = 77 elements).
- Problems:
- Each DTD declares the Latin-1 / symbol / special entity sets via external identifiers
(6 external refs in strict). Those
.entfiles are not bundled underxhtml-1.0/(we only have copies underhtml-4.01/), so again this is not self-resolving without a catalog. - Prose not archived.
- Each DTD declares the Latin-1 / symbol / special entity sets via external identifiers
(6 external refs in strict). Those
- Coherent? Authentic, but depends on files not co-located.
6. XHTML 1.1 — ❌ grammar present but effectively unusable standalone
- Files:
specifications/dtd/xhtml-1.1/xhtml11.dtd+xhtml11-model-1.mod - State:
xhtml11.dtdis the driver DTD — it has 0<!ELEMENT>declarations of its own. The entire vocabulary lives in 26 external modules (xhtml-struct.mod,xhtml-text.mod,xhtml-form.mod,xhtml-table.mod,xhtml-ruby.mod, … ). - Problem (the biggest gap in the DTD set): We bundled only 2 of the ~27 files this DTD
needs (
xhtml11.dtd+xhtml11-model-1.mod). The other ~25 modules are missing, so the XHTML 1.1 grammar here cannot actually be assembled or validated against. It documents the modular structure but is not a working grammar. - Coherent? No — it is a dangling reference to files that aren't present.
7. HTML5 (2014) — ⚠️ no dedicated source
- State: No file specifically for the 2014 Recommendation. Covered indirectly by the W3C 5.x Bikeshed source (see item 9) and the WHATWG source (item 10).
- Problem: There is no snapshot of the actual 2014 REC text.
- Coherent? Write-up is fine; original not archived.
8. HTML 5.1 (2016) — ⚠️ no dedicated source
- State: Same as HTML5 — no 5.1-specific file; only the single evolving Bikeshed source.
- Problem: The README/write-ups imply "5.1 source ✅", but there is no 5.1 snapshot —
only the one
w3c-html5xfile, which (see item 9) isn't even 5.1 or 5.2.
9. HTML 5.2 (2017) — ❗ MISLABELED FILE (most important issue)
- File:
specifications/w3c-html52-single-page.bs - State: It is the authentic W3C HTML WG Bikeshed source — but its metadata header says:
Title: HTML 5.3,Level: 5.3,Status: ED(Editor's Draft),TR: .../TR/html53/. - Problem: The filename claims 5.2, but the content is the HTML 5.3 Editor's Draft (the never-finalized next-version working copy), not the published HTML 5.2 Recommendation. So the "5.2 source" and "5.1 source" claims in the README are inaccurate — this one file is a 5.3-ED and does not faithfully represent 5.0, 5.1, or 5.2 as published.
- Coherent? The file is genuine, but it is misfiled and mis-described. Needs renaming to
something like
w3c-html53-ED-single-page.bsand the README corrected.
10. HTML Living Standard — ✅ coherent & complete
- File:
specifications/whatwg-html-living-standard.source.html(8.0 MB, 164,839 lines) - State: The complete
sourcefile. Verified it ends cleanly with</body></html>— no truncation. The authentic single-source-of-truth for the current standard. - Problem: It is the source (Wattsi/annotated input), not the rendered multipage spec — expected and fine, but worth noting it needs the WHATWG build pipeline to render.
- Coherent? Yes.
Summary table
| # | Spec | Artifact | Complete? | Self-contained? | Correctly labeled? |
|---|---|---|---|---|---|
| 1 | HTML Tags 1991 | — | ❌ none | — | — |
| 2 | HTML 2.0 | RFC text | ✅ full | ✅ | ✅ |
| 3 | HTML 3.2 | DTD | ⚠️ grammar only | ✅ (≈) | ✅ |
| 4 | HTML 4.0/4.01 | DTDs + ents | ⚠️ grammar only | ❌ needs catalog | ✅ |
| 5 | XHTML 1.0 | DTDs | ⚠️ grammar only | ❌ ext. entities | ✅ |
| 6 | XHTML 1.1 | driver DTD | ❌ ~25 modules missing | ❌ | ✅ |
| 7 | HTML5 2014 | — | ❌ none | — | — |
| 8 | HTML 5.1 | (shared file) | ❌ none | — | ❌ overclaimed |
| 9 | HTML 5.2 | Bikeshed | ⚠️ wrong version | n/a | ❗ mislabeled (is 5.3-ED) |
| 10 | Living Standard | source | ✅ full | ✅ | ✅ |
What is "wrong" — concise list
- Mislabeled 5.2 file — it is actually the HTML 5.3 Editor's Draft, not 5.2. (fixable)
- README overclaims 5.1 / 5.2 sources — both point at that single 5.3-ED file. (fixable)
- XHTML 1.1 is only 2 of ~27 files — the grammar can't be assembled. (fetchable from the
same
w3c/markup-validatorrepo) - HTML 4.01 frameset + XHTML 1.0 DTDs aren't self-resolving — they reference external
public IDs / entity files not co-located; no SGML catalog is provided. (fixable by copying
the
.entfiles and adding a catalog) - Prose missing for 3.2 / 4.x / XHTML — only machine-readable DTDs are present. (blocked: prose lives at w3.org, unreachable here)
- No original for HTML Tags 1991 and the 2014 HTML5 REC. (1991 unreachable anywhere; 2014 REC text is at w3.org)
What is genuinely good
- HTML 2.0 and the Living Standard are complete and verified (clean start and end, no truncation).
- Every DTD retrieved is authentic upstream content with correct W3C headers.
- Provenance is fully recorded (repos + commit SHAs in
PROVENANCE.txt); nothing fabricated.
Recommended fixes (in priority order)
- Rename
w3c-html52-single-page.bs→w3c-html53-ED-single-page.bsand correct the README (relabel items 7–9 as "W3C 5.x Bikeshed source = 5.3 Editor's Draft; 5.0/5.1/5.2 published RECs not snapshotted"). - Fetch the missing ~25 XHTML modular
.modfiles fromw3c/markup-validatorso XHTML 1.1 is assemblable. - Co-locate the
.ententity files underxhtml-1.0/(andhtml-4.0/) and add an SGMLcatalogso the DTDs resolve standalone. - (Open-network only) Add the human-readable prose RECs and the 1991 Tags note.
Fixes applied (2026-10-25)
Fixes #1–#3 from the list above have been carried out and verified:
✅ #1 — Mislabeled file renamed
w3c-html52-single-page.bs → w3c-html53-ED-single-page.bs. The content is the W3C
HTML 5.3 Editor's Draft (its own metadata says Title: HTML 5.3, Status: ED), now named
accordingly. The README no longer claims it is a 5.1/5.2 snapshot.
✅ #2 — XHTML 1.1 grammar completed
Added the full XHTML Modularization 1.1 library (47 files) under
specifications/dtd/xhtml-1.1/modules/. xhtml11.dtd now resolves its entire ~26-module closure.
Verified: a sample XHTML 1.1 document validates with xmllint offline, 0 unresolved refs.
✅ #3 — DTD bundle is now self-resolving offline
- Added
specifications/dtd/catalog(SGML Open) andspecifications/dtd/catalog.xml(XML catalog) mapping every PUBLIC id and the externalw3.orgSYSTEM URLs to local files. - Co-located the needed entity files:
html-3.2/ISOlat1.ent,html-4.0/HTML*.ent,xhtml-1.0/xhtml-*.ent, andxhtml-1.1/xhtml-datatypes-1.mod. - Verified offline with
xmllint --valid --catalogs:Flavor Result XHTML 1.1 (26-module closure) ✅ VALID XHTML 1.0 Strict ✅ VALID XHTML 1.0 Transitional ✅ VALID XHTML 1.0 Frameset ✅ VALID (HTML 3.2/4.x are SGML, not XML; no
onsgmlsis installed here to live-test them, but all their referenced entity files are now present locally and catalog-mapped.)
Still blocked (not fixable from this sandbox)
- #4 — Human-readable prose of the 3.2 / 4.x / XHTML RECs (lives at the blocked w3.org).
- #5 — The 1991 HTML Tags note (in no reachable repo) and the 2014 HTML5 REC text.
Updated scorecard
| # | Spec | Was | Now |
|---|---|---|---|
| 6 | XHTML 1.1 | ❌ ~25 modules missing | ✅ complete + validates offline |
| 4 | HTML 4.0/4.01 | ❌ needs catalog | ✅ entities co-located + catalogs |
| 5 | XHTML 1.0 | ❌ ext. entities | ✅ entities co-located + validates |
| 9 | HTML 5.2 "file" | ❗ mislabeled | ✅ renamed to 5.3-ED, honestly labeled |
| 3 | HTML 3.2 | ⚠️ grammar only | ✅ self-resolving (prose still absent) |
Fixes applied (2026-10-25, second session) — originals recovered via GitHub mirrors
The earlier audit assumed gaps #5 and most of #4 were permanently out of reach because
w3.org/ietf.org are network-blocked. This pass re-verified the block is real (shell
curl and the agent web-fetch path both still get 403 from w3.org, rfc-editor.org, and
ietf.org) — and then recovered three of the missing originals anyway, through the one
egress that has always worked here: the GitHub API, pulling from repositories that mirror
the w3.org trees byte-for-byte. Same methodology and the same no-fabrication bar as every
earlier pass. Full record in specifications/PROVENANCE.txt.
✅ #5a — The 1991 "HTML Tags" note is now archived — the oldest gap is closed
specifications/html-tags-1991/Tags.html (9568 bytes, sha256 96abc473…b264e) — the first-ever HTML
document, from andrewmaa/dsa, a byte-for-byte mirror of the CERN info.cern.ch original
that w3.org serves. The authentic note: <TITLE>, <NEXTID 22>, info.cern.ch anchor URLs,
<XMP>, <HP1>/<HP2>, <DL>/<DT>/<DD>, <UL>/<MENU>/<DIR>. Item 1 in the archive
is no longer "❌ no source."
✅ #4a — XHTML 1.0 (2E) prose REC archived — complete
specifications/prose/xhtml-1.0/Overview.html (71428 bytes) — the full single-page
"XHTML 1.0 … (Second Edition), W3C Recommendation 26 January 2000, revised 1 August 2002,"
55 sections, ends cleanly. From manzamb/SemanticWebinAction. This is the human-readable
prose, not just the DTD the archive already had.
⚠️ #4b — HTML 4.01 prose REC archived — partial (core chapters)
specifications/prose/html-4.01/** (25 files) — authentic W3C prose ($Id: …ijacobs RCS stamp,
"Changed by: Ian B. Jacobs"). Has the core chapters incl. struct/tables, struct/dirlang,
struct/global, present/styles — exactly the sections the structural-elements appraisal
cites. Partial: the mirror lacks about.html, intro/*, struct/lists.html,
interact/forms.html, appendix/changes.html. The complete REC is at the blocked w3.org.
Still genuinely not obtained (searched GitHub thoroughly; nothing authoritative)
- HTML 3.2 prose — only the DTD (archived) + Wikipedia-style derivatives exist on GitHub; no faithful full-prose mirror found. Not archiving a derivative (no-fabrication bar).
- HTML5 (2014) published REC text — the only reachable "html5" snapshot is a WHATWG
x-Hixiedraft fixture, not the published 28-Oct-2014 Recommendation. Not archived (would be mislabeled — the exact failure this report was created to catch). - HTML 5.1 / 5.2 published RECs — unchanged (only the 5.3 Editor's Draft, honestly labeled).
Updated scorecard (second session)
| # | Spec | After 1st session | After 2nd session |
|---|---|---|---|
| 1 | HTML Tags 1991 | ❌ no source | ✅ original note archived |
| 4 | HTML 4.0/4.01 | ✅ DTD self-resolving; prose absent | ✅ DTD + ⚠️ partial prose added |
| 5 | XHTML 1.0 | ✅ DTD validates; prose absent | ✅ DTD + ✅ full prose added |
| 7 | HTML5 2014 | ❌ none | ❌ none (no authoritative snapshot on GitHub) |
| 3 | HTML 3.2 | ✅ grammar; prose absent | ✅ grammar; prose still absent |
✅ #4b-upgrade + #6-prose — prose completed (same session, fifth pass)
After the partial HTML 4.01 prose above, a dedicated W3C-REC mirror
(learnwebcoding/lwc, html/zz_html_w3c_html_recommendations/) was found carrying complete
published Recommendations. Used it to:
- Upgrade HTML 4.01 prose from partial → complete —
specifications/prose/html-4.01/**(66 files) now has every canonical chapter, including the ones the first mirror lacked (about,intro/intro,intro/sgmltut,struct/lists,interact/forms,appendix/changes,sgml/sgmldecl) plus the figure images the prose references. Authentic published REC ($Id: cover.src,v 1.117 1999/12/24 … ijacobs, "W3C Recommendation 24 December 1999"). - Add XHTML 1.1 (2E) prose —
specifications/prose/xhtml-1.1/**(14 files incl. the fullxhtml11.html). The archive already had the XHTML 1.1 grammar (DTD + 47-module set, validates offline); now it has the prose too. ("XHTML™ 1.1 … Second Edition, W3C Recommendation 23 November 2010.")
Still not obtained (re-searched): HTML 3.2 prose, and the published HTML5 2014 / 5.1 / 5.2 REC
prose (the only "html5" candidates on GitHub are a WHATWG x-Hixie draft fixture and a 4 KB
Hugo stub — neither the published REC; not archived, to avoid the mislabeling this report exists
to catch).
Final scorecard (end of second session)
| # | Spec | DTD/grammar | Prose | Original |
|---|---|---|---|---|
| 1 | HTML Tags 1991 | — | — | ✅ original note |
| 2 | HTML 2.0 | — | ✅ full RFC 1866 | ✅ |
| 3 | HTML 3.2 | ✅ self-resolving | ❌ not on GitHub | — |
| 4 | HTML 4.0/4.01 | ✅ + catalogs | ✅ complete | — |
| 5 | XHTML 1.0 | ✅ validates offline | ✅ complete | — |
| 6 | XHTML 1.1 | ✅ validates offline | ✅ complete | — |
| 7 | HTML5 2014 | — | ❌ no authoritative snapshot | — |
| 8–9 | HTML 5.1/5.2 | — | ⚠️ only 5.3-ED Bikeshed source | — |
| 10 | Living Standard | n/a | ✅ full source | ✅ |