Ulysses design philosophy → tsumugu-ed

Verdict: tsumugu-ed already carries the spine of the Ulysses philosophy — a semantic single-source layer (entry@1, richer than Markdown XL), an html[data-*] toggle matrix that treats appearance as a swappable layer over one canonical source, and a content-first client search. The fault line is authoring versus reading: Ulysses protects an uninterrupted writer, while a learner’s interruptions (lookup, grade, reread) are the product, and our visual encodings (role color, ruby, status ramp, Hán-Việt rail) carry meaning — stripping them strips the lesson. We keep Ulysses’ recede-the-chrome and content/presentation-separation instincts and drop its strip-all-meaning minimalism. Six moves are actioned; print-as-export and a “why it’s free” manifesto are out.

Provenance: a 7-agent, ~408k-token grounded workflow — Ulysses’ philosophy web-verified, the live repos read directly (/Users/n1/Projects/tsumugu{,-ed,-wiki}), four mapping lenses, one adversarial critique that filtered the cargo-cult.

The hinge

Ulysses’ conviction is that the editor is the product — roughly a year of an 18-month v1 cycle spent on the writing surface, because that is where the work lives. Its principles follow: plain text + semantic markup (WYSIWYM), one source rendered into many themeable Export Styles, a unified Library you retrieve by content, calm focus offered rather than enforced, opinionated minimalism over a surface you furnish yourself.

We already built the half that transfers:

• entry@1 is more semantic than Markdown XL. Fields are definition / form / story / meanings / hanViet with role-tagged composition (義/聲/形/空) and grounding tags, where Ulysses has only “heading / emphasis.”
• The toggle matrix (theme · gloss · reading · 繁/简) over a stdlib idempotent renderer is Ulysses’ “appearance is a swappable layer,” done as CSS show/hide with OpenCC precomputed offline.
• The sharded CJK search retrieves by meaning, sound, or shape — the Library’s find-by-content, already shipped.

Where it inverts: Ulysses strips formatting because appearance is deferred to export; our appearance is the lesson, decided at render time. A learner rereads, looks up, and grades non-linearly, so total flow-protection works against comprehension here.

Decision ledger

Actioning (the subset Wedge committed to):

Move	Why it earns the pass	Cost
VI translations of Simplified Stories	Closes a VI-zero cell for the exact audience the bridge targets. The 982 authored Simplified stories carry EN + 簡明中文 and 0 Vietnamese. The one move that adds depth instead of re-presenting it.	Authoring capacity (loom)
Static family pages	Renders the component/phonetic graph as a navigable library instead of burying the sound-family inside one FORM card. by-component.json (2,949 keys) + facet-phonetic.json already on disk.	Medium, one render pass
Realm switcher	Three nav lines + a “full entry →” link from the reader hover popup to the stable /c/{char}.html URLs. Cheapest path to “one product” via wayfinding, no codebase or palette merge.	Small
Cmd/Ctrl-K to focus search	Matches the muscle memory the Quartz wiki already uses; the box is on every page.	Small, one listener
Facets → URL hash	activeFacet is in-memory only — reflect it into the URL and rehydrate, and every facet becomes a shareable saved view (Ulysses Filters). Redeems the browse H1 that already promises “browse by family.”	Medium
Reader Calm	A calmer reading surface; see the printing note below. Wedge is considering a white/light background for simplicity and printability, or a minimalist color experiment.	Small–medium

Out:

• Print study sheets as an export feature — exploitation risk. A “print every entry” leg makes wholesale scraping of the corpus into a redistributable compilation trivial, which is the wrong tradeoff for a free resource we have invested heavily in. The single-page printing need is met incidentally by the Reader Calm light theme (a learner prints the one page they are studying via the browser).
• Surfacing a “why it’s free” manifesto — the least interesting part. No manifestos; the values show through the content and the work itself.

Rejected as cargo-cult (from the critique, endorsed):

• ePub / Anki export legs — they advertise the content gaps (word examples 26/7,001) and break the learning round-trip; the reader already does Anki-with-audio over richer reading content.
• Token refactor to merge dictionary (Paper & Ink) and reader (wnac/Copilot) skins — large effort to erase a useful cognitive distinction (bright lookup versus immersive reading).
• A dark “ink” skin on the dictionary — solves no learner problem.
• Dictionary-side mark-known state — a second source of truth for “what does Wedge know,” exactly the duplicate/stale-state pain; the reader owns that model.
• Daily targets / streaks — anti-fit for a reference tool, and it violates “protect the student’s attention.”
• Live in-page “save as…” export preview — previews exports that should not ship.

Agreed in principle, not yet scheduled:

• Promote the mission-defining toggles (VI gloss, 繁/简, Hán-Việt) out of the 設⚙ popover into first sight — closes the discoverability gap without breaking the calm.
• A maker-facing coverage filter (words missing examples, simplified missing VI) that points the facet machinery inward to drive the loom. This is the deepest read (below).

Why VI Simplified Stories is the pick that matters

Ulysses’ real lesson for a maker: invest in the editor — the authoring pipeline — because presentation only redistributes content. Our biggest verified gap is content asymmetry (word examples 26/7,001 ≈ 0.4%, collocations ~0%, 982 Simplified stories at 0 VI). Of the six actioned moves, VI translations of Simplified Stories is the only one that adds depth rather than re-presenting it, and it fills a Vietnamese-zero cell for the audience the whole bridge exists to serve. The rejected export legs would have fanned that thinness onto more surfaces; this fills one.

Reader Calm and the printing stance

A clean light or white reader theme answers the printing need that the rejected print-export would have served: a learner prints the single page in front of them through the browser, with no feature that turns the whole corpus into an easy bulk grab. The reader is already two-layer (--wnac-* raw → --tsg-* semantic), so a warm or white variant is a token swap under one selector with no component changes. Open thread: whether the experiment lands on a warm paper tone (bridging toward the dictionary’s Paper & Ink) or a plain white minimalist surface.

Open decisions

1. Reader Calm palette: warm paper (cross-surface continuity) versus plain white (maximum minimalism / print legibility) — to be settled by experiment.
2. Sequence the six: the three small moves (realm switcher, Cmd/Ctrl-K, facets→URL) are a single low-risk pass; family pages and VI Simplified Stories are the larger bets; VI stories compete with the character backlog for loom capacity.
3. Whether the maker-facing coverage filter graduates from “agreed” to scheduled — it serves the learner indirectly but more than any export format would.

Links

• Hán-Việt bridge deep dive — the same authoring-over-presentation discipline applied to the Sino-Vietnamese bridge.
• VI translation campaign — the loom that will carry the Simplified-stories VI pass.
• Simplified edition + VN toggle — where the 982 Simplified stories were authored (EN + 簡明中文, VI left blank).
• Tsumugu Encoding Dictionary