.gitignore

@@ -16,17 +16,6 @@ data/backlinks.json
 data/build-stats.json
 data/build-start.txt
 data/last-build-seconds.txt
-data/semantic-index.bin
-data/semantic-meta.json
 
 # IGNORE.txt is for the local build and need not be synced.
 IGNORE.txt
-
-# Model files for client-side semantic search (~22 MB binary artifacts).
-# Download with: make download-model
-static/models/
-
-# Generated WebP companions (produced by tools/convert-images.sh at build time).
-# To intentionally commit a WebP, use: git add -f path/to/file.webp
-static/**/*.webp
-content/**/*.webp

HOMEPAGE.md

@@ -1,129 +0,0 @@
-I think this is exactly right. Let me lay out the concrete plan.
-
-The insight driving your approach is: **the personality is the filter, and that's intentional.** The kind of collaborators, advisors, and readers you want are precisely the ones who encounter a homepage that opens with intellectual warmth and curiosity and think *yes, this is someone I want to talk to*. The professional information is there for those who need it, but it's not the first thing — it's the second or third, placed where someone who's already interested can find it instantly.
-
-Here's how I'd structure the page, top to bottom:
-
----
-
-### The Plan
-
-**1. Your name.** Large, in Spectral. Not centered — left-aligned, consistent with how essay titles appear on the rest of the site. No subtitle, no description line. Just the name. The site header and nav sit above it as usual.
-
-**2. The introduction.** Two to four paragraphs of prose. This is the heart of the homepage and the thing that does the most work. It should be written the way you'd introduce yourself to someone interesting at a dinner party — not "I am a computer scientist" but something closer to what's already there, except tighter and more specific. The current introduction is good but it has two problems: the first sentence ("You have reached the website of...") is a web convention from 2003 that undersells you, and the middle veers into site-navigation explanation ("This website is organized broadly by Portals and Tags...") which is functional information masquerading as prose. Strip the navigational explanation entirely — readers will figure out portals and tags by using them, and if they don't, the search works. Replace it with something that communicates what you actually care about, what you're working on right now, what this place *is*. Your Me page demonstrates that you write about yourself with genuine flair. Channel that here, compressed.
-
-The Latin welcome at the end of the current introduction (*Te accipio, hospes benignus*) is a lovely touch and should stay — it's exactly the kind of earned ornament that signals personality. But it should be the capstone of the introduction, not buried after a navigation guide.
-
-**3. The professional row.** A single horizontal line of compact links, visually quiet, in Fira Sans at a slightly smaller size than body text. Something like:
-
-> Biography · CV · Email · GitHub · ORCID · GPG
-
-No labels explaining what these are. Anyone who needs your CV knows what "CV" means. This row is for the professor who just reviewed your paper and wants to know more, the potential collaborator who met you at a conference, the PhD committee member checking your background. They get what they need in a glance, without the homepage making a big deal about it.
-
-Styled as a single line with middot separators, perhaps in smallcaps or with slightly muted color. It should be clearly present but visually subordinate to the prose above and the portals below. Think of it as a utility strip.
-
-**4. The curiosities row.** A second horizontal line, same visual treatment as the professional row, linking to the unusual features of the site — the things that make levineuwirth.org distinctive and that signal to a certain kind of visitor that they're in the right place:
-
-> Memento Mori · Build · Commonplace · Colophon · Library · Random
-
-These are the easter eggs promoted to the front door. Someone who sees "Memento Mori" and "Build Telemetry" on a personal homepage and clicks them is exactly the kind of person you want reading your site. These links serve as a personality signal as much as a navigation element — they say *this site has layers, and the author thinks about infrastructure, mortality, and curation*.
-
-This row should be visually parallel to the professional row — same size, same treatment — so that together they read as two lines of a compact directory. The professional row is "here's how to reach me in the conventional world." The curiosities row is "here's what makes this place unusual."
-
-**5. Portal links.** Below both rows, after some breathing room, the portals. Not as cards. As a simple vertical list or a wrapped horizontal line, each portal as a plain text link. If you want brief annotations (and I think you should, because several portal names are ambiguous to a first-time visitor), they should be very short — five to ten words, in muted text:
-
-> Research — formal inquiry and open problems
-> Nonfiction — essays, criticism, living documents
-> Fiction — stories and a novel in progress
-> Poetry — verse, formal and free
-> Music — compositions, scores, and recordings
-> AI — on intelligence, artificial and otherwise
-> Tech — systems, tools, and craft
-> Miscellany — everything that defies category
-
-Each annotation is just enough to tell a new visitor what they'll find behind the door, without turning the portal list into a card grid. The annotations should be in your voice — "everything that defies category" is better than "blog posts and other content."
-
-**6. A "Recently" section (optional, add when corpus supports it).** Below the portals, 3–5 most recently published or substantially revised items. Title, date, portal tag. Auto-populated by Hakyll `recentFirst`. This is Direction B from your HOMEPAGE.md — the heartbeat that rewards returning visitors. I'd defer this until you have enough content that the list changes meaningfully between visits (probably 8–10 published pieces). When it's ready, it goes here.
-
-**7. Footer.** As it currently is — the build timestamp, the license, the sig. No change needed.
-
----
-
-### What's removed
-
-- The card grid. Gone entirely.
-- The "How to navigate this site" collapsible (already removed, stays removed).
-- The contact row as a separate visual element at the bottom. The contact links are absorbed into the professional row (item 3).
-- The "Random Page" card. Random moves to the curiosities row, where it fits perfectly.
-- The "About" card. The Me page is linked from the nav and from the Biography link in the professional row.
-
-### What's new
-
-- The professional row (compact, one line, visually quiet).
-- The curiosities row (compact, one line, same treatment).
-- Portal annotations (very short, in your voice).
-- The introduction is rewritten to be tighter and more personal.
-
-### What's unchanged
-
-- Your name at the top.
-- The nav bar and portal dropdown.
-- The footer.
-- The overall monochrome palette and typographic system.
-
----
-
-### Visual rhythm of the page
-
-```
-[nav bar with portals dropdown and settings]
-
-Levi Neuwirth                              ← large, Spectral
-
-Two to four paragraphs of introduction     ← body text, Spectral
-in your voice. What you care about,        
-what you're working on, what this place    
-is. Te accipio, hospes benignus.           
-
-Biography · CV · Email · GitHub ·          ← compact row, Fira Sans, muted
-ORCID · GPG                                
-
-Memento Mori · Build · Commonplace ·       ← compact row, same treatment
-Colophon · Library · Random                
-
-                                           ← breathing room
-
-Research — formal inquiry                  ← portal list, Fira Sans
-Nonfiction — essays and living documents   
-Fiction — stories and a novel              
-Poetry — verse, formal and free            
-Music — compositions and scores            
-AI — on intelligence                       
-Tech — systems and craft                   
-Miscellany — everything else               
-
-                                           ← (future: Recently section)
-
-[footer: license, build, sig]
-```
-
-The entire page fits on one screen at desktop width (assuming 3–4 paragraphs of introduction). There is nothing to scroll past, nothing to decode, no interface to learn. A reader arrives, encounters your voice, sees where to find professional information, notices some intriguing links, and chooses a portal. The page takes about 45 seconds to read, which is exactly right for a homepage.
-
----
-
-### On the transition toward G
-
-This design degrades gracefully into Direction G over time. As your reputation grows and the corpus deepens, you can:
-
-1. Shorten the introduction to one paragraph, then to one sentence, then to nothing.
-2. Drop the portal annotations as readers learn what each portal contains.
-3. Eventually remove the professional row (it moves to the Me page permanently).
-4. What remains: your name, the curiosities row (which becomes a signature element of the site), and the portal list.
-
-Each step is a subtraction, and each subtraction signals growing confidence. The infrastructure supports all of these states without any engineering changes — it's just editing `index.md` and the homepage template.
-
----
-
-### The one thing I'd encourage you to write first
-
-The introduction. Not a draft — the real thing. Sit down, write it as if you're explaining to a curious stranger at a dinner party what this website is and why it exists, and what's on your mind right now. Don't worry about whether it's "good enough for a homepage." The Me page proves you can write this kind of thing with warmth and depth. The introduction is the same skill, compressed. Once that prose exists, the rest of the homepage design falls into place around it — it's just CSS and template work.

Makefile

@@ -1,4 +1,4 @@
-.PHONY: build deploy sign download-model convert-images pdf-thumbs watch clean dev
+.PHONY: build deploy sign watch clean dev
 
 # Source .env for GITHUB_TOKEN and GITHUB_REPO if it exists.
 # .env format: KEY=value (one per line, no `export` prefix, no quotes needed).
@@ -9,8 +9,6 @@ build:
 	@git add content/
 	@git diff --cached --quiet || git commit -m "auto: $$(date -u +%Y-%m-%dT%H:%M:%SZ)"
 	@date +%s > data/build-start.txt
-	@./tools/convert-images.sh
-	@$(MAKE) -s pdf-thumbs
 	cabal run site -- build
 	pagefind --site _site
 	@if [ -d .venv ]; then \
@@ -26,33 +24,6 @@ build:
 sign:
 	@./tools/sign-site.sh
 
-# Download the quantized ONNX model for client-side semantic search.
-# Run once; files are gitignored. Safe to re-run (skips existing files).
-download-model:
-	@./tools/download-model.sh
-
-# Convert JPEG/PNG images to WebP companions (also runs automatically in build).
-# Requires cwebp: pacman -S libwebp  /  apt install webp
-convert-images:
-	@./tools/convert-images.sh
-
-# Generate first-page thumbnails for PDFs in static/papers/ (also runs in build).
-# Requires pdftoppm: pacman -S poppler  /  apt install poppler-utils
-# Thumbnails are written as static/papers/foo.thumb.png alongside each PDF.
-# Skipped silently when pdftoppm is not installed or static/papers/ is empty.
-pdf-thumbs:
-	@if command -v pdftoppm >/dev/null 2>&1; then \
-	  find static/papers -name '*.pdf' 2>/dev/null | while read pdf; do \
-	    thumb="$${pdf%.pdf}.thumb"; \
-	    if [ ! -f "$${thumb}.png" ] || [ "$$pdf" -nt "$${thumb}.png" ]; then \
-	      echo "  pdf-thumb $$pdf"; \
-	      pdftoppm -r 100 -f 1 -l 1 -png -singlefile "$$pdf" "$$thumb"; \
-	    fi; \
-	  done; \
-	else \
-	  echo "pdf-thumbs: pdftoppm not found — install poppler (skipping)"; \
-	fi
-
 deploy: build sign
 	@if [ -z "$(GITHUB_TOKEN)" ] || [ -z "$(GITHUB_REPO)" ]; then \
 	  echo "Skipping GitHub push: set GITHUB_TOKEN and GITHUB_REPO in .env"; \

WRITING.md

@@ -47,9 +47,6 @@ authors:                  # optional; overrides the default "Levi Neuwirth" link
   - "Levi Neuwirth | /me.html"
   - "Collaborator | https://their.site"
   - "Plain Name"          # URL optional; omit for plain-text credit
-affiliation:              # optional; shown below author in metadata block
-  - "Brown University | https://cs.brown.edu"
-  - "Some Research Lab"   # URL optional; scalar string also accepted
 further-reading:          # optional; see Citations section
   - someKey
   - anotherKey
@@ -272,59 +269,11 @@ not derivable from the page title.
 
 ---
 
-## PDF Embeds
-
-Embed a hosted PDF in a full PDF.js viewer (page navigation, zoom, text
-selection) using the `{{pdf:...}}` directive on its own line:
-
-```markdown
-{{pdf:/papers/smith2023.pdf}}
-{{pdf:/papers/smith2023.pdf#5}}        ← open at page 5 (bare integer)
-{{pdf:/papers/smith2023.pdf#page=5}}   ← explicit form, same result
-```
-
-The directive produces an iframe pointing at the vendored PDF.js viewer at
-`/pdfjs/web/viewer.html?file=...`.  The PDF must be served from the same
-origin (i.e. it must be a file you host).
-
-**Storing papers.** Drop PDFs in `static/papers/`; Hakyll's static rule copies
-everything under `static/` to `_site/` unchanged.  Reference them as
-`/papers/filename.pdf`.
-
-**One-time vendor setup.** PDF.js is not included in the repo.  Install it once:
-
-```bash
-npm install pdfjs-dist
-mkdir -p static/pdfjs
-cp -r node_modules/pdfjs-dist/web   static/pdfjs/
-cp -r node_modules/pdfjs-dist/build static/pdfjs/
-rm -rf node_modules package-lock.json
-```
-
-Then commit `static/pdfjs/` (it is static and changes only when you want to
-upgrade PDF.js).  Hakyll's existing `static/**` rule copies it through without
-any new build rules.
-
-**Optional: disable the "Open file" button** in the viewer.  Edit
-`static/pdfjs/web/viewer.html` and set `AppOptions.set('disableOpenFile', true)`
-in the `webViewerLoad` callback, or add a thin CSS rule to `viewer.css`:
-
-```css
-#openFile, #secondaryOpenFile { display: none !important; }
-```
-
----
-
 ## Math
 
-Pandoc parses LaTeX math and wraps it in `class="math inline"` / `class="math display"`
+KaTeX renders client-side from raw LaTeX. CSS and JS are loaded conditionally
-spans. KaTeX CSS is loaded conditionally on pages that contain math — this styles the
+on pages that have `math: true` set in their context (all essays and posts have
-pre-rendered output. Client-side KaTeX JS rendering is not yet loaded; complex math
+this by default).
-will appear as LaTeX source. Build-time server-side rendering is planned but not yet
-implemented. Simple math currently renders through Pandoc's built-in KaTeX span output.
-
-`math: true` is auto-set for all essays and blog posts. Standalone pages that use
-math must set it explicitly in frontmatter to load the KaTeX CSS.
 
 | Syntax | Usage |
 |--------|-------|
@@ -722,19 +671,19 @@ The [CPU]{.smallcaps} handles this.
 
 Wrap any paragraph in a `::: dropcap` fenced div to get a drop cap regardless
 of its position in the document. The first line is automatically rendered in
-small caps via `::first-line { font-variant-caps: small-caps }`.
+small caps.
 
 ```markdown
 ::: dropcap
-A personal website is not a publication. It is a position — something you
+COMPOSITION [IS]{.smallcaps} PERHAPS MORE THAN ANYTHING ELSE THE PRACTICE OF MY
-inhabit, argue from, and occasionally revise in public.
+LIFE. I say these strong words because I feel strongly about this process.
 :::
 ```
 
-Write in normal mixed case. The CSS applies `font-variant-caps: small-caps` to
+The opening word (or words, before a space) should be written in ALL CAPS in
-the entire first rendered line, converting lowercase letters to small-cap glyphs.
+source — they will render as small caps via `::first-line`. The `[IS]{.smallcaps}`
-Use `[WORD]{.smallcaps}` spans to force specific words into small-caps anywhere
+span is not strictly necessary but can force specific words into the smallcaps
-in the paragraph.
+run if needed.
 
 A paragraph that immediately follows a `::: dropcap` block will be indented
 correctly (`text-indent: 1.5em`), matching the paragraph-after-paragraph rule.

build/Contexts.hs

@@ -28,37 +28,6 @@ import SimilarLinks (similarLinksField)
 import Stability    (stabilityField, lastReviewedField, versionHistoryField)
 import Tags       (tagLinksField)
 
--- ---------------------------------------------------------------------------
--- Affiliation field
--- ---------------------------------------------------------------------------
-
--- | Parses the @affiliation@ frontmatter key and exposes each entry as
---   @affiliation-name@ / @affiliation-url@ pairs.
---
---   Accepts a scalar string or a YAML list. Each entry may use pipe syntax:
---     @"Brown University | https://cs.brown.edu"@
---   Entries without a URL still produce a row; @affiliation-url@ fails
---   (evaluates to noResult), so @$if(affiliation-url)$@ works in templates.
---
---   Usage:
---     $for(affiliation-links)$
---       $if(affiliation-url)$<a href="$affiliation-url$">$affiliation-name$</a>
---       $else$$affiliation-name$$endif$$sep$ · $endfor$
-affiliationField :: Context a
-affiliationField = listFieldWith "affiliation-links" ctx $ \item -> do
-    meta <- getMetadata (itemIdentifier item)
-    let entries = case lookupStringList "affiliation" meta of
-                    Just xs -> xs
-                    Nothing -> maybe [] (:[]) (lookupString "affiliation" meta)
-    return $ map (Item (fromFilePath "") . parseEntry) entries
-  where
-    ctx = field "affiliation-name" (return . fst . itemBody)
-       <> field "affiliation-url"  (\i -> let u = snd (itemBody i)
-                                           in if null u then noResult "no url" else return u)
-    parseEntry s = case break (== '|') s of
-        (name, '|' : url) -> (trim name, trim url)
-        (name, _)         -> (trim name, "")
-
 -- ---------------------------------------------------------------------------
 -- Build time field
 -- ---------------------------------------------------------------------------
@@ -177,37 +146,11 @@ confidenceTrendField = field "confidence-trend" $ \item -> do
                         | otherwise  -> return "\x2192"   -- →
                     _ -> return "\x2192"
 
--- | @$overall-score$@: weighted composite of confidence (50 %),
---   evidence quality (30 %), and importance (20 %), expressed as an
---   integer on a 0–100 scale.
---   Returns @noResult@ when any contributing field is absent, so
---   @$if(overall-score)$@ guards the template safely.
---
---   Formula:  raw   = conf/100·0.5 + ev/5·0.3 + imp/5·0.2   (0–1)
---             score = clamp₀₋₁₀₀(round(raw · 100))
-overallScoreField :: Context String
-overallScoreField = field "overall-score" $ \item -> do
-    meta <- getMetadata (itemIdentifier item)
-    let readInt s = readMaybe s :: Maybe Int
-    case ( readInt =<< lookupString "confidence" meta
-         , readInt =<< lookupString "evidence"   meta
-         , readInt =<< lookupString "importance" meta
-         ) of
-        (Just conf, Just ev, Just imp) ->
-            let raw :: Double
-                raw   = fromIntegral conf / 100.0 * 0.5
-                      + fromIntegral ev  / 5.0   * 0.3
-                      + fromIntegral imp / 5.0   * 0.2
-                score = max 0 (min 100 (round (raw * 100.0) :: Int))
-            in  return (show score)
-        _ -> fail "overall-score: confidence, evidence, or importance not set"
-
 -- | All epistemic context fields composed.
 epistemicCtx :: Context String
 epistemicCtx =
     dotsField "importance-dots" "importance"
     <> dotsField "evidence-dots" "evidence"
-    <> overallScoreField
     <> confidenceTrendField
     <> stabilityField
     <> lastReviewedField
@@ -219,7 +162,6 @@ epistemicCtx =
 essayCtx :: Context String
 essayCtx =
     authorLinksField
-    <> affiliationField
     <> snapshotField "toc"          "toc"
     <> snapshotField "word-count"   "word-count"
     <> snapshotField "reading-time" "reading-time"
@@ -242,7 +184,6 @@ essayCtx =
 postCtx :: Context String
 postCtx =
     authorLinksField
-    <> affiliationField
     <> backlinksField
     <> similarLinksField
     <> dateField "date"     "%-d %B %Y"
@@ -255,7 +196,7 @@ postCtx =
 -- ---------------------------------------------------------------------------
 
 pageCtx :: Context String
-pageCtx = authorLinksField <> affiliationField <> siteCtx
+pageCtx = authorLinksField <> siteCtx
 
 -- ---------------------------------------------------------------------------
 -- Reading contexts (fiction + poetry)

build/Filters.hs

@@ -15,8 +15,6 @@ import qualified Filters.Smallcaps  as Smallcaps
 import qualified Filters.Dropcaps   as Dropcaps
 import qualified Filters.Math       as Math
 import qualified Filters.Wikilinks  as Wikilinks
-import qualified Filters.Transclusion  as Transclusion
-import qualified Filters.EmbedPdf      as EmbedPdf
 import qualified Filters.Code       as Code
 import qualified Filters.Images     as Images
 
@@ -34,7 +32,6 @@ applyAll
     . Images.apply
 
 -- | Apply source-level preprocessors to the raw Markdown string.
---   Order matters: EmbedPdf must run before Transclusion, because the
+--   Run before 'readPandocWith'.
---   transclusion parser would otherwise treat {{pdf:...}} as a broken slug.
 preprocessSource :: String -> String
-preprocessSource = Transclusion.preprocess . EmbedPdf.preprocess . Wikilinks.preprocess
+preprocessSource = Wikilinks.preprocess

build/Filters/EmbedPdf.hs

@@ -1,81 +0,0 @@
-{-# LANGUAGE GHC2021 #-}
--- | Source-level preprocessor for inline PDF embeds.
---
---   Rewrites block-level @{{pdf:...}}@ directives to raw HTML that renders the
---   named file inside a vendored PDF.js viewer iframe.
---
---   Syntax (must be the sole content of a line after trimming):
---
--- > {{pdf:/papers/foo.pdf}}          — embed from page 1
--- > {{pdf:/papers/foo.pdf#5}}        — start at page 5  (bare integer)
--- > {{pdf:/papers/foo.pdf#page=5}}   — start at page 5  (explicit form)
---
---   The file path must be root-relative (begins with @/@).
---   PDF.js is expected to be vendored at @/pdfjs/web/viewer.html@.
-module Filters.EmbedPdf (preprocess) where
-
-import Data.Char (isDigit)
-import Data.List (isPrefixOf, isSuffixOf)
-
--- | Apply PDF-embed substitution to the raw Markdown source string.
-preprocess :: String -> String
-preprocess = unlines . map processLine . lines
-
-processLine :: String -> String
-processLine line =
-    case parseDirective (trim line) of
-        Nothing                    -> line
-        Just (filePath, pageHash)  -> renderEmbed filePath pageHash
-
--- | Parse a @{{pdf:/path/to/file.pdf}}@ or @{{pdf:/path.pdf#N}}@ directive.
---   Returns @(filePath, pageHash)@ where @pageHash@ is either @""@ or @"#page=N"@.
-parseDirective :: String -> Maybe (String, String)
-parseDirective s
-    | not ("{{pdf:" `isPrefixOf` s) = Nothing
-    | not ("}}"     `isSuffixOf` s) = Nothing
-    | otherwise =
-        let inner        = take (length s - 2) (drop 6 s)  -- strip "{{pdf:" and "}}"
-            (path, frag) = break (== '#') inner
-        in  if null path
-                then Nothing
-                else Just (path, parsePageHash frag)
-
--- | Convert the fragment part of the directive (e.g. @#5@ or @#page=5@) to a
---   PDF.js-compatible @#page=N@ hash, or @""@ if absent/invalid.
-parsePageHash :: String -> String
-parsePageHash ('#' : rest)
-    | "page=" `isPrefixOf` rest =
-        let n = takeWhile isDigit (drop 5 rest)
-        in  if null n then "" else "#page=" ++ n
-    | all isDigit rest && not (null rest) = "#page=" ++ rest
-parsePageHash _ = ""
-
--- | Render the HTML for a PDF embed.
-renderEmbed :: String -> String -> String
-renderEmbed filePath pageHash =
-    let viewerUrl = "/pdfjs/web/viewer.html?file=" ++ encodeQueryValue filePath ++ pageHash
-    in  "<div class=\"pdf-embed-wrapper\">"
-     ++ "<iframe class=\"pdf-embed\""
-     ++ " src=\"" ++ viewerUrl ++ "\""
-     ++ " title=\"PDF document\""
-     ++ " loading=\"lazy\""
-     ++ " allowfullscreen></iframe>"
-     ++ "</div>"
-
--- | Percent-encode characters that would break a query-string value.
---   Slashes are left unencoded so root-relative paths remain readable and
---   work correctly with PDF.js's internal fetch.
-encodeQueryValue :: String -> String
-encodeQueryValue = concatMap enc
-  where
-    enc ' ' = "%20"
-    enc '&' = "%26"
-    enc '?' = "%3F"
-    enc '+' = "%2B"
-    enc '"' = "%22"
-    enc c   = [c]
-
--- | Strip leading and trailing spaces.
-trim :: String -> String
-trim = f . f
-  where f = reverse . dropWhile (== ' ')

build/Filters/Images.hs

@@ -1,23 +1,23 @@
 {-# LANGUAGE GHC2021 #-}
 {-# LANGUAGE OverloadedStrings #-}
--- | Image filter: lazy loading, lightbox markers, and WebP <picture> wrappers.
+-- | Image attribute filter.
 --
---   For local raster images (JPG, JPEG, PNG, GIF), emits a @<picture>@ element
+--   Walks all @Image@ inlines and:
---   with a WebP @<source>@ and the original format as the @<img>@ fallback.
+--   * Adds @loading="lazy"@ to every image.
---   tools/convert-images.sh produces the companion .webp files at build time.
+--   * Adds @data-lightbox="true"@ to images that are NOT already wrapped in
+--     a @Link@ inline (i.e. the image is not itself a hyperlink).
 --
---   SVG files and external URLs are passed through with only lazy loading
+--   The wrapping-link check is done by walking the document with two passes:
---   (and lightbox markers for standalone images).
+--   a block-level walk that handles the common @Link [Image …] …@ pattern,
+--   and a plain image walk that stamps @loading="lazy"@ on everything else.
 module Filters.Images (apply) where
 
-import           Data.Char            (toLower)
 import           Data.Text            (Text)
 import qualified Data.Text            as T
-import           System.FilePath      (replaceExtension)
 import           Text.Pandoc.Definition
 import           Text.Pandoc.Walk     (walk)
 
--- | Apply image attribute injection and WebP wrapping to the entire document.
+-- | Apply image attribute injection to the entire document.
 apply :: Pandoc -> Pandoc
 apply = walk transformInline
 
@@ -25,118 +25,32 @@ apply = walk transformInline
 -- Core transformation
 -- ---------------------------------------------------------------------------
 
+-- | Process a single inline node.
+--
+--   * @Link … [Image …] …@  — image inside a link: add only @loading="lazy"@.
+--   * @Image …@             — standalone image: add both @loading="lazy"@ and
+--                             @data-lightbox="true"@.
+--   * Anything else         — pass through unchanged.
 transformInline :: Inline -> Inline
 transformInline (Link lAttr ils lTarget) =
-    -- Recurse into link contents; images inside a link get no lightbox marker.
+    -- Recurse into link contents, but mark any images inside as linked
-    Link lAttr (map wrapLinkedImg ils) lTarget
+    -- (so they receive lazy loading only, no lightbox marker).
+    Link lAttr (map (addLazyOnly) ils) lTarget
   where
-    wrapLinkedImg (Image iAttr alt iTarget) = renderImg iAttr alt iTarget False
+    addLazyOnly (Image iAttr alt iTarget) =
-    wrapLinkedImg x                         = x
+        Image (addAttr "loading" "lazy" iAttr) alt iTarget
+    addLazyOnly x = x
 transformInline (Image attr alt target) =
-    renderImg attr alt target True
+    Image (addAttr "data-lightbox" "true" (addAttr "loading" "lazy" attr)) alt target
 transformInline x = x
 
--- | Dispatch on image type:
---   * Local raster → @<picture>@ with WebP @<source>@
---   * Everything else → plain @<img>@ with loading/lightbox attrs
-renderImg :: Attr -> [Inline] -> Target -> Bool -> Inline
-renderImg attr alt target@(src, _) lightbox
-    | isLocalRaster (T.unpack src) =
-        RawInline (Format "html") (renderPicture attr alt target lightbox)
-    | otherwise =
-        Image (addLightbox lightbox (addAttr "loading" "lazy" attr)) alt target
-  where
-    addLightbox True  a = addAttr "data-lightbox" "true" a
-    addLightbox False a = a
-
 -- ---------------------------------------------------------------------------
--- <picture> rendering
+-- Attribute helpers
 -- ---------------------------------------------------------------------------
 
--- | Emit a @<picture>@ element with a WebP @<source>@ and an @<img>@ fallback.
+-- | Prepend a key=value pair to an @Attr@'s key-value list (if not already
-renderPicture :: Attr -> [Inline] -> Target -> Bool -> Text
+--   present, to avoid duplicating attributes that come from Markdown).
-renderPicture (ident, classes, kvs) alt (src, title) lightbox =
-    T.concat
-        [ "<picture>"
-        , "<source srcset=\"", T.pack webpSrc, "\" type=\"image/webp\">"
-        , "<img"
-        , attrId ident
-        , attrClasses classes
-        , " src=\"",     esc src,                "\""
-        , attrAlt alt
-        , attrTitle title
-        , " loading=\"lazy\""
-        , if lightbox then " data-lightbox=\"true\"" else ""
-        , renderKvs passedKvs
-        , ">"
-        , "</picture>"
-        ]
-  where
-    webpSrc   = replaceExtension (T.unpack src) ".webp"
-    -- Strip attrs we handle explicitly so they don't appear twice.
-    passedKvs = filter (\(k, _) -> k `notElem` ["loading", "data-lightbox"]) kvs
-
-attrId :: Text -> Text
-attrId t = if T.null t then "" else " id=\"" <> esc t <> "\""
-
-attrClasses :: [Text] -> Text
-attrClasses [] = ""
-attrClasses cs = " class=\"" <> T.intercalate " " (map esc cs) <> "\""
-
-attrAlt :: [Inline] -> Text
-attrAlt ils = let t = stringify ils
-              in  if T.null t then "" else " alt=\"" <> esc t <> "\""
-
-attrTitle :: Text -> Text
-attrTitle t = if T.null t then "" else " title=\"" <> esc t <> "\""
-
-renderKvs :: [(Text, Text)] -> Text
-renderKvs = T.concat . map (\(k, v) -> " " <> k <> "=\"" <> esc v <> "\"")
-
--- ---------------------------------------------------------------------------
--- Helpers
--- ---------------------------------------------------------------------------
-
--- | True for local (non-URL) images with a raster format we can convert.
-isLocalRaster :: FilePath -> Bool
-isLocalRaster src = not (isUrl src) && lowerExt src `elem` [".jpg", ".jpeg", ".png", ".gif"]
-
-isUrl :: String -> Bool
-isUrl s = any (`isPrefixOf` s) ["http://", "https://", "//", "data:"]
-  where isPrefixOf pfx str = take (length pfx) str == pfx
-
--- | Extension of a path, lowercased (e.g. ".JPG" → ".jpg").
-lowerExt :: FilePath -> String
-lowerExt = map toLower . reverse . ('.' :) . takeWhile (/= '.') . tail . dropWhile (/= '.') . reverse
-
--- | Prepend a key=value pair if not already present.
 addAttr :: Text -> Text -> Attr -> Attr
-addAttr k v (i, cs, kvs)
+addAttr k v (ident, classes, kvs)
-    | any ((== k) . fst) kvs = (i, cs, kvs)
+    | any ((== k) . fst) kvs = (ident, classes, kvs)
-    | otherwise               = (i, cs, (k, v) : kvs)
+    | otherwise               = (ident, classes, (k, v) : kvs)
-
--- | Plain-text content of a list of inlines (for alt text).
-stringify :: [Inline] -> Text
-stringify = T.concat . map go
-  where
-    go (Str t)         = t
-    go Space           = " "
-    go SoftBreak       = " "
-    go LineBreak       = " "
-    go (Emph ils)      = stringify ils
-    go (Strong ils)    = stringify ils
-    go (Code _ t)      = t
-    go (Link _ ils _)  = stringify ils
-    go (Image _ ils _) = stringify ils
-    go (Span _ ils)    = stringify ils
-    go _               = ""
-
--- | HTML-escape a text value for use in attribute values.
-esc :: Text -> Text
-esc = T.concatMap escChar
-  where
-    escChar '&'  = "&amp;"
-    escChar '<'  = "&lt;"
-    escChar '>'  = "&gt;"
-    escChar '"'  = "&quot;"
-    escChar c    = T.singleton c

build/Filters/Links.hs

@@ -16,25 +16,8 @@ import           Text.Pandoc.Definition
 import           Text.Pandoc.Walk     (walk)
 
 -- | Apply link classification to the entire document.
---   Two passes: PDF links first (rewrites href to viewer URL), then external
---   link classification (operates on http/https, so no overlap).
 apply :: Pandoc -> Pandoc
-apply = walk classifyLink . walk classifyPdfLink
+apply = walk classifyLink
-
--- | Rewrite root-relative PDF links to open via the vendored PDF.js viewer.
---   Preserves the original path in @data-pdf-src@ so the popup thumbnail
---   provider can locate the corresponding @.thumb.png@ file.
---   Skips links that are already pointing at the viewer (idempotent).
-classifyPdfLink :: Inline -> Inline
-classifyPdfLink (Link (ident, classes, kvs) ils (url, title))
-    | "/" `T.isPrefixOf` url
-    , ".pdf" `T.isSuffixOf` T.toLower url
-    , "pdf-link" `notElem` classes =
-        let viewerUrl = "/pdfjs/web/viewer.html?file=" <> encodeQueryValue url
-            classes'  = classes ++ ["pdf-link"]
-            kvs'      = kvs ++ [("data-pdf-src", url)]
-        in  Link (ident, classes', kvs') ils (viewerUrl, title)
-classifyPdfLink x = x
 
 classifyLink :: Inline -> Inline
 classifyLink (Link (ident, classes, kvs) ils (url, title))
@@ -66,16 +49,3 @@ domainIcon url
     | "doi.org"       `T.isInfixOf` url = "doi"
     | "github.com"    `T.isInfixOf` url = "github"
     | otherwise                          = "external"
-
--- | Percent-encode characters that would break a @?file=@ query-string value.
---   Slashes are intentionally left unencoded so root-relative paths remain
---   readable and work correctly with PDF.js's internal fetch.
-encodeQueryValue :: Text -> Text
-encodeQueryValue = T.concatMap enc
-  where
-    enc ' ' = "%20"
-    enc '&' = "%26"
-    enc '?' = "%3F"
-    enc '+' = "%2B"
-    enc '"' = "%22"
-    enc c   = T.singleton c

build/Filters/Transclusion.hs

@@ -1,60 +0,0 @@
-{-# LANGUAGE GHC2021 #-}
--- | Source-level transclusion preprocessor.
---
---   Rewrites block-level {{slug}} and {{slug#section}} directives to raw
---   HTML placeholders that transclude.js resolves at runtime.
---
---   A directive must be the sole content of a line (after trimming) to be
---   replaced — this prevents accidental substitution inside prose or code.
---
---   Examples:
---     {{my-essay}}              → full-page transclusion of /my-essay.html
---     {{essays/deep-dive}}      → /essays/deep-dive.html (full body)
---     {{my-essay#introduction}} → section "introduction" of /my-essay.html
-module Filters.Transclusion (preprocess) where
-
-import Data.List (isSuffixOf, isPrefixOf, stripPrefix)
-
--- | Apply transclusion substitution to the raw Markdown source string.
-preprocess :: String -> String
-preprocess = unlines . map processLine . lines
-
-processLine :: String -> String
-processLine line =
-    case parseDirective (trim line) of
-        Nothing             -> line
-        Just (url, secAttr) ->
-            "<div class=\"transclude\" data-src=\"" ++ url ++ "\""
-            ++ secAttr ++ "></div>"
-
--- | Parse a {{slug}} or {{slug#section}} directive.
---   Returns (absolute-url, section-attribute-string) or Nothing.
-parseDirective :: String -> Maybe (String, String)
-parseDirective s = do
-    inner <- stripPrefix "{{" s >>= stripSuffix "}}"
-    case break (== '#') inner of
-        ("", _)            -> Nothing
-        (slug, "")         -> Just (slugToUrl slug, "")
-        (slug, '#' : sec)
-            | null sec     -> Just (slugToUrl slug, "")
-            | otherwise    -> Just (slugToUrl slug,
-                                    " data-section=\"" ++ sec ++ "\"")
-        _                  -> Nothing
-
--- | Convert a slug (possibly with leading slash, possibly with path segments)
---   to a root-relative .html URL.
-slugToUrl :: String -> String
-slugToUrl slug
-    | "/" `isPrefixOf` slug = slug ++ ".html"
-    | otherwise              = "/" ++ slug ++ ".html"
-
--- | Strip a suffix from a string, returning Nothing if not present.
-stripSuffix :: String -> String -> Maybe String
-stripSuffix suf str
-    | suf `isSuffixOf` str = Just (take (length str - length suf) str)
-    | otherwise             = Nothing
-
--- | Strip leading and trailing spaces.
-trim :: String -> String
-trim = f . f
-  where f = reverse . dropWhile (== ' ')

build/Site.hs

@@ -18,14 +18,6 @@ import Tags       (buildAllTags, applyTagRules)
 import Pagination (blogPaginateRules)
 import Stats      (statsRules)
 
--- Poems inside collection subdirectories, excluding their index pages.
-collectionPoems :: Pattern
-collectionPoems = "content/poetry/*/*.md" .&&. complement "content/poetry/*/index.md"
-
--- All poetry content (flat + collection), excluding collection index pages.
-allPoetry :: Pattern
-allPoetry = "content/poetry/*.md" .||. collectionPoems
-
 feedConfig :: FeedConfiguration
 feedConfig = FeedConfiguration
     { feedTitle       = "Levi Neuwirth"
@@ -152,17 +144,9 @@ rules = do
             >>= loadAndApplyTemplate "templates/default.html"     commonplaceCtx
             >>= relativizeUrls
 
-    match "content/colophon.md" $ do
-        route   $ constRoute "colophon.html"
-        compile $ essayCompiler
-            >>= loadAndApplyTemplate "templates/essay.html"   essayCtx
-            >>= loadAndApplyTemplate "templates/default.html" essayCtx
-            >>= relativizeUrls
-
     match ("content/*.md"
             .&&. complement "content/index.md"
-            .&&. complement "content/commonplace.md"
+            .&&. complement "content/commonplace.md") $ do
-            .&&. complement "content/colophon.md") $ do
         route   $ gsubRoute "content/" (const "")
                   `composeRoutes` setExtension "html"
         compile $ pageCompiler
@@ -197,7 +181,6 @@ rules = do
     -- ---------------------------------------------------------------------------
     -- Poetry
     -- ---------------------------------------------------------------------------
-    -- Flat poems (e.g. content/poetry/sonnet-60.md)
     match "content/poetry/*.md" $ do
         route   $ gsubRoute "content/poetry/" (const "poetry/")
                   `composeRoutes` setExtension "html"
@@ -207,24 +190,6 @@ rules = do
             >>= loadAndApplyTemplate "templates/default.html"  poetryCtx
             >>= relativizeUrls
 
-    -- Collection poems (e.g. content/poetry/shakespeare-sonnets/sonnet-1.md)
-    match collectionPoems $ do
-        route   $ gsubRoute "content/poetry/" (const "poetry/")
-                  `composeRoutes` setExtension "html"
-        compile $ poetryCompiler
-            >>= saveSnapshot "content"
-            >>= loadAndApplyTemplate "templates/reading.html"  poetryCtx
-            >>= loadAndApplyTemplate "templates/default.html"  poetryCtx
-            >>= relativizeUrls
-
-    -- Collection index pages (e.g. content/poetry/shakespeare-sonnets/index.md)
-    match "content/poetry/*/index.md" $ do
-        route   $ gsubRoute "content/poetry/" (const "poetry/")
-                  `composeRoutes` setExtension "html"
-        compile $ pageCompiler
-            >>= loadAndApplyTemplate "templates/default.html"  pageCtx
-            >>= relativizeUrls
-
     -- ---------------------------------------------------------------------------
     -- Fiction
     -- ---------------------------------------------------------------------------
@@ -326,7 +291,7 @@ rules = do
                     essays  <- loadAll ("content/essays/*.md"  .&&. hasNoVersion)
                     posts   <- loadAll ("content/blog/*.md"    .&&. hasNoVersion)
                     fiction <- loadAll ("content/fiction/*.md" .&&. hasNoVersion)
-                    poetry  <- loadAll (allPoetry             .&&. hasNoVersion)
+                    poetry  <- loadAll ("content/poetry/*.md"  .&&. hasNoVersion)
                     filtered <- filterM (hasPortal p) (essays ++ posts ++ fiction ++ poetry)
                     recentFirst filtered
 
@@ -372,7 +337,7 @@ rules = do
                                 (   (    "content/essays/*.md"
                                     .||. "content/blog/*.md"
                                     .||. "content/fiction/*.md"
-                                    .||. allPoetry
+                                    .||. "content/poetry/*.md"
                                     .||. "content/music/*/index.md"
                                     )
                                 .&&. hasNoVersion

build/Stats.hs

@@ -8,14 +8,12 @@ module Stats (statsRules) where
 
 import Control.Exception          (IOException, catch)
 import Control.Monad              (forM)
-import Data.List                  (find, isSuffixOf, sort, sortBy)
+import Data.List                  (find, isSuffixOf, sortBy)
 import qualified Data.Map.Strict  as Map
 import Data.Maybe                 (catMaybes, fromMaybe, isJust, listToMaybe)
 import Data.Ord                   (comparing, Down (..))
 import qualified Data.Set         as Set
-import Data.Time                  (getCurrentTime, formatTime, defaultTimeLocale,
+import Data.Time                  (getCurrentTime, formatTime, defaultTimeLocale)
-                                   Day, parseTimeM, utctDay, addDays, diffDays)
-import Data.Time.Calendar         (toGregorian, dayOfWeek)
 import System.Directory           (doesDirectoryExist, getFileSize, listDirectory)
 import System.Exit                (ExitCode (..))
 import System.FilePath            (takeExtension, (</>))
@@ -111,197 +109,6 @@ normUrl u
     | ".html" `isSuffixOf` u = take (length u - 5) u
     | otherwise              = u
 
-pad2 :: (Show a, Integral a) => a -> String
-pad2 n = if n < 10 then "0" ++ show n else show n
-
--- | Median of a non-empty list; returns 0 for empty.
-median :: [Int] -> Int
-median [] = 0
-median xs = let s = sort xs in s !! (length s `div` 2)
-
--- ---------------------------------------------------------------------------
--- Date helpers (for /stats/ page)
--- ---------------------------------------------------------------------------
-
-parseDay :: String -> Maybe Day
-parseDay = parseTimeM True defaultTimeLocale "%Y-%m-%d"
-
--- | First Monday on or before 'day' (start of its ISO week).
-weekStart :: Day -> Day
-weekStart day = addDays (fromIntegral (negate (fromEnum (dayOfWeek day)))) day
-
--- | Intensity class for the heatmap (hm0 … hm4).
-heatClass :: Int -> String
-heatClass 0             = "hm0"
-heatClass n | n < 500  = "hm1"
-heatClass n | n < 2000 = "hm2"
-heatClass n | n < 5000 = "hm3"
-heatClass _             = "hm4"
-
-shortMonth :: Int -> String
-shortMonth m = case m of
-    1 -> "Jan"; 2 -> "Feb";  3 -> "Mar"; 4 -> "Apr"
-    5 -> "May"; 6 -> "Jun";  7 -> "Jul"; 8 -> "Aug"
-    9 -> "Sep"; 10 -> "Oct"; 11 -> "Nov"; 12 -> "Dec"
-    _ -> ""
-
--- ---------------------------------------------------------------------------
--- Heatmap SVG
--- ---------------------------------------------------------------------------
-
--- | 52-week writing activity heatmap (inline SVG, CSS-variable colors).
-renderHeatmap :: Map.Map Day Int -> Day -> String
-renderHeatmap wordsByDay today =
-    let cellSz  = 10 :: Int
-        gap     = 2  :: Int
-        step    = cellSz + gap
-        hdrH    = 22 :: Int       -- vertical space for month labels
-        nWeeks  = 52
-        -- First Monday of the 52-week window
-        startDay = addDays (fromIntegral (-(nWeeks - 1)) * 7) (weekStart today)
-        nDays    = diffDays today startDay + 1
-        allDays  = [addDays i startDay | i <- [0 .. nDays - 1]]
-        weekOf d = fromIntegral (diffDays d startDay `div` 7) :: Int
-        dowOf  d = fromEnum (dayOfWeek d)        -- Mon=0..Sun=6
-        svgW     = (nWeeks - 1) * step + cellSz
-        svgH     = 6 * step + cellSz + hdrH
-
-        -- Month labels: one per first-of-month day
-        monthLbls = concatMap (\d ->
-            let (_, mo, da) = toGregorian d
-            in  if da == 1
-                then "<text class=\"hm-lbl\" x=\"" ++ show (weekOf d * step)
-                  ++ "\" y=\"14\">" ++ shortMonth mo ++ "</text>"
-                else "") allDays
-
-        -- One rect per day
-        cells = concatMap (\d ->
-            let wc         = fromMaybe 0 (Map.lookup d wordsByDay)
-                (yr, mo, da) = toGregorian d
-                x          = weekOf d * step
-                y          = dowOf d * step + hdrH
-                tip        = show yr ++ "-" ++ pad2 mo ++ "-" ++ pad2 da
-                          ++ if wc > 0 then ": " ++ commaInt wc ++ " words" else ""
-            in  "<rect class=\"" ++ heatClass wc ++ "\""
-             ++ " x=\"" ++ show x ++ "\" y=\"" ++ show y ++ "\""
-             ++ " width=\"" ++ show cellSz ++ "\" height=\"" ++ show cellSz ++ "\""
-             ++ " rx=\"2\"><title>" ++ tip ++ "</title></rect>") allDays
-
-        -- Inline legend (five sample rects)
-        legendW = 5 * step - gap
-        legendSvg =
-            "<svg width=\"" ++ show legendW ++ "\" height=\"" ++ show cellSz ++ "\""
-         ++ " viewBox=\"0 0 " ++ show legendW ++ " " ++ show cellSz ++ "\""
-         ++ " style=\"display:inline;vertical-align:middle\">"
-         ++ concatMap (\i ->
-                "<rect class=\"hm" ++ show i ++ "\""
-             ++ " x=\"" ++ show (i * step) ++ "\" y=\"0\""
-             ++ " width=\"" ++ show cellSz ++ "\" height=\"" ++ show cellSz ++ "\""
-             ++ " rx=\"2\"/>") [0..4]
-         ++ "</svg>"
-
-    in  "<figure class=\"stats-heatmap\">"
-     ++ "<svg width=\"" ++ show svgW ++ "\" height=\"" ++ show svgH ++ "\""
-     ++ " viewBox=\"0 0 " ++ show svgW ++ " " ++ show svgH ++ "\""
-     ++ " class=\"heatmap-svg\" role=\"img\""
-     ++ " aria-label=\"52-week writing activity heatmap\">"
-     ++ "<style>"
-     ++ ".hm0{fill:var(--hm-0)}.hm1{fill:var(--hm-1)}.hm2{fill:var(--hm-2)}"
-     ++ ".hm3{fill:var(--hm-3)}.hm4{fill:var(--hm-4)}"
-     ++ ".hm-lbl{font-size:9px;fill:var(--text-faint);font-family:sans-serif}"
-     ++ "</style>"
-     ++ monthLbls ++ cells
-     ++ "</svg>"
-     ++ "<figcaption class=\"heatmap-legend\">"
-     ++ "Less\xA0" ++ legendSvg ++ "\xA0More"
-     ++ "</figcaption>"
-     ++ "</figure>"
-
--- ---------------------------------------------------------------------------
--- Stats page sections
--- ---------------------------------------------------------------------------
-
-renderMonthlyVolume :: Map.Map Day Int -> String
-renderMonthlyVolume wordsByDay =
-    section "volume" "Monthly volume" $
-    let byMonth = Map.fromListWith (+)
-            [ ((y, m), wc)
-            | (day, wc) <- Map.toList wordsByDay
-            , let (y, m, _) = toGregorian day
-            ]
-    in  if Map.null byMonth
-        then "<p><em>No dated content yet.</em></p>"
-        else
-            let maxWC = max 1 $ maximum $ Map.elems byMonth
-                bar (y, m) =
-                    let wc  = fromMaybe 0 (Map.lookup (y, m) byMonth)
-                        pct = if wc == 0 then 0 else max 2 (wc * 100 `div` maxWC)
-                        lbl = shortMonth m ++ " \x2019" ++ drop 2 (show y)
-                    in  "<div class=\"build-bar-row\">"
-                     ++ "<span class=\"build-bar-label\">" ++ lbl ++ "</span>"
-                     ++ "<span class=\"build-bar-wrap\"><span class=\"build-bar\" style=\"width:"
-                     ++ show pct ++ "%\"></span></span>"
-                     ++ "<span class=\"build-bar-count\">"
-                     ++ (if wc > 0 then commaInt wc else "") ++ "</span>"
-                     ++ "</div>"
-            in  "<div class=\"build-bars\">" ++ concatMap bar (Map.keys byMonth) ++ "</div>"
-
-renderCorpus :: [TypeRow] -> [PageInfo] -> String
-renderCorpus typeRows allPIs =
-    section "corpus" "Corpus" $ concat
-    [ dl [ ("Total words",       commaInt totalWords)
-         , ("Total pages",       commaInt (length allPIs))
-         , ("Total reading time", rtStr totalWords)
-         , ("Average length",    commaInt avgWC ++ " words")
-         , ("Median length",     commaInt medWC ++ " words")
-         ]
-    , table ["Type", "Pages", "Words", "Reading time"]
-        (map row typeRows)
-        (Just ["Total", commaInt (sum (map trCount typeRows))
-              , commaInt totalWords, rtStr totalWords])
-    ]
-  where
-    hasSomeWC  = filter (\p -> piWC p > 0) allPIs
-    totalWords = sum (map trWords typeRows)
-    avgWC = if null hasSomeWC then 0 else totalWords `div` length hasSomeWC
-    medWC = median (map piWC hasSomeWC)
-    row r = [trLabel r, commaInt (trCount r), commaInt (trWords r), rtStr (trWords r)]
-
-renderNotable :: [PageInfo] -> String
-renderNotable allPIs =
-    section "notable" "Notable" $ concat
-    [ "<p><strong>Longest</strong></p>"
-    , pageList (take 5 (sortBy (comparing (Down . piWC)) hasSomeWC))
-    , "<p><strong>Shortest</strong></p>"
-    , pageList (take 5 (sortBy (comparing piWC) hasSomeWC))
-    ]
-  where
-    hasSomeWC = filter (\p -> piWC p > 50) allPIs
-    pageList ps = "<ol class=\"build-page-list\">"
-        ++ concatMap (\p -> "<li>" ++ link (piUrl p) (piTitle p)
-                         ++ " \x2014 " ++ commaInt (piWC p) ++ " words</li>") ps
-        ++ "</ol>"
-
-renderStatsTags :: [(String, Int)] -> Int -> String
-renderStatsTags topTags uniqueCount =
-    section "tags" "Tags" $ concat
-    [ dl [("Unique tags", commaInt uniqueCount)]
-    , table ["Tag", "Items"] (map row topTags) Nothing
-    ]
-  where row (t, n) = [link ("/" ++ t ++ "/") t, show n]
-
-statsTOC :: String
-statsTOC = "<ol>\n" ++ concatMap item entries ++ "</ol>\n"
-  where
-    item (i, t) = "<li><a href=\"#" ++ i ++ "\" data-target=\"" ++ i ++ "\">"
-               ++ t ++ "</a></li>\n"
-    entries = [ ("activity", "Writing activity")
-              , ("volume",   "Monthly volume")
-              , ("corpus",   "Corpus")
-              , ("notable",  "Notable")
-              , ("tags",     "Tags")
-              ]
-
 -- ---------------------------------------------------------------------------
 -- IO: output directory walk
 -- ---------------------------------------------------------------------------
@@ -573,11 +380,7 @@ pageTOC = "<ol>\n" ++ concatMap item sections ++ "</ol>\n"
 -- ---------------------------------------------------------------------------
 
 statsRules :: Tags -> Rules ()
-statsRules tags = do
+statsRules tags =
-
-  -- -------------------------------------------------------------------------
-  -- Build telemetry page (/build/)
-  -- -------------------------------------------------------------------------
     create ["build/index.html"] $ do
         route idRoute
         compile $ do
@@ -727,73 +530,3 @@ statsRules tags = do
                 >>= loadAndApplyTemplate "templates/essay.html"   ctx
                 >>= loadAndApplyTemplate "templates/default.html" ctx
                 >>= relativizeUrls
-
-  -- -------------------------------------------------------------------------
-  -- Writing statistics page (/stats/)
-  -- -------------------------------------------------------------------------
-  create ["stats/index.html"] $ do
-        route idRoute
-        compile $ do
-            essays  <- loadAll ("content/essays/*.md"      .&&. hasNoVersion)
-            posts   <- loadAll ("content/blog/*.md"        .&&. hasNoVersion)
-            poems   <- loadAll ("content/poetry/*.md"      .&&. hasNoVersion)
-            fiction <- loadAll ("content/fiction/*.md"     .&&. hasNoVersion)
-            comps   <- loadAll ("content/music/*/index.md" .&&. hasNoVersion)
-
-            essayWCs   <- mapM loadWC essays
-            postWCs    <- mapM loadWC posts
-            poemWCs    <- mapM loadWC poems
-            fictionWCs <- mapM loadWC fiction
-            compWCs    <- mapM loadWC comps
-
-            let allItems = essays ++ posts ++ poems ++ fiction ++ comps
-                typeRows =
-                    [ TypeRow "Essays"       (length essays)  (sum essayWCs)
-                    , TypeRow "Blog posts"   (length posts)   (sum postWCs)
-                    , TypeRow "Poems"        (length poems)   (sum poemWCs)
-                    , TypeRow "Fiction"      (length fiction) (sum fictionWCs)
-                    , TypeRow "Compositions" (length comps)   (sum compWCs)
-                    ]
-
-            allPIs <- catMaybes <$> mapM loadPI allItems
-
-            -- Build wordsByDay: for each item with a parseable `date`, map that
-            -- day to the item's word count (summing if multiple items share a date).
-            datePairs <- fmap catMaybes $ forM allItems $ \item -> do
-                meta <- getMetadata (itemIdentifier item)
-                wc   <- loadWC item
-                return $ case lookupString "date" meta >>= parseDay of
-                    Nothing -> Nothing
-                    Just d  -> Just (d, wc)
-            let wordsByDay = Map.fromListWith (+) datePairs
-
-            let tagFreqs   = map (\(t, ids) -> (t, length ids)) (tagsMap tags)
-                topTags    = take 15 (sortBy (comparing (Down . snd)) tagFreqs)
-                uniqueTags = length tagFreqs
-
-            today <- unsafeCompiler (utctDay <$> getCurrentTime)
-
-            let content = concat
-                    [ section "activity" "Writing activity" (renderHeatmap wordsByDay today)
-                    , renderMonthlyVolume wordsByDay
-                    , renderCorpus typeRows allPIs
-                    , renderNotable allPIs
-                    , renderStatsTags topTags uniqueTags
-                    ]
-                plainText = stripHtmlTags content
-                wc        = length (words plainText)
-                rt        = readingTime plainText
-                ctx       = constField "toc"          statsTOC
-                         <> constField "word-count"   (show wc)
-                         <> constField "reading-time" (show rt)
-                         <> constField "title"        "Writing Statistics"
-                         <> constField "abstract"     "Writing activity, corpus breakdown, \
-                                                      \and tag distribution — computed at build time."
-                         <> constField "build"        "true"
-                         <> authorLinksField
-                         <> siteCtx
-
-            makeItem content
-                >>= loadAndApplyTemplate "templates/essay.html"   ctx
-                >>= loadAndApplyTemplate "templates/default.html" ctx
-                >>= relativizeUrls

cabal.project.freeze

@@ -6,7 +6,7 @@ constraints: any.Glob ==0.10.2,
              any.Only ==0.1,
              any.QuickCheck ==2.15.0.1,
              any.StateVar ==1.2.2,
-             any.aeson ==2.2.1.0,
+             any.aeson ==2.2.0.0,
              any.aeson-pretty ==0.8.10,
              any.ansi-terminal ==1.1,
              any.ansi-terminal-types ==1.1,

content/colophon.md

@@ -10,8 +10,10 @@ abstract: On the design, tools, and philosophy of this site — and by extension
 ::: dropcap
 A personal website is not a publication. It is a position. A publication presents work
 in a finalized, immutable state, and carries with it some sort of declaration - "this is my most polished and prized work!"; 
-a position is something you inhabit, argue from, and continuously
+a position is something you inhabit, argue from, and occasionally
-revise in public. This page explains the design decisions forming my broader **position** and why they took the form they did.
+revise in public. Every design decision on this site — the typeface, the build system,
+the absence of tracking, the presence of confidence scores — encodes a value. This page
+explains those values, and why they took the form they did.
 
 What follows is a colophon in the grand old sense: a printer's note at the end of the book,
 recording how it was made, who made it, etc. The difference: here, the
@@ -24,83 +26,117 @@ of argument, but *the only* form of argument permitted.
 ## Typography
 
 ::: dropcap
-You are reading this sentence in SPECTRAL, which is not only a font with particular personal importance to me, but also an exceedingly pleasing font to read.  [OpenType]{.smallcaps} features — `smcp`, `onum`, `liga`, `calt` are used throughout the website, which necessitates our self-hosting setup.^[Google Fonts strips OpenType features during
+You are reading this sentence in SPECTRAL, which is not only a font with particular personal importance to me, but also an exceedingly pleasing font to read.
+:::
+
+[TODO: The [OpenType]{.smallcaps} features — `smcp`, `onum`, `liga`, `calt` — and why
+self-hosting is required to use them.]^[Google Fonts strips OpenType features during
 subsetting for bandwidth. Self-hosting with `pyftsubset` preserves `smcp` (small
 capitals), `onum` (old-style figures), `liga` (common ligatures), and the full optical
 size range. The difference is visible: old-style figures sit on the baseline rather
 than hanging above it, small capitals are drawn to match the x-height rather than being
 shrunken full caps, and ligatures prevent collisions in letter pairs like *fi* and *fl*.]
-:::
 
-The UI and headers are Fira Sans. Variation is good, and moreover, humanist sans are rather ubiquitous (we have Frutiger to thank for this fact!) - perhaps I am making some type of statement by not choosing one of the more corporation variations of it, like the dreaded Calibri and Tahoma you might recognize from Microslop (formerly known as *Microsoft*) products. Code uses Jetbrains Mono, which is simply the font that I use within my editor. Code should look like code, simple as that.
+[TODO: Fira Sans for [UI]{.smallcaps} and headers. What humanist warmth means in a
+sans-serif and why it pairs well with Spectral. JetBrains Mono for code — code should
+look like code.]
 
-The monochrome palette is an application of restraint grounded in my studies of Tools for Thought. 
+The monochrome palette is a discipline. Color is often used to do work that typography
-Color is often used to do work that typography
+should do: distinguishing hierarchy, marking emphasis, signaling state. When those
-should do, such as demonstrating hierarchy, creating emphasis, etc. When those
+functions are handled by weight, size, and spacing instead, color becomes available
-functions are handled by weight, size, and spacing instead^[Color and saturation/hue are actually well known to be less effective than other means of distinguishment. I refer you to Tufte's *The Visual Display of Quantitative Information* for more.], color becomes available for the things it cannot be substituted for — and on a site with no data visualizations
+for the things it cannot be substituted for — and on a site with no data visualizations
 requiring color encoding, those things turn out to be very few.^[The one exception is
 the heatmap on the statistics page, which uses a four-step greyscale scale. Even there,
 the encoding is luminance rather than hue.]
+
 ---
 
 ## The Build
 
-This is a [Static Website](https://en.wikipedia.org/wiki/Static_web_page). For the purposes of this website, though the content is highly dynamic and iterated upon, the medium of expression is rather stable. There are numerous advantages to using a static webpage, many of which are focused at the Hetzner box from which this webpage is served. I use [Hakyll](https://jaspervdj.be/hakyll/) for reasons of performance, extensibility, and, of course given the underlying language Haskell, elegance! I had been wanting to do a project in Haskell ever since I took my undergraduate programming languages course,^[The programming languages course at Brown, somewhat infamous, is taught entirely with Racket, which is essentially a dialect of Lisp. The course itself is extremely focused within the functional paradigm as far as implementations go. I am aware that Racket itself, curiously, has some means by which static webpages can be built - the course infrastructure was produced this way, and this website has in a few places taken minor inspiration from it!] and Hakyll was more extensible and thus suitable than the alternative I was looking at most strongly, Hugo (in Go, a language with which I am intimately familiar). The philosophy of a static website is that the website is a program and the content is the source code^[Source code here **chiefly distinct** from mere markup language, like HTML.]. The step of compilation present in Haskell, which is outlined below, means that what you have here received in your browser is not merely a runtime rendering decision, but rather a deterministic artifact. By this step of compilation, the [Markdown](https://en.wikipedia.org/wiki/Markdown) in which I write these webpages is transformed to exactly what you currently see.
+::: dropcap
+[TODO: Hakyll and Pandoc. Why Haskell. The compile step as philosophy — this site is
+a program and the content is the source code. The analogy to compiled software: readers
+receive a deterministic artifact, not a runtime rendering decision.]
+:::
 
-The [AST](https://en.wikipedia.org/wiki/Abstract_syntax_tree) we use is heavily customized and modified. The chain is roughly markdown -> pandoc -> citations -> wikilinks -> preprocessing -> sidenotes -> smallcaps and dropcaps -> links -> images -> math. Mathematics with LaTeX requires a second pass and is rendered at build-time with KaTeX - no math rendering occurs in your browser. Samples from music are displayed as SVGs, generally typeset with Lilypond through some helper scripts I wrote to automate the process.
+[TODO: What Pandoc [AST]{.smallcaps} filters do. The filter chain: wikilinks →
+preprocessor, then sidenotes, smallcaps, dropcaps, links, images, math. All
+transformation before the reader arrives.]^[The full pipeline: Markdown source →
+Pandoc [AST]{.smallcaps} → citeproc → filter chain (sidenotes, smallcaps, dropcaps,
+wikilinks, external link classification, responsive images) → HTML. Complex math goes
+through a second pass: KaTeX renders server-side at build time, producing static
+[HTML]{.smallcaps}+MathML. No math rendering happens in the browser.]
 
-The semantic search model is a particularly intriguing aspect of the website. The model used is self-hosted, with weights served from the same origin. There are NO external API calls when you use this, in contrast to just about every other similar feature on other websites. This is essential for the privacy model that this site strives to achieve - see **Design Decisions** for more.
+[TODO: The semantic search model — self-hosted [ONNX]{.smallcaps} weights served
+same-origin, no external [API]{.smallcaps} calls. Why this matters for the site's
+privacy posture.]
 
-A full accounting of what this build process has actually produced is available at the [[Build]] page. It is generated automatically at each compile: corpus word counts, length distributions, tag frequencies, link analysis, epistemic coverage, repository metrics, and build timing — all computed from the same source tree that produces the content. Think of it as the build system reporting on itself.
+[TODO: Build telemetry — pointer to `/build/` and `/stats/` for the empirical picture.
-
+This section covers intent; those pages cover state.]
-{{build}}
 
 ---
 
 ## The Computing Environment
 
 ::: dropcap
-Every tool I use was chosen rather than accepted. This distinction matters: the default path through the software ecosystem, particularly for developers, routes through choices made by convenience, employer mandate, or inertia rather than deliberation. I have consistently refused that path — not out of contrarianism, but because I find that the tools one uses become constitutive of the work one produces. This site is downstream of those choices in every particular.
+[TODO: The organizing principle for this section: configuration is code, privacy is
+a first principle, tools should be earned rather than merely used. These are not
+preferences but positions, and this site is downstream of them.]
 :::
 
 ### Desktop
 
-My primary desktop is a rig I built myself, running **Gentoo Linux** with **Hyprland** and a custom shell, *Levshell*, implemented with [Quickshell](https://github.com/quickshell-mirror/quickshell). The reasons for Gentoo are worth stating explicitly, since the choice is frequently met with bewilderment:
+[TODO: Gentoo Linux, Hyprland, Levshell (custom shell via Quickshell). Why Gentoo:
-
+compilation for performance, fine-grained [USE]{.smallcaps} flag control, the
-- Compiling software from source delivers measurable performance increases — not marginal ones.
+community. Why Hyprland: tiling window management as a productivity commitment.
-- It provides fine-grained control over software configuration via [USE]{.smallcaps} flags, linking options, and the like. This matters for a machine tuned to a high degree of specificity.
+[AMD]{.smallcaps} hardware throughout.]^[The personal computing setup is documented
-- It is, in my experience, the best-maintained Linux distribution I have ever used.
+in greater depth on the [[Me]] page. This section focuses on what is directly
-- The community is phenomenal.
+relevant to how this site is built and maintained.]
-
-I have strong hardware preferences to match: [AMD]{.smallcaps} processors and graphics throughout. This is not tribal loyalty — I have used multiple distinct products from each major vendor and formed the preference empirically.
 
 ### Laptop
 
-For mobile computing, I use a [P]{.smallcaps}-series ThinkPad running **Arch Linux** — the same Hyprland environment, the same *Levshell*, the same muscle memory, but without the compilation overhead. Gentoo is impractical on battery-constrained hardware: compilation times are simply too long, and running the compiler continuously on battery is not a reasonable tradeoff. Arch is a sound alternative. Portage is better than pacman, but pacman on a laptop is entirely correct.
+[TODO: Arch Linux on ThinkPad P-series. Why Arch rather than Gentoo on battery-
-
+constrained hardware. The continuity of the Hyprland environment across machines —
-The Hyprland environment runs identically on both machines. This is deliberate — I do not want to maintain two workflows. Every keybinding, every visual detail, every tool behaves the same. The continuity is itself a productivity commitment.
+the same keybindings, the same visual language, the same muscle memory.]
 
 ### Editor
 
-Everything on this site — every word of prose, every line of Haskell, every [CSS]{.smallcaps} rule — was written in **Emacs**. I use Emacs for the same reason I use Gentoo: it is configurable to a degree that other editors treat as pathological. Running `make watch` alongside an Emacs buffer gives something close to a [WYSIWYG]{.smallcaps} experience for this workflow: save the file, the Hakyll watcher recompiles, the browser updates. The loop is tight enough that the editor and the rendered output feel nearly continuous.
+[TODO: Emacs. Everything written and coded in Emacs. The relationship between the
+editor and the content — writing Markdown in Emacs with a Hakyll watcher running is
+as close to a [WYSIWYG]{.smallcaps} experience as the workflow gets.]
 
 ### Privacy-First Computing
 
-My email and [VPN]{.smallcaps} are self-hosted; I use Thunderbird as a client for the former. For browsing I use [LibreWolf](https://librewolf.net/), not Firefox: the Chromium monopoly and Mozilla's evident incompetence at browser development are, to me, equally concerning developments, and LibreWolf is the most coherent response to both. My phone runs [GrapheneOS](https://grapheneos.org/) — the only reasonably secure and private option for a mobile device, and one whose restrictions are, frankly, a feature rather than a limitation.
+[TODO: Self-hosted email and [VPN]{.smallcaps}. LibreWolf over Firefox (the Chromium
-
+monopoly and Mozilla's drift). GrapheneOS. The principle: privacy as a first principle
-The principle underlying all of these choices is the same one underlying the site's **No Tracking** policy: privacy is an architectural decision, not a settings toggle. You build the infrastructure first. Bolting on privacy after the fact, whether in a browser or on a website, is not privacy — it is the appearance of privacy.
+means building the infrastructure first, not bolting on settings after. The same
+principle applies to this site — no tracking is not a policy decision, it is an
+architectural one.]
 
 ---
 
 ## [AI]{.smallcaps} and This Site
 
 ::: dropcap
-I will never use AI to write, whether for my personal communications with anyone or for pieces on this website. I take this extremely seriously - writing is religious in severity to me. The writing on this website is wholly human and wholly my own, to the extent that any writing can be.
+This site was built in collaboration with Claude. The Pandoc filters, the [CSS]{.smallcaps},
+the JavaScript, the Hakyll build system — much of the infrastructure emerged from an
+iterative dialogue between author and model. This is worth stating plainly, because
+the site's content argues for a specific position on [AI]{.smallcaps} use, and that
+position is most legible when the colophon is honest about where it was and was not applied.
 :::
 
-Much of the code that comprises the build system of this website was created in collaboration with AI. Rather than "vibe coding" proper, this was the result of an intensive engineering process where AI and I were equals in collaboration. Notably, all of the major architectural choices, design decisions, idiosyncracies, and elements of the tech stack were chosen entirely by me, and AI systems were only used to automate production of some (but not all) of the code that was required. 
+[TODO: The augmentation/automation distinction made concrete. Building the
+infrastructure that hosts original thought is different from generating the thought.
+The Pandoc filter that transforms footnotes into sidenotes was written with Claude;
+the footnotes themselves were not. The CSS that sets the typeface was written with
+Claude; the prose set in that typeface was not. The distinction is the point.]
 
-The commit history, of course, is available for you to view and licensed accordingly - see **No Tracking** for more.
+[TODO: What this means for the site's claims to originality. The content — the essays,
+the compositions, the poetry, the research — is entirely Levi's. The medium was built
+collaboratively. This is the same relationship a writer has with an editor, a compositor,
+or a press operator: the vehicle is shared work; the voice is not.]^[The model used
+throughout is Claude (Anthropic). The full git history is publicly available and
+records what was changed when, which is a more honest account than any prose summary.]
 
 ---
 
@@ -108,58 +144,61 @@ The commit history, of course, is available for you to view and licensed accordi
 
 ### Sidenotes
 
-The sidenotes are provided by a JavaScript file that was forked from the website of Gwern Branwen and authored by
+[TODO: The Tufte influence — *Envisioning Information*, the three-column layout, the
-Said Achmiz; I have simplified the script to fit the needs of this website and made some minor modifications.
+argument that footnotes interrupt reading in a way that marginalia does not. The
-
+sidenote as a first-class typographic element rather than a concession to citation
+convention. Credit: Said Achmiz for the sidenote positioning [JS]{.smallcaps} (the
+implementation used here is a simplified fork).]
 
 ### No Tracking
 
 The site has no analytics, no visit counters, no fingerprinting, and no third-party
 scripts.^[This is enforced at the nginx level via [CSP]{.smallcaps} headers, not just
 by omission. The Content Security Policy prevents any script not explicitly whitelisted
-from executing. The whitelist is short.] The Hetzner VPS that provides this content runs
+from executing. The whitelist is short.] The reason is not only privacy, though privacy
-only open source software, and my machines use *almost exclusively*^[It is nearly impossible to run an entirely free system, but in approximation, it is actually wonderfully easy.] the same. The code is licensed under MIT and hosted
+is sufficient. The deeper reason is that knowing your readership changes what you write
-on a public repository at this domain with a [GitHub mirror](https://github.com/levineuwirth/levineuwirth.org); you are welcome
+for it — gradually, invisibly, in the direction of whatever the metrics reward. A site
-to inspect it, fork it, or, more broadly, do whatever you please with it. 
+without metrics cannot optimize for them. This is a feature.
 
 ### Living Documents
 
-The dominant convention of academic and professional publication is that a document, once released, is finished. It carries an implicit claim: *this is what I think, full stop.*^[This is particularly problematic in academia, where there is a long tradition of researchers whose work was eventually disproven taking an extreme defensive stance, usually rooted in [confirmation bias](https://en.wikipedia.org/wiki/Confirmation_bias).] I find this convention dishonest in proportion to how seldom it is actually true. Thinking is continuous; positions shift; evidence accumulates; people change their minds and rarely say so in public. This site operates under a different premise, one that I strive to operate all of my life under.
+[TODO: The epistemic metadata system — `status`, `confidence`, `modified`, `importance`.
-
+What each field means. Why marking `status: Ongoing` on a document matters: it is an
-Every essay and post on this site carries an **epistemic footer** — a structured block that reports my current relationship to the work. The footer only appears when a `status` field is set in the document's frontmatter; standalone pages and very short items omit it. The full set of fields:
+honest claim about the document's relationship to the author's current thinking, rather
-
+than the implicit permanence that publication usually signals. The modified date is
-- **Status** — a controlled vocabulary describing where the work stands: *Draft*, *Working model*, *Durable*, *Refined*, *Superseded*, or *Deprecated*. A document marked *Working model* is not just unfinished — it is a position I currently hold but would not stake much on. A document marked *Durable* is something I expect to hold up. *Superseded* means I wrote a better version; *Deprecated* means I no longer endorse it.
+not decorative; it is the reader's primary signal for whether to treat a document as
-
+settled or in flux.]
-- **Confidence** — an integer from 0–100, representing my credence in the central thesis. Not false precision: a rough honest assessment is more useful than no assessment at all. When a `confidence-history` list is present, a trend arrow (↑ ↓ →) is derived automatically from the last two entries — so you can see not just *what* I think but whether I am growing more or less confident over time.
-
-- **Importance** — how much I think this matters, on a 1–5 dot scale (●●●○○). Useful for orienting a reader who has limited time.
-
-- **Evidence** — how well-evidenced the claims are, on the same 1–5 scale. An essay with high importance and low evidence is a speculative position and should be read accordingly.
-
-- **Overall score** — a single 0–100 integer derived automatically from the three quantitative fields above: confidence weighted at 50%, evidence at 30%, importance at 20%. It appears adjacent to the Epistemic link in the top metadata row and is not entered manually.
-
-- **Scope**, **Novelty**, **Practicality** — optional qualitative fields in the expanded footer. *Scope* ranges from *personal* to *civilizational*; *novelty* from *conventional* to *innovative*; *practicality* from *abstract* to *exceptional*. These are not ratings — they are orientations.
-
-- **Stability** — auto-computed at every build from `git log --follow`. The heuristic: very new or barely-touched documents are *volatile*; actively-revised documents are *revising*; older documents with more commits settle into *fairly stable*, *stable*, or *established*. This requires no manual maintenance — the build reads the repository history and makes the inference.
-
-The version history block, directly above the epistemic footer, uses a three-tier fallback: authored `history:` notes when they exist (written by me when the git log alone would not convey what changed), then the raw git log, then the `date:` frontmatter field as a creation record. `make build` auto-commits any changed content files before the Hakyll compilation runs, so the git log is always current.
-
-The point of all this is simple: when you read something on this site, you should know what kind of claim I am making. The date a document was last modified is not decorative. A 40% confidence rating is not self-deprecation. The system is an attempt to make explicit something that most writing leaves implicit — where the author actually stands.
 
 ---
 
 ## Influences
 
-The amount of influences on this website is immense, and cannot be detailed in the fullest extent. Every other webpage that I have visited, whether beautiful or pitiful, has evoked some type of reaction or response in me, and that response has played some role, even if minute, in the design of this website. I can point to Tufte's influence on many of my design choices, and for the introduction to Tufte, I am thankful to CSCI1377 at Brown. I am thankful to the many other courses I took in my undergrad that influenced how I interact or ideologically view visualizations, networks, systems, etc.
+[TODO: gwern.net — the primary architectural and philosophical ancestor. What was
+taken (sidenotes, living documents, epistemic metadata, semantic zoom, the long-form
+personal essay as the primary unit). What was changed (visual register — gwern's
+monochrome is denser; this site is more spacious — and scope).]
 
-The tradition of the personal website is one that is built on a sense of community and interaction. I am thankful to everyone else who has a personal website and shares their content with the world. I am also particularly greatful to the open source and broader open culture movements, who have given me and the world so much. This website would not exist without you - and I wouldn't be the person I am without your influence - what a role model! 
+[TODO: Tufte — *Envisioning Information* and *The Visual Display of Quantitative
+Information*. The three-column layout, the sidenote tradition, the principle that
+good typography is invisible.]
+
+[TODO: The personal website tradition more broadly — the early web's culture of
+idiosyncratic personal pages before social media collapsed everyone into the same
+templates. This site is partly an argument that that culture was worth having.]
 
 ---
 
 ## The Future
 
 This site is unfinished. Several portals have no content yet. The annotated bibliography
-is sparse. I am in the progress of migrating content, so stay tuned!
+is sparse. Some pages are marked `status: Stub` for good reason. This is acknowledged
+rather than concealed, because a site that presents itself as complete when it is not
+is making a false claim about the relationship between the author and the work.
+
+[TODO: Specific things that are in progress — content migration, the library portal,
+the fiction and research portals as the corpus grows. The Recently section on the
+homepage, deferred until the corpus supports it.]
 
 The colophon itself is a living document. When the site changes substantially, this
-page will change with it. The git history functions as an authoritative record; this page is my personal annotated summary.
+page will change with it. The git history is the authoritative record; this page is
+the annotated summary.

content/essays/empirical-musings-spaced-repetition.md

@@ -1,22 +0,0 @@
----
-title: "Empirical Musings on Spaced Repetition for Learning Beyond Memorization"
-date: 2026-03-26
-abstract: >
-  A compendium of years of informal, empirical experiments aimed at extending the efficacy of Anki beyond rote memorization to more intricate levels of learning.
-tags:
-  - nonfiction
-  - research
-  - research/informal
-  - miscellany
-authors:
-  - "Levi Neuwirth | /me.html"
-status: "Draft"
-confidence: 65
-importance: 3
-evidence: 2
-scope: broad
-novelty: idiosyncratic
-practicality: high
-confidence history:
-   - 65
----

content/essays/test-essay.md

@@ -3,7 +3,6 @@ title: A Test Essay
 date: 2026-03-14
 abstract: A comprehensive end-to-end exercise of the Hakyll pipeline — typography, code, math, sidenotes, filters, tables, exhibits, and annotations.
 tags: [meta]
-affiliation: "Department of Imaginary Systems, University of Nowhere | https://example.com"
 status: Working model
 confidence: 72
 importance: 3

content/gpg.md

@@ -4,15 +4,17 @@ title: GPG Key
 
 Public key for [ln@levineuwirth.org](mailto:ln@levineuwirth.org).
 
-**Key ID:** `B01A01AD1B5C9663`
+**Key ID:** `XXXXXXXXXXXXXXXX`
 
 **Fingerprint:**
 ```
-CD90 AE96 383B BAF4 15A1  D740 B01A 01AD 1B5C 9663
+XXXX XXXX XXXX XXXX XXXX  XXXX XXXX XXXX XXXX XXXX
 ```
 
 Download: [pubkey.asc](/gpg/pubkey.asc)
 
+Also available on [keys.openpgp.org](https://keys.openpgp.org).
+
 ## Verification
 
 ```bash

content/me/index.md

@@ -132,5 +132,27 @@ My sport of choice these days is distance running. I am particularly fond of tra
 
 I don't use strava, nor an electronic watch, nor do I really pay much attention to my times. This takes away from what is to me the great joy in running - being disconnected, in tune with my thoughts and my bodily perceptions, and enjoying the only moment that there is - NOW. 
 
+---
 
+## Technology I use
 
+### Desktop
+I have used many different setups during my life. My earliest vague technological memories are of a family laptop that ran Windows XP. Windows 7 was used in the computer lab at my elementary school, a place that classes ventured to once or twice a year. I came into possession of my first true computer in middle school - it (regrettably) ran Windows 10, and although I did basic programming on it, I didn't make it much farther than simple Python, text-based games, and occasional Scratch. I have since superceded this hardware; my primary desktop is now a rig that I built, running Gentoo Linux with Hyprland and a custom shell ("Levshell") implemented with [Quickshell](https://github.com/quickshell-mirror/quickshell). I use Gentoo on my desktop for several reasons:
+
+- I care deeply about performance, and compiling my software delivers measurable performance increases.
+- Compilation also gives me fine-grained control over my software configuration, linking, etc.
+- I find it is the best maintained Linux distribution that I have ever used.
+- The community is phenomenal.
+
+I have strong opinions about all aspects of my computing, not just the operating system (and variant thereof). I have a strong preference for AMD processors over Intel, and AMD graphics cards over Nvidia. (I have used at least 2 distinct products from each company, so the sample size is *good enough*.) My current desktop uses all AMD components, as does my current laptop. In theory, I would absolutely love to move to a RISC-V machine, but I have no idea when this will be feasible. My email and VPN are self-hosted, and I use Thunderbird as a client for the former. Somewhat similarly, I use [LibreWolf](https://librewolf.net/) for my browser. I find the Chromium monopoly and the incompetence of Mozilla at browser development to be simultaneously and equally concerning, so LibreWolf was the best solution. I use Emacs for editing most things, including this website. I've played around with many other IDES, and generally have strong opinions on them. Neovim, for instance, seems wonderful - but I absolutely cannot stand vim keybindings, much preferring the Emacs sort. VSCode is pathetic slop - does anything more need to be said? Zed looked promising, but seems to be going in the wrong direction.
+
+### Laptop
+As for laptops, I originally used a Macbook, but this proved too little power for my computing needs. After my sophomore year at Brown, before I began an intensive summer of AI research work, I ditched my Macbook for a P-series Thinkpad. On it I run Arch Linux with Hyprland and a custom shell ("Levshell"). I use Arch in favor of Gentoo because the compilation times are simply too long on my laptop, and due to portability concerns, constant compilation on battery is far from ideal. I deeply love both Arch and Gentoo, but Gentoo does remain my preference. Portage *is* better than pacman. I do not *necessarily* have anything outright against MacOS or Macs in general - in fact, the Apple Silicon hardware seems to be very good, and ARM as an architecture is certainly superior to amd64 - but I have invested so much time in creating a hyper-optimized setup on Linux that is custom tailored to the finest details of my workflow that I am unlikely^[This is still a pretty good outlook for MacOS - I can say with **absolute utmost certainty** that I will **never** use Windows again!] to ever return to the land of Mac. I do not think I can productively live without my hyprland keybinds and modifications - I need a tiling window manager that behaves exactly how I expect!  
+
+### Phone
+I use [GrapheneOS](https://grapheneos.org/) on my phone, and for multiple reasons. Not only is it the only reasonably secure and private option for a mobile device, but the restrictions that it imposes help me to use my phone only in meaningful ways. I have my four most important apps pinned - AnkiDroid, Phone, Signal, and Wikipedia - and very little else installed.
+
+### Artificial Intelligence
+I believe that AI is a great power that brings with it great responsibility. I thus use AI very cautiously, as someone who worked on one of the frontier models as a summer job^[No, I won't tell *you* which, at least not here and not yet. What I *will* say is that working in such a capacity **greatly** changes how you look at these tools, in a juxtaposing sort of way.]. I refuse to delegate or outsource all of my original thought, programming, and especially writing to AI models. I will use AIs to collaborate on code (indeed, this website has been the result of such collaboration), and in more limited sense for preliminary literature searches and random ideation. I will **never** use an AI model to write for me, and urge you to also never do this. There has never been a more crucial moment in the history of writing to write for oneself! 
+
+I do not have a binding preference for any model. I currently use the Claude family of models more than any proprietary other, but if they are no longer the best models for the types of tasks I would like to do, then I will switch at moment's notice. As for open models, I tend to prioritize these - my desktop is more than capable of running open models - and the best open models are very quickly closing the gap with the proprietary frontier models. I hope to be a part of the push for an open victory over proprietary during my upcoming years as a graduate student!  

levineuwirth.cabal

@@ -32,7 +32,6 @@ executable site
         Filters.Dropcaps
         Filters.Smallcaps
         Filters.Wikilinks
-        Filters.Transclusion
         Filters.Links
         Filters.Math
         Filters.Code

pyproject.toml

@@ -13,6 +13,7 @@ dependencies = [
     "faiss-cpu>=1.9",
     "numpy>=2.0",
     "beautifulsoup4>=4.12",
+    "einops>=0.8",
     # CPU-only torch — avoids pulling ~3 GB of CUDA libraries
     "torch>=2.5",
 ]

spec.md

@@ -1,7 +1,7 @@
-# levineuwirth.org — Design Specification v11
+# levineuwirth.org — Design Specification v9
 
 **Author:** Levi Neuwirth
-**Date:** March 2026 (v11: 21 March 2026)
+**Date:** March 2026 (v9: 20 March 2026)
 **Status:** LIVING DOCUMENT — Updated as implementation progresses.
 
 ---
@@ -178,12 +178,8 @@ build:
 	@git add content/
 	@git diff --cached --quiet || git commit -m "auto: $(date -u +%Y-%m-%dT%H:%M:%SZ)"
 	@date +%s > data/build-start.txt
-	@./tools/convert-images.sh        # WebP conversion (skipped if cwebp absent)
 	cabal run site -- build
 	pagefind --site _site
-	@if [ -d .venv ]; then \
-	  uv run python tools/embed.py || echo "Warning: embedding failed"; \
-	fi
 	> IGNORE.txt          # clear stability pins after each build
 	@BUILD_END=$(date +%s); BUILD_START=$(cat data/build-start.txt); \
 	  echo $((BUILD_END - BUILD_START)) > data/last-build-seconds.txt
@@ -199,12 +195,6 @@ watch:
 
 clean:
 	cabal run site -- clean
-
-download-model:
-	@./tools/download-model.sh   # fetch quantized ONNX model to static/models/ (once per machine)
-
-convert-images:
-	@./tools/convert-images.sh   # manual trigger; also runs in build
 ```
 
 ### Hosting Timeline
@@ -224,9 +214,7 @@ server {
     ssl_certificate /etc/letsencrypt/live/levineuwirth.org/fullchain.pem;
     ssl_certificate_key /etc/letsencrypt/live/levineuwirth.org/privkey.pem;
 
-    # cdn.jsdelivr.net required for transformers.js (semantic search library).
+    add_header Content-Security-Policy "default-src 'self'; script-src 'self'; style-src 'self' 'unsafe-inline'; img-src 'self'; font-src 'self';" always;
-    # Model weights served same-origin from /models/ — connect-src stays 'self'.
-    add_header Content-Security-Policy "default-src 'self'; script-src 'self' https://cdn.jsdelivr.net; style-src 'self' 'unsafe-inline'; img-src 'self'; font-src 'self';" always;
 
     gzip on;
     gzip_types text/html text/css application/javascript application/json image/svg+xml;
@@ -297,14 +285,11 @@ levineuwirth.org/
 │   │   ├── annotations.js          # localStorage highlight/annotation engine (UI deferred)
 │   │   ├── score-reader.js         # Score reader: page-turn, movement jumps, deep linking
 │   │   ├── viz.js                  # Vega-Lite render + dark mode re-render via MutationObserver
-│   │   ├── semantic-search.js      # Client-side semantic search: transformers.js + Float32Array cosine ranking
 │   │   ├── search.js               # Pagefind UI init + ?q= pre-fill + search timing (#search-timing)
 │   │   └── prism.min.js            # Syntax highlighting
 │   ├── fonts/                      # Self-hosted WOFF2 (subsetted with OT features)
 │   ├── gpg/
 │   │   └── pubkey.asc              # Ed25519 signing subkey public key (master: CD90AE96…; subkey: C9A42A6F…)
-│   ├── models/                     # Self-hosted ONNX model (gitignored; run: make download-model)
-│   │   └── all-MiniLM-L6-v2/      # ~22 MB quantized — served at /models/ for semantic-search.js
 │   └── images/
 │       └── link-icons/             # SVG icons for external link classification
 │           ├── external.svg
@@ -346,7 +331,7 @@ levineuwirth.org/
 │   │   ├── Links.hs                # External link classification + data-link-icon attributes
 │   │   ├── Math.hs                 # Simple LaTeX → Unicode conversion
 │   │   ├── Code.hs                 # Prepend language- prefix for Prism.js
-│   │   ├── Images.hs               # Lazy loading, lightbox data-attributes, WebP <picture> wrapper for local raster images
+│   │   ├── Images.hs               # Lazy loading, lightbox data-attributes
 │   │   ├── Score.hs                # Score fragment SVG inlining + currentColor replacement
 │   │   └── Viz.hs                  # Visualization IO filter: runs Python scripts, inlines SVG / Vega-Lite JSON
 │   ├── Authors.hs                  # Author-as-tag system (slugify, authorLinksField, author pages)
@@ -365,10 +350,7 @@ levineuwirth.org/
 │   ├── subset-fonts.sh
 │   ├── viz_theme.py                # Matplotlib monochrome helpers (apply_monochrome, save_svg, LINESTYLE_CYCLE)
 │   ├── sign-site.sh                # Detach-sign every _site/**/*.html → .html.sig (called by `make sign`)
-│   ├── preset-signing-passphrase.sh  # Cache signing subkey passphrase in gpg-agent (run once per boot)
+│   └── preset-signing-passphrase.sh  # Cache signing subkey passphrase in gpg-agent (run once per boot)
-│   ├── download-model.sh           # Fetch quantized ONNX model to static/models/ (run once per machine)
-│   ├── convert-images.sh           # Convert JPEG/PNG → WebP companions via cwebp (runs automatically in build)
-│   └── embed.py                    # Build-time embedding pipeline: similar-links + semantic search index
 ├── levineuwirth.cabal
 ├── cabal.project
 ├── cabal.project.freeze
@@ -412,10 +394,10 @@ levineuwirth.org/
 - [x] Selection popup — `selection-popup.js` / `selection-popup.css`; context-aware toolbar appears 450 ms after text selection; see Implementation Notes
 - [x] Print stylesheet — `print.css` (media="print"); single-column, light colors, sidenotes as indented blocks, external URLs shown
 - [x] Current page (`/current.html`) — now-page; added to primary nav
-- [x] Annotations — `annotations.js` / `annotations.css`; localStorage storage, text re-anchoring, highlight marks, tooltip with delete; color-picker UI in selection popup (four swatches + optional note field)
+- [~] Annotations — `annotations.js` / `annotations.css`; localStorage infrastructure + highlight re-anchoring written; UI (button in selection popup) deferred
 
 ### Phase 4: Creative Content & Polish
-- [x] Image handling (lazy load, lightbox, figures, WebP `<picture>` wrapper for local raster images)
+- [x] Image handling (lazy load, lightbox, figures)
 - [x] Homepage (replaces standalone index; gateway + curated recent content)
 - [x] Poetry typesetting — codex reading mode (`reading.html`, `reading.css`, `reading.js`); `poetryCompiler` with `Ext_hard_line_breaks`; narrower measure, stanza spacing, drop-cap suppressed
 - [x] Fiction reading mode — same codex layout; `fictionCompiler`; chapter drop caps + smallcaps lead-in via `h2 + p::first-letter`; reading mode infrastructure shared with poetry
@@ -427,15 +409,14 @@ levineuwirth.org/
 
 ### Phase 5: Infrastructure & Advanced
 - [x] **Arch Linux VPS + nginx + certbot + DNS migration** — Hetzner VPS provisioned, Arch Linux installed, nginx configured (config in §III), TLS cert via certbot, DNS migrated from DreamHost. `make deploy` pushes to GitHub and rsyncs to VPS.
-- [x] **Semantic embedding pipeline** — Implemented. See Phase 6 "Embedding-powered similar links" and "Full-text semantic search".
+- [ ] **Semantic embedding pipeline** — Superseded by Phase 6 "Embedding-powered similar links" (local model, no API cost).
 - [x] **Backlinks with context** — Two-pass build-time system (`build/Backlinks.hs`). Pass 1: `version "links"` compiles each page lightly (wikilinks preprocessed, links + context extracted, serialised as JSON). Pass 2: `create ["data/backlinks.json"]` inverts the map. `backlinksField` in `essayCtx` / `postCtx` loads the JSON and renders `<details>`-collapsible per-entry lists. `popups.js` excludes `.backlink-source` links from the preview popup. Context paragraph uses `runPure . writeHtml5String` on the surrounding `Para` block. See Implementation Notes.
 - [ ] **Link archiving** — For all external links in `data/bibliography.bib` and in page bodies, check availability and save snapshots (Wayback Machine `save` API or local archivebox instance). Store archive URLs in `data/link-archive.json`; `Filters.Links` injects `data-archive-url` attributes; `popups.js` falls back to the archive if the live URL returns 404.
 - [ ] **Self-hosted git (Forgejo)** — Run Forgejo on the VPS. Mirror the build repo. Link from the colophon. Not essential; can remain on GitHub indefinitely.
 - [ ] **Reader mode** — Distraction-free reading overlay: hides nav, TOC, sidenotes; widens the body column to ~70ch; activated via a keyboard shortcut or settings panel toggle. Distinct from focus mode (which affects the nav) — reader mode affects the content layout.
-- [ ] **HTTP/3 + QUIC** — nginx 1.25+ supports HTTP/3 via `listen 443 quic reuseport` + `http3 on` + `Alt-Svc` header. Requires UDP 443 open in Hetzner's Cloud Firewall. Deferred: latency is currently geographic RTT, not server processing; gains would be modest for a static site from a single DC. CDN alternatives (Bunny.net, multi-region Hetzner with GeoDNS) would address the root cause but raise ethical or operational complexity concerns. Revisit if latency becomes a real user complaint. Server-side improvements (brotli pre-compression, `open_file_cache`) are a lower-cost step first.
 
 ### Phase 6: Deferred Features
-- [x] **Annotation UI** — `annotations.js` / `annotations.css`: localStorage storage, text-stream re-anchoring, four highlight colors (amber/sage/steel/rose), hover tooltip with delete. Selection popup "Annotate" button triggers a color-swatch + optional note picker; Enter or "Highlight" button commits; Escape cancels. Picker positioned above the selection, same inverted style as the tooltip. Settings panel includes a "Clear Annotations" button (with confirmation) that wipes all annotations site-wide via `Annotations.clearAll()`.
+- [ ] **Annotation UI** — The `annotations.js` / `annotations.css` infrastructure exists (localStorage storage, re-anchoring on load, four highlight colors, hover tooltip). The selection popup "Annotate" button was removed pending a design decision on the color-picker and note-entry UX. Revisit: a popover with four color swatches and an optional text field, triggered from the selection popup.
 - [~] **Visualization pipeline** — Implemented as a Pandoc IO filter (`Filters.Viz`), not a per-slug Hakyll rule. See Phase 4 entry and Implementation Notes. Infrastructure complete; production content pending.
 - [x] **Music catalog page** — `/music/` index listing all compositions grouped by instrumentation category (orchestral → chamber → solo → vocal → choral → electronic → other), with an optional Featured section. Auto-generated from composition frontmatter by `build/Catalog.hs`; renders HTML in Haskell (same pattern as backlinks). Category, year, duration, instrumentation, and ◼/♫ indicators for score/recording availability. `content/music/index.md` provides prose intro + abstract. Template: `templates/music-catalog.html`. CSS: `static/css/catalog.css`. Context: `musicCatalogCtx` (provides `catalog: true` flag, `featured-works`, `has-featured`, `catalog-by-category`).
 - [x] **Score reader swipe gestures** — `touchstart`/`touchend` listeners on `#score-reader-stage` with passive: true. Threshold: ≥ 50 px horizontal, < 30 px vertical drift. Left swipe → next page; right swipe → previous page.
@@ -447,12 +428,10 @@ levineuwirth.org/
 - [x] **Epistemic profile** — Replaces the old `certainty` / `importance` fields with a richer multi-axis system. **Compact** (always visible in footer): status chip · confidence % · importance dots · evidence dots. **Expanded** (`<details>`): stability (auto) · scope · novelty · practicality · last reviewed · confidence trend. Auto-calculation in `build/Stability.hs` via `git log --follow`; `IGNORE.txt` pins overrides. See Metadata section and Implementation Notes for full schema and vocabulary.
 - [ ] **Writing statistics dashboard** — A `/stats` page computed entirely at build time from the corpus. Contents: total word count across all content types, essay/post/poem count, words written per month rendered as a GitHub-style contribution heatmap (SVG generated by Haskell or a Python script), average and median essay length, longest essay, most-cited essay (by backlink count), tag distribution as a treemap, reading-time histogram, site growth over time (cumulative word count by date). All data collected during the Hakyll build from compiled items and their snapshots; serialized to `data/stats.json` and rendered into a dedicated `stats.html` template.
 - [x] **Memento mori** — Implemented at `/memento-mori/` as a full standalone page. 90×52 grid of weeks anchored to birthday anniversaries (nested year/week loop via `setFullYear`; week 52 stretched to eve of next birthday to absorb 365th/366th days). Week popup shows dynamic day-count and locale-derived day names. Score fragment (bassoon, `content/memento-mori/scores/bsn.svg`) inlined via `Filters.Score`. Linked from footer (MM).
-- [x] **Embedding-powered similar links** — `tools/embed.py` encodes every `#markdownBody` page with `all-MiniLM-L6-v2` (384 dims, unit-normalised), builds a FAISS `IndexFlatIP`, queries top-5 neighbours per page (cosine ≥ 0.30), writes `data/similar-links.json`. `build/SimilarLinks.hs` provides `similarLinksField` in `essayCtx`/`postCtx` with Hakyll dependency tracking; "Related" section rendered in `page-footer.html`. Staleness check skips re-embedding when JSON is newer than all HTML. Called by `make build` via `uv run`; non-fatal if `.venv` absent. See Implementation Notes.
+- [ ] **Embedding-powered similar links** — Precompute dense vector embeddings for every page using a local model (e.g. `nomic-embed-text` or `gte-large` via `ollama` or `llama.cpp`) on personal hardware — no API dependency, no per-call cost. At build time, a Python script reads `_site/` HTML, embeds each page, computes top-N cosine neighbors, and writes `data/similar-links.json` (slug → [{slug, title, score}]). Hakyll injects this into each page's context (via `Metadata.hs` reading the JSON); template renders a "Related" section in the page footer. Analogous to gwern's `GenerateSimilar.hs` but model-agnostic and self-hosted. Note: supersedes the Phase 5 "Semantic embedding pipeline" stub — that stub should be replaced by this when implemented.
 - [x] **Bidirectional backlinks with context** — See Phase 5 above; implemented with full context-paragraph extraction. Merged with the Phase 5 stub.
 - [x] **Signed pages / content integrity** — `make sign` (called by `make deploy`) runs `tools/sign-site.sh`: walks `_site/**/*.html`, produces a detached ASCII-armored `.sig` per page. Signing uses a dedicated Ed25519 subkey isolated in `~/.gnupg-signing/` (master `sec#` stub + `ssb` signing subkey). Passphrase cached 24 h in the signing agent via `tools/preset-signing-passphrase.sh` + `gpg-preset-passphrase`; `~/.gnupg-signing/gpg-agent.conf` sets `allow-preset-passphrase`. Footer "sig" link points to `$url$.sig`; hovering shows the ASCII armor via `popups.js` `sigContent` provider. Public key at `static/gpg/pubkey.asc` → served at `/gpg/pubkey.asc`. Fingerprints: master `CD90AE96…B5C9663`; signing subkey `C9A42A6F…2707066` (keygrip `619844703EC398E70B0045D7150F08179CFEEFE3`). See Implementation Notes.
-- [x] **Self-hosted semantic search model** — `tools/download-model.sh` fetches the quantized ONNX model (`all-MiniLM-L6-v2`, ~22 MB, 5 files) from HuggingFace into `static/models/all-MiniLM-L6-v2/` (gitignored). `semantic-search.js` sets `env.localModelPath = '/models/'` and `env.allowRemoteModels = false` before calling `pipeline()`, so all model weight fetches are same-origin. The CDN import of transformers.js itself still requires `cdn.jsdelivr.net` in `script-src`; `connect-src` stays `'self'`. `make download-model` is a one-time setup step per machine. See Implementation Notes.
+- [ ] **Full-text semantic search** — A secondary search mode alongside Pagefind's keyword index. Precompute embeddings for every paragraph (same pipeline as similar links). Store as a compact binary or JSON index. At query time, either: (a) compute the query embedding client-side using a small WASM model (e.g. `transformers.js` with a quantized MiniLM) and run cosine similarity against the stored paragraph vectors, or (b) use a precomputed query-expansion table (top-K words → relevant slugs, offline). Surfaced as a "Semantic search" toggle on `/search.html`. Returns paragraphs rather than pages as the result unit, with the source page title and a link to the specific section. This finds conceptually related content even when exact keywords differ — searching "the relationship between music and mathematics" surfaces relevant essays regardless of vocabulary.
-- [x] **Responsive images (WebP)** — `tools/convert-images.sh` walks `static/` and `content/`, calls `cwebp -q 85` to produce `.webp` companions alongside every JPEG/PNG (skips existing; exits gracefully if `cwebp` absent). `make build` runs it before Hakyll so WebP files are present when `static/**` is copied. `build/Filters/Images.hs` detects local raster images and emits `RawInline "html"` `<picture>` elements with a `<source srcset="…webp" type="image/webp">` and an `<img>` fallback; SVG, external URLs, and `data:` URIs pass through as plain `<img>`. Generated `.webp` files gitignored via `static/**/*.webp` / `content/**/*.webp`. See Implementation Notes.
-- [x] **Full-text semantic search** — `tools/embed.py` also produces paragraph-level embeddings (same `all-MiniLM-L6-v2` model, same pass): walks `<p>/<li>/<blockquote>` in `#markdownBody`, tracks nearest preceding heading for context, writes `data/semantic-index.bin` (raw Float32, N×384) + `data/semantic-meta.json` ([{url, title, heading, excerpt}]). Client: `static/js/semantic-search.js` dynamically imports `@xenova/transformers@2` from CDN, embeds the query, brute-force cosine-ranks all paragraph vectors (fast at <5k paragraphs in JS), renders top-8 results as title + section heading + excerpt. Surfaced on `/search.html` as a **Keyword / Semantic** tab strip; active tab persists in `localStorage` (keyword default on first visit). Both tabs on same page; `semantic-search.js` loaded within existing `$if(search)$` block. See Implementation Notes.
 
 ---
 
@@ -689,39 +668,6 @@ Implemented across `build/Stability.hs`, `build/Contexts.hs`, `templates/partial
 
 **nginx:** `.sig` files need no special handling — they're served as static files alongside `.html`. The `try_files` directive handles `$uri` directly.
 
-### Embedding pipeline + semantic search
-
-**Model unification:** Both similar-links (page-level) and semantic search (paragraph-level) use `all-MiniLM-L6-v2` (384 dims). This is a deliberate simplification: the same model runs at build time (Python/sentence-transformers) and query time (browser/transformers.js `Xenova/all-MiniLM-L6-v2` quantized), guaranteeing that query vectors and corpus vectors are in the same embedding space.
-
-**Build-time (`tools/embed.py`):** One HTML parse pass per file extracts both the full-page text (for similar-links) and individual paragraphs (for semantic search). Model is loaded once and both encoding jobs run sequentially. Outputs: `data/similar-links.json`, `data/semantic-index.bin` (raw `float32`, shape `[N_paragraphs, 384]`), `data/semantic-meta.json`. All three are gitignored (generated). Staleness check: skips the entire run if all three outputs are newer than all `_site/` HTML.
-
-**Binary index format:** `para_vecs.tobytes()` writes a flat, little-endian `float32` array. In JS: `new Float32Array(arrayBuffer)`. No header, no framing — row `i` starts at byte offset `i × 384 × 4`. This is the simplest possible format and avoids a numpy/npy parser in the browser.
-
-**Client-side search (`semantic-search.js`):** Dynamically imports `@xenova/transformers@2` from jsDelivr CDN on first query (lazy — no load cost on pages that never use semantic search). Fetches binary index + metadata JSON (also lazy, browser-cached). Brute-force dot product over a `Float32Array` in a tight JS loop — fast enough at <5k paragraphs; revisit with a WASM FAISS binding if the corpus grows beyond ~20k paragraphs. Vectors are unit-normalised at build time, so dot product = cosine similarity.
-
-**Tab default + localStorage:** Keyword (Pagefind) is the default on first visit — zero cold-start. User's last-used tab is stored under `search-tab` in `localStorage` and restored on load, so returning users who prefer semantic always land there. If `localStorage` is unavailable (private browsing restrictions), falls back silently to keyword.
-
-**Self-hosted model:** `semantic-search.js` sets `mod.env.localModelPath = '/models/'` and `mod.env.allowRemoteModels = false` immediately after the CDN import. transformers.js then resolves model files as `GET /models/all-MiniLM-L6-v2/{file}`, which are same-origin. `make download-model` (= `tools/download-model.sh`) fetches 5 files from HuggingFace: `config.json`, `tokenizer.json`, `tokenizer_config.json`, `special_tokens_map.json`, `onnx/model_quantized.onnx`. The files live in `static/models/` (gitignored) and are copied to `_site/models/` by the existing `static/**` Hakyll rule.
-
-**CSP:** `script-src` requires `https://cdn.jsdelivr.net` for the transformers.js library import. `connect-src` stays `'self'` — all model weight fetches are same-origin after `allowRemoteModels = false`. The binary index and meta JSON are also same-origin.
-
-### Responsive images — WebP `<picture>` wrapping
-
-`build/Filters/Images.hs` inspects each `Image` inline's `src`. If it is a local raster (not starting with `http://`, `https://`, `//`, or `data:`; extension `.jpg`/`.jpeg`/`.png`/`.gif`), it emits `RawInline (Format "html")` containing a `<picture>` element:
-
-```html
-<picture>
-  <source srcset="/images/foo.webp" type="image/webp">
-  <img src="/images/foo.jpg" alt="…" loading="lazy" data-lightbox="true">
-</picture>
-```
-
-The WebP `srcset` is computed at build time by `System.FilePath.replaceExtension`. No IO is needed in the filter — the `<source>` is always emitted; browsers silently ignore it if the file doesn't exist (falling back to `<img>`). SVG, external URLs, and `data:` URIs remain plain `<img>` tags. Images inside `<a>` links get no `data-lightbox` marker (same as before).
-
-`tools/convert-images.sh` performs the actual conversion: `cwebp -q 85` per file. Runs before Hakyll in `make build` and is also available as a standalone `make convert-images` target. Exits 0 with a notice if `cwebp` is not installed, so the build never fails on machines without `libwebp`. Generated `.webp` files are gitignored; `git add -f` to commit an authored WebP.
-
-**Quality note:** `-q 85` is a good default for photographic images. For pixel-art, diagrams, or images that are already highly compressed, `-lossless` or a higher quality setting may be appropriate (edit the script).
-
 ### Annotations (infrastructure only)
 `annotations.js` stores annotations as JSON in `localStorage` under `site-annotations`, scoped per `location.pathname`. On `DOMContentLoaded`, `applyAll()` re-anchors saved annotations via a `TreeWalker` text-stream search (concatenates all text nodes in `#markdownBody`, finds exact match by index, builds a `Range`, wraps with `<mark>`). Cross-element ranges use `extractContents()` + `insertNode()` fallback. Four highlight colors (amber / sage / steel / rose) defined in `annotations.css` as `rgba` overlays with `box-decoration-break: clone`. Hover tooltip shows note, date, and delete button. Public API: `window.Annotations.add(text, color, note)` / `.remove(id)`. The selection-popup "Annotate" button is currently removed pending a UI revision.
 

static/css/annotations.css

@@ -30,14 +30,6 @@ mark.user-annotation:hover { filter: brightness(0.80); }
 [data-theme="dark"] mark.user-annotation.user-annotation--steel { background-color: rgba(112, 150, 184, 0.42); }
 [data-theme="dark"] mark.user-annotation.user-annotation--rose  { background-color: rgba(200, 116, 116, 0.42); }
 
-/* System dark mode fallback (for prefers-color-scheme without explicit data-theme) */
-@media (prefers-color-scheme: dark) {
-    :root:not([data-theme="light"]) mark.user-annotation.user-annotation--amber { background-color: rgba(245, 158,  11, 0.40); }
-    :root:not([data-theme="light"]) mark.user-annotation.user-annotation--sage  { background-color: rgba(107, 158, 120, 0.42); }
-    :root:not([data-theme="light"]) mark.user-annotation.user-annotation--steel { background-color: rgba(112, 150, 184, 0.42); }
-    :root:not([data-theme="light"]) mark.user-annotation.user-annotation--rose  { background-color: rgba(200, 116, 116, 0.42); }
-}
-
 /* ============================================================
    ANNOTATION TOOLTIP
    Appears on hover over a mark. Inverted colours like the
@@ -103,113 +95,9 @@ mark.user-annotation:hover { filter: brightness(0.80); }
 
 .ann-tooltip-delete:hover { opacity: 1; }
 
-/* ============================================================
-   ANNOTATE PICKER
-   Appears above the selection when "Annotate" is clicked.
-   Inverted colors (dark bg, light text) — same as the tooltip.
-   ============================================================ */
-
-.ann-picker {
-    position: absolute;
-    z-index: 760;
-    background: var(--text);
-    color: var(--bg);
-    border-radius: 5px;
-    box-shadow: 0 2px 12px rgba(0, 0, 0, 0.28);
-    padding: 0.55rem 0.6rem;
-    font-family: var(--font-sans);
-    font-size: 0.75rem;
-    width: 13rem;
-
-    opacity: 0;
-    visibility: hidden;
-    transition: opacity 0.12s ease, visibility 0.12s ease;
-    pointer-events: none;
-}
-
-.ann-picker.is-visible {
-    opacity: 1;
-    visibility: visible;
-    pointer-events: auto;
-}
-
-.ann-picker-swatches {
-    display: flex;
-    gap: 0.45rem;
-    margin-bottom: 0.5rem;
-    align-items: center;
-}
-
-.ann-picker-swatch {
-    width: 1.05rem;
-    height: 1.05rem;
-    border-radius: 50%;
-    border: 2px solid transparent;
-    cursor: pointer;
-    padding: 0;
-    flex-shrink: 0;
-    outline: none;
-    transition: transform 0.1s ease, border-color 0.1s ease;
-}
-
-.ann-picker-swatch--amber { background: rgba(245, 158,  11, 0.9); }
-.ann-picker-swatch--sage  { background: rgba(107, 158, 120, 0.9); }
-.ann-picker-swatch--steel { background: rgba(112, 150, 184, 0.9); }
-.ann-picker-swatch--rose  { background: rgba(200, 116, 116, 0.9); }
-
-.ann-picker-swatch.is-selected,
-.ann-picker-swatch:focus-visible {
-    border-color: var(--bg);
-    transform: scale(1.2);
-}
-
-.ann-picker-note {
-    display: block;
-    width: 100%;
-    box-sizing: border-box;
-    background: rgba(255, 255, 255, 0.1);
-    border: 1px solid rgba(255, 255, 255, 0.2);
-    border-radius: 3px;
-    color: inherit;
-    font-family: var(--font-sans);
-    font-size: 0.72rem;
-    padding: 0.28rem 0.4rem;
-    margin-bottom: 0.45rem;
-    outline: none;
-}
-
-.ann-picker-note::placeholder { opacity: 0.5; }
-
-.ann-picker-note:focus {
-    border-color: rgba(255, 255, 255, 0.45);
-}
-
-.ann-picker-actions {
-    display: flex;
-    justify-content: flex-end;
-}
-
-.ann-picker-submit {
-    background: rgba(255, 255, 255, 0.15);
-    border: 1px solid rgba(255, 255, 255, 0.2);
-    border-radius: 3px;
-    color: inherit;
-    font-family: var(--font-sans);
-    font-size: 0.72rem;
-    padding: 0.22rem 0.55rem;
-    cursor: pointer;
-    line-height: 1.4;
-    transition: background 0.1s ease;
-}
-
-.ann-picker-submit:hover {
-    background: rgba(255, 255, 255, 0.28);
-}
-
 /* ============================================================
    REDUCE MOTION
    ============================================================ */
 
 [data-reduce-motion] mark.user-annotation { transition: none; }
 [data-reduce-motion] .ann-tooltip         { transition: none; }
-[data-reduce-motion] .ann-picker          { transition: none; }

static/css/base.css

@@ -120,13 +120,6 @@
 
     /* Transitions */
     --transition-fast: 0.15s ease;
-
-    /* Writing activity heatmap (light mode) */
-    --hm-0: #e8e8e4;   /* empty cell */
-    --hm-1: #b4b4b0;   /* < 500 words */
-    --hm-2: #787874;   /* 500–1999 words */
-    --hm-3: #424240;   /* 2000–4999 words */
-    --hm-4: #1a1a1a;   /* 5000+ words */
 }
 
 
@@ -153,13 +146,6 @@
 
     --selection-bg:    #d4d0c8;
     --selection-text:  #121212;
-
-    /* Writing activity heatmap (dark mode) */
-    --hm-0: #252524;
-    --hm-1: #484844;
-    --hm-2: #6e6e6a;
-    --hm-3: #9e9e9a;
-    --hm-4: #d4d0c8;
 }
 
 /* System dark mode fallback */
@@ -182,12 +168,6 @@
 
         --selection-bg:    #d4d0c8;
         --selection-text:  #121212;
-
-        --hm-0: #252524;
-        --hm-1: #484844;
-        --hm-2: #6e6e6a;
-        --hm-3: #9e9e9a;
-        --hm-4: #d4d0c8;
     }
 }
 
@@ -208,9 +188,6 @@ html {
     line-height: var(--line-height);
     -webkit-text-size-adjust: 100%;
     scroll-behavior: smooth;
-    /* clip (not hidden) — prevents horizontal scroll at the viewport level
-       without creating a scroll container, so position:sticky still works. */
-    overflow-x: clip;
 }
 
 body {

static/css/build.css

@@ -107,41 +107,3 @@
     margin: 0;
     font-variant-numeric: tabular-nums;
 }
-
-/* ============================================================
-   Writing statistics page (/stats/)
-   ============================================================ */
-
-/* Heatmap figure */
-.stats-heatmap {
-    margin: 1.5rem 0 1rem;
-    overflow-x: auto;
-}
-
-.heatmap-svg {
-    display: block;
-    max-width: 100%;
-}
-
-/* Legend row below heatmap */
-.heatmap-legend {
-    display: flex;
-    align-items: center;
-    gap: 0.4rem;
-    margin-top: 0.5rem;
-    font-size: 0.75em;
-    font-family: var(--font-ui, var(--font-sans));
-    color: var(--text-faint);
-}
-
-/* Ordered lists on the Notable section */
-.build-page-list {
-    font-size: 0.95em;
-    padding-left: 1.5rem;
-    margin: 0.25rem 0 1rem;
-}
-
-.build-page-list li {
-    margin: 0.3rem 0;
-    font-variant-numeric: tabular-nums;
-}

static/css/components.css

@@ -206,13 +206,6 @@ nav.site-nav {
     text-align: center;
 }
 
-.settings-btn--danger {
-    color: rgba(180, 80, 80, 0.85);
-}
-.settings-btn--danger:hover {
-    color: rgba(180, 80, 80, 1);
-}
-
 .settings-col {
     display: flex;
     flex-direction: column;
@@ -252,63 +245,6 @@ nav.site-nav {
     scroll-behavior:          auto   !important;
 }
 
-/* ── Mobile nav (≤540px) ─────────────────────────────────────────────
-   Controls are position:absolute on desktop (out of flex flow, pinned
-   right). On narrow viewports they collide with the primary links.
-   Fix: bring them into flow, wrap the row, add a separator line.
-   ──────────────────────────────────────────────────────────────────── */
-
-@media (max-width: 540px) {
-    .nav-row-primary {
-        flex-wrap: wrap;
-        justify-content: center;
-        padding: 0.4rem 0.75rem;
-        gap: 0.25rem 0;
-    }
-
-    .nav-primary {
-        flex-wrap: wrap;
-        justify-content: center;
-        font-size: 0.72rem;
-    }
-
-    /* Larger vertical padding → usable tap targets without extra markup */
-    .nav-primary a {
-        padding: 0.35rem 0.55rem;
-    }
-
-    /* border-left separators look broken on wrapped rows (the first link
-       of a new row still gets a left border). Replace with a gap. */
-    .nav-primary a + a {
-        border-left: none;
-        margin-left: 0.1rem;
-    }
-
-    /* Pull controls out of absolute positioning, span full width, center */
-    .nav-controls {
-        position: static;
-        width: 100%;
-        justify-content: center;
-        padding-top: 0.3rem;
-        border-top: 1px solid var(--border);
-        margin-top: 0.15rem;
-    }
-
-    .nav-portal-toggle {
-        padding: 0.3rem 0;
-    }
-
-    /* Portal row: tighter side padding on narrow screens */
-    .nav-portals {
-        padding-left: 0.75rem;
-        padding-right: 0.75rem;
-    }
-
-    .nav-portals a {
-        padding: 0.3rem 0.55rem;
-    }
-}
-
 /* Row 2: portal links — hidden until nav.js adds .is-open */
 .nav-portals {
     display: none;
@@ -564,26 +500,6 @@ nav.site-nav {
     font-variant-caps: normal;
 }
 
-/* Affiliation: institution name below author, in sans to contrast the serif author row */
-.meta-affiliation {
-    font-family: var(--font-sans);
-    font-size: 0.7rem;
-    font-variant-caps: all-small-caps;
-    letter-spacing: 0.05em;
-    color: var(--text-faint);
-    text-align: center;
-}
-
-.meta-affiliation a {
-    color: var(--text-faint);
-    text-decoration: none;
-    transition: color var(--transition-fast);
-}
-
-.meta-affiliation a:hover {
-    color: var(--text-muted);
-}
-
 /* Authors: "by" label + name */
 .meta-authors {
     font-size: 0.78rem;
@@ -612,15 +528,6 @@ nav.site-nav {
     text-decoration: underline;
     text-underline-offset: 2px;
 }
-/* When the score box is present, suppress the link-level underline and
-   restore it only on the label text — leaves the gap and box undecorated. */
-.meta-pagelinks a:has(.meta-overall-score):hover {
-    text-decoration: none;
-}
-.meta-pagelinks a:has(.meta-overall-score):hover .ep-link-label {
-    text-decoration: underline;
-    text-underline-offset: 2px;
-}
 .meta-pagelinks a + a::before {
     content: " · ";
     color: var(--border);
@@ -876,24 +783,6 @@ details[open] > .ep-summary::before { content: "▾\00a0"; }
 }
 
 
-/* Overall score percentage in the top metadata nav (adjacent to Epistemic link).
-   Renders as a compact numeric chip: "72%" in small-caps sans, separated from
-   the link text by a faint centerdot. */
-.meta-overall-score {
-    font-family: var(--font-sans);
-    font-size: 0.68rem;
-    font-variant-numeric: tabular-nums;
-    letter-spacing: 0.03em;
-    color: var(--text-muted);
-    display: inline-block;
-    text-decoration: none;
-    border: 1px solid var(--border-muted);
-    border-radius: 2px;
-    padding: 0.05em 0.5em;
-    margin-left: 0.5em;
-    vertical-align: 0.05em;
-}
-
 /* ============================================================
    PAGINATION NAV
    ============================================================ */
@@ -1086,96 +975,6 @@ h3:hover .section-toggle,
     min-height: 1.2em; /* reserve space to prevent layout shift */
 }
 
-/* Search tabs (Keyword / Semantic toggle) */
-.search-tabs {
-    display: flex;
-    gap: 0;
-    margin-bottom: 0.1rem;
-    border-bottom: 1px solid var(--border);
-}
-
-.search-tab {
-    background: none;
-    border: none;
-    border-bottom: 2px solid transparent;
-    margin-bottom: -1px;
-    padding: 0.45rem 1rem;
-    font-family: var(--font-sans);
-    font-size: var(--text-size-small);
-    color: var(--text-faint);
-    cursor: pointer;
-    transition: color 0.1s, border-color 0.1s;
-}
-
-.search-tab:hover       { color: var(--text-muted); }
-.search-tab.is-active   { color: var(--text); border-bottom-color: var(--text); }
-
-/* Search panels (keyword / semantic) — only active panel visible */
-.search-panel           { display: none; }
-.search-panel.is-active { display: block; }
-
-/* Semantic query input — mirrors Pagefind UI style */
-.semantic-query-input {
-    width: 100%;
-    margin-top: 1.65rem;
-    padding: 0.5rem 0.75rem;
-    font-family: var(--font-sans);
-    font-size: 0.85rem;
-    color: var(--text);
-    background: var(--bg);
-    border: 1px solid var(--border);
-    border-radius: 2px;
-    box-sizing: border-box;
-    outline: none;
-}
-
-.semantic-query-input:focus { border-color: var(--text-muted); }
-
-.semantic-status {
-    font-family: var(--font-mono);
-    font-size: var(--text-size-small);
-    color: var(--text-faint);
-    margin-top: 0.5rem;
-    min-height: 1.2em;
-}
-
-/* Semantic results list */
-.semantic-results-list {
-    list-style: none;
-    margin: 1.2rem 0 0;
-    padding: 0;
-}
-
-.semantic-result {
-    padding: 0.85rem 0;
-    border-bottom: 1px solid var(--border);
-}
-
-.semantic-result:last-child { border-bottom: none; }
-
-.semantic-result-title {
-    font-family: var(--font-sans);
-    font-weight: 600;
-    font-size: 0.9rem;
-    color: var(--text);
-    text-decoration: none;
-}
-
-.semantic-result-title:hover { text-decoration: underline; }
-
-.semantic-result-heading {
-    font-family: var(--font-sans);
-    font-size: var(--text-size-small);
-    color: var(--text-faint);
-}
-
-.semantic-result-excerpt {
-    margin: 0.3rem 0 0;
-    font-size: var(--text-size-small);
-    color: var(--text-muted);
-    line-height: 1.5;
-}
-
 /* ============================================================
    COMPOSITION LANDING PAGE
    ============================================================ */
@@ -1282,176 +1081,3 @@ h3:hover .section-toggle,
     margin-top: 0.5rem;
     accent-color: var(--text-muted);
 }
-
-
-/* ============================================================
-   TRANSCLUSION
-   States for <div class="transclude"> placeholders injected by
-   Filters.Transclusion and populated by transclude.js.
-   ============================================================ */
-
-/* Loading placeholder — subtle pulse so the page doesn't look broken
-   while the fetch is in flight. */
-.transclude--loading {
-    min-height: 1.65rem;
-    background: var(--bg-offset);
-    border-radius: 2px;
-    animation: transclude-pulse 1.2s ease-in-out infinite;
-}
-
-@keyframes transclude-pulse {
-    0%, 100% { opacity: 1; }
-    50%       { opacity: 0.45; }
-}
-
-/* Loaded — transparent wrapper; content reads as native. */
-.transclude--loaded {
-    /* No visual treatment: transcluded content is visually native. */
-}
-
-/* Optional: uncomment to show a subtle left rule on loaded transclusions.
-.transclude--loaded {
-    border-left: 2px solid var(--border);
-    padding-left: 1rem;
-    margin-left: -1rem;
-}
-*/
-
-/* Error state — faint inline notice. */
-.transclude--error {
-    font-family: var(--font-mono);
-    font-size: var(--text-size-small);
-    color: var(--text-faint);
-    padding: 0.4rem 0;
-}
-
-/* Content wrapper — display:contents so the injected nodes participate
-   in normal document flow without adding a structural block. */
-.transclude--content {
-    display: contents;
-}
-
-/* ============================================================
-   COPY BUTTON
-   Injected by copy.js into every <pre> block. Fades in on hover.
-   ============================================================ */
-
-.copy-btn {
-    position: absolute;
-    top: 0.5rem;
-    right: 0.5rem;
-    font-family: var(--font-sans);
-    font-size: 0.72rem;
-    font-weight: 500;
-    letter-spacing: 0.03em;
-    text-transform: uppercase;
-    color: var(--text-faint);
-    background: var(--bg);
-    border: 1px solid var(--border-muted);
-    border-radius: 3px;
-    padding: 0.2em 0.55em;
-    cursor: pointer;
-    opacity: 0;
-    transition: opacity 0.15s, color 0.15s, border-color 0.15s;
-    user-select: none;
-    line-height: 1.6;
-}
-
-pre:hover .copy-btn,
-.copy-btn:focus-visible {
-    opacity: 1;
-}
-
-.copy-btn:hover {
-    color: var(--text-muted);
-    border-color: var(--border);
-}
-
-.copy-btn[data-copied] {
-    color: var(--text-muted);
-    border-color: var(--border);
-    opacity: 1;
-}
-
-/* ============================================================
-   TOUCH DEVICES
-   (hover: none) and (pointer: coarse) reliably targets
-   touchscreen-primary devices (phones, tablets) without
-   relying on viewport width or user-agent sniffing.
-
-   Goals:
-     1. Suppress :hover states that "stick" after a tap
-     2. Ensure tap targets meet ~44px minimum
-     3. (popup disable is handled in popups.js)
-   ============================================================ */
-
-@media (hover: none) and (pointer: coarse) {
-    /* Nav: ensure adequate vertical tap area even on wider tablets
-       where the ≤540px query doesn't fire */
-    .nav-primary a {
-        padding-top: 0.35rem;
-        padding-bottom: 0.35rem;
-    }
-
-    .nav-portal-toggle {
-        padding-top: 0.3rem;
-        padding-bottom: 0.3rem;
-    }
-
-    /* Settings gear: expand hit area without visual change */
-    .settings-toggle {
-        padding: 0.5rem;
-        margin: -0.5rem;
-    }
-
-    /* Portal row links */
-    .nav-portals a {
-        padding-top: 0.3rem;
-        padding-bottom: 0.3rem;
-    }
-
-    /* Suppress stuck hover states — reset to their non-hover values */
-    .nav-primary a:hover           { color: var(--text-muted); }
-    .nav-portals a:hover           { color: var(--text-faint); }
-    .nav-portal-toggle:hover       { color: var(--text-faint); }
-    .settings-toggle:hover         { color: var(--text-faint); }
-    .toc-nav a:hover               { color: var(--text-faint); }
-    .toc-nav > ol > li > a:hover   { color: var(--text-muted); }
-    .hp-pro-row a:hover            { color: var(--text-muted); }
-    .hp-curiosity-row a:hover      { color: var(--text-muted); }
-    .hp-portal-name:hover          { text-decoration-color: var(--border-muted); }
-    .meta-tag:hover                { color: var(--text-faint); }
-    .backlink-source:hover         { color: var(--text-muted); }
-    .section-toggle:hover          { color: var(--text-faint); }
-
-    /* Copy button: always visible on touch (no persistent hover) */
-    .copy-btn { opacity: 1; }
-}
-
-/* ============================================================
-   PDF EMBED
-   Inline PDF.js viewer iframes rendered from {{pdf:...}} directives.
-   ============================================================ */
-
-.pdf-embed-wrapper {
-    margin: 1.5rem 0;
-    border: 1px solid var(--border);
-    border-radius: 2px;
-    overflow: hidden;
-    background: var(--bg-subtle);
-}
-
-.pdf-embed {
-    display: block;
-    width: 100%;
-    height: 70vh;
-    min-height: 420px;
-    border: none;
-}
-
-@media (max-width: 600px) {
-    .pdf-embed {
-        height: 60vh;
-        min-height: 300px;
-    }
-}

static/css/home.css

@@ -1,106 +1,191 @@
 /* home.css — Homepage-specific styles (loaded only on index.html) */
 
 /* ============================================================
-   UTILITY ROWS
+   CONTACT ROW
-   Professional row and curiosities row: compact Fira Sans
+   Professional links: Email · CV · About · GitHub · GPG · ORCID
-   link strips separated by middots. Two rows, visually equal.
    ============================================================ */
 
-.hp-pro-row,
+.contact-row {
-.hp-curiosity-row {
     display: flex;
     flex-wrap: wrap;
-    align-items: center;
+    justify-content: space-evenly;
-    justify-content: center;
+    gap: 0.35rem 1rem;
+    margin: 1.75rem 0 0;
     font-family: var(--font-sans);
     font-size: var(--text-size-small);
-    margin-top: 0.75rem;
 }
 
-/* First row: extra top margin since the hr is gone — breathing
+.contact-row a {
-   room between the intro block and the utility rows. */
+    display: inline-flex;
-.hp-pro-row {
+    align-items: center;
-    margin-top: 2rem;
+    gap: 0.3em;
-}
-
-.hp-pro-row a,
-.hp-curiosity-row a {
     color: var(--text-muted);
     text-decoration: none;
     transition: color var(--transition-fast);
-    padding: 0.1rem 0.15rem;
 }
 
-.hp-pro-row a:hover,
+.contact-row a:hover {
-.hp-curiosity-row a:hover {
     color: var(--text);
 }
 
-.hp-sep {
+.contact-row a[data-contact-icon]::before {
-    color: var(--text-faint);
+    content: '';
-    padding: 0 0.2em;
+    display: inline-block;
-    user-select: none;
+    width: 0.85em;
-    pointer-events: none;
+    height: 0.85em;
+    flex-shrink: 0;
+    background-color: currentColor;
+    mask-size: contain;
+    mask-repeat: no-repeat;
+    mask-position: center;
+    -webkit-mask-size: contain;
+    -webkit-mask-repeat: no-repeat;
+    -webkit-mask-position: center;
+}
+
+.contact-row a[data-contact-icon="email"]::before {
+    mask-image: url('/images/link-icons/email.svg');
+    -webkit-mask-image: url('/images/link-icons/email.svg');
+}
+.contact-row a[data-contact-icon="document"]::before {
+    mask-image: url('/images/link-icons/document.svg');
+    -webkit-mask-image: url('/images/link-icons/document.svg');
+}
+.contact-row a[data-contact-icon="person"]::before {
+    mask-image: url('/images/link-icons/person.svg');
+    -webkit-mask-image: url('/images/link-icons/person.svg');
+}
+.contact-row a[data-contact-icon="github"]::before {
+    mask-image: url('/images/link-icons/github.svg');
+    -webkit-mask-image: url('/images/link-icons/github.svg');
+}
+.contact-row a[data-contact-icon="key"]::before {
+    mask-image: url('/images/link-icons/key.svg');
+    -webkit-mask-image: url('/images/link-icons/key.svg');
+}
+.contact-row a[data-contact-icon="orcid"]::before {
+    mask-image: url('/images/link-icons/orcid.svg');
+    -webkit-mask-image: url('/images/link-icons/orcid.svg');
 }
 
 /* ============================================================
-   LATIN BENEDICTION
+   SITE GUIDE (expandable <details>)
-   Right-aligned closing salutation, set in lang="la" for
-   correct hyphenation and screen-reader pronunciation.
    ============================================================ */
 
-.hp-latin {
+.site-guide {
-    text-align: right;
+    margin: 1.5rem 0 0;
-    margin-top: 1.5rem;
+    border: 1px solid var(--border);
+    border-radius: 3px;
 }
 
-.hp-latin p {
+.site-guide summary {
-    margin: 0;
-    color: var(--text-muted);
-}
-
-/* ============================================================
-   PORTAL LIST
-   Annotated portal directory, eight items, one per line.
-   Name link — em dash — short description in faint text.
-   ============================================================ */
-
-.hp-portals {
-    margin-top: 2.5rem;
-    font-family: var(--font-sans);
-}
-
-.hp-portal-item {
     display: flex;
-    align-items: baseline;
+    align-items: center;
-    gap: 0.4em;
+    gap: 0.5rem;
-    padding: 0.2rem 0;
+    padding: 0.6rem 0.9rem;
-}
+    font-family: var(--font-sans);
-
-.hp-portal-name {
-    color: var(--text);
-    text-decoration: underline;
-    text-decoration-color: var(--border-muted);
-    text-decoration-thickness: 0.15em;
-    text-underline-offset: 0.2em;
-    text-decoration-skip-ink: auto;
-    transition: text-decoration-color var(--transition-fast), color var(--transition-fast);
-    flex-shrink: 0;
-}
-
-.hp-portal-name:hover {
-    text-decoration-color: var(--link-hover-underline);
-}
-
-.hp-portal-dash {
-    color: var(--text-faint);
-    padding: 0 0.15em;
-    user-select: none;
-    flex-shrink: 0;
-}
-
-.hp-portal-desc {
     font-size: var(--text-size-small);
-    color: var(--text-faint);
+    color: var(--text-muted);
-    line-height: 1.4;
+    cursor: pointer;
+    list-style: none;
+    user-select: none;
+    transition: color var(--transition-fast);
+}
+
+.site-guide summary::-webkit-details-marker { display: none; }
+
+.site-guide summary::before {
+    content: '▶';
+    font-size: 0.6rem;
+    transition: transform var(--transition-fast);
+    flex-shrink: 0;
+}
+
+.site-guide[open] summary::before {
+    transform: rotate(90deg);
+}
+
+.site-guide summary:hover {
+    color: var(--text);
+}
+
+.site-guide-body {
+    padding: 0.75rem 1rem 1rem;
+    font-family: var(--font-sans);
+    font-size: var(--text-size-small);
+    color: var(--text-muted);
+    line-height: 1.6;
+    border-top: 1px solid var(--border);
+}
+
+.site-guide-body p  { margin: 0 0 0.6rem; }
+.site-guide-body p:last-child { margin-bottom: 0; }
+
+/* ============================================================
+   CURATED GRID
+   Hand-picked entry points, one per portal.
+   ============================================================ */
+
+.curated-grid {
+    display: grid;
+    grid-template-columns: 1fr 1fr;
+    gap: 1rem;
+    margin-top: 2.5rem;
+}
+
+@media (max-width: 580px) {
+    .curated-grid {
+        grid-template-columns: 1fr;
+    }
+}
+
+.curated-card {
+    display: flex;
+    flex-direction: column;
+    gap: 0.3rem;
+    padding: 0.9rem 1.1rem;
+    border: 1px solid var(--border);
+    border-radius: 3px;
+    text-decoration: none;
+    color: inherit;
+    transition: border-color var(--transition-fast);
+}
+
+.curated-card:hover {
+    border-color: var(--text-muted);
+}
+
+.curated-portal {
+    font-family: var(--font-sans);
+    font-size: 0.68rem;
+    font-variant: small-caps;
+    letter-spacing: 0.07em;
+    color: var(--text-faint);
+}
+
+.curated-title {
+    font-family: var(--font-serif);
+    font-size: 1rem;
+    line-height: 1.35;
+    color: var(--text);
+}
+
+/* Reset <button> to look like a card */
+button.curated-card {
+    background: none;
+    font: inherit;
+    text-align: left;
+    cursor: pointer;
+    width: 100%;
+}
+
+button.curated-card:hover {
+    border-color: var(--text-muted);
+}
+
+.curated-desc {
+    font-family: var(--font-sans);
+    font-size: var(--text-size-small);
+    color: var(--text-muted);
+    line-height: 1.45;
 }

static/css/layout.css

@@ -164,16 +164,22 @@ body > footer {
     text-align: right;
 }
 
-.footer-build-link {
+.footer-build {
     font-size: 0.72rem;
     color: var(--text-faint);
     font-variant-numeric: tabular-nums;
+}
+
+.footer-build-link {
+    margin-left: 0.3em;
+    font-size: 0.72rem;
+    color: var(--text-faint);
     text-decoration: none;
+    opacity: 0.6;
 }
 
 .footer-build-link:hover {
-    text-decoration: underline;
+    opacity: 1;
-    text-underline-offset: 2px;
 }
 
 
@@ -215,23 +221,6 @@ body > footer {
         grid-column: 1;
         width: 100%;
     }
-
-    /* Footer: stack vertically so three sections don't fight for width. */
-    body > footer {
-        flex-direction: column;
-        align-items: center;
-        gap: 0.3rem;
-        padding: 0.9rem 1rem;
-        text-align: center;
-    }
-
-    .footer-left {
-        justify-content: center;
-    }
-
-    .footer-right {
-        text-align: center;
-    }
 }
 
 /*  Below ~900px: body spans full width (TOC hidden, no phantom column). */

static/css/popups.css

@@ -104,31 +104,6 @@
     display: none;
 }
 
-/* Epistemic preview popup — content cloned from #epistemic; ep-* styles inherited */
-.popup-epistemic {
-    min-width: 230px;
-}
-
-/* PDF thumbnail popup — first-page image generated by pdftoppm at build time */
-.link-popup:has(.popup-pdf) {
-    padding: 0;
-    overflow: hidden;
-}
-
-.popup-pdf {
-    line-height: 0; /* collapse whitespace gap below <img> */
-}
-
-.popup-pdf-thumb {
-    display: block;
-    width: 320px;
-    max-width: 100%;
-    height: auto;
-    max-height: 420px;
-    object-fit: contain;
-    object-position: top left;
-}
-
 /* PGP signature popup */
 .popup-sig pre {
     margin: 0;

static/css/typography.css

@@ -215,8 +215,8 @@ abbr {
 sup { font-variant-position: super; line-height: 1; }
 sub { font-variant-position: sub; line-height: 1; }
 
-/* Realistic Ink Highlighting — excludes user annotation marks */
+/* Realistic Ink Highlighting */
-#markdownBody mark:not(.user-annotation) {
+#markdownBody mark {
     background-color: transparent;
     background-image: linear-gradient(
         104deg, 
@@ -231,7 +231,7 @@ sub { font-variant-position: sub; line-height: 1; }
     padding: 0 0.2em;
     color: inherit;
 }
-[data-theme="dark"] #markdownBody mark:not(.user-annotation) {
+[data-theme="dark"] #markdownBody mark {
     background-image: linear-gradient(
         104deg, 
         rgba(100, 130, 180, 0) 0%, 
@@ -240,17 +240,6 @@ sub { font-variant-position: sub; line-height: 1; }
         rgba(80, 110, 160, 0) 100%
     );
 }
-@media (prefers-color-scheme: dark) {
-    :root:not([data-theme="light"]) #markdownBody mark:not(.user-annotation) {
-        background-image: linear-gradient(
-            104deg,
-            rgba(100, 130, 180, 0) 0%,
-            rgba(100, 130, 180, 0.4) 2%,
-            rgba(80, 110, 160, 0.5) 98%,
-            rgba(80, 110, 160, 0) 100%
-        );
-    }
-}
 
 
 /* ============================================================
@@ -268,7 +257,6 @@ code, kbd, samp {
 }
 
 pre {
-    position: relative;
     font-family: var(--font-mono);
     font-size: 0.88em;
     font-feature-settings: 'liga' 1, 'calt' 1;

static/js/annotations.js

@@ -234,17 +234,6 @@
             return ann;
         },
         remove: removeById,
-        clearAll: function () {
-            saveAll([]);
-            document.querySelectorAll('mark.user-annotation').forEach(function (mark) {
-                var parent = mark.parentNode;
-                if (!parent) return;
-                while (mark.firstChild) parent.insertBefore(mark.firstChild, mark);
-                parent.removeChild(mark);
-                parent.normalize();
-            });
-            hideTooltip(true);
-        },
     };
 
     document.addEventListener('DOMContentLoaded', function () {

static/js/collapse.js

@@ -2,16 +2,21 @@
    Self-guards via #markdownBody check; no-ops on non-essay pages.
    Persists collapsed state per heading in localStorage.
    Retriggered sidenote positioning after each transition via window resize.
-
-   Exposes window.reinitCollapse(container) for use by transclude.js when
-   newly injected content contains collapsible headings.
 */
 (function () {
     'use strict';
 
     var PREFIX = 'section-collapsed:';
 
-    function initHeading(heading) {
+    document.addEventListener('DOMContentLoaded', function () {
+        var body = document.getElementById('markdownBody');
+        if (!body) return;
+        if (body.hasAttribute('data-no-collapse')) return;
+
+        var headings = Array.from(body.querySelectorAll('h2[id], h3[id]'));
+        if (!headings.length) return;
+
+        headings.forEach(function (heading) {
             var level   = parseInt(heading.tagName[1], 10);
             var content = [];
             var node    = heading.nextElementSibling;
@@ -93,23 +98,6 @@
                 }
                 window.dispatchEvent(new Event('resize'));
             });
-    }
-
-    document.addEventListener('DOMContentLoaded', function () {
-        var body = document.getElementById('markdownBody');
-        if (!body) return;
-        if (body.hasAttribute('data-no-collapse')) return;
-
-        var headings = Array.from(body.querySelectorAll('h2[id], h3[id]'));
-        if (!headings.length) return;
-
-        headings.forEach(initHeading);
         });
-
+    });
-    // Public entry point for transclude.js: initialize collapse toggles on
-    // headings inside a newly injected fragment.
-    window.reinitCollapse = function (container) {
-        Array.from(container.querySelectorAll('h2[id], h3[id]'))
-            .forEach(initHeading);
-    };
 }());

static/js/copy.js

@@ -1,45 +0,0 @@
-/* copy.js — Copy-to-clipboard button for <pre> code blocks.
- *
- * Injects a .copy-btn into every <pre> element on the page.
- * The button is visually hidden until the block is hovered (CSS handles this).
- * On click: copies the text content of the block, shows "copied" briefly.
- */
-
-(function () {
-    'use strict';
-
-    var RESET_DELAY = 1800; /* ms before label reverts to "copy" */
-
-    function attachButton(pre) {
-        var btn = document.createElement('button');
-        btn.className = 'copy-btn';
-        btn.textContent = 'copy';
-        btn.setAttribute('aria-label', 'Copy code to clipboard');
-
-        btn.addEventListener('click', function () {
-            var text = pre.querySelector('code')
-                ? pre.querySelector('code').innerText
-                : pre.innerText;
-
-            navigator.clipboard.writeText(text).then(function () {
-                btn.textContent = 'copied';
-                btn.setAttribute('data-copied', '');
-                setTimeout(function () {
-                    btn.textContent = 'copy';
-                    btn.removeAttribute('data-copied');
-                }, RESET_DELAY);
-            }).catch(function () {
-                btn.textContent = 'error';
-                setTimeout(function () {
-                    btn.textContent = 'copy';
-                }, RESET_DELAY);
-            });
-        });
-
-        pre.appendChild(btn);
-    }
-
-    document.addEventListener('DOMContentLoaded', function () {
-        document.querySelectorAll('pre').forEach(attachButton);
-    });
-}());

static/js/popups.js

@@ -38,10 +38,6 @@
     ------------------------------------------------------------------ */
 
     function init() {
-        // Hover popups are meaningless on touch-primary devices and interfere
-        // with tap navigation (first tap = hover, second tap = follow link).
-        if (window.matchMedia('(hover: none) and (pointer: coarse)').matches) return;
-
         popup = document.createElement('div');
         popup.className = 'link-popup';
         popup.setAttribute('aria-live', 'polite');
@@ -62,11 +58,6 @@
             bind(el, citationContent);
         });
 
-        /* Epistemic jump link — preview of the status/confidence/dot block */
-        root.querySelectorAll('a[href="#epistemic"]').forEach(function (el) {
-            bind(el, epistemicContent);
-        });
-
         /* Internal links — absolute (/foo) and relative (../../foo) same-origin hrefs.
            relativizeUrls in Hakyll makes index-page links relative, so we must match both. */
         root.querySelectorAll('a[href^="/"], a[href^="./"], a[href^="../"]').forEach(function (el) {
@@ -76,16 +67,10 @@
             if (!inAuthors && !isBacklink) {
                 if (el.closest('nav, #toc, footer, .page-meta-footer, .metadata')) return;
                 if (el.classList.contains('cite-link') || el.classList.contains('meta-tag')) return;
-                if (el.classList.contains('pdf-link')) return;
             }
             bind(el, internalContent);
         });
 
-        /* PDF links — rewritten to viewer URL by Links.hs; thumbnail on hover */
-        root.querySelectorAll('a.pdf-link[data-pdf-src]').forEach(function (el) {
-            bind(el, pdfContent);
-        });
-
         /* PGP signature links in footer */
         root.querySelectorAll('a.footer-sig-link').forEach(function (el) {
             bind(el, sigContent);
@@ -561,48 +546,6 @@
         return html;
     }
 
-    /* Epistemic jump link — reads the #epistemic section already in the DOM
-       and renders a compact summary: status/confidence/dots + expanded DL.
-       All ep-* classes are already styled via components.css. */
-    function epistemicContent() {
-        var section = document.getElementById('epistemic');
-        if (!section) return Promise.resolve(null);
-
-        var html = '<div class="popup-epistemic">';
-
-        var compact = section.querySelector('.ep-compact');
-        if (compact) {
-            html += '<div class="ep-compact">' + compact.innerHTML + '</div>';
-        }
-
-        var expanded = section.querySelector('.ep-expanded');
-        if (expanded) {
-            html += '<dl class="ep-expanded">' + expanded.innerHTML + '</dl>';
-        }
-
-        html += '</div>';
-        return Promise.resolve(html || null);
-    }
-
-    /* Local PDF — shows the build-time first-page thumbnail (.thumb.png).
-       Returns null (no popup) if the thumbnail file does not exist. */
-    function pdfContent(target) {
-        var src   = target.dataset.pdfSrc;
-        if (!src) return Promise.resolve(null);
-        var thumb = src.replace(/\.pdf$/i, '.thumb.png');
-        if (cache[thumb]) return Promise.resolve(cache[thumb]);
-        /* HEAD request: verify thumbnail exists before committing to a popup. */
-        return fetch(thumb, { method: 'HEAD', credentials: 'same-origin' })
-            .then(function (r) {
-                if (!r.ok) return null;
-                return store(thumb,
-                    '<div class="popup-pdf">'
-                    + '<img class="popup-pdf-thumb" src="' + esc(thumb) + '" alt="PDF first page">'
-                    + '</div>');
-            })
-            .catch(function () { return null; });
-    }
-
     /* PGP signature — fetch the .sig file and display ASCII armor */
     function sigContent(target) {
         var href = target.getAttribute('href');

static/js/random.js

@@ -1,12 +1,14 @@
-/* random.js — "Random Page" for the homepage.
+/* random.js — "Random Page" button for the homepage.
    Fetches /random-pages.json (essays + blog posts, generated at build time)
-   and navigates to a uniformly random entry on click.
+   and navigates to a uniformly random entry on click. */
-   Attaches to: #random-page-btn (legacy button) or [data-random] (link row). */
 (function () {
     'use strict';
 
-    function goRandom(e) {
+    document.addEventListener('DOMContentLoaded', function () {
-        e.preventDefault();
+        var btn = document.getElementById('random-page-btn');
+        if (!btn) return;
+
+        btn.addEventListener('click', function () {
             fetch('/random-pages.json')
                 .then(function (r) { return r.json(); })
                 .then(function (pages) {
@@ -14,12 +16,6 @@
                     window.location.href = pages[Math.floor(Math.random() * pages.length)];
                 })
                 .catch(function () {});
-    }
-
-    document.addEventListener('DOMContentLoaded', function () {
-        var els = document.querySelectorAll('#random-page-btn, [data-random]');
-        els.forEach(function (el) {
-            el.addEventListener('click', goRandom);
         });
     });
 }());

static/js/selection-popup.js

@@ -5,8 +5,10 @@
      code  (known lang) → Copy · [MDN / Hoogle / Docs…]
      code  (unknown)    → Copy
      math               → Copy · nLab · OEIS · Wolfram
-     prose (multi-word) → Annotate · BibTeX · Copy · DuckDuckGo · Here · Wikipedia
+     prose (multi-word) → Annotate* · BibTeX · Copy · DuckDuckGo · Here · Wikipedia
-     prose (one word)   → Annotate · BibTeX · Copy · Define · DuckDuckGo · Here · Wikipedia
+     prose (one word)   → Annotate* · BibTeX · Copy · Define · DuckDuckGo · Here · Wikipedia
+
+   (* = placeholder, not yet wired)
 */
 (function () {
     'use strict';
@@ -15,8 +17,6 @@
 
     var popup     = null;
     var showTimer = null;
-    var picker    = null;
-    var pickerColor = 'amber';
 
     /* ------------------------------------------------------------------
        Documentation providers keyed by Prism language identifier.
@@ -58,7 +58,7 @@
         document.addEventListener('keyup',     onKeyUp);
         document.addEventListener('mousedown', onMouseDown);
         document.addEventListener('keydown',   onKeyDown);
-        window.addEventListener('scroll', function () { hide(); hidePicker(); }, { passive: true });
+        window.addEventListener('scroll',      hide, { passive: true });
     }
 
     /* ------------------------------------------------------------------
@@ -80,13 +80,11 @@
 
     function onMouseDown(e) {
         if (popup.contains(e.target)) return;
-        if (picker && picker.classList.contains('is-visible') && picker.contains(e.target)) return;
         hide();
-        hidePicker();
     }
 
     function onKeyDown(e) {
-        if (e.key === 'Escape') { hide(); hidePicker(); }
+        if (e.key === 'Escape') hide();
     }
 
     /* ------------------------------------------------------------------
@@ -151,7 +149,7 @@
 
         position(rect);
         popup.style.visibility = '';
-        bindActions(text, rect);
+        bindActions(text);
     }
 
     function hide() {
@@ -213,9 +211,8 @@
                  + btn('wolfram', 'Wolfram');
         }
 
-        /* Prose: Annotate · BibTeX · Copy · [Define] · DuckDuckGo · Here · Wikipedia */
+        /* Prose — alphabetical: BibTeX · Copy · [Define] · DuckDuckGo · Here · Wikipedia */
-        return btn('annotate',  'Annotate')
+        return btn('cite',      'BibTeX')
-             + btn('cite',      'BibTeX')
              + btn('copy',      'Copy')
              + (oneWord ? btn('define', 'Define') : '')
              + btn('search',    'DuckDuckGo')
@@ -243,11 +240,11 @@
        Action bindings
     ------------------------------------------------------------------ */
 
-    function bindActions(text, rect) {
+    function bindActions(text) {
         popup.querySelectorAll('[data-action]').forEach(function (el) {
             if (el.getAttribute('aria-disabled') === 'true') return;
             el.addEventListener('click', function () {
-                dispatch(el.getAttribute('data-action'), text, el, rect);
+                dispatch(el.getAttribute('data-action'), text, el);
                 hide();
             });
         });
@@ -328,7 +325,7 @@
        Action dispatch
     ------------------------------------------------------------------ */
 
-    function dispatch(action, text, el, rect) {
+    function dispatch(action, text, el) {
         var q = encodeURIComponent(text);
         if (action === 'search') {
             window.open('https://duckduckgo.com/?q=' + q, '_blank', 'noopener,noreferrer');
@@ -365,97 +362,8 @@
         } else if (action === 'here') {
             /* Site search via Pagefind — opens search page with query pre-filled. */
             window.open('/search.html?q=' + q, '_blank', 'noopener,noreferrer');
-
-        } else if (action === 'annotate') {
-            showAnnotatePicker(text, rect);
         }
     }
 
-    /* ------------------------------------------------------------------
-       Annotate picker — color swatches + optional note input
-    ------------------------------------------------------------------ */
-
-    function showAnnotatePicker(text, selRect) {
-        if (!picker) {
-            picker = document.createElement('div');
-            picker.className = 'ann-picker';
-            picker.setAttribute('role', 'dialog');
-            picker.setAttribute('aria-label', 'Annotate selection');
-            document.body.appendChild(picker);
-        }
-
-        pickerColor = 'amber';
-        picker.innerHTML =
-            '<div class="ann-picker-swatches">'
-            + swatchBtn('amber') + swatchBtn('sage') + swatchBtn('steel') + swatchBtn('rose')
-            + '</div>'
-            + '<input class="ann-picker-note" type="text" placeholder="Note (optional)" maxlength="200">'
-            + '<div class="ann-picker-actions">'
-            + '<button class="ann-picker-submit">Highlight</button>'
-            + '</div>';
-
-        picker.style.visibility = 'hidden';
-        picker.classList.add('is-visible');
-        positionPicker(selRect);
-        picker.style.visibility = '';
-
-        var note      = picker.querySelector('.ann-picker-note');
-        var submitBtn = picker.querySelector('.ann-picker-submit');
-
-        picker.querySelector('.ann-picker-swatch[data-color="amber"]').classList.add('is-selected');
-
-        picker.querySelectorAll('.ann-picker-swatch').forEach(function (sw) {
-            sw.addEventListener('click', function () {
-                picker.querySelectorAll('.ann-picker-swatch').forEach(function (s) {
-                    s.classList.remove('is-selected');
-                });
-                sw.classList.add('is-selected');
-                pickerColor = sw.getAttribute('data-color');
-            });
-        });
-
-        function commit() {
-            if (window.Annotations) {
-                window.Annotations.add(text, pickerColor, note.value.trim());
-            }
-            hidePicker();
-        }
-
-        submitBtn.addEventListener('click', commit);
-        note.addEventListener('keydown', function (e) {
-            if (e.key === 'Enter')  { e.preventDefault(); commit(); }
-            if (e.key === 'Escape') { hidePicker(); }
-        });
-
-        setTimeout(function () { note.focus(); }, 0);
-    }
-
-    function hidePicker() {
-        if (picker) picker.classList.remove('is-visible');
-    }
-
-    function positionPicker(rect) {
-        var pw  = picker.offsetWidth;
-        var ph  = picker.offsetHeight;
-        var GAP = 8;
-        var sy  = window.scrollY;
-        var sx  = window.scrollX;
-        var vw  = window.innerWidth;
-
-        var left = rect.left + sx + rect.width / 2 - pw / 2;
-        left = Math.max(sx + GAP, Math.min(left, sx + vw - pw - GAP));
-
-        var top = rect.top + sy - ph - GAP;
-        if (top < sy + GAP) top = rect.bottom + sy + GAP;
-
-        picker.style.left = left + 'px';
-        picker.style.top  = top  + 'px';
-    }
-
-    function swatchBtn(color) {
-        return '<button class="ann-picker-swatch ann-picker-swatch--' + color
-            + '" data-color="' + color + '" aria-label="' + color + '"></button>';
-    }
-
     document.addEventListener('DOMContentLoaded', init);
 }());

static/js/semantic-search.js

@@ -1,220 +0,0 @@
-/* semantic-search.js — Client-side semantic search using paragraph embeddings.
- *
- * At build time, tools/embed.py produces:
- *   /data/semantic-index.bin   raw Float32Array (N_paragraphs × 384 dims)
- *   /data/semantic-meta.json   [{url, title, heading, excerpt}, ...]
- *
- * At query time, transformers.js embeds the user's query with all-MiniLM-L6-v2
- * (same model used at build time) and ranks paragraphs by cosine similarity.
- * All computation is client-side; no server required.
- *
- * Model: Xenova/all-MiniLM-L6-v2 (~22 MB quantized, cached by browser after first load)
- * Model files served from /models/all-MiniLM-L6-v2/ (same-origin; run tools/download-model.sh)
- * Index format: raw little-endian Float32, shape [N, 384], unit-normalized
- *
- * CSP: requires cdn.jsdelivr.net in script-src (transformers.js library).
- *      connect-src stays 'self' — model weights are served same-origin.
- */
-(function () {
-    'use strict';
-
-    var MODEL      = 'all-MiniLM-L6-v2';          /* local name, no Xenova/ prefix */
-    var MODEL_PATH = '/models/';                   /* served same-origin */
-    var DIM        = 384;
-    var TOP_K      = 8;
-    var CDN        = 'https://cdn.jsdelivr.net/npm/@xenova/transformers@2';
-
-    var extractor = null;   /* loaded lazily on first search */
-    var vectors   = null;   /* Float32Array, shape [N, DIM] */
-    var meta      = null;   /* [{url, title, heading, excerpt}] */
-    var indexReady = false;
-
-    var queryEl   = document.getElementById('semantic-query');
-    var statusEl  = document.getElementById('semantic-status');
-    var resultsEl = document.getElementById('semantic-results');
-
-    if (!queryEl) return;   /* not on the search page */
-
-    /* ------------------------------------------------------------------
-       Index loading — fetch once, lazily
-    ------------------------------------------------------------------ */
-
-    function loadIndex() {
-        if (indexReady) return Promise.resolve();
-
-        return Promise.all([
-            fetch('/data/semantic-index.bin').then(function (r) {
-                if (!r.ok) throw new Error('semantic-index.bin not found');
-                return r.arrayBuffer();
-            }),
-            fetch('/data/semantic-meta.json').then(function (r) {
-                if (!r.ok) throw new Error('semantic-meta.json not found');
-                return r.json();
-            }),
-        ]).then(function (results) {
-            vectors   = new Float32Array(results[0]);
-            meta      = results[1];
-            indexReady = true;
-        });
-    }
-
-    /* ------------------------------------------------------------------
-       Model loading — dynamic import from CDN, lazy
-    ------------------------------------------------------------------ */
-
-    function loadModel() {
-        if (extractor) return Promise.resolve(extractor);
-        setStatus('Loading model…');
-        return import(CDN).then(function (mod) {
-            /* Point transformers.js at our self-hosted model files. */
-            mod.env.localModelPath   = MODEL_PATH;
-            mod.env.allowRemoteModels = false;
-            return mod.pipeline('feature-extraction', MODEL, { quantized: true });
-        }).then(function (pipe) {
-            extractor = pipe;
-            return extractor;
-        });
-    }
-
-    /* ------------------------------------------------------------------
-       Search
-    ------------------------------------------------------------------ */
-
-    function cosineSims(queryVec) {
-        /* queryVec is already unit-normalized; dot product = cosine similarity */
-        var N       = meta.length;
-        var scores  = new Float32Array(N);
-        for (var i = 0; i < N; i++) {
-            var dot = 0;
-            var off = i * DIM;
-            for (var d = 0; d < DIM; d++) dot += queryVec[d] * vectors[off + d];
-            scores[i] = dot;
-        }
-        return scores;
-    }
-
-    function topK(scores) {
-        var indices = Array.from({ length: meta.length }, function (_, i) { return i; });
-        indices.sort(function (a, b) { return scores[b] - scores[a]; });
-        return indices.slice(0, TOP_K).map(function (i) {
-            return { idx: i, score: scores[i] };
-        });
-    }
-
-    function runSearch(query) {
-        query = query.trim();
-        if (!query) { clearResults(); return; }
-
-        setStatus('Searching…');
-
-        var indexPromise = loadIndex().catch(function (err) {
-            setStatus('Semantic index not available — run make build first.');
-            throw err;
-        });
-        var modelPromise = loadModel();
-
-        Promise.all([indexPromise, modelPromise]).then(function (results) {
-            var pipe = results[1];
-            return pipe(query, { pooling: 'mean', normalize: true });
-        }).then(function (output) {
-            var queryVec = output.data;   /* Float32Array, length 384 */
-            var scores   = cosineSims(queryVec);
-            var hits     = topK(scores);
-            renderResults(hits);
-            setStatus(hits.length ? '' : 'No results found.');
-        }).catch(function (err) {
-            if (err.message && err.message.indexOf('not available') === -1) {
-                setStatus('Search error — see console for details.');
-                console.error('semantic-search:', err);
-            }
-        });
-    }
-
-    /* ------------------------------------------------------------------
-       Rendering
-    ------------------------------------------------------------------ */
-
-    function renderResults(hits) {
-        if (!hits.length) { clearResults(); return; }
-
-        var html = '<ol class="semantic-results-list">';
-        for (var i = 0; i < hits.length; i++) {
-            var h = hits[i];
-            var m = meta[h.idx];
-            var sameHeading = m.heading === m.title;
-            html += '<li class="semantic-result">'
-                 + '<a class="semantic-result-title" href="' + esc(m.url) + '">'
-                 + esc(m.title) + '</a>';
-            if (!sameHeading) {
-                html += '<span class="semantic-result-heading"> § ' + esc(m.heading) + '</span>';
-            }
-            html += '<p class="semantic-result-excerpt">' + esc(m.excerpt) + '</p>'
-                 + '</li>';
-        }
-        html += '</ol>';
-        resultsEl.innerHTML = html;
-    }
-
-    function clearResults() {
-        resultsEl.innerHTML = '';
-    }
-
-    function setStatus(msg) {
-        statusEl.textContent = msg;
-    }
-
-    function esc(s) {
-        return String(s)
-            .replace(/&/g, '&amp;')
-            .replace(/</g, '&lt;')
-            .replace(/>/g, '&gt;')
-            .replace(/"/g, '&quot;');
-    }
-
-    /* ------------------------------------------------------------------
-       Tab switching — persists choice in localStorage
-    ------------------------------------------------------------------ */
-
-    var STORAGE_KEY = 'search-tab';
-
-    function activateTab(target) {
-        document.querySelectorAll('.search-tab').forEach(function (b) {
-            var active = b.dataset.tab === target;
-            b.classList.toggle('is-active', active);
-            b.setAttribute('aria-selected', active ? 'true' : 'false');
-        });
-        document.querySelectorAll('.search-panel').forEach(function (p) {
-            p.classList.toggle('is-active', p.dataset.panel === target);
-        });
-        try { localStorage.setItem(STORAGE_KEY, target); } catch (e) {}
-    }
-
-    document.querySelectorAll('.search-tab').forEach(function (btn) {
-        btn.addEventListener('click', function () { activateTab(btn.dataset.tab); });
-    });
-
-    /* Restore last-used tab (falls back to keyword if unset or unrecognised) */
-    var saved = null;
-    try { saved = localStorage.getItem(STORAGE_KEY); } catch (e) {}
-    if (saved === 'semantic') activateTab('semantic');
-
-    /* ------------------------------------------------------------------
-       Input handling — debounced, 400 ms
-    ------------------------------------------------------------------ */
-
-    var debounceTimer = null;
-    queryEl.addEventListener('input', function () {
-        clearTimeout(debounceTimer);
-        var q = queryEl.value.trim();
-        if (!q) { clearResults(); setStatus(''); return; }
-        debounceTimer = setTimeout(function () { runSearch(q); }, 400);
-    });
-
-    /* Pre-fill from ?q= on load — mirror keyword search behaviour */
-    var params = new URLSearchParams(window.location.search);
-    var initial = params.get('q');
-    if (initial) {
-        queryEl.value = initial;
-        runSearch(initial);
-    }
-}());

static/js/settings.js

@@ -89,13 +89,6 @@
         else if (action === 'focus-mode')    toggleDataAttr('focus-mode',    'data-focus-mode');
         else if (action === 'reduce-motion') toggleDataAttr('reduce-motion', 'data-reduce-motion');
         else if (action === 'print')         { setOpen(false); window.print(); }
-        else if (action === 'clear-annotations') { clearAnnotations(); }
-    }
-
-    function clearAnnotations() {
-        if (!confirm('Remove all highlights and annotations across every page?')) return;
-        if (window.Annotations) window.Annotations.clearAll();
-        setOpen(false);
     }
 
     /* Theme ----------------------------------------------------------- */

static/js/transclude.js

@@ -1,139 +0,0 @@
-/* transclude.js — Client-side lazy transclusion.
- *
- * Authored in Markdown as a standalone line:
- *   {{slug}}              — embed full body of /slug.html
- *   {{slug#section-id}}   — embed one section by heading id
- *   {{path/to/page}}      — sub-path pages work the same way
- *
- * The Haskell preprocessor (Filters.Transclusion) converts these at build
- * time to placeholder divs:
- *   <div class="transclude" data-src="/slug.html"
- *        data-section="section-id"></div>
- *
- * This script finds those divs, fetches the target page, extracts the
- * requested content, rewrites cross-page fragment hrefs, injects the
- * content inline, and retriggers layout-dependent JS (sidenotes, collapse).
- */
-
-(function () {
-    'use strict';
-
-    /* Shared fetch cache — one network request per URL regardless of how
-     * many transclusions reference the same page. */
-    var cache = {};
-
-    function fetchPage(url) {
-        if (!cache[url]) {
-            cache[url] = fetch(url).then(function (r) {
-                if (!r.ok) throw new Error('HTTP ' + r.status);
-                return r.text();
-            });
-        }
-        return cache[url];
-    }
-
-    function parseDoc(html) {
-        return new DOMParser().parseFromString(html, 'text/html');
-    }
-
-    /* Extract a named section: the heading element with id=sectionId plus
-     * all following siblings until the next heading at the same or higher
-     * level (lower number), or end of parent. */
-    function extractSection(doc, sectionId) {
-        var anchor = doc.getElementById(sectionId);
-        if (!anchor) return null;
-
-        var level = parseInt(anchor.tagName[1], 10);
-        if (!level) return null;
-
-        var nodes = [anchor.cloneNode(true)];
-        var el    = anchor.nextElementSibling;
-
-        while (el) {
-            if (/^H[1-6]$/.test(el.tagName) &&
-                parseInt(el.tagName[1], 10) <= level) break;
-            nodes.push(el.cloneNode(true));
-            el = el.nextElementSibling;
-        }
-
-        return nodes.length ? nodes : null;
-    }
-
-    /* Extract the full contents of #markdownBody. */
-    function extractBody(doc) {
-        var body = doc.getElementById('markdownBody');
-        if (!body) return null;
-        var nodes = Array.from(body.children).map(function (el) {
-            return el.cloneNode(true);
-        });
-        return nodes.length ? nodes : null;
-    }
-
-    /* Rewrite href="#fragment" → href="srcUrl#fragment" so in-page anchor
-     * links from the source page remain valid when embedded elsewhere. */
-    function rewriteFragmentHrefs(nodes, srcUrl) {
-        nodes.forEach(function (node) {
-            node.querySelectorAll('a[href^="#"]').forEach(function (a) {
-                a.setAttribute('href', srcUrl + a.getAttribute('href'));
-            });
-        });
-    }
-
-    /* After injection, retrigger layout-dependent subsystems. */
-    function reinitFragment(container) {
-        /* sidenotes.js repositions on resize — dispatch to trigger it. */
-        window.dispatchEvent(new Event('resize'));
-
-        /* collapse.js exposes reinitCollapse for newly added headings. */
-        if (typeof window.reinitCollapse === 'function') {
-            window.reinitCollapse(container);
-        }
-
-        /* gallery.js can expose reinitGallery when needed. */
-        if (typeof window.reinitGallery === 'function') {
-            window.reinitGallery(container);
-        }
-    }
-
-    function loadTransclusion(el) {
-        var src     = el.dataset.src;
-        var section = el.dataset.section || null;
-        if (!src) return;
-
-        el.classList.add('transclude--loading');
-
-        fetchPage(src)
-            .then(function (html) {
-                var doc   = parseDoc(html);
-                var nodes = section
-                    ? extractSection(doc, section)
-                    : extractBody(doc);
-
-                if (!nodes) {
-                    el.classList.replace('transclude--loading', 'transclude--error');
-                    el.textContent = '[transclusion not found: '
-                        + src + (section ? '#' + section : '') + ']';
-                    return;
-                }
-
-                rewriteFragmentHrefs(nodes, src);
-
-                var wrapper = document.createElement('div');
-                wrapper.className = 'transclude--content';
-                nodes.forEach(function (n) { wrapper.appendChild(n); });
-
-                el.classList.replace('transclude--loading', 'transclude--loaded');
-                el.appendChild(wrapper);
-
-                reinitFragment(el);
-            })
-            .catch(function (err) {
-                el.classList.replace('transclude--loading', 'transclude--error');
-                console.warn('transclude: failed to load', src, err);
-            });
-    }
-
-    document.addEventListener('DOMContentLoaded', function () {
-        document.querySelectorAll('div.transclude').forEach(loadTransclusion);
-    });
-}());

templates/default.html

@@ -9,7 +9,6 @@ $partial("templates/partials/nav.html")$
 $if(search)$
 <script src="/pagefind/pagefind-ui.js"></script>
 <script src="/js/search.js" defer></script>
-<script src="/js/semantic-search.js" defer></script>
 $endif$
 $body$
 $partial("templates/partials/footer.html")$

templates/partials/footer.html

@@ -7,7 +7,7 @@
         <span class="footer-license">CC BY-SA-NC 4.0 · <a href="https://github.com/levineuwirth/levineuwirth.org">MIT</a> · <a href="/memento-mori.html" class="footer-mm">MM</a></span>
     </div>
     <div class="footer-right">
-        <a href="/build/" class="footer-build-link" aria-label="Build telemetry">build</a> $build-time$
+        <span class="footer-build">build $build-time$</span><a href="/build/" class="footer-build-link" aria-label="Build telemetry">→</a>
         · <a href="$url$.sig" class="footer-sig-link" aria-label="PGP signature for this page" title="Ed25519 signing subkey C9A42A6F AD444FBE 566FD738 531BDC1C C2707066 · public key at /gpg/pubkey.asc">sig</a>
     </div>
 </footer>

templates/partials/head.html

@@ -39,5 +39,3 @@ $if(viz)$
 <script src="/js/viz.js" defer></script>
 $endif$
 <script src="/js/collapse.js" defer></script>
-<script src="/js/transclude.js" defer></script>
-<script src="/js/copy.js" defer></script>

templates/partials/metadata.html

@@ -12,14 +12,9 @@
     <div class="meta-row meta-authors">
         <span class="meta-label">by</span>$if(poet)$$poet$$else$$for(author-links)$<a href="$author-url$">$author-name$</a>$sep$, $endfor$$endif$
     </div>
-    $if(affiliation-links)$
-    <div class="meta-row meta-affiliation">
-        $for(affiliation-links)$$if(affiliation-url)$<a href="$affiliation-url$">$affiliation-name$</a>$else$$affiliation-name$$endif$$sep$ · $endfor$
-    </div>
-    $endif$
     <nav class="meta-row meta-pagelinks" aria-label="Page sections">
         <a href="#version-history">History</a>
-        $if(status)$<a href="#epistemic">$if(overall-score)$<span class="ep-link-label">Epistemic</span><span class="meta-overall-score" title="Overall score: $overall-score$/100">$overall-score$%</span>$else$Epistemic$endif$</a>$endif$
+        $if(status)$<a href="#epistemic">Epistemic</a>$endif$
         $if(bibliography)$<a href="#bibliography">Bibliography</a>$endif$
         $if(backlinks)$<a href="#backlinks">Backlinks</a>$endif$
     </nav>

templates/partials/nav.html

@@ -41,7 +41,6 @@
                         </div>
                         <div class="settings-section">
                             <button class="settings-btn settings-btn--full" data-action="print">Print</button>
-                            <button class="settings-btn settings-btn--full settings-btn--danger" data-action="clear-annotations">Clear Annotations</button>
                         </div>
                     </div>
                 </div>

templates/partials/page-footer.html

@@ -34,7 +34,7 @@
 
         $if(status)$
         <div class="meta-footer-section meta-footer-epistemic" id="epistemic">
-            <h3><a href="/colophon.html#living-documents">Epistemic</a></h3>
+            <h3>Epistemic</h3>
             <div class="ep-compact">
                 <span class="ep-status">$status$</span>
                 $if(confidence)$<span class="ep-confidence" title="Confidence">$confidence$%</span>$endif$

tools/convert-images.sh

@@ -1,39 +0,0 @@
-#!/usr/bin/env bash
-# convert-images.sh — Produce WebP companions for every local raster image.
-#
-# Walks static/ and content/ for JPEG and PNG files and calls cwebp to produce
-# a .webp file alongside each one.  Existing .webp files are skipped (safe to
-# re-run).  If cwebp is not found the script exits 0 so the build continues.
-#
-# Requires: cwebp (libwebp) — pacman -S libwebp  /  apt install webp
-#
-# Quality: -q 85 is a good default for photographic content.  For images that
-# are already highly compressed, -lossless avoids further degradation.
-
-set -euo pipefail
-
-if ! command -v cwebp >/dev/null 2>&1; then
-    echo "convert-images: cwebp not found — skipping WebP conversion." >&2
-    echo "  Install: pacman -S libwebp  (or: apt install webp)" >&2
-    exit 0
-fi
-
-REPO_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
-
-converted=0
-skipped=0
-
-while IFS= read -r -d '' img; do
-    webp="${img%.*}.webp"
-    if [ -f "$webp" ]; then
-        skipped=$((skipped + 1))
-    else
-        echo "  webp  ${img#"$REPO_ROOT/"}"
-        cwebp -quiet -q 85 "$img" -o "$webp"
-        converted=$((converted + 1))
-    fi
-done < <(find "$REPO_ROOT/static" "$REPO_ROOT/content" \
-              \( -name "*.jpg" -o -name "*.jpeg" -o -name "*.png" \) \
-              -print0 2>/dev/null)
-
-echo "convert-images: ${converted} converted, ${skipped} already present."

tools/download-model.sh

@@ -1,38 +0,0 @@
-#!/usr/bin/env bash
-# download-model.sh — Download the quantized ONNX model for client-side semantic search.
-#
-# Downloads Xenova/all-MiniLM-L6-v2 (quantized ONNX, ~22 MB total) from HuggingFace
-# into static/models/all-MiniLM-L6-v2/ so it can be served same-origin, keeping the
-# nginx CSP to 'self' for connect-src.
-#
-# Run once before deploying. Files are gitignored (binary artifacts).
-# Re-running is safe — existing files are skipped.
-
-set -euo pipefail
-
-REPO_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
-MODEL_DIR="$REPO_ROOT/static/models/all-MiniLM-L6-v2"
-BASE_URL="https://huggingface.co/Xenova/all-MiniLM-L6-v2/resolve/main"
-
-mkdir -p "$MODEL_DIR/onnx"
-
-fetch() {
-    local src="$1" dst="$2"
-    if [ -f "$dst" ]; then
-        echo "  skip  $src (already present)"
-        return
-    fi
-    echo "  fetch $src"
-    curl -fsSL --progress-bar "$BASE_URL/$src" -o "$dst"
-}
-
-echo "Downloading all-MiniLM-L6-v2 to $MODEL_DIR ..."
-
-fetch "config.json"                  "$MODEL_DIR/config.json"
-fetch "tokenizer.json"               "$MODEL_DIR/tokenizer.json"
-fetch "tokenizer_config.json"        "$MODEL_DIR/tokenizer_config.json"
-fetch "special_tokens_map.json"      "$MODEL_DIR/special_tokens_map.json"
-fetch "onnx/model_quantized.onnx"    "$MODEL_DIR/onnx/model_quantized.onnx"
-
-echo "Done. static/models/all-MiniLM-L6-v2/ is ready."
-echo "Run 'make build' to copy the model files into _site/."

tools/embed.py

@@ -1,21 +1,22 @@
 #!/usr/bin/env python3
 """
-embed.py — Build-time embedding pipeline.
+embed.py — Build-time similar-links generator.
 
-Produces two outputs from _site/**/*.html:
+Reads _site/**/*.html, embeds each page with nomic-embed-text-v1.5,
+builds a FAISS IndexFlatIP, and writes data/similar-links.json:
 
-  data/similar-links.json       Page-level similarity (for "Related" footer section)
+  { "/path/to/page/": [{"url": "...", "title": "...", "score": 0.87}, ...] }
-  data/semantic-index.bin       Paragraph vectors as raw Float32 array (N × DIM)
-  data/semantic-meta.json       Paragraph metadata: [{url, title, heading, excerpt}]
 
-Both use all-MiniLM-L6-v2 (384 dims) — the same model shipped to the browser
+Called by `make build` when .venv exists. Failures are non-fatal (make prints
-via transformers.js for query-time semantic search.
+a warning and continues). Run `uv sync` first to provision the environment.
 
-Called by `make build` when .venv exists. Failures are non-fatal.
+Staleness check: skips re-embedding if data/similar-links.json is newer than
-Staleness check: skips if all output files are newer than every HTML in _site/.
+every HTML file in _site/ — so content-only rebuilds that don't touch HTML
+won't re-embed.
 """
 
 import json
+import os
 import re
 import sys
 from pathlib import Path
@@ -31,116 +32,83 @@ from sentence_transformers import SentenceTransformer
 
 REPO_ROOT   = Path(__file__).parent.parent
 SITE_DIR    = REPO_ROOT / "_site"
-SIMILAR_OUT    = REPO_ROOT / "data" / "similar-links.json"
+OUT_FILE    = REPO_ROOT / "data" / "similar-links.json"
-SEMANTIC_BIN   = REPO_ROOT / "data" / "semantic-index.bin"
+MODEL_NAME  = "nomic-ai/nomic-embed-text-v1.5"
-SEMANTIC_META  = REPO_ROOT / "data" / "semantic-meta.json"
+TOP_N       = 5
-
+MIN_SCORE   = 0.30   # cosine similarity threshold; discard weak matches
-MODEL_NAME     = "all-MiniLM-L6-v2"
+# Pages to exclude from both indexing and results (exact URL paths)
-DIM            = 384
-
-TOP_N          = 5      # similar-links: neighbours per page
-MIN_SCORE      = 0.30   # similar-links: discard weak matches
-MIN_PARA_CHARS = 80     # semantic: skip very short paragraphs
-MAX_PARA_CHARS = 1000   # semantic: truncate before embedding
-
 EXCLUDE_URLS = {"/search/", "/build/", "/404.html", "/feed.xml", "/music/feed.xml"}
 
-STRIP_SELECTORS = [
-    "nav", "footer", "#toc", ".link-popup", "script", "style",
-    ".page-meta-footer", ".metadata", "[data-pagefind-ignore]",
-]
-
 # ---------------------------------------------------------------------------
 # Staleness check
 # ---------------------------------------------------------------------------
 
 def needs_update() -> bool:
-    outputs = [SIMILAR_OUT, SEMANTIC_BIN, SEMANTIC_META]
+    """Return True if similar-links.json is missing or older than any _site HTML."""
-    if not all(p.exists() for p in outputs):
+    if not OUT_FILE.exists():
         return True
-    oldest = min(p.stat().st_mtime for p in outputs)
+    json_mtime = OUT_FILE.stat().st_mtime
-    return any(html.stat().st_mtime > oldest for html in SITE_DIR.rglob("*.html"))
+    for html in SITE_DIR.rglob("*.html"):
+        if html.stat().st_mtime > json_mtime:
+            return True
+    return False
 
 # ---------------------------------------------------------------------------
-# HTML parsing helpers
+# HTML → text extraction
 # ---------------------------------------------------------------------------
 
-def _url_from_path(html_path: Path) -> str:
+def extract(html_path: Path) -> dict | None:
+    """
+    Parse an HTML file and extract:
+      - url:     root-relative URL path  (e.g. "/essays/my-essay/")
+      - title:   page <title> text
+      - text:    plain text of the page body (nav/footer/TOC stripped)
+    Returns None for pages that should not be indexed.
+    """
+    raw = html_path.read_text(encoding="utf-8", errors="replace")
+    soup = BeautifulSoup(raw, "html.parser")
+
+    # Derive root-relative URL from file path
     rel = html_path.relative_to(SITE_DIR)
     if rel.name == "index.html":
         url = "/" + str(rel.parent) + "/"
-        return url.replace("//", "/")
+        url = url.replace("//", "/")   # root index.html → "/"
-    return "/" + str(rel)
+    else:
-
+        url = "/" + str(rel)
-def _clean_soup(soup: BeautifulSoup) -> None:
-    for sel in STRIP_SELECTORS:
-        for el in soup.select(sel):
-            el.decompose()
-
-def _title(soup: BeautifulSoup, url: str) -> str:
-    h1 = soup.find("h1")
-    if h1:
-        return h1.get_text(" ", strip=True)
-    tag = soup.find("title")
-    raw = tag.get_text(" ", strip=True) if tag else url
-    return re.split(r"\s+[—–-]\s+", raw)[0].strip()
-
-# ---------------------------------------------------------------------------
-# Page-level extraction  (for similar-links)
-# ---------------------------------------------------------------------------
-
-def extract_page(html_path: Path) -> dict | None:
-    raw  = html_path.read_text(encoding="utf-8", errors="replace")
-    soup = BeautifulSoup(raw, "html.parser")
-    url  = _url_from_path(html_path)
 
     if url in EXCLUDE_URLS:
         return None
+
+    # Only index actual content pages — skip index/tag/feed/author pages
+    # that have no prose body.
     body = soup.select_one("#markdownBody")
     if body is None:
         return None
 
-    title = _title(soup, url)
+    # Title: prefer <h1>, fall back to <title> (strip " — Site Name" suffix)
-    _clean_soup(soup)
+    h1 = soup.find("h1")
+    if h1:
+        title = h1.get_text(" ", strip=True)
+    else:
+        title_tag = soup.find("title")
+        raw_title = title_tag.get_text(" ", strip=True) if title_tag else url
+        title = re.split(r"\s+[—–-]\s+", raw_title)[0].strip()
 
-    text = re.sub(r"\s+", " ", body.get_text(" ", strip=True)).strip()
+    # Remove elements that aren't content
-    if len(text) < 100:
+    for sel in ["nav", "footer", "#toc", ".link-popup", "script", "style",
+                ".page-meta-footer", ".metadata", "[data-pagefind-ignore]"]:
+        for el in soup.select(sel):
+            el.decompose()
+
+    text = body.get_text(" ", strip=True)
+    # Collapse runs of whitespace
+    text = re.sub(r"\s+", " ", text).strip()
+
+    if len(text) < 100:   # too short to embed meaningfully
         return None
 
-    return {"url": url, "title": title, "text": text}
+    # Feed title + text to the model so title is part of the representation
-
+    return {"url": url, "title": title, "text": f"search_document: {title}\n\n{text}"}
-# ---------------------------------------------------------------------------
-# Paragraph-level extraction  (for semantic search)
-# ---------------------------------------------------------------------------
-
-def extract_paragraphs(html_path: Path, url: str, title: str) -> list[dict]:
-    raw  = html_path.read_text(encoding="utf-8", errors="replace")
-    soup = BeautifulSoup(raw, "html.parser")
-    body = soup.select_one("#markdownBody")
-    if body is None:
-        return []
-
-    _clean_soup(soup)
-
-    paras    = []
-    heading  = title  # track current section heading
-
-    for el in body.find_all(["h1", "h2", "h3", "h4", "p", "li", "blockquote"]):
-        if el.name in ("h1", "h2", "h3", "h4"):
-            heading = el.get_text(" ", strip=True)
-            continue
-        text = re.sub(r"\s+", " ", el.get_text(" ", strip=True)).strip()
-        if len(text) < MIN_PARA_CHARS:
-            continue
-        paras.append({
-            "url":     url,
-            "title":   title,
-            "heading": heading,
-            "excerpt": text[:200] + ("…" if len(text) > 200 else ""),
-            "text":    text[:MAX_PARA_CHARS],
-        })
-
-    return paras
 
 # ---------------------------------------------------------------------------
 # Main
@@ -152,82 +120,67 @@ def main() -> int:
         return 0
 
     if not needs_update():
-        print("embed.py: all outputs up to date — skipping")
+        print("embed.py: similar-links.json is up to date — skipping")
         return 0
 
-    # --- Extract pages + paragraphs in one pass ---
     print("embed.py: extracting pages…")
     pages = []
-    paragraphs = []
-
     for html in sorted(SITE_DIR.rglob("*.html")):
-        page = extract_page(html)
+        page = extract(html)
-        if page is None:
+        if page:
-            continue
             pages.append(page)
-        paragraphs.extend(extract_paragraphs(html, page["url"], page["title"]))
 
     if not pages:
         print("embed.py: no indexable pages found", file=sys.stderr)
         return 0
 
-    # --- Load model once for both tasks ---
+    print(f"embed.py: embedding {len(pages)} pages with {MODEL_NAME}…")
-    print(f"embed.py: loading {MODEL_NAME}…")
+    model = SentenceTransformer(MODEL_NAME, trust_remote_code=True)
-    model = SentenceTransformer(MODEL_NAME)
 
-    # --- Similar-links (page level) ---
+    texts = [p["text"] for p in pages]
-    print(f"embed.py: embedding {len(pages)} pages…")
+    # nomic requires a task prefix; we used "search_document:" above for the
-    page_vecs = model.encode(
+    # corpus. For queries we'd use "search_query:" — but here both corpus and
-        [p["text"] for p in pages],
+    # query are the same documents, so we use "search_document:" throughout.
-        normalize_embeddings=True,
+    embeddings = model.encode(
+        texts,
+        normalize_embeddings=True,   # unit vectors → inner product == cosine
         show_progress_bar=True,
-        batch_size=64,
+        batch_size=32,
-    ).astype(np.float32)
+    )
+    embeddings = np.array(embeddings, dtype=np.float32)
 
-    index = faiss.IndexFlatIP(page_vecs.shape[1])
+    print("embed.py: building FAISS index…")
-    index.add(page_vecs)
+    dim   = embeddings.shape[1]
-    scores_all, indices_all = index.search(page_vecs, TOP_N + 1)
+    index = faiss.IndexFlatIP(dim)   # exact inner product; fine for < 10k pages
+    index.add(embeddings)
 
-    similar: dict[str, list] = {}
+    print("embed.py: querying nearest neighbours…")
+    # Query all at once: returns (n_pages, TOP_N+1) — +1 because self is #1
+    scores_all, indices_all = index.search(embeddings, TOP_N + 1)
+
+    result: dict[str, list] = {}
     for i, page in enumerate(pages):
         neighbours = []
         for rank in range(TOP_N + 1):
-            j, score = int(indices_all[i, rank]), float(scores_all[i, rank])
+            j     = int(indices_all[i, rank])
-            if j == i or score < MIN_SCORE:
+            score = float(scores_all[i, rank])
-                continue
+            if j == i:
-            neighbours.append({"url": pages[j]["url"], "title": pages[j]["title"],
+                continue                      # skip self
-                                "score": round(score, 4)})
+            if score < MIN_SCORE:
+                continue                      # skip weak matches
+            neighbours.append({
+                "url":   pages[j]["url"],
+                "title": pages[j]["title"],
+                "score": round(score, 4),
+            })
             if len(neighbours) == TOP_N:
                 break
         if neighbours:
-            similar[page["url"]] = neighbours
+            result[page["url"]] = neighbours
 
-    SIMILAR_OUT.parent.mkdir(parents=True, exist_ok=True)
+    OUT_FILE.parent.mkdir(parents=True, exist_ok=True)
-    SIMILAR_OUT.write_text(json.dumps(similar, ensure_ascii=False, indent=2))
+    OUT_FILE.write_text(json.dumps(result, ensure_ascii=False, indent=2))
-    print(f"embed.py: wrote {len(similar)} similar-links entries")
+    print(f"embed.py: wrote {len(result)} entries to {OUT_FILE.relative_to(REPO_ROOT)}")
-
-    # --- Semantic index (paragraph level) ---
-    if not paragraphs:
-        print("embed.py: no paragraphs extracted — skipping semantic index")
-        return 0
-
-    print(f"embed.py: embedding {len(paragraphs)} paragraphs…")
-    para_vecs = model.encode(
-        [p["text"] for p in paragraphs],
-        normalize_embeddings=True,
-        show_progress_bar=True,
-        batch_size=64,
-    ).astype(np.float32)
-
-    SEMANTIC_BIN.write_bytes(para_vecs.tobytes())
-
-    meta = [{"url": p["url"], "title": p["title"],
-             "heading": p["heading"], "excerpt": p["excerpt"]}
-            for p in paragraphs]
-    SEMANTIC_META.write_text(json.dumps(meta, ensure_ascii=False))
-
-    print(f"embed.py: wrote {len(paragraphs)} paragraphs to semantic index "
-          f"({SEMANTIC_BIN.stat().st_size // 1024} KB)")
     return 0
 
 

tools/import-poetry.py

@@ -1,401 +0,0 @@
-#!/usr/bin/env python3
-"""
-import-poetry.py — Import a poetry collection from a Project Gutenberg plain-text file.
-
-Produces:
-  content/poetry/{collection-slug}/index.md          Collection index page
-  content/poetry/{collection-slug}/{poem-slug}.md    One file per poem
-
-Usage:
-  python tools/import-poetry.py gutenberg.txt \\
-      --poet "William Shakespeare" \\
-      --collection "Sonnets" \\
-      --date 1609 \\
-      --title-prefix "Sonnet" \\
-      --tags poetry,english \\
-      [--slug shakespeare-sonnets] \\
-      [--interactive] \\
-      [--dry-run] \\
-      [--overwrite]
-
-The --title-prefix controls per-poem title generation:
-  "Sonnet" → "Sonnet 1", "Sonnet 2", ..., slug "sonnet-1", "sonnet-2"
-  If omitted, defaults to the singular of --collection (strips trailing 's').
-"""
-
-import argparse
-import re
-import sys
-from pathlib import Path
-from typing import Optional
-
-REPO_ROOT   = Path(__file__).parent.parent
-POETRY_DIR  = REPO_ROOT / "content" / "poetry"
-
-# ---------------------------------------------------------------------------
-# Roman numeral conversion
-# ---------------------------------------------------------------------------
-
-_ROMAN_VALS = [
-    ("M", 1000), ("CM", 900), ("D", 500), ("CD", 400),
-    ("C",  100), ("XC",  90), ("L",  50), ("XL",  40),
-    ("X",   10), ("IX",   9), ("V",   5), ("IV",   4), ("I", 1),
-]
-
-def roman_to_int(s: str) -> Optional[int]:
-    s = s.upper().strip()
-    i, result = 0, 0
-    for numeral, value in _ROMAN_VALS:
-        while s[i : i + len(numeral)] == numeral:
-            result += value
-            i += len(numeral)
-    return result if i == len(s) and result > 0 else None
-
-# Matches a line that is *solely* a Roman numeral (with optional period/trailing space).
-# Anchored; leading/trailing whitespace stripped by caller.
-_ROMAN_RE = re.compile(
-    r"^(M{0,4}(?:CM|CD|D?C{0,3})(?:XC|XL|L?X{0,3})(?:IX|IV|V?I{0,3}))\.?$",
-    re.IGNORECASE,
-)
-
-# ---------------------------------------------------------------------------
-# Slug generation
-# ---------------------------------------------------------------------------
-
-def slugify(s: str) -> str:
-    s = s.lower()
-    s = re.sub(r"[^\w\s-]", "", s)
-    s = re.sub(r"[\s_]+", "-", s)
-    s = re.sub(r"-+", "-", s)
-    return s.strip("-")
-
-# ---------------------------------------------------------------------------
-# Gutenberg parsing
-# ---------------------------------------------------------------------------
-
-_START_RE = re.compile(r"\*\*\* START OF THE PROJECT GUTENBERG", re.IGNORECASE)
-_END_RE   = re.compile(r"\*\*\* END OF THE PROJECT GUTENBERG",   re.IGNORECASE)
-
-def strip_gutenberg(text: str) -> tuple[str, str]:
-    """Return (header, body) where body is the text between the PG markers."""
-    lines = text.splitlines()
-    start = 0
-    end   = len(lines)
-    for i, line in enumerate(lines):
-        if _START_RE.search(line):
-            start = i + 1
-            break
-    for i, line in enumerate(lines):
-        if _END_RE.search(line):
-            end = i
-            break
-    header = "\n".join(lines[:start])
-    body   = "\n".join(lines[start:end])
-    return header, body
-
-def parse_gutenberg_meta(header: str) -> dict:
-    meta: dict = {}
-    for line in header.splitlines():
-        for field in ("Title", "Author", "Release date", "Release Date"):
-            if line.startswith(field + ":"):
-                meta[field.lower().replace(" ", "-")] = line.split(":", 1)[1].strip()
-    return meta
-
-# ---------------------------------------------------------------------------
-# Poem splitting
-# ---------------------------------------------------------------------------
-
-def split_poems(body: str) -> list[dict]:
-    """
-    Split body text into individual poems using Roman-numeral headings as
-    boundaries. Returns a list of dicts:
-        { number: int, roman: str, lines: list[str] }
-
-    Lines are raw — call normalize_stanzas() before writing.
-    """
-    lines = body.splitlines()
-    poems: list[dict] = []
-    current: Optional[dict] = None
-
-    for line in lines:
-        stripped = line.strip()
-        m = _ROMAN_RE.match(stripped)
-        if m and stripped:  # empty stripped means blank line, not a heading
-            number = roman_to_int(m.group(1))
-            if number is not None:
-                if current is not None and _has_content(current["lines"]):
-                    poems.append(current)
-                current = {"number": number, "roman": m.group(1).upper(), "lines": []}
-                continue
-        if current is not None:
-            current["lines"].append(line)
-
-    if current is not None and _has_content(current["lines"]):
-        poems.append(current)
-
-    return poems
-
-def _has_content(lines: list[str], min_words: int = 4) -> bool:
-    text = " ".join(l.strip() for l in lines if l.strip())
-    return len(text.split()) >= min_words
-
-# ---------------------------------------------------------------------------
-# Stanza normalization
-# ---------------------------------------------------------------------------
-
-def normalize_stanzas(raw: list[str]) -> list[str]:
-    """
-    Strip common indentation, remove leading/trailing blank lines, collapse
-    runs of more than one blank line to a single blank line (stanza break).
-    """
-    lines = [l.rstrip() for l in raw]
-
-    # Trim leading/trailing blank lines
-    while lines and not lines[0].strip():
-        lines.pop(0)
-    while lines and not lines[-1].strip():
-        lines.pop()
-
-    # Determine and strip common leading whitespace on content lines
-    content = [l for l in lines if l.strip()]
-    if content:
-        indent = min(len(l) - len(l.lstrip()) for l in content)
-        lines = [l[indent:] if len(l) >= indent else l for l in lines]
-
-    # Collapse multiple consecutive blank lines to one
-    out: list[str] = []
-    prev_blank = False
-    for l in lines:
-        blank = not l.strip()
-        if blank and prev_blank:
-            continue
-        out.append(l)
-        prev_blank = blank
-
-    # Final trim
-    while out and not out[0].strip():
-        out.pop(0)
-    while out and not out[-1].strip():
-        out.pop()
-
-    return out
-
-def first_content_line(lines: list[str]) -> str:
-    for l in lines:
-        if l.strip():
-            return l.strip()
-    return ""
-
-# ---------------------------------------------------------------------------
-# YAML helpers
-# ---------------------------------------------------------------------------
-
-def yaml_str(s: str) -> str:
-    """Quote a string for YAML if it needs it."""
-    needs_quote = (
-        not s
-        or s[0] in " \t"
-        or s[-1] in " \t"
-        or any(c in s for c in ':{}[]|>&*!,#?@`\'"')
-    )
-    if needs_quote:
-        return '"' + s.replace("\\", "\\\\").replace('"', '\\"') + '"'
-    return s
-
-# ---------------------------------------------------------------------------
-# File generation
-# ---------------------------------------------------------------------------
-
-def make_poem_file(
-    poem:            dict,
-    title_prefix:    str,
-    poet:            str,
-    collection:      str,
-    collection_slug: str,
-    date:            str,
-    tags:            list[str],
-) -> tuple[str, str]:
-    """Return (filename_stem, markdown_content)."""
-    title     = f"{title_prefix} {poem['number']}"
-    slug      = slugify(title)
-    norm      = normalize_stanzas(poem["lines"])
-    abstract  = first_content_line(norm)
-    tag_yaml  = "[" + ", ".join(tags) + "]"
-    col_url   = f"/poetry/{collection_slug}/"
-
-    fm = f"""\
----
-title: {yaml_str(title)}
-number: {poem['number']}
-poet: {yaml_str(poet)}
-collection: {yaml_str(collection)}
-collection-url: {col_url}
-date: {date}
-tags: {tag_yaml}
-abstract: {yaml_str(abstract)}
----
-
-"""
-    content = fm + "\n".join(norm) + "\n"
-    return slug, content
-
-def make_collection_index(
-    collection:      str,
-    poet:            str,
-    date:            str,
-    tags:            list[str],
-    collection_slug: str,
-    title_prefix:    str,
-    poems:           list[dict],
-) -> str:
-    tag_yaml   = "[" + ", ".join(tags) + "]"
-    count      = len(poems)
-    abstract   = f"{count} poem{'s' if count != 1 else ''}"
-
-    poem_links = "\n".join(
-        f"- [{title_prefix} {p['number']}](./{slugify(title_prefix + ' ' + str(p['number']))}.html)"
-        for p in sorted(poems, key=lambda p: p["number"])
-    )
-
-    return f"""\
----
-title: {yaml_str(collection)}
-poet: {yaml_str(poet)}
-date: {date}
-tags: {tag_yaml}
-abstract: {yaml_str(abstract)}
----
-
-*{poet}* · {date}
-
-{poem_links}
-"""
-
-# ---------------------------------------------------------------------------
-# Interactive review
-# ---------------------------------------------------------------------------
-
-def interactive_review(poems: list[dict], title_prefix: str) -> list[dict]:
-    approved: list[dict] = []
-    total = len(poems)
-    for idx, poem in enumerate(poems, 1):
-        title   = f"{title_prefix} {poem['number']}"
-        preview = first_content_line(normalize_stanzas(poem["lines"]))
-        n_lines = sum(1 for l in poem["lines"] if l.strip())
-
-        print(f"\n{'─' * 60}")
-        print(f"  [{idx}/{total}]  {poem['roman']}.  →  {title}")
-        print(f"  First line : {preview}")
-        print(f"  Body lines : {n_lines}")
-        print()
-        resp = input("  [Enter] include   s skip   q quit: ").strip().lower()
-        if resp == "q":
-            print("Stopped at user request.")
-            break
-        elif resp == "s":
-            print(f"  Skipped {title}.")
-            continue
-        approved.append(poem)
-
-    return approved
-
-# ---------------------------------------------------------------------------
-# Main
-# ---------------------------------------------------------------------------
-
-def main() -> None:
-    parser = argparse.ArgumentParser(
-        description="Import a Gutenberg poetry collection into content/poetry/."
-    )
-    parser.add_argument("source",            help="Path to the Gutenberg .txt file")
-    parser.add_argument("--poet",            required=True, help='e.g. "William Shakespeare"')
-    parser.add_argument("--collection",      required=True, help='e.g. "Sonnets"')
-    parser.add_argument("--date",            required=True, help="Publication year, e.g. 1609")
-    parser.add_argument("--title-prefix",    help='Per-poem title prefix, e.g. "Sonnet". Defaults to singular of --collection.')
-    parser.add_argument("--tags",            default="poetry", help="Comma-separated tags (default: poetry)")
-    parser.add_argument("--slug",            help="Override collection directory slug")
-    parser.add_argument("--interactive",     action="store_true", help="Review each poem before writing")
-    parser.add_argument("--dry-run",         action="store_true", help="Show what would be written; write nothing")
-    parser.add_argument("--overwrite",       action="store_true", help="Overwrite existing files")
-    args = parser.parse_args()
-
-    source = Path(args.source)
-    if not source.exists():
-        print(f"error: file not found: {source}", file=sys.stderr)
-        sys.exit(1)
-
-    # Defaults
-    title_prefix     = args.title_prefix or args.collection.rstrip("s")
-    collection_slug  = args.slug or slugify(f"{args.poet}-{args.collection}")
-    tags             = [t.strip() for t in args.tags.split(",")]
-    out_dir          = POETRY_DIR / collection_slug
-
-    text = source.read_text(encoding="utf-8", errors="replace")
-    header, body = strip_gutenberg(text)
-
-    if not body.strip():
-        print("warning: Gutenberg markers not found — treating entire file as body", file=sys.stderr)
-        body = text
-
-    poems = split_poems(body)
-
-    if not poems:
-        print("No poems detected. The file may not use Roman-numeral headings.", file=sys.stderr)
-        print("First 50 lines of body:", file=sys.stderr)
-        for ln in body.splitlines()[:50]:
-            print(f"  {repr(ln)}", file=sys.stderr)
-        sys.exit(1)
-
-    print(f"Detected {len(poems)} poems  ·  collection: {args.collection}  ·  poet: {args.poet}")
-
-    if args.interactive:
-        poems = interactive_review(poems, title_prefix)
-        print(f"\n{len(poems)} poem(s) approved for import.")
-
-    if not poems:
-        print("Nothing to write.")
-        return
-
-    # Build file map
-    files: dict[Path, str] = {}
-    for poem in poems:
-        slug, content = make_poem_file(
-            poem, title_prefix, args.poet, args.collection,
-            collection_slug, args.date, tags,
-        )
-        files[out_dir / f"{slug}.md"] = content
-
-    files[out_dir / "index.md"] = make_collection_index(
-        args.collection, args.poet, args.date, tags,
-        collection_slug, title_prefix, poems,
-    )
-
-    # Dry run
-    if args.dry_run:
-        print(f"\nDry run — {len(files)} file(s) → {out_dir.relative_to(REPO_ROOT)}/")
-        for path in sorted(files):
-            marker = " (exists)" if path.exists() else ""
-            print(f"  {path.name}{marker}")
-        print(f"\nSample — first poem:\n{'─'*60}")
-        first_content = next(v for k, v in files.items() if k.name != "index.md")
-        print(first_content[:800])
-        return
-
-    # Write
-    out_dir.mkdir(parents=True, exist_ok=True)
-    written = skipped = 0
-    for path, content in sorted(files.items()):
-        if path.exists() and not args.overwrite:
-            print(f"  skip  {path.name}")
-            skipped += 1
-        else:
-            path.write_text(content, encoding="utf-8")
-            print(f"  write {path.name}")
-            written += 1
-
-    print(f"\n{written} written, {skipped} skipped → {out_dir.relative_to(REPO_ROOT)}/")
-    if not args.dry_run:
-        print("Next: make clean && make build")
-
-if __name__ == "__main__":
-    main()

tools/refreeze.sh

@@ -1,21 +0,0 @@
-#!/usr/bin/env bash
-# refreeze.sh — Regenerate cabal.project.freeze after a pacman -Syu updates
-# Haskell libraries. Run from anywhere inside the repo.
-set -euo pipefail
-
-REPO_ROOT="$(git -C "$(dirname "$0")" rev-parse --show-toplevel)"
-FREEZE="$REPO_ROOT/cabal.project.freeze"
-
-cd "$REPO_ROOT"
-
-echo "==> Removing stale freeze file..."
-rm -f "$FREEZE"
-
-echo "==> Resolving dependencies and writing new freeze file..."
-cabal freeze
-
-echo "==> Verifying build..."
-cabal build
-
-echo ""
-echo "Done. cabal.project.freeze updated."

uv.lock

@@ -171,6 +171,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/e7/05/c19819d5e3d95294a6f5947fb9b9629efb316b96de511b418c53d245aae6/cycler-0.12.1-py3-none-any.whl", hash = "sha256:85cef7cff222d8644161529808465972e51340599459b8ac3ccbac5a854e0d30", size = 8321, upload-time = "2023-10-07T05:32:16.783Z" },
 ]
 
+[[package]]
+name = "einops"
+version = "0.8.2"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/2c/77/850bef8d72ffb9219f0b1aac23fbc1bf7d038ee6ea666f331fa273031aa2/einops-0.8.2.tar.gz", hash = "sha256:609da665570e5e265e27283aab09e7f279ade90c4f01bcfca111f3d3e13f2827", size = 56261, upload-time = "2026-01-26T04:13:17.638Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/2a/09/f8d8f8f31e4483c10a906437b4ce31bdf3d6d417b73fe33f1a8b59e34228/einops-0.8.2-py3-none-any.whl", hash = "sha256:54058201ac7087911181bfec4af6091bb59380360f069276601256a76af08193", size = 65638, upload-time = "2026-01-26T04:13:18.546Z" },
+]
+
 [[package]]
 name = "faiss-cpu"
 version = "1.13.2"
@@ -492,6 +501,7 @@ source = { virtual = "." }
 dependencies = [
     { name = "altair" },
     { name = "beautifulsoup4" },
+    { name = "einops" },
     { name = "faiss-cpu" },
     { name = "matplotlib" },
     { name = "numpy" },
@@ -504,6 +514,7 @@ dependencies = [
 requires-dist = [
     { name = "altair", specifier = ">=5.4" },
     { name = "beautifulsoup4", specifier = ">=4.12" },
+    { name = "einops", specifier = ">=0.8" },
     { name = "faiss-cpu", specifier = ">=1.9" },
     { name = "matplotlib", specifier = ">=3.9" },
     { name = "numpy", specifier = ">=2.0" },