okf 1.10.0 → 1.12.0

data/lib/okf/bundle/folder.rb

@@ -42,6 +42,10 @@ module OKF
         @bundle.graph(minimal: minimal, body: body)
       end
 
+      def skeleton
+        @bundle.skeleton
+      end
+
       def catalog
         @bundle.catalog
       end
@@ -64,13 +68,28 @@ module OKF
         end
       end
 
-      # Human-readable "parent/dir" name — the default HTML title.
-      # The bundle's display label, "parent/dir" — path arithmetic, no disk. It
-      # is a class method so a caller that only wants the label (the registry
-      # naming an entry) can have it without a Reader.read of every file.
+      # The bundle's display label — path arithmetic, no disk. It is a class
+      # method so a caller that only wants the label (the registry naming an
+      # entry) can have it without a Reader.read of every file.
+      #
+      # "parent/dir", because a bundle directory's own name is rarely unique
+      # enough to name it by — except when that name is `.okf`, the conventional
+      # container, and then the parent carries the whole answer on its own. A
+      # registry of eight projects is eight rows reading `…/.okf`, which is the
+      # one word that tells none of them apart; `repo/.okf` is read as "repo" by
+      # anyone looking at it anyway.
+      #
+      # A directory with no parent to borrow (`/.okf`) keeps its own name: the
+      # parent is `/`, which names nothing. That case used to compose into
+      # `//.okf`.
       def self.label(root)
         pathname = Pathname.new(root)
-        "#{pathname.parent.basename}/#{pathname.basename}"
+        parent = pathname.parent.basename.to_s
+        base = pathname.basename.to_s
+        return base if [ "/", "." ].include?(parent)
+        return parent if base == ".okf"
+
+        "#{parent}/#{base}"
       end
 
       def name

data/lib/okf/bundle/linter.rb

@@ -324,7 +324,7 @@ module OKF
       end
 
       def concepts_in(dir)
-        @concepts.select { |concept| File.dirname("#{concept.id}.md") == dir }
+        @concepts.select { |concept| OKF.dir_of(concept.id) == dir }
       end
 
       def index_path_for(dir)

data/lib/okf/bundle/search/index.rb

@@ -12,8 +12,9 @@ module OKF
       #
       # Matching is by *token*: a term matches a whole word or a word it prefixes
       # ("dedup" reaches "deduplication"), and `fuzzy:` opts into typo tolerance.
-      # The index is built per call — see .okf/capabilities/search.md for why that
-      # ceiling stands and what lifts it.
+      # The index builds per call unless the caller holds a Search::Corpus, which
+      # builds it once through .prepare and reuses it — the server does, the
+      # one-shot CLI cannot. See .okf/capabilities/search.md.
       module Index
         CAPABILITIES = %i[fuzzy prefix].freeze
 
@@ -37,9 +38,18 @@ module OKF
           # can neither match nor be credited. The hit's `terms` are MiniFTS's
           # matched *document* terms — already lowercased, and present in the text
           # verbatim even when the query only prefixed them.
-          def call(documents, terms, fields:, fuzzy: false, **_options)
+          # The expensive half, separated so a long-lived caller can hold it. A
+          # Corpus calls this once; a one-shot call still goes through #call and
+          # builds inline, which is why this is an addition and not a new
+          # requirement on the engine contract.
+          def prepare(documents)
             index = MiniFTS.new(fields: FIELDS, id_field: "key")
             index.add_all(documents)
+            index
+          end
+
+          def call(documents, terms, fields:, fuzzy: false, prepared: nil, **_options)
+            index = prepared || prepare(documents)
 
             options = { combine_with: "AND", prefix: true, boost: WEIGHTS, fields: fields }
             options[:fuzzy] = FUZZY_DISTANCE if fuzzy

data/lib/okf/bundle/search.rb

@@ -186,12 +186,89 @@ module OKF
         new(bundles, terms, fields: fields, regexp: regexp, fuzzy: fuzzy, engine: engine, engines: engines).results
       end
 
+      # The searchable text of one concept, by field. Here rather than on an
+      # instance because a Corpus builds documents with no query in hand.
+      def self.field_texts(concept)
+        {
+          "id" => concept.id,
+          "title" => concept.title.to_s,
+          "type" => concept.type.to_s,
+          "description" => concept.description.to_s,
+          "tags" => Array(concept.tags).join(" "),
+          "body" => concept.body
+        }
+      end
+
+      # A corpus prepared once and queried many times: the documents, the key →
+      # concept map, and each engine's built index.
+      #
+      # This is the asymmetry the engine choice was always argued from. A CLI
+      # process loads a bundle, asks one question and exits, so an index build has
+      # exactly one query to amortize over and the scan wins. A server is the
+      # other case — the build is ~95% of the index path's cost, and paying it per
+      # request made every search re-read the whole corpus. Held once, it is paid
+      # once.
+      #
+      # Pure: it holds concepts, never disk. Which is also the cost — the corpus
+      # is a snapshot, so a body edited after it was built is searchable only
+      # after the holder drops it. That matches the graph, which is memoized the
+      # same way and for the same reason.
+      class Corpus
+        attr_reader :bundles, :documents, :sources
+
+        def initialize(bundles)
+          @bundles = bundles
+          @documents = []
+          @sources = {}
+          @indexes = {}
+          bundles.each do |slug, bundle|
+            bundle.concepts.each do |concept|
+              key = "#{slug}#{KEY_SEPARATOR}#{concept.id}"
+              @sources[key] = [ slug, concept ]
+              @documents << Search.field_texts(concept).merge("key" => key)
+            end
+          end
+        end
+
+        # nil for an engine with nothing to prebuild — the scan reads raw text and
+        # has no index to hold — so the option only ever reaches one that declared
+        # it can. Memoized per engine id: two engines over one corpus is legal.
+        def index_for(engine)
+          return nil unless engine.respond_to?(:prepare)
+
+          @indexes[engine.id] ||= engine.prepare(@documents)
+        end
+      end
+
+      # Prepare a corpus for a long-lived caller. Hand the result back to .with
+      # for every query.
+      #
+      # `engine:` builds that engine's index *now* rather than on the first query.
+      # Without it the corpus holds only the documents, and the expensive half —
+      # the index — is still built lazily, which puts the whole cost on whoever
+      # searches first. A server knows its engine at boot, so it can pay there.
+      def self.prepare(bundles, engine: nil, engines: nil)
+        corpus = Corpus.new(bundles)
+        return corpus if OKF.blank?(engine)
+
+        corpus.index_for(engine_for([], engines: engines || self.engines, name: engine))
+        corpus
+      end
+
+      # Query a prepared corpus. Same rows as .across, without rebuilding what the
+      # corpus already holds.
+      def self.with(corpus, terms, fields: nil, regexp: false, fuzzy: false, engine: nil, engines: nil)
+        new(corpus.bundles, terms, fields: fields, regexp: regexp, fuzzy: fuzzy,
+          engine: engine, engines: engines, corpus: corpus).results
+      end
+
       # Raises RegexpError on an invalid pattern with `regexp: true`, and
       # UnsupportedQuery when no engine can answer — the caller owns turning
       # either into a usage error. `engines:` overrides the registry, which is how
       # the "nothing qualifies" path stays reachable without an addon installed.
-      def initialize(bundles, terms, fields: nil, regexp: false, fuzzy: false, engine: nil, engines: nil)
+      def initialize(bundles, terms, fields: nil, regexp: false, fuzzy: false, engine: nil, engines: nil, corpus: nil)
         @bundles = bundles
+        @corpus = corpus
         @terms = Array(terms).reject { |term| OKF.blank?(term) }.map(&:to_s)
         @fields = fields.nil? || fields.empty? ? FIELDS : fields
         @regexp = regexp
@@ -202,7 +279,7 @@ module OKF
       end
 
       # Ranked match rows, catalog-style identity plus where the terms hit:
-      # [{ slug:, id:, title:, type:, area:, tags:, matched: [field, …], score:, snippet: }, …]
+      # [{ slug:, id:, title:, type:, dir:, top_dir:, tags:, matched: [field, …], score:, snippet: }, …]
       # ordered by score descending, then slug, then id. `slug` is present only
       # when searching across bundles. No terms means no matches.
       def results
@@ -242,6 +319,7 @@ module OKF
       # what it can act on, so there is nothing left for it to ignore.
       def engine_options(chosen)
         options = { fields: @fields }
+        options[:prepared] = @corpus.index_for(chosen) if @corpus
         ROUTABLE.each do |capability|
           options[capability] = requested[capability] if chosen.capabilities.include?(capability)
         end
@@ -255,6 +333,13 @@ module OKF
       # Every concept as an indexable document, keyed uniquely across bundles.
       # @sources keeps the way back, so the index stores no fields of its own.
       def documents
+        # The corpus already walked every concept and kept the map that turns a
+        # hit back into a row; taking its sources is what makes that reuse whole.
+        if @corpus
+          @sources = @corpus.sources
+          return @corpus.documents
+        end
+
         docs = []
         @bundles.each do |slug, bundle|
           bundle.concepts.each do |concept|
@@ -269,14 +354,7 @@ module OKF
       # { field => original-case text } for every searchable field. The index reads
       # all of them; `fields:` narrows the search, not the document.
       def field_texts(concept)
-        {
-          "id" => concept.id,
-          "title" => concept.title.to_s,
-          "type" => concept.type.to_s,
-          "description" => concept.description.to_s,
-          "tags" => Array(concept.tags).join(" "),
-          "body" => concept.body
-        }
+        Search.field_texts(concept)
       end
 
       # `slug` leads the row so a merged result reads bundle-first, and drops
@@ -288,7 +366,8 @@ module OKF
           id: concept.id,
           title: (concept.title || concept.id).to_s,
           type: concept.type.to_s,
-          area: area_of(concept.id),
+          dir: OKF.dir_of(concept.id),
+          top_dir: top_dir_of(concept.id),
           tags: Array(concept.tags).map(&:to_s),
           matched: matched,
           score: score.round(4),
@@ -342,8 +421,9 @@ module OKF
         end
       end
 
-      # A concept's top-level area, mirroring the catalog's definition.
-      def area_of(id)
+      # A concept's top-level dir, mirroring the catalog's definition — the first
+      # path segment. OKF.dir_of keeps the levels this one rolls up.
+      def top_dir_of(id)
         id.include?("/") ? id.split("/").first : "(root)"
       end
     end

data/lib/okf/bundle/skeleton.rb

@@ -0,0 +1,241 @@
+# frozen_string_literal: true
+
+module OKF
+  class Bundle
+    # The graph reduced to what a reader can hold in their head. Pure — built from
+    # an OKF::Bundle::Graph's nodes and edges, does no I/O, and decides nothing
+    # about how any of it is drawn.
+    #
+    # A dense bundle is not dense in the way a hub-and-spoke picture is. Measured
+    # on a 47-concept bundle with 227 links: the top hub takes 13 inbound and the
+    # median takes 4, so there is no 80/20 to exploit — dropping two thirds of the
+    # concepts still leaves 53 edges. The density lives *between directories*:
+    # 173 of those 227 links (76%) cross a directory boundary, and they collapse
+    # into 50 directory-to-directory arcs of which the top ten carry half the mass.
+    # That is the reduction worth drawing, and it is why #arcs exists at all.
+    #
+    # ── the two things it produces, and who reads them ──
+    #
+    #   dirs + arcs   the reduction, as counts: one row per directory, one
+    #                 weighted arc per ordered pair. Printed by
+    #                 `okf graph --traffic`, with cohesion derived from them.
+    #   edges         every link with the cut it survives (`keep_at`). Nothing
+    #                 prints these — they are what lets the graph page lay a
+    #                 large bundle out on its strongest links first.
+    #
+    # Both are emitted *unthresholded*, and #suggested_cut names where to cut
+    # rather than cutting, so a caller narrows the picture without this class
+    # having to know what a picture is.
+    #
+    # Directories come off the concept *id* (OKF.dir_of), not the file path, so
+    # this agrees with #catalog, #hubs and the `--dir` filter. Bundle#directory_index
+    # groups by path instead — deliberately, since an index.md is a physical
+    # listing — so the two disagree for a concept whose frontmatter `id` moves it.
+    # This side follows the id because the edges do.
+    class Skeleton
+      attr_reader :dirs, :arcs, :edges
+
+      def self.build(bundle)
+        graph = bundle.graph(minimal: true)
+        ids = graph.nodes.map { |node| node[:id] }
+        pairs = graph.edges.map { |edge| [ edge[:source], edge[:target] ] }
+        dir_by_id = ids.map { |id| [ id, OKF.dir_of(id) ] }.to_h
+
+        new(
+          dirs: dirs_for(dir_by_id, pairs),
+          arcs: arcs_for(dir_by_id, pairs),
+          edges: edges_for(pairs, neighbours_for(ids, pairs))
+        )
+      end
+
+      # ── the cuts (see above) ──
+
+      def self.arcs_above(arcs, weight)
+        arcs.select { |arc| arc[:weight] >= weight }
+      end
+
+      def self.edges_within(edges, keep_at)
+        edges.select { |edge| edge[:keep_at] <= keep_at }
+      end
+
+      # ── the directories ──
+
+      # One row per directory that holds a concept, plus every ancestor up to the
+      # root, so the tree stays connected through a directory that holds nothing
+      # directly. `count` is direct and `subtree` is at-or-below — the same pair
+      # `okf dirs` prints, and for the same reason: a direct count alone cannot
+      # say where the mass is once the listing is cut off at a depth.
+      def self.dirs_for(dir_by_id, pairs)
+        direct = Hash.new(0)
+        dir_by_id.each_value { |dir| direct[dir] += 1 }
+        internal = Hash.new(0)
+        pairs.each do |source, target|
+          from = dir_by_id[source]
+          internal[from] += 1 if from == dir_by_id[target]
+        end
+
+        every_dir(direct.keys).map do |dir|
+          { dir: dir, parent: parent_of(dir), count: direct[dir],
+            subtree: direct.reduce(0) { |sum, (other, n)| under?(other, dir) ? sum + n : sum },
+            internal: internal[dir] }
+        end
+      end
+
+      # The cross-directory link mass, aggregated. Directed and never a self-arc:
+      # a directory's internal links are carried on its own row (`internal`),
+      # because an arc from a box to itself is a loop the eye has to untangle to
+      # learn a number the box could simply have carried.
+      def self.arcs_for(dir_by_id, pairs)
+        weights = Hash.new(0)
+        pairs.each do |source, target|
+          from = dir_by_id[source]
+          to = dir_by_id[target]
+          weights[[ from, to ]] += 1 unless from == to
+        end
+
+        weights.map { |(from, to), weight| { source: from, target: to, weight: weight } }
+               .sort_by { |arc| [ -arc[:weight], arc[:source], arc[:target] ] }
+      end
+
+      # ── where to cut ──
+
+      # The arc cut that leaves a readable picture, chosen from the *shape* of
+      # the bundle rather than fixed. A fixed cut cannot work: measured at
+      # weight 3 across ten bundles it left 2 arcs on one and 136 on another —
+      # too tight to be a picture at one end, no reduction at all at the other.
+      #
+      # What stays constant when a bundle grows is not the arc count but the
+      # arcs *per box*: a node-link diagram reads at roughly one to two edges
+      # per node regardless of size. So the target is 1.5 arcs per directory,
+      # and the cut is whatever weight delivers it — 22 arcs over 13 directories
+      # on one bundle, 191 over 95 on another, both about the same density.
+      #
+      # The floor of 8 is for the small end, where 1.5-per-box would cut a
+      # ten-arc bundle down to something that no longer shows how it is joined
+      # up. Ties are kept rather than broken, so the result is *at least* the
+      # target — the alternative is dropping one of two arcs that weigh the
+      # same, which is an arbitrary choice presented as a threshold.
+      def self.suggested_cut(arcs, dir_count)
+        target = [ (dir_count * 1.5).ceil, 8 ].max
+        return 1 if arcs.empty? || arcs.size <= target
+
+        arcs[target - 1][:weight]
+      end
+
+      # ── the edges, each with the cut it survives ──
+
+      # The local-degree sparsifier (Lindner et al.): every concept keeps its own
+      # most-connected neighbours, and an edge survives if *either* end kept it.
+      # The union is the point — it is what stops a sparsifier from stranding the
+      # quiet half of the bundle, which a global "drop the weakest edges" rule
+      # does immediately.
+      #
+      # Not the disparity filter, the usual name in this territory: that one reads
+      # an edge's weight against its endpoint's total, and every link here weighs
+      # exactly 1, which makes every proportion identical and the filter a coin
+      # toss. Weighted-graph tools do not transfer to an unweighted graph just
+      # because both are graphs.
+      #
+      # `keep_at` is the smallest cut (0–100) at which the edge appears, so the
+      # consumer's whole job is `keep_at <= n`. An edge to a node's single most
+      # connected neighbour keeps at 0 and is therefore never cut away: the
+      # skeleton always spans every linked concept.
+      def self.edges_for(pairs, neighbours)
+        order = ordered_neighbours(neighbours)
+        pairs.map do |source, target|
+          { source: source, target: target,
+            keep_at: [ keep_at(order, neighbours, source, target),
+                       keep_at(order, neighbours, target, source) ].min }
+        end
+      end
+
+      # Where `other` sits in `id`'s neighbours (1 = most connected), turned into
+      # the smallest cut that reaches it. Position 1 needs no budget at all; past
+      # that, a cut of n keeps ceil(degree ** n/100) neighbours, so the answer is
+      # the least n satisfying that — solved directly rather than searched, and
+      # pinned against the definition it came from in the unit test.
+      def self.keep_at(order, neighbours, id, other)
+        position = order[id].index(other) + 1
+        return 0 if position == 1
+
+        degree = neighbours[id].size
+        (100 * Math.log(position - 1) / Math.log(degree)).floor + 1
+      end
+
+      def self.ordered_neighbours(neighbours)
+        neighbours.each_with_object({}) do |(id, set), out|
+          out[id] = set.sort_by { |other| [ -neighbours[other].size, other ] }
+        end
+      end
+
+      # Undirected neighbour sets. A link is structure regardless of which end
+      # authored it, and a concept every page cites but that cites nothing back
+      # is as central as one that does the citing.
+      def self.neighbours_for(ids, pairs)
+        sets = ids.map { |id| [ id, Set.new ] }.to_h
+        pairs.each do |source, target|
+          sets[source] << target if sets.key?(source) && sets.key?(target)
+          sets[target] << source if sets.key?(source) && sets.key?(target)
+        end
+        sets
+      end
+
+      # ── path arithmetic (pure; the bundle root is ".") ──
+
+      def self.every_dir(dirs)
+        seen = {}
+        dirs.each do |dir|
+          current = dir
+          loop do
+            seen[current] = true
+            break if current == "."
+
+            current = parent_of(current)
+          end
+        end
+        seen.keys.sort_by { |dir| dir == "." ? "" : dir }
+      end
+
+      def self.parent_of(dir)
+        return nil if dir == "."
+
+        File.dirname(dir)
+      end
+
+      # At or below — the same rule `--dir` is answered against, so a row's
+      # `subtree` and what `--dir <that row>` returns can never disagree.
+      def self.under?(dir, ancestor)
+        return true if ancestor == "."
+
+        dir == ancestor || dir.start_with?("#{ancestor}/")
+      end
+
+      def initialize(dirs:, arcs:, edges:)
+        @dirs = dirs
+        @arcs = arcs
+        @edges = edges
+      end
+
+      def suggested_cut
+        @suggested_cut ||= self.class.suggested_cut(arcs, dirs.size)
+      end
+
+      # The `keep_at` of each edge in +edges+, in *its* order — so a caller
+      # holding an OKF::Bundle::Graph can line the two up. Matched on the pair
+      # rather than by index: both lists derive from the same graph and so are
+      # already parallel today, and an index-aligned read would go wrong in
+      # silence on the day one of them stops being built that way.
+      def cuts_for(graph_edges)
+        by_pair = edges.each_with_object({}) { |edge, out| out[[ edge[:source], edge[:target] ]] = edge[:keep_at] }
+        graph_edges.map { |edge| by_pair[[ edge[:source], edge[:target] ]] || 0 }
+      end
+
+      # What the drawn views read. `edges` stays out: it is per-link data whose
+      # only consumer needs it at boot, before any fetch could answer — so it
+      # rides inline with the graph instead (see OKF::Render::Graph#edge_cuts_json).
+      def to_h
+        { dirs: dirs, arcs: arcs, suggested_cut: suggested_cut }
+      end
+    end
+  end
+end

data/lib/okf/bundle.rb

@@ -95,6 +95,10 @@ module OKF
       Graph.build(self, minimal: minimal, body: body)
     end
 
+    def skeleton
+      Skeleton.build(self)
+    end
+
     # Rich per-concept metadata the catalog / files / stats consumers want but the
     # lean graph omits — the descriptive frontmatter fields plus in/out link degree
     # taken from the graph edges. Pure: derived from the concepts and their links,
@@ -118,29 +122,29 @@ module OKF
           timestamp: concept.timestamp&.to_s,
           status: concept.frontmatter["status"]&.to_s,
           backlog_ref: concept.frontmatter["backlog_ref"]&.to_s,
-          dir: File.dirname("#{id}.md"),
-          area: area_of(id),
+          dir: OKF.dir_of(id),
+          top_dir: top_dir_of(id),
           links_out: out_degree[id],
           links_in: in_degree[id]
         }
       end.sort_by { |entry| entry[:id] }
     end
 
-    # Concepts ranked by inbound link degree, each with the areas its inbound
-    # links come from — the evidence for "is this hub well-homed?": a hub whose
-    # inbound majority is foreign to its own area is a move candidate, one with
-    # a single dominant foreign area already names its better home. Only
-    # concepts with at least one inbound link appear. Pure: derived from the
-    # graph edges. Shared by the `okf graph --hubs` view.
+    # Concepts ranked by inbound link degree, each with the top-level dirs its
+    # inbound links come from — the evidence for "is this hub well-homed?": a hub
+    # whose inbound majority is foreign to its own top-level dir is a move
+    # candidate, one with a single dominant foreign dir already names its better
+    # home. Only concepts with at least one inbound link appear. Pure: derived
+    # from the graph edges. Shared by the `okf graph --hubs` view.
     def hubs
       inbound = {}
       graph(minimal: true).edges.each do |edge|
-        (inbound[edge[:target]] ||= Hash.new(0))[area_of(edge[:source])] += 1
+        (inbound[edge[:target]] ||= Hash.new(0))[top_dir_of(edge[:source])] += 1
       end
 
       inbound.map do |id, sources|
-        by_area = sources.sort_by { |area, count| [ -count, area ] }.to_h
-        { id: id, area: area_of(id), inbound: by_area.values.reduce(0, :+), by_area: by_area }
+        by_top_dir = sources.sort_by { |top_dir, count| [ -count, top_dir ] }.to_h
+        { id: id, top_dir: top_dir_of(id), inbound: by_top_dir.values.reduce(0, :+), by_top_dir: by_top_dir }
       end.sort_by { |row| [ -row[:inbound], row[:id] ] }
     end
 
@@ -189,9 +193,10 @@ module OKF
 
     private
 
-    # A concept's top-level area, derived from its id — the same derivation the
-    # catalog exposes, so every grouped view labels the bundle root "(root)".
-    def area_of(id)
+    # A concept's top-level dir, derived from its id — the first path segment, the
+    # same derivation the catalog exposes, so every grouped view labels the bundle
+    # root "(root)". OKF.dir_of keeps the levels this one rolls up.
+    def top_dir_of(id)
       id.include?("/") ? id.split("/").first : "(root)"
     end
 

data/lib/okf/cli/catalog.rb

@@ -2,8 +2,8 @@
 
 module OKF
   class CLI
-    # Every concept with its metadata, grouped by area. The widest of the read
-    # views, and the one the others narrow down from.
+    # Every concept with its metadata, grouped by top-level dir. The widest of the
+    # read views, and the one the others narrow down from.
     class Catalog < Command
       def self.id
         :catalog
@@ -15,14 +15,14 @@ module OKF
 
       def self.help_rows
         [
-          [ "catalog   <dir|@slug> [--json] [filters]", "list concepts with metadata, by area" ]
+          [ "catalog   <dir|@slug> [--json] [filters]", "list concepts with metadata, by top-level dir" ]
         ]
       end
 
       def call(argv)
         options = { json: false }
         parser = OptionParser.new do |o|
-          o.banner = "Usage: okf catalog <dir|@slug> [--type T] [--area A] [--tag T] [--json]"
+          o.banner = "Usage: okf catalog <dir|@slug> [--type T] [--dir D] [--tag T] [--json]"
           json_flags(o, options, "emit the catalog as JSON")
           projection_flags(o, options)
           filter_flags(o, options, :type, :area, :tag)
@@ -44,9 +44,9 @@ module OKF
 
       def print_catalog(dir, entries, total)
         @out.puts "Catalog — #{bundle_label(dir)} (#{counted(entries.size, total, "concept")})"
-        entries.group_by { |entry| entry[:area] }.sort_by(&:first).each do |area, group|
+        entries.group_by { |entry| entry[:top_dir] }.sort_by(&:first).each do |top_dir, group|
           @out.puts
-          @out.puts "  #{area == "(root)" ? "(root)" : "#{area}/"} (#{group.size})"
+          @out.puts "  #{top_dir == "(root)" ? "(root)" : "#{top_dir}/"} (#{group.size})"
           group.each do |entry|
             links = entry[:links_out] + entry[:links_in]
             meta = [ entry[:type], (links.positive? ? "↳#{links}" : nil), entry[:status] ].compact.join("  ·  ")