About

Rights note — read before publishing anything

The site is public. The repository is private, but everything that reaches the site is published. Three lines govern what may go up.

1. Ulysses (1922) is in the public domain. Quotation from the novel is unrestricted. Quote it freely and at length.

2. The apparatus around it is not. Declan Kiberd's introduction and notes in the Penguin Modern Classics edition are in copyright, as are Gifford's Ulysses Annotated, the Slote/Mamigonian/Turner Annotations, and every other secondary work this project cites. Paraphrase them in our own words and cite the source. Do not reproduce their text, and do not reconstruct a source's argument point by point in a way that substitutes for reading it.

3. Dimension 16's sources are reviews of a 2026 film. The film, its stills and its script are in copyright, and so is every review the comparison rests on. Paraphrase and cite; publish no stills without a licence; quote no review at length.

4. Emily Wilson's 2017 translation is in copyright (W.W. Norton). Dimension 16 compares it directly, which needs short parallel extracts — ordinary critical practice. Keep extracts brief, always name Wilson and the translation, and publish no continuous run of it. The same holds for any other translation the project quotes.

5. Commentary generated by other models is not our own writing. The documents in library/ were generated by a third model and supplied by the author. They are held as a cited source, never republished wholesale, and anything drawn from them is attributed to that source and marked for what it is — sourced fact, second-hand report, or the commentator's interpretation (anchor A5, catch C-23).

6. The PDF register embeds its typefaces, so the typefaces had to be free to embed. A downloadable PDF carries a subset of every font it renders, which makes typography a rights question rather than a taste question — the system serif this site uses on screen is a commercially licensed face, and embedding it in a public download is a licence question the project does not need to have. The PDFs are therefore set in DejaVu Serif and DejaVu Sans Mono, under the Bitstream Vera / DejaVu licence, which grants the right to reproduce and distribute the fonts — embedding included — on one condition: the copyright and permission notice travels with them. It does: design/fonts/LICENSE_DEJAVU.txt in the repository, and a line in every PDF's colophon. The fonts are build-time only and are never served by the site.

7. Dimension 14, the Atlas, is built from freely licensed material and nothing else. Rescoped by the author on 19 July 2026: publicly available sources only, and if a rights problem could not be avoided the dimension was to be struck. It can be avoided, so these are the rules it is built under (catch C-32).

• Every image carries a named licence at a named URL on the credits page. No image enters the Atlas without one. "Probably old enough" is not a licence.

• An institution's terms are not the copyright. The National Library of Ireland and Tailte Éireann both assert re-use conditions over their own scans while holding no copyright in the underlying works. Source the same material from a repository that imposes no such terms — Wikimedia Commons, per file. Do not source from nli.ie or its Flickr stream, whose terms are private-research-only.

• Attribution lives on the credits page, never on a panel. CC BY 4.0 §3(a)(2) expressly permits this, and Creative Commons' own guidance names the case. There is a trap attached: CC BY-SA 2.0 requires credit at least as prominent as any other authorship credit that appears, so the colophon is lawful only while no panel carries any credit at all. Warburg's refusal of the caption and the licence compliance are one decision. Adding a single signature or caption to a panel would breach every share-alike licence on it simultaneously.

• Do not crop share-alike images to fit the grid — pad them uncropped, which keeps the panels clear of the adaptation boundary. Licence the panels themselves CC BY-SA 4.0.

• A watermark is a credit on the panel, so a watermarked file cannot be used. Added 25 July 2026, from the place-layer survey, which found a correctly licensed CC BY 2.0 photograph of the Martello tower carrying "Alessandro Ambrosetti | 2015" burned into the pixels. The no-caption rule above is what makes the colophon lawful, and it is satisfied vacuously only while no panel carries any credit at all — a credit in the image is still a credit, whoever put it there. The escape that looks obvious is closed: removing a watermark is an alteration of a licensed work, which a share-alike or attribution licence does not permit us to make silently. So the file is refused instead. Commons flags these with {{watermark}} in the wikitext, which is how the survey caught it, but the flag is applied by volunteers and its absence proves nothing — look at the pixels. tools/atlas-licence.mjs already declared this blind spot in its own header ("it reads TEXT; a file whose page is correct and whose pixels carry a watermark passes"). The limit was known and unenforced; this is the rule that enforces it, at survey time, by eye.

• "No known copyright restrictions" is not a licence, and the National Library of Ireland's Flickr stream is where that trap lives. Recorded 25 July 2026 because the place survey hit it four times in one place: the finest period images of Dalkey — Robert French, Lawrence Collection, exactly the class clause 7 clears — carry {{Flickr-no known copyright restrictions}} and no public-domain tag of any kind, sourced from nli.ie's stream. Both grounds already appear above; what is new is how good the rejected images are. The temptation is real and it should be expected rather than resisted by memory.

• A REGION OF A MAP SHEET IS SELECTING THE DOCUMENT, NOT CROPPING TO FIT THE GRID. Ruled by the author, 27 July 2026, on the open question may-a-map-sheet-be-cropped. The no-crop bullet above was written for photographs, and it still binds every one of them: a share-alike photograph is padded uncropped, which is what catch C-77 was about and what it stays about. A map sheet is a different object. A six-inch sheet covers about eight kilometres and a panel slot renders at about a hundred pixels, so the whole sheet in a slot is a grey rectangle — and the whole sheet was never the document the slot names, any more than the whole of Thom's is the document when the slot names one address on page 1482. The line, and it is where the clause always meant to draw it: cropping to fit the grid is an aesthetic act and is refused; selecting the region that contains the subject is documentary and is admitted. Two conditions, both part of the ruling. (i) THE FRAME IS POSITIONED BY THE MAP'S OWN CONTENT — it must contain the sheet's own lettering naming the place, never be trimmed to make a composition. This is machine-checked already: check.mjs §11ad requires that lettering to be quoted, so a frame that does not contain the name cannot be recorded as establishing anything. (ii) THE COLOPHON SAYS SO. Every map frame is disclosed as a region of a named sheet, with the sheet number and the licence. CC BY 4.0 §3(a)(1)(B) requires a modification to be indicated, and a caption-free panel cannot indicate anything — so the colophon carries it, as it carries every other thing the form forbids a panel to say. Why the licences differ here and it matters: the NLS sheets are CC-BY, not CC-BY-SA, so the share-alike prominence condition that makes the colophon load-bearing for the photographs is not engaged by the maps at all. The condition on the maps is attribution and indication of change, and both live in the colophon.

• CORROBORATION IS ADMISSIBLE WHERE THE SOURCE STRUCTURALLY DOES NOT NAME. Ruled by the author the same day, on drawn-but-not-named, scoped. Rule 4(i) admits by naming, and Sandymount Strand on the six-inch sheet is an unbroken field of the Ordnance Survey sand stipple with no lettering in it at all — the cartographer drew the strand and did not label it, because a beach is not a street. The place is unmistakably depicted and permanently unnamed, and Thom's cannot rescue it either: the survey records that there is no street-directory entry for the Strand itself. So: where a source structurally does not name a class of subject, the position may be established by corroboration — adjacent labels, the coastline's own shape, the sheet number — and the corroboration is recorded and quoted like any other evidence, marked apart from a naming (A5). THE SCOPE IS THE RULING. It does not reach a subject the source does name and a search has not yet found. Episode 9's Kildare Street is printed on that sheet somewhere; five frames were read and it was not reached, and that slot stays not established. "The search has not got there yet" is not "the source does not do this", and admitting an inference because a search was tiring is the failure this project has logged more often than any other (learning L-38).

• A RIGHTS STATEMENT IS NOT A LICENCE — AND IT IS NOT A REFUSAL EITHER. Ruled by the author, 27 July 2026, on build panel session 08's author item 4, which was the last gate on the document layer. The bullet above refuses "no known copyright restrictions", and the Thom's Official Directory 1904 scans this project needs come from the University of Victoria, whose item metadata states a rights statement — rightsstatements.org/vocab/NKC/1.0/, "No Known Copyright" — and not a licence. Read literally, the clause refused half the document layer. The ruling separates two things the clause had run together: (i) The INSTITUTION'S STATEMENT clears nothing. It is the institution's opinion about somebody else's work, and clause 7's opening rule — an institution's terms are not the copyright — cuts both ways: their terms cannot bind us and their reassurance cannot clear us. (ii) The UNDERLYING WORK'S STATUS is what a clearance rests on, and it is established from the work, not from the host. Thom's 1904 is a corporate publication of 1904 and its own age establishes it; the evidence goes on the record with the slot, as it does for every object clearance. Why the National Library of Ireland's Flickr stream still fails, and it fails for the right reason rather than by grandfathering: those images carried no public-domain tag of any kind, and nothing established the underlying work's status. There was nothing to rest on. The Robert French photographs cleared elsewhere in this clause are cleared on his death in 1917, not on anybody's statement about them. What this does NOT relax: nothing is admitted on a host's say-so, and "it is obviously old" remains what this clause already calls it — not a licence.

• RULE 4 — ADMISSIBILITY, ORDERED, FIRST FAILURE DISPOSING. Adopted 27 July 2026 from build panel session 07, on the author's delegation "You need to resolve the goals yourself." It is the standing test for all fifty-four slots and every future frame, and it runs in this order, stopping at the first failure: (i) IDENTIFICATION. The held record, quoted, must establish that the frame shows the thing the slot names. A surveyor's hedge is not a clearance — the slot asserts identity by position, so a frame that "does not claim to be" the subject cannot be placed as it. This test runs backwards over what is already acquired, not only forwards over candidates; run backwards on 27 July it failed three of six acquired place slots (catch C-78). (ii) STRUCTURAL SURVIVAL. Street width, building line, ground plane, circulation and use still 1904's. Demolition, bombardment, fire or pedestrianisation fails absolutely. (iii) FIRST-READER MISDATING. Could a reader meeting this frame with no caption take it for 16 June 1904? Two tests are killed as standalone grounds: self-dating (a frame is not admissible merely because it announces its own period) and most-legible-object. The reason is Kiberd's inversion, which is the finding worth keeping: a monochrome 1946 street is more dangerous than a 2010 one, because bicycles and long coats read as period and invite the misdating, while a Burger King fascia cannot be mistaken for anything.

• THE REACH OF RATIFIED RULE 3, fixed permanently. A building being itself is allowed; a nameplate that does a caption's work is refused (C-64) needed a boundary, and this is it. A nameplate does a caption's work when it supplies THE BOOK'S IDENTIFICATION — Dlugacz, Davy Byrne's, the Freeman's Journal, the names Joyce put in the day. Lettering that supplies only a DATE does not engage the rule at all — Burger King, eircom, Woolworth's name nothing in the book and merely say how far we stand from it. And lettering cast into a building's own fabric stays allowed even when it names the institution the episode happens in, because the alternative is an atlas that can never show a named building, which would empty the place layer of every institution in the day. Applied against the project's own preference the first time it ran: it refused the Duke Street frame the panel's minority wanted — the only cleared frame showing the street the book names — because Davy Byrne's fascia sits legible in it, and admitted its companion instead, which shows Grafton Street and not Duke Street. Episode 8's panel therefore carries half its named subject, recorded rather than smoothed over.

• A WITHDRAWN CLEARANCE LEAVES THE TREE AND LEAVES A TOMBSTONE. Ruled by the author, 5 August 2026 — "The tombstone" — on a question the closing audit raised the same day by putting the Atlas's image family under the URL contract. If a clearance is ever contradicted after acquisition, the licensed file comes out of the tree, exactly as this clause has always required and as check.mjs §11p enforces; and because the address is contracted and docs/url-contract.md withdraws no address ever, the address keeps answering with a page naming what stood there, that its clearance was withdrawn, and the date. The rights rule and the contract stop colliding without either yielding: what a published address promises is that it answers, not that it answers with the same bytes forever. Scoped: this reaches a rights fact arriving from outside about a file already published. It is not a route for a session that has changed its mind about an image, and not a precedent for retiring a page. Nothing is built under it today — no clearance has failed.

• Cleared: period photographs attributed to Robert French (d. 1917; public domain in Ireland and the EU since 1988); contemporary photographs under CC0, CC BY or CC BY-SA, with Irish freedom of panorama (CRRA 2000 s.93) clearing the architecture; Thom's 1904 directory; the 1922 text.

• Not cleared, and dropped: any page scan of a 16 June 1904 Dublin newspaper — no free source exists, and the digitisers hold the scans even though the 1904 text is out of copyright; Joyce's handwriting and any manuscript image — the Rosenbach expressly neither grants nor denies permission, and unpublished Joyce remains in copyright in the UK until 2039; photographic portraits of Joyce (Abbott and Freund both have active licensors); the separate tram map.

• The map layer is settled, and the licence was read rather than inferred (19 July 2026, catch C-33). The Ordnance Survey six-inch Ireland sheets come from the National Library of Scotland, whose per-image tag on the Dublin sheets reads Re-use: CC-BY (NLS) and whose terms page resolves that to CC BY 4.0 — not NC, not ND, not SA. The required credit, verbatim: "Reproduced with the permission of the National Library of Scotland", plus a link to maps.nls.uk for online use. County Dublin is held in a revision surveyed 1906–1909 and printed 1909–1912, which is the layer this project wants. Two doors that look right and are not: the Digital Repository of Ireland applies its own CC BY-NC-ND reuse licence, which fails this bar, and links back to NLS anyway; and Wikimedia Commons holds no OS Ireland Dublin sheets at all. Go to NLS directly. Build constraint: the NLS IIIF server caps a request at roughly half a megapixel, so there is no whole-sheet download — the Atlas must tile at 512×512 or smaller. Corrected 25 July 2026, by testing it instead of repeating it (L-31). The conclusion holds and the number was wrong by nearly half. The server declares "maxArea": 270000 in its own info.json and enforces it to the pixel: 600×450 (270,000) returns 200, and 601×450 (270,450) returns 403. So the cap is about 0.27 megapixels, not 0.5. A 512×512 tile is 262,144 pixels — 97.1% of the cap, not half of it. The tiling conclusion survives by three per cent rather than by a factor of two, and a session that had trusted "roughly half a megapixel" would have sized its tiles at around 700×700 and collected 403s. info.json also declares tiles: [{width: 512, height: 512, …}], so 512 is the server's own tile size and departing from it buys at most 34 requests out of 759 while losing the tile-aligned cache. One sheet is 33 × 23 = 759 tiles, roughly 14 MB and about nine minutes at a polite request rate. The whole-sheet claim is confirmed: the only pre-rendered sizes offered are three thumbnails, the largest 521 px wide.

• The NLS front door is shut to any tool, and that has a consequence for this clause. maps.nls.uk sits behind an AWS WAF: every non-browser request returns HTTP 405 "Human Verification", and a real browser gets a security-check interstitial. So the per-image Re-use: CC-BY (NLS) tag recorded above could not be re-read on 25 July 2026. It was read once, on 19 July 2026, by the survey that produced catch C-33; nothing since has confirmed it, the Wayback Machine holds no capture of any Irish six-inch sheet's view page, and the archived copyright page of 3 December 2025 agrees with what is written here but is not the per-image tag. Stated as a limit rather than left implied (L-18): this clause's licence line is currently unverifiable by tool, and someone should open https://maps.nls.uk/view/247667874 by hand and read the box. Nothing is being sourced from NLS until that is done.

  THE BOX WAS READ, 26 July 2026 (catch C-75). The author opened the item page in his own browser, cleared the WAF's human-verification challenge, and read the line where this survey predicted it would sit — directly beneath the page's "Search: [Type a keyword…]" box. It reads Re-use: CC-BY (NLS). — the tag recorded above, unchanged, with CC-BY rendered as a hyperlink and NLS's own "Order this map" bar below it. Catch C-63's freeze is lifted, and it is lifted by a record rather than by this sentence: the reading lives in data/atlas-document-survey.json under osSheets.licenceReadings, every document slot's outstanding list derives from it, and §11w fails the build in both directions — if a reading is current and any slot still announces the freeze, or if a slot announces a reading no record supports. Prose thaws when somebody remembers to edit it in eighteen places, which is catch C-42; a record thaws by existing.

  Three limits travel with that reading and are part of it (A5, L-17). (1) The evidence is a screenshot crop carrying no URL bar and no sheet number, so it does not itself establish which sheet — the author was asked to open Sheet 18 at /view/247667874 and his sending the box back is his report that he read it there; that report is the source, and it is marked as his report. (2) CC-BY is a link and the link target is not in the crop, so the reading confirms the tag and not that it still resolves to 4.0 rather than another version; the 4.0 above rests on the 19 July reading and the archived copyright page, which agree. (3) The tag is per image and the Atlas would draw on ten sheets — one reading clears one sheet, and no reading is asserted for the other nine.

  A tool still cannot read it. The WAF is unchanged: re-tested 26 July 2026 from a real browser engine against both maps.nls.uk and map-view.nls.uk, and both serve the human-verification interstitial rather than the page. That is why the ruling named a person.

  And the map layer is still gated, on something other than its licence. The institutional-marks question (catch C-72, learning L-34) is unruled and reaches every institutional scan. An Ordnance Survey sheet photographed by a library is an institutional scan, and if that question goes the literal way these sheets fail on their marks whatever this clause says about their licence. Corrected 5 August 2026 by the closing audit — the question was RULED 26 July 2026 (catch C-76) and this paragraph went on announcing the gate for ten days. The author's words: "We have to accept limitations and mention them where needed and continue." An institutional mark — ownership stamp, accession number, barcode, shelfmark — on a scan of a period document is admitted and disclosed wherever it bites; a photographer's watermark stays refused, because the line the ruling draws is between a mark of ownership and a mark of authorship. The document layer was then built under that ruling (C-100), and the disclosure lives on the Atlas's own colophon. The acquisition route itself does not need the front door, and is recorded in data/atlas-document-survey.json: geoserver.nls.uk's WFS sheet index → the sheet's IMAGE id → map-view.nls.uk's IIIF endpoint. Both are ungated. A public S3 XYZ tileset of the same edition also exists at mapseries-tilesets.s3.amazonaws.com/os/dublin_2nd/{z}/{x}/{y}.png. Neither route changes the licence question; they only mean the WAF is not the obstacle.

• The object layer cleared 17 of 18 (catch C-33), each with a named licence at a named URL. The panel that could not be cleared (episode 17) is ruled to take a period document in that slot instead, which is still a second layer. Superseded 26 July 2026 (catch C-76), and this clause went on asserting the withdrawn ruling for a day. The author ruled the other way: the Buchanan report moves into the DOCUMENT layer, where it belongs, and episode 17's OBJECT slot stays empty and says so — "cleanest bookkeeping: a document recorded as a document". Leaving it in the object slot as a declared substitute would have had the manifest calling a document an object, which is the class of mislabel this project keeps catching (A5). Two live contradictory rulings sat in this file and in the manifest at once; found by build panel session 07, which read the governing document rather than the summary of it. Stated exactly, 25 July 2026: that is a ruling, not an acquisition. The manifest records a candidate and a preference among three options; no file has been fetched, and this line previously said the slot "carries" the document, in the present indicative, for a thing nobody had yet obtained.

• The object layer is acquired: 17 of 17 cleared slots hold a file. Episodes 4, 7, 8, 10, 13, 15 and 16 were cleared on 19 July and never fetched; they were fetched on 25 July, and every one of the seventeen now carries the clearance-comparison's own verdict in artifacts/atlas-of-one-day/provenance.json — the licence Commons reports today against the licence the rights survey recorded, agreeing on family and, where both state one, on version. Before that run the gate had never fired on a real image. Clearance and acquisition are two different facts and this note keeps them apart.

• Where a photograph would mislead, use a document instead. 7 Eccles Street was demolished in 1967. In a caption-free atlas a convincing photograph of the surviving terrace would read as Bloom's door and be the wrong house, and there would be no caption to say so. The Thom's 1904 entry for the address is the honest element, and the better one.

Practical rule for the build: anything on the site that is not Joyce's 1922 text or our own writing carries an attribution line. If it can't be attributed, it doesn't go up.