AI Summary
One week to ACM Hypertext. Both of Frode’s papers were rejected (he accepts the reviews; the work wasn’t ready), so he’s there for a demo only — “Tim Berners-Lee was also just doing a demo.” ACM meeting tomorrow on official Origami Text EPUB versions of the proceedings; the committee has said yes in principle. Platform spread: Mac, visionOS (Apple reviewers couldn’t open an EPUB this weekend — fixed), iOS imminent, Safari plugin, web reader at futuretextlab.info, Chrome plugin likely. No Lab call Monday 14 Sept.
Mark’s ACM HT dataset. 38 years of proceedings as JSON (migrated from CSV this year), exported from his Tinderbox master; every full/short paper with at least one intra-conference citation, BibTeX nested per item, a separate author file keyed to longest-form names. Two views at shoantel.com (list with cites-left / cited-by-right on hover; TAB switches to a walkable hyperlist). Mark’s thesis, repeated three times in different forms: what breaks information systems isn’t the coding but the absence of informational expertise — nobody codes linkages nobody has identified, source-record errors from the 1990s are still live, and data work is treated as “somebody else’s job.” Frode live-imported the dataset into Origami Text via code Claude wrote; Mark: he’d waited six years for someone to do that, and it proves the constraint was data collection, not engineering.
Authoring and AI. Eric cited Engelbart’s Authorship Provisions in AUGMENT (1984). Mark’s “accidental” HT2026 paper came from noticing that Roam/Obsidian namecheck early hypertext but drop link types and link bases; his point is cherry-picking old ideas, not venerating them. Eric wants persistent 3D index cards and admitted conflict about AI writing better than he does. Peter: style ≠ authority; Ward Cunningham’s federated wiki had to partition search indexes to keep human content findable under AI-generated pages — “mad cow disease for the web.” Frode proposed a first-class AI notetype in Author: author-approved, dated, explicitly not claimed as the author’s words. Mark: LLMs have no provenance or edit history, and are “painfully inarticulate about how they work,” which is what limits his trust. Peter’s chat line: Dunning-Kruger affects LLMs too. (A small live example: Claude told Frode shoantel.com blocked automated access; Mark called it “a really interesting lie” — the data URLs were sitting in the page’s JS.)
Origami Text demo. All HT2026 TAPS bundles loaded; concept/author extraction with pin / set aside (“your collection has to be human-curated, otherwise it is not yours”); overview, inline citations, horizontal, focus, poetic and AI re-paragraphing modes; colour-by-meaning/argument as deliberate provocations; three AI tabs (proposals, issues — which Frode thinks should go, summary with names/keywords that jump to in-text occurrences); Acquire for cited-but-not-held documents. Mark: sort keywords case-insensitively; interest is a gradient, not in/out (spatial hypertext); acquisition has two levels — a properly resolved reference vs the artifact. Frode: “sometimes a digital thing really is just a pointer.”
Gabo’s Seed demo. Collaboration without central servers, Xanadu + Engelbart as north star. Key/vault identity portable across devices and spaces; every node stores both ends of a link so links can’t break and the more-cited becomes more permanent (“local collective memory”); CRDT changes; immutable versions plus dynamic latest-version links; transclusion with attribution and backlinks; republish/fork with full signed history; the protocol is the messages (“blobs”) and apps interpret them freely. Apache licensed, ~20 users. Peter raised licensing early (federated wiki forces CC BY-SA 4.0 to make forking safe); Mark: copyright is print-shaped and the real need is to quote and cite without republishing or stealing traffic. Astral asked about GPL-style backporting of modifications; Gabo: possible via backlinks, owner chooses to merge or not. Eric: Seed as a transport layer for Origami documents instead of email, with “for attention of.”
The interop proposal. Frode’s BibTeX snippet: DOI as format-independent identity of the work, then vm-source-epub, vm-source-seed, vm-source-pdf, vm-source-web so a reader chooses which edition to resolve, at paragraph resolution. Mark: annotations must be federated objects with a copy in the EPUB, not prisoners of the file; Frode: EPUBs should be immutable like PDF. Gabo’s question — how does metadata consensus work, and where does Origami-required metadata live if ACM ever fixes an official schema? Answer arrived at: as external triplet blobs/backlinks from non-owners, which Seed still needs to build. Frode’s spatial model: map: nodes / connections / views[{ref, x, y, z}], nothing else — no typeface, no orientation, stored as a list at the back of the EPUB. Multiple people’s views per document, toggleable (teacher → student headset scenario; Mariusz asked the same for iPad). Peter and Mark both pushed for relationships/semantic position over coordinates; Mark: elicitation vs elucidation, spatial hypertext belongs to the former, keep links note-to-note and flexible. Headset demo: HT2026 papers gridded, Mark’s paper with citations tailing into a floor-locked timeline.
Actions. ACM meeting Tue 8 Sept. Frode–Gabo call Tue 9:00 UK, same channel, on the VM-source fields and Seed’s API. Frode to encode Map spatial coordinates into documents this week and post the proposal for Eric (Seed). Peter to broker JSON-level interop with Ward Cunningham’s federated wiki; Gabo offered schemas and GitHub. Frode’s demo-day poster: “Origami Text — try it now.” Homework: think selfishly about how you want to author — plurality only comes from that.
Analysis
- The real outcome is a three-layer addressing model — work (DOI) → edition (
vm-source-*) → anchor (#p7/ block ID) — and it dissolves the “document vs distributed database” incompatibility Frode named at the start. It’s the graceful-degradation-of-addressability idea from 2016 finally given a wire format. Two things are unresolved: the anchor layer (the EPUB paragraph-ID ↔ Seed block-ID mapping from the 1 Sept meeting still has no rule), and whether the VM-source fields belong in BibTeX at all or only in Visual-Meta JSON — Mark’s chat note that BibTeX is a print-era format whose escaping confuses JSON readers is the warning here. - Gabo’s consensus question got a better answer than it looks. External triplet blobs mean Origami never needs ACM’s blessing for its fields, and Eric spotted that they double as per-person spatial views. That’s the first concrete mechanism for the plurality Frode keeps asking for — and it maps directly onto the
views[]array. Worth making that equivalence explicit in tomorrow’s call. - Mark’s thesis is the session’s spine and a strategic claim about the Lab. If the bottleneck is curated data and schema agreements rather than code, then Claude-in-Xcode collapses the cost of the easy part and makes the Lab’s comparative advantage the hard part: Mark’s dataset, the metadata conventions, the citation-format discipline. The demo-vs-data line (“demos no longer need to be superficial”) should probably become a stated principle.
- Coordinates vs relationships is a live disagreement, not a settled one. Frode’s minimal
{ref,x,y,z}won the room on parsimony, but Peter (“up and to the left” as semantics) and Mark (capture the relationship, not the position) are making the same objection from different angles, and Frode admitted he hasn’t thought about encoding relations.connectionsexists in the model but is untyped. Given Mark’s link-types point, an optionaltypeon connections is the cheapest concession. - Immutability tension is only half resolved. “EPUBs should not change” plus Seed’s mutable CRDT documents works if the EPUB is understood as one edition pinned to a Seed immutable-version link — but nobody said that sentence. It should go in the ACM conversation too: what ACM distributes is the frozen edition; the living graph lives elsewhere.
- Licensing was raised and dropped. Peter’s point — decide the licence before you scale — matters more once Origami EPUBs are an official ACM channel and transcluded on Seed. Mark’s rule (upload OA papers, link the rest) is the current de facto policy; it isn’t written anywhere.
- Authoring itself got one idea: the AI-note annotation type. It’s a good one — it makes provenance a document primitive rather than an etiquette — and it lines up with Peter’s index-partitioning precedent and Astral’s Internet Archive note about tracking model drift. The rest of the authoring question was deferred to homework, which is fine, but it’s the third session running where it’s been the title and not the content.