diff --git a/.claude/memory/pr-screenshots-workflow.md b/.claude/memory/pr-screenshots-workflow.md
index 69a3ea52..4ab9c80b 100644
--- a/.claude/memory/pr-screenshots-workflow.md
+++ b/.claude/memory/pr-screenshots-workflow.md
@@ -34,4 +34,6 @@ metadata:
 
 **Emulating a "before" shot and drawing an open select (spec 065, 2026-11-10):** when the old code is the new code minus a collaborator's output (spec 065's pickers listed places only), take "before" in the same throwaway system spec by stubbing the new collaborator (`allow_any_instance_of(Location::Choices).to receive(:decks).and_return([])`; fine in an uncommitted spec) and say in `verification.md` that "before" is emulated, not built from `main`. Headless Firefox doesn't capture an open native `<select>`, so a second "-options" shot sets `select.setAttribute('size', select.options.length)` plus `height:auto !important` (the design system's fixed 28/36px height clips the list) in the page and inside the narrow frame; it distorts the neighbouring row, so judge layout from the closed shots. 48 JPEGs at 1000px width came to 1.3 MB.
 
+- **Tab bar in narrow-frame shots (spec 077, 2026-11-11):** the fixed tab bar sits at the frame's bottom: a frame taller than the window (e.g. `open_in_narrow_frame` with `height: 1500`) keeps it out of an element shot of a panel; a full-page phone shot needs a frame no taller than the window's viewport to show it (spec 074).
+
 **A frame taller than the window's viewport loses the fixed tab bar (spec 074, 2026-11-11):** a narrow frame is clipped to the window, and the phone tab bar is fixed to the frame's bottom, so the picture shows no bar. Reproduced: an 815px frame in a 1280×800 window (a 715px viewport) gave an 815px shot with no bar, in a 1280×900 window (815px viewport) the same shot had it, and a 1700px frame in this suite's 1400×1400 window (1315px viewport) gave a 1315px shot with no bar. Spec 071's shots 03, 04, 05 and others look like this; its window can't be read back, so the cause is reproduced, not confirmed. Make the window taller than the frame (`page.current_window.resize_to(…)`, and the after hook restores it) or shoot in a frame no taller than the viewport, and look for the bar in every phone shot.
diff --git a/app/assets/stylesheets/collector/additions.css b/app/assets/stylesheets/collector/additions.css
index 05e71ed6..c049d1d9 100644
--- a/app/assets/stylesheets/collector/additions.css
+++ b/app/assets/stylesheets/collector/additions.css
@@ -444,10 +444,15 @@ body:has(.c-locations) .c-appbar__back { min-height:var(--touch-floor); min-widt
 .c-curve__avg b { font:600 13px/18px var(--font-mono); color:var(--ink); }
 
 /* ---------- DeckShape card types (spec 060): a name, a count, and the proxies in a muted suffix ---------- */
-.c-types { margin:0; padding:0; list-style:none; border-top:1px solid var(--line); }
-.c-types li { display:flex; align-items:baseline; gap:var(--space-2); min-height:32px; padding:var(--space-1) 0; border-bottom:1px solid var(--line); font:400 14px/20px var(--font-sans); }
-.c-types__n { margin-left:auto; font:400 14px/20px var(--font-mono); }
-.c-types__proxy { min-width:4.5rem; text-align:right; font:400 12px/16px var(--font-mono); color:var(--ink-muted); } /* reserved even when empty, so the counts line up */
+/* Spec 077 (#320): the list is a grid of three columns and every row takes them as its own (subgrid), so the count column is as
+   wide as the widest count and the suffix column as wide as the widest suffix, and every count ends at one x whatever a suffix
+   says. The suffix cell keeps its 4.5rem floor so a list of short suffixes looks as it did; the name cell may wrap, even mid-word,
+   so a five-digit count and suffix at a 320px viewport never push a row past the list. */
+.c-types { display:grid; grid-template-columns:minmax(0, 1fr) auto auto; column-gap:var(--space-2); margin:0; padding:0; list-style:none; border-top:1px solid var(--line); }
+.c-types li { display:grid; grid-template-columns:subgrid; grid-column:1 / -1; align-items:baseline; min-height:32px; padding:var(--space-1) 0; border-bottom:1px solid var(--line); font:400 14px/20px var(--font-sans); }
+.c-types li > :first-child { min-width:0; overflow-wrap:anywhere; }
+.c-types__n { margin-left:auto; font:400 14px/20px var(--font-mono); } /* in a grid cell an auto margin right-aligns it, as in the old flex row */
+.c-types__proxy { min-width:4.5rem; text-align:right; font:400 12px/16px var(--font-mono); color:var(--ink-muted); } /* reserved even when empty, so a list with short suffixes keeps its look */
 
 /* ---------- DeckShape mana bar (spec 060): one bar split by colour, the only chart that may use the mana fills ---------- */
 /* README Charts rule: every segment has its letter and count (the caption carries a segment under 10%, which the markup
@@ -463,6 +468,7 @@ body:has(.c-locations) .c-appbar__back { min-height:var(--touch-floor); min-widt
 .c-manabar__seg--g { background:var(--mana-g); color:var(--on-mana-g); }
 .c-manabar__seg--narrow span { display:none; }
 .c-manacap { margin:0; font:400 13px/18px var(--font-sans); color:var(--ink-muted); }
+.c-manacap__item { white-space:nowrap; } /* spec 077 (#320): a caption line breaks only between colours, so a name stays with its count and no line starts with a dot */
 
 /* ---------- Deck page layout and the panel's two copies (spec 060) ---------- */
 /* The shape panel is rendered twice, an aside and a closed details, and each hides the other: a closed details hides its
diff --git a/app/models/mtg/deck_shape.rb b/app/models/mtg/deck_shape.rb
index 1eda2504..0b2aaa01 100644
--- a/app/models/mtg/deck_shape.rb
+++ b/app/models/mtg/deck_shape.rb
@@ -121,6 +121,13 @@ def unvalued_note
   # The five colours in the order the bar draws them, with the words the caption and label use.
   COLOURS = { "W" => "white", "U" => "blue", "B" => "black", "R" => "red", "G" => "green" }.freeze
 
+  # Counts below this are written in full; from it a count in a slot of fixed width takes a unit, and each unit is this many of the one
+  # before (spec 077 decision 5). The abbreviation is uniform, not width-driven: the server cannot know the width, and a count that
+  # fits at a 360px viewport still overflows at 320px.
+  SHORT_COUNT_STEP = 1_000
+  # The units of an abbreviated count: thousands, millions, billions.
+  SHORT_COUNT_UNITS = %w[k m b].freeze
+
   # The segments of the mana bar for the colours that have a symbol (decision 8), in W, U, B, R, G order.
   # @return [Array<Segment>]
   def symbols
@@ -135,10 +142,50 @@ def symbols
     end
   end
 
-  # "White 2 · 25% · Blue 4 · 50%": every colour present with its count and percentage.
+  # "White 2 · 25% · Blue 4 · 50%": every colour present with its count and percentage, exact (the abbreviation is for a slot of
+  # fixed width and this is prose). Built from {#caption_items}.
   # @return [String, nil] nil when there is no coloured symbol
-  def caption
-    symbols.map { |s| "#{COLOURS[s.colour].capitalize} #{delimited(s.count)} · #{s.percent}%" }.join(" · ").presence
+  def caption = caption_items.join(" ").presence
+
+  # The caption as one string per colour, "White 2 · 25% ·" and, last, "Blue 4 · 50%": every item but the last ends in the separator, so a
+  # view that sets each item as an unbreakable unit lets a line break only between items and never before a "·" (spec 077 decision 4,
+  # #320). Joined with one space they are {#caption}, character for character.
+  # @return [Array<String>] empty when there is no coloured symbol
+  def caption_items
+    items = symbols.map { |s| "#{COLOURS[s.colour].capitalize} #{delimited(s.count)} · #{s.percent}%" }
+    items.each_with_index.map { |item, index| index == items.size - 1 ? item : "#{item} ·" }
+  end
+
+  # A count written for a slot of fixed width, the curve's column count and a mana segment's label (spec 077 decisions 1, 2 and 5, #320).
+  # Below 1,000 it is the number as it is. From there it takes a unit: one decimal under 10 of the unit with a trailing ".0" dropped
+  # ("1.4k", "2k"), whole from 10 ("12k", "123k"), rounded half up on exact integers (1,350 is "1.4k", 1,349 "1.3k"). A value that
+  # rounds to 1,000 of its unit moves to the next unit (999,500 is "1m"), and one that rounds to 10.0 is written whole (9,999 is "10k"),
+  # so nothing is longer than four characters through "999m". It is not `number_to_human`, which keeps two significant digits (123,456
+  # would read "120k"). The exact number stays in every `aria-label`, the legend, the caption and the types.
+  #
+  # @example
+  #   short_count(999)     # => "999"
+  #   short_count(1_357)   # => "1.4k"
+  #   short_count(9_999)   # => "10k"
+  #   short_count(999_500) # => "1m"
+  # @param number [Integer] a count, 0 or more
+  # @return [String]
+  def short_count(number)
+    return delimited(number) if number < SHORT_COUNT_STEP
+
+    value = Rational(number, SHORT_COUNT_STEP)
+    unit = 0
+    while value.round >= SHORT_COUNT_STEP && unit < SHORT_COUNT_UNITS.size - 1
+      value /= SHORT_COUNT_STEP
+      unit += 1
+    end
+    tenths = (value * 10).round
+    body = if value < 10 && tenths < 100
+      (tenths % 10).zero? ? (tenths / 10).to_s : format("%d.%d", *tenths.divmod(10))
+    else
+      value.round.to_s
+    end
+    "#{body}#{SHORT_COUNT_UNITS[unit]}"
   end
 
   # "Coloured mana symbols in Main: white 2, blue 4" for the bar's `aria-label`.
diff --git a/app/views/mtg/deck_shapes/_parts.html.erb b/app/views/mtg/deck_shapes/_parts.html.erb
index 078504a2..b2e12aec 100644
--- a/app/views/mtg/deck_shapes/_parts.html.erb
+++ b/app/views/mtg/deck_shapes/_parts.html.erb
@@ -1,10 +1,10 @@
-<%# The three parts of the deck shape panel for Magic (spec 069, DeckShape.md): a part of zero has no element. %>
+<%# The three parts of the deck shape panel for Magic (spec 069, DeckShape.md): a part of zero has no element. A curve column's count and a mana segment's label are in slots of fixed width, so they are abbreviated from 1,000 (spec 077, MTG::DeckShape#short_count); every other number is exact. %>
 <div class="c-shape__part"><h3 class="c-shape__heading">Mana curve</h3>
   <div class="c-curve">
     <div role="img" aria-label="<%= figures.curve_label %>">
       <div class="c-curve__cols" aria-hidden="true">
         <% figures.curve.each do |column| %>
-          <div class="c-curve__col"><span class="c-curve__count"><%= number_with_delimiter(column.total) %></span>
+          <div class="c-curve__col"><span class="c-curve__count"><%= figures.short_count(column.total) %></span>
             <div class="c-curve__bar">
               <% if column.proxy.positive? %><div class="c-curve__proxy" style="--proxy:<%= column.proxy %>;--max:<%= figures.curve_max %>"></div><% end %>
               <% if column.owned.positive? %><div class="c-curve__owned" style="--owned:<%= column.owned %>;--max:<%= figures.curve_max %>"></div><% end %>
@@ -36,10 +36,10 @@
   <% if figures.symbols.any? %>
     <ul class="c-manabar" role="img" aria-label="<%= figures.symbols_label %>">
       <% figures.symbols.each do |segment| %>
-        <li class="<%= class_names("c-manabar__seg", "c-manabar__seg--#{segment.colour.downcase}", "c-manabar__seg--narrow": segment.narrow) %>" style="--n:<%= segment.count %>"><span><%= segment.colour %> <%= number_with_delimiter(segment.count) %></span></li>
+        <li class="<%= class_names("c-manabar__seg", "c-manabar__seg--#{segment.colour.downcase}", "c-manabar__seg--narrow": segment.narrow) %>" style="--n:<%= segment.count %>"><span><%= segment.colour %> <%= figures.short_count(segment.count) %></span></li>
       <% end %>
     </ul>
-    <p class="c-manacap"><%= figures.caption %></p>
+    <p class="c-manacap"><%= safe_join(figures.caption_items.map { |item| tag.span(item, class: "c-manacap__item") }, " ") %></p>
   <% else %>
     <p class="c-manacap">No coloured mana symbols.</p>
   <% end %>
diff --git a/docs/architecture/collection-and-lots.md b/docs/architecture/collection-and-lots.md
index 0c9def8a..4ba69ce7 100644
--- a/docs/architecture/collection-and-lots.md
+++ b/docs/architecture/collection-and-lots.md
@@ -134,7 +134,7 @@ The Decks section is the second link of the header and of the tab bar, after Col
 
 The page's collectible part is two optional answers of the collecting vocabulary, after its first, `formats`: `deck_card(entry, format_key)` returns a `Catalog::DeckCards::Answer` (group, position, legality word, whether a `deck_cost` partial draws the cost); the contract is written in `Catalog::DeckCards`' comment, and a vocabulary that answers nothing gets one flat table, a muted dash and no marker (spec 066 decision 5, AC-6.1). Magic groups by the first face's type line: a Creature is a Creature, otherwise a Land is a Land, otherwise the first of Planeswalkers, Battles, Instants, Sorceries, Artifacts, Enchantments, then Other (decision 6, `MTG::DeckGroups`); the cost is the first face's (decision 11, `mtg/printings/_deck_cost`); the legality marker is the printing's stored legality in the deck's own format, nothing for Casual or an unregistered format, "Not legal" for a missing entry, written twice and shown once by width (decision 8, `MTG::Collecting::LEGALITY_FORMATS`). The third answer is `deck_shape(zones)` (spec 069 decision 1, #256), below.
 
-The shape panel is the collecting vocabulary's third answer, after `formats` and `deck_card`: `deck_shape(zones)` takes `Deck::Page#zones` as the page built them and returns a `Catalog::DeckShape::Answer` (`partial`, `figures`, `scope`, `facts`) or nil, and `DecksController#show` asks once through `Catalog::DeckShape.answer`, one line after building the page; a vocabulary without it, a collectible type no longer registered and a nil answer draw no panel and leave the second column empty from 992px; the contract is written in `Catalog::DeckShape`'s comment, which names no collectible, format or zone (spec 069 decision 1, FR-1, AC-1.4; spec 060 AC-3.12). The core partial `decks/_shape` draws each copy's frame (the `aside`'s `section`, `h2` and scope line; the `details`' summary with the facts line and its `c-shape--bare` box) and renders the extension's partial inside, in the `details` copy, then the `aside` copy, before the zones; the figures are built once and rendered twice (decisions 1, 12, AC-1.1, AC-6.3). Magic's answer is `MTG::Collecting.deck_shape` over the figures object `MTG::DeckShape` and its partial `mtg/deck_shapes/_parts`. It counts the Commander and Main zones only (`MTG::DeckShape::COUNTED_ZONES`, in the deck's zone order; Sideboard and Considering never count, and the extension, not the core, names them), so the scope line reads "Main · 60 cards" or "Commander + Main · 100 cards" ("1 card"), a deck with neither zone has no panel, a deck of a format no longer registered keeps its Main zone's panel, and a deck with no card shows the panel's empty states (decisions 2, 10, 14, AC-1.5, AC-1.8, AC-2.8). A card is one copy or one proxy, so owned and proxy count together; a row of another collectible's card (page answer `Catalog::DeckCards::NONE`, no position) is skipped by every part, the scope line's number included (decision 3, AC-3.5). Card types: one type per card, the zone tables' own group (`row.answer.position`, `MTG::DeckGroups`) in the order of `MTG::DeckGroups::NAMES`, only the groups that have a card, each with its count and "N proxy" when it has proxies, so the counts add up to the scope line's number (decision 3, AC-3.1–AC-3.4). The curve counts nonland cards (not in Lands, so Dryad Arbor counts and an Artifact Land does not) whose card has a mana value, from `mtg_cards.mana_value` through the printing's identity (Scryfall's number, spec 058); column `n` from 0 to 5 holds the values from `n` up to but not including `n + 1` (floor, so 0.5 is in 0 and 2.5 in 2) and 6+ holds every value from 6, owned and proxy kept apart (decision 4, AC-2.1–AC-2.3, AC-2.6). The average is the exact mean of the curve's cards (rational arithmetic, so a mean of 0.15 shows 0.2), shown to one decimal rounded half up with the trailing zero kept ("3.0") and thousands delimited, and left out when no card is in the curve (decision 5, AC-2.4). A nonland card with no mana value is left out of the curve, the average and the legend, never counted as 0, and a second muted line under the average says "N nonland cards have no mana value yet." ("1 nonland card has no mana value yet."); the card is still in its type and the scope line (decision 6, AC-2.5). The mana bar counts coloured symbols of the first face's `mana_cost` on the loaded printing, the cost the zone tables show, never the joined `mtg_cards.mana_cost`, for every card in the counted zones, Lands included, each symbol times the card's count: `{W}`, `{U}`, `{B}`, `{R}` and `{G}` count one toward their colour; a hybrid (`{W/U}`) counts one toward each colour; a Phyrexian symbol (`{G/P}`) counts one toward its colour and a Phyrexian hybrid (`{G/W/P}`) one toward each; `{2/W}` and `{C/W}` count one toward the colour only; generic, `{X}`, `{C}`, snow and half symbols (`{HW}`) are never counted (decisions 7, 8, AC-4.1–AC-4.5). Each colour present shows its count and its share of all coloured symbols, each percentage rounded half up on its own so a caption may add to 99 or 101, in W, U, B, R, G order; a segment under 10% of the symbols (`count × 100 < total × 10`) is narrow and shows no label inside, its count staying in the caption; with no coloured symbol the part says "No coloured mana symbols." and an empty Card types part says "No cards yet." (decisions 9, 14, AC-4.6–AC-4.8). **A split card's second half adds no symbol to the bar, and that is deliberate:** a card is read through its first face for its type, cost and symbols, as the zone tables read it, while its mana value is Scryfall's number for the card, both halves added, so a split card is in the curve column of that sum while its symbols are the first half's only (spec 069 Non-Goals, decision 7, AC-4.5). The panel costs one statement of its own, `MTG::Card` for the identity ids of the counted zones' printings, none when those zones hold no card, and reads everything else from the rows `Deck::Page` loaded, so a page's count is the same however large the deck is (decision 11, AC-6.1, AC-6.2); it stores nothing and the export is unchanged (decision 13).
+The shape panel is the collecting vocabulary's third answer, after `formats` and `deck_card`: `deck_shape(zones)` takes `Deck::Page#zones` as the page built them and returns a `Catalog::DeckShape::Answer` (`partial`, `figures`, `scope`, `facts`) or nil, and `DecksController#show` asks once through `Catalog::DeckShape.answer`, one line after building the page; a vocabulary without it, a collectible type no longer registered and a nil answer draw no panel and leave the second column empty from 992px; the contract is written in `Catalog::DeckShape`'s comment, which names no collectible, format or zone (spec 069 decision 1, FR-1, AC-1.4; spec 060 AC-3.12). The core partial `decks/_shape` draws each copy's frame (the `aside`'s `section`, `h2` and scope line; the `details`' summary with the facts line and its `c-shape--bare` box) and renders the extension's partial inside, in the `details` copy, then the `aside` copy, before the zones; the figures are built once and rendered twice (decisions 1, 12, AC-1.1, AC-6.3). Magic's answer is `MTG::Collecting.deck_shape` over the figures object `MTG::DeckShape` and its partial `mtg/deck_shapes/_parts`. It counts the Commander and Main zones only (`MTG::DeckShape::COUNTED_ZONES`, in the deck's zone order; Sideboard and Considering never count, and the extension, not the core, names them), so the scope line reads "Main · 60 cards" or "Commander + Main · 100 cards" ("1 card"), a deck with neither zone has no panel, a deck of a format no longer registered keeps its Main zone's panel, and a deck with no card shows the panel's empty states (decisions 2, 10, 14, AC-1.5, AC-1.8, AC-2.8). A card is one copy or one proxy, so owned and proxy count together; a row of another collectible's card (page answer `Catalog::DeckCards::NONE`, no position) is skipped by every part, the scope line's number included (decision 3, AC-3.5). Card types: one type per card, the zone tables' own group (`row.answer.position`, `MTG::DeckGroups`) in the order of `MTG::DeckGroups::NAMES`, only the groups that have a card, each with its count and "N proxy" when it has proxies, so the counts add up to the scope line's number (decision 3, AC-3.1–AC-3.4). The curve counts nonland cards (not in Lands, so Dryad Arbor counts and an Artifact Land does not) whose card has a mana value, from `mtg_cards.mana_value` through the printing's identity (Scryfall's number, spec 058); column `n` from 0 to 5 holds the values from `n` up to but not including `n + 1` (floor, so 0.5 is in 0 and 2.5 in 2) and 6+ holds every value from 6, owned and proxy kept apart (decision 4, AC-2.1–AC-2.3, AC-2.6). The average is the exact mean of the curve's cards (rational arithmetic, so a mean of 0.15 shows 0.2), shown to one decimal rounded half up with the trailing zero kept ("3.0") and thousands delimited, and left out when no card is in the curve (decision 5, AC-2.4). A nonland card with no mana value is left out of the curve, the average and the legend, never counted as 0, and a second muted line under the average says "N nonland cards have no mana value yet." ("1 nonland card has no mana value yet."); the card is still in its type and the scope line (decision 6, AC-2.5). The mana bar counts coloured symbols of the first face's `mana_cost` on the loaded printing, the cost the zone tables show, never the joined `mtg_cards.mana_cost`, for every card in the counted zones, Lands included, each symbol times the card's count: `{W}`, `{U}`, `{B}`, `{R}` and `{G}` count one toward their colour; a hybrid (`{W/U}`) counts one toward each colour; a Phyrexian symbol (`{G/P}`) counts one toward its colour and a Phyrexian hybrid (`{G/W/P}`) one toward each; `{2/W}` and `{C/W}` count one toward the colour only; generic, `{X}`, `{C}`, snow and half symbols (`{HW}`) are never counted (decisions 7, 8, AC-4.1–AC-4.5). Each colour present shows its count and its share of all coloured symbols, each percentage rounded half up on its own so a caption may add to 99 or 101, in W, U, B, R, G order; a segment under 10% of the symbols (`count × 100 < total × 10`) is narrow and shows no label inside, its count staying in the caption; with no coloured symbol the part says "No coloured mana symbols." and an empty Card types part says "No cards yet." (decisions 9, 14, AC-4.6–AC-4.8). The visible number in a curve column's count and in a mana segment's label is abbreviated from 1,000 (`MTG::DeckShape#short_count`: 1,357 is "1.4k", 12,345 "12k", 1,250,000 "1.3m", half up on integers, at most four characters through 999m), because both sit in a slot of fixed width; every other number and every `aria-label` stays exact and delimited; the caption is a run of unbreakable items (`caption_items`, one per colour, so a line never starts with "·"); and the Card types list is a grid whose rows are subgrids, so the counts line up whatever the suffix (spec 077 decisions 1 to 5, #320; it supersedes, in part, spec 060 decision 2, AC-2.2, AC-2.4, AC-2.5 and AC-2.6 (the written-in-full numbers, the fixed 4.5rem minimum, the 279px) and spec 069 decision 14 for those two numbers). A mana segment narrower than its label (labels of four, five and six characters need about 14%, 17% and 20% of the bar at a 320px viewport; the 10% rule is spec 060 decision 4) is a known limit, [#330](https://github.com/plainprogrammer/Collector/issues/330) (spec 077 decision 2, ruled option A). **A split card's second half adds no symbol to the bar, and that is deliberate:** a card is read through its first face for its type, cost and symbols, as the zone tables read it, while its mana value is Scryfall's number for the card, both halves added, so a split card is in the curve column of that sum while its symbols are the first half's only (spec 069 Non-Goals, decision 7, AC-4.5). The panel costs one statement of its own, `MTG::Card` for the identity ids of the counted zones' printings, none when those zones hold no card, and reads everything else from the rows `Deck::Page` loaded, so a page's count is the same however large the deck is (decision 11, AC-6.1, AC-6.2); it stores nothing and the export is unchanged (decision 13).
 
 | From | Action | Result |
 |---|---|---|
@@ -246,6 +246,8 @@ Search and the printings page show each owned printing's quantity ("×3") on its
 | `MTG::DeckShape::COUNTED_ZONES` | `commander`, `main` | `app/models/mtg/deck_shape.rb` | The zone keys the shape panel counts, in the deck's zone order; Sideboard and Considering are left out (the issue design's decision 4) | spec 069 decision 2; #256 |
 | `MTG::DeckShape::TOP_COLUMN` | 6 | `app/models/mtg/deck_shape.rb` | The curve's last column, "6+", holds every mana value from 6; the columns 0 to 5 floor | spec 069 decision 4; #256 |
 | The narrow threshold | 10% of the coloured symbols (`count × 100 < total × 10`) | `MTG::DeckShape#symbols` (no constant) | A mana bar segment under it shows no label, the caption carrying its count | spec 069 decision 9, AC-4.6; #256 |
+| `MTG::DeckShape::SHORT_COUNT_STEP` | 1,000 | `app/models/mtg/deck_shape.rb` | A count in a slot of fixed width is written in full below it and takes a unit from it, each unit 1,000 of the one before | spec 077 decision 5; #320 |
+| `MTG::DeckShape::SHORT_COUNT_UNITS` | `k`, `m`, `b` | `app/models/mtg/deck_shape.rb` | The units of an abbreviated count | spec 077 decision 5; #320 |
 | `Lot::FULL_MESSAGE` | "One lot can hold at most 9,999 copies." | `app/models/lot.rb` | The form and number-entry message at the cap | spec 004 AC-9.3; spec 026 AC-4.4 |
 | `Catalog::QuickAddsController::FULL_LOT` | "You already have the most copies one lot can hold (9,999)." | `app/controllers/catalog/quick_adds_controller.rb` | The quick add's cap message | spec 004 AC-7.3a |
 | `CollectionGrid::PER_PAGE`, `CollectionTable::PER_PAGE` | 120 | `app/models/collection_grid.rb`, `collection_table.rb` | Tiles and rows per page | spec 004 AC-11.3; spec 006 AC-2.4 |
@@ -311,7 +313,7 @@ Measured bars, not constants (recorded in each spec's `verification.md` where on
 - The card page's "Your copies" shows its Location column, every row "No location", for an account with no location; whether it should hide as the grid and table do is open for the maintainer (spec 032 `verification.md`, Implementation review).
 - An unknown sort direction with a known column sorts that column ascending, not the default order spec 006 AC-3.7 names (`CollectionTable::Sort.parse`).
 - No sort by location (spec 037 Non-Goals); a move's merged lots stay merged when moved back (spec 041 decision 1); no partial lots in bulk (decision 2).
-- [#50](https://github.com/plainprogrammer/Collector/issues/50) built the card page's Free to use and In decks (spec 073). [#49](https://github.com/plainprogrammer/Collector/issues/49) built the deck itself and [#252](https://github.com/plainprogrammer/Collector/issues/252) put copies into it: a copy reaches a zone from the lot form, Move copies, bulk Move to…, number entry and the scanner, and the filter finds a zone or a whole deck (spec 065). The deck page is built (#253, spec 066), with proxies (#254, spec 068), Add to deck from the card page (#255, spec 071), replacing a proxy with a copy (#257, spec 070: "Use my copy" takes only free copies, so a copy in another deck's zone or another zone of this deck is moved out first with Move copies) and its shape panel (#256, spec 069); a deck's format changes in the app since spec 064 ([#251](https://github.com/plainprogrammer/Collector/issues/251)), and the "Free to use" rule for Considering and the shape panel's counted zones are under "A deck" above (spec 062 Non-Goals, superseded by spec 064 for the format). Open on the panel: its layout at extreme counts, [#320](https://github.com/plainprogrammer/Collector/issues/320) (a four-digit 6+ count in a 360px browser frame, crowded mana bar labels, a long proxy suffix in the Card types rows: spec 060's CSS, spec 069 verification "Findings").
+- [#50](https://github.com/plainprogrammer/Collector/issues/50) built the card page's Free to use and In decks (spec 073). [#49](https://github.com/plainprogrammer/Collector/issues/49) built the deck itself and [#252](https://github.com/plainprogrammer/Collector/issues/252) put copies into it: a copy reaches a zone from the lot form, Move copies, bulk Move to…, number entry and the scanner, and the filter finds a zone or a whole deck (spec 065). The deck page is built (#253, spec 066), with proxies (#254, spec 068), Add to deck from the card page (#255, spec 071), replacing a proxy with a copy (#257, spec 070: "Use my copy" takes only free copies, so a copy in another deck's zone or another zone of this deck is moved out first with Move copies) and its shape panel (#256, spec 069); a deck's format changes in the app since spec 064 ([#251](https://github.com/plainprogrammer/Collector/issues/251)), and the "Free to use" rule for Considering and the shape panel's counted zones are under "A deck" above (spec 062 Non-Goals, superseded by spec 064 for the format). Open on the panel: its mana labels in segments narrower than their label, [#330](https://github.com/plainprogrammer/Collector/issues/330) (spec 077 decision 2; the other limits of [#320](https://github.com/plainprogrammer/Collector/issues/320) are fixed by spec 077).
 - The card page's `Add to deck` is hidden for a retired printing or a token, even for My copies (spec 071 decision 2).
 - `Deck#zones` is ordered by id, which is the format's order only while every zone was written with the deck: a zone the import's repair adds later, and now a zone a change of format adds (Commander to Modern lists Sideboard after Considering), gets a later id, so it shows last there. The deck page orders what it reads by the format's registered order (spec 066 decision 3, `Deck::Page`), so it reads Main, Sideboard, Considering; `Deck#zones` itself is unchanged (spec 062 verification, open items; spec 064 "Supersedes and updates").
 - `POST /collection/selection` accepts any location id as a stored filter (spec 037 decision 7), so any id can be stored, a deck's own among them, although no page offers one (a zone's and `deck:<id>` are values the filter offers since spec 065, a deck's own id is not); `spec/requests/bulk_zone_filter_spec.rb` pins that each bulk path then acts on the account's own lots in that zone, keeps a lot in its zone unless the collector moves it, and never unplaces one by accident (spec 062 decision 6).
@@ -349,6 +351,7 @@ Measured bars, not constants (recorded in each spec's `verification.md` where on
 - [Spec 069](../specs/069-deck-shape-panel/spec.md) (issue [#256](https://github.com/plainprogrammer/Collector/issues/256)): all (the deck page's shape panel: the seam `Catalog::DeckShape` and Magic's answer `MTG::Collecting.deck_shape` over `MTG::DeckShape`, the counted zones, the types, the curve, the average, the missing mana value, the symbols, the percentages, the split-card limit and the statement count). It supersedes spec 066 AC-1.6's clause "it has no shape panel", AC-5.1's "the second column is empty" for a deck that draws a panel, and its Goals and Non-Goals lines that leave the right column empty (spec 066 is not edited; spec 060 AC-3.12 stands for a deck that draws none).
 - [Spec 070](../specs/070-replace-proxy-with-copy/spec.md) (issue [#257](https://github.com/plainprogrammer/Collector/issues/257)): all (`Lot.free`, `Lot::MoveIntoZone`, `Deck::Proxy::UseCopies` and `AddCopy`, `Deck::Proxy#lower_by!`, the two replacement pages and their controllers, the row menu's two new items, the notices and sentences); the pages' 404 rule is [tenancy and accounts](tenancy-and-accounts.md)'. It supersedes spec 068 decision 5 ("proxy items only"), AC-2.6, AC-2.7's absence of the two items, and in part decision 11 and AC-4.2 (the two replacement pages are writers that move or add a copy and lower the proxy), and records a departure from spec 006 AC-8.1 for lot ids in Use my copy's form.
 - [Spec 073](../specs/073-card-page-deck-stats/spec.md) (issue [#50](https://github.com/plainprogrammer/Collector/issues/50)): decisions 1 to 10, Stories 1 to 6 and FR-1 to FR-3 (the card page's stats and In decks section, the card as the scope, the reads, `Lot.free` as the one definition of free). It supersedes spec 004 FR-9's single stat, its Non-Goal and Out of Scope entries for "In decks" and "Free to use" and AC-8.1's order in part, and fulfils spec 071 AC-1.4 and the Non-Goals of specs 029, 032, 033, 057 and 070 that name #50 (spec 073 "Supersedes and updates"; no spec but 073 is edited).
+- [Spec 077](../specs/077-deck-shape-extreme-counts/spec.md) (issues [#320](https://github.com/plainprogrammer/Collector/issues/320) and [#330](https://github.com/plainprogrammer/Collector/issues/330)): all (the deck shape panel at extreme counts: `MTG::DeckShape#short_count` and the abbreviated curve count and segment label, the Card types grid, the caption's unbreakable items, `caption_items`, the mana-label band as a documented limit, ruled option A). It supersedes in part spec 060 decision 2 and AC-2.2, AC-2.4 to AC-2.6 and spec 069 decision 14 and its Findings 1 to 4 (neither is edited).
 - [Spec 057](../specs/057-deck-design-rules/spec.md): decision 12 (the design system is the project's own, so the notes that said an override waits for a re-export are retired).
 - Cited for a neighbour's fact only: [spec 009](../specs/009-card-scanner-confirm-flow/spec.md) AC-8.2 (the collection page's "Scan cards"), [spec 023](../specs/023-signed-out-catalog-browsing/spec.md) FR-3 (a visitor holds nothing), [spec 028](../specs/028-apply-import/spec.md) decision 7 (an apply clears working state).
 - ADRs [0019](../adr/0019-lot-as-the-unit-of-ownership.md), [0025](../adr/0025-a-lots-location-is-part-of-its-identity.md), [0028](../adr/0028-decks-are-locations-with-zones.md); [0023](../adr/0023-locations-make-export-format-version-2.md) as far as locations and their names use it; [0017](../adr/0017-row-level-tenancy-one-user-per-account.md) and [0018](../adr/0018-collectible-agnostic-catalog-core.md) as far as lots use them; [0015](../adr/0015-admin-progress-by-polling.md) for number entry's refresh.
@@ -381,3 +384,4 @@ Measured bars, not constants (recorded in each spec's `verification.md` where on
 | 2026-11-11 | Spec 070 (replacing a proxy with a copy, #257: `Lot.free` and the glossary's "Free", `Lot::MoveIntoZone`, Use my copy (`UseCopies`) and Add a new copy (`AddCopy`), the row menu's two items and their order, `Deck::Proxy.replaceable?` and `#lower_by!`, the `shown` fields that make a replay refuse, the notices and sentences; the invariant that a proxy changes only by hand now names the two replacement pages; supersedes spec 068 decision 5, AC-2.6, AC-2.7 and AC-4.2 in part and decision 11's writers; the deck guard's count): scope, glossary, invariants, the deck paragraphs and the proxy table, sources | [#323](https://github.com/plainprogrammer/Collector/pull/323) |
 | 2026-11-11 | Spec 071 (add to a deck from the card page, #255: the card page's head action, the Add to deck page and `Deck::Addition`, the free-copies reading `Lot::FreeCopies` over `Lot.free` and the take order, the "My copies" sentence, a proxy per finish, the all-or-nothing take through `Lot::MoveIntoZone`, the refusals and the result on the card page; the glossary's free copy, constants, decisions, the open-issues line and sources): scope, glossary, the card-page paragraph, constants, decisions and sources updated | [#325](https://github.com/plainprogrammer/Collector/pull/325) |
 | 2026-11-11 | Spec 073 (the card page's "Free to use" and "In decks", #50: the three-stat strip, the one definition of free read through `Lot::FreeCopies` with "In decks" as its complement, the "In decks" section from `Deck::CardUses` and its head count, the card as the scope with the Proxy column counting any printing, a visitor's overview reading nothing, the constant number of statements, the three scoped CSS rules and "Your copies"' place class; the Considering and card-page paragraphs, scope, sources and the open-issues line; supersedes spec 004 FR-9's single stat and fulfils spec 071 AC-1.4) | spec 073's pull request |
+| 2026-11-11 | Spec 077 (the deck shape panel at extreme counts, #320: `MTG::DeckShape#short_count` and `#caption_items`, the curve column's count and the mana segment's label abbreviated from 1,000 and every other number exact, the Card types list a grid of subgrid rows, the caption's unbreakable items, the mana-label band recorded as a limit and offered as #330; the two `SHORT_COUNT_*` Named constants; the Issues bullet; specs 060 and 069 superseded in part) | spec 077's pull request |
diff --git a/docs/design-system/README.md b/docs/design-system/README.md
index e9fcde28..584df9c5 100644
--- a/docs/design-system/README.md
+++ b/docs/design-system/README.md
@@ -136,7 +136,7 @@ App additions (spec 048, in `collector/additions.css`): `c-table__place-name` (t
 
 App additions (spec 057, in `collector/additions.css`): the label and option rules of a grouped select (documented in `Form` and `FilterSelect`), `c-select--wide` (`FilterSelect`), `c-stats--three` (`ItemPage`) and the logo mark's stand-in for the wordmark from 640px to 719px while the header holds five links (`AppHeader`).
 
-App additions (spec 060, in `collector/additions.css`; spec 068's `AddProxy` adds no class, only `.c-decks .c-pagehead__title`): `ProxyTag` (`c-proxy`), `DeckShape` (`c-shape`, `c-curve`, `c-types`, `c-manabar`, `c-manacap`), `DeckPage` (`c-deck`, `c-shapefold`, `c-zone`), `DecksIndex` (`c-decks`) and the pick list (`c-pick`) of `AddToDeck` and `UseMyCopy`.
+App additions (spec 060, in `collector/additions.css`; spec 068's `AddProxy` adds no class, only `.c-decks .c-pagehead__title`): `ProxyTag` (`c-proxy`), `DeckShape` (`c-shape`, `c-curve`, `c-types`, `c-manabar`, `c-manacap`, `c-manacap__item`), `DeckPage` (`c-deck`, `c-shapefold`, `c-zone`), `DecksIndex` (`c-decks`) and the pick list (`c-pick`) of `AddToDeck` and `UseMyCopy`.
 
 App additions (spec 062 and 064, in `collector/additions.css`): `DecksIndex`'s row actions (`c-decks__actions`: Rename, Change format and Remove on one line on a wide screen, wrapping onto lines of their own below 640px, spec 064), the removal pages' wrapping rule (`.c-decks .c-confirm`) and the Rename and Change format heads' `.c-decks .c-pagehead > div` (`min-width: 0`, spec 064). Spec 065 (#252) removed `c-field__static`, the read-only field's rule that spec 062 added for the Edit copy form's Location, which no view uses any more; the grouped picker and "Deck (all zones)" use spec 057's rules, and the app adds no CSS for them. Spec 065 adds one rule to `collector/additions.css`: `.c-pagehead__stats` breaks inside a word (`overflow-wrap:anywhere`) on every page head, so a 60-character place or deck name in a stats line (Move copies', the Decks pages') stays inside a 360px frame; it is the one wrap rule, and spec 064's copy scoped to `.c-decks` is gone.
 
diff --git a/docs/design-system/components/DeckShape.md b/docs/design-system/components/DeckShape.md
index c94c056d..ac1c03a8 100644
--- a/docs/design-system/components/DeckShape.md
+++ b/docs/design-system/components/DeckShape.md
@@ -24,7 +24,7 @@ A panel that answers what a deck does from the catalog alone: its mana curve, it
       <ul class="c-types"><li><span>Lands</span><span class="c-types__n">24</span><span class="c-types__proxy">2 proxy</span></li><li><span>Artifacts</span><span class="c-types__n">4</span><span class="c-types__proxy"></span></li></ul></div>
     <div class="c-shape__part"><h3 class="c-shape__heading">Mana symbols</h3>
       <ul class="c-manabar" role="img" aria-label="Coloured mana symbols in Main: blue 44, red 16"><li class="c-manabar__seg c-manabar__seg--u" style="--n:44"><span>U 44</span></li><li class="c-manabar__seg c-manabar__seg--r" style="--n:16"><span>R 16</span></li></ul>
-      <p class="c-manacap">Blue 44 · 73% · Red 16 · 27%</p></div>
+      <p class="c-manacap"><span class="c-manacap__item">Blue 44 · 73% ·</span> <span class="c-manacap__item">Red 16 · 27%</span></p></div>
   </section>
 </aside>
 ```
@@ -57,7 +57,7 @@ The `details` copy, shown below 640px of the page container, holds the same thre
 
 **The curve** (`c-curve`, spec 060 decisions 1 and 20, AC-2.2, AC-2.3).
 
-- A column is a count above a bar of at most 96px, with the proxy part on top of the owned part, and one baseline in `line-strong` under the seven columns. The mana values 0 to 6+ are labelled under it. The count above a column is owned plus proxy, in mono.
+- A column is a count above a bar of at most 96px, with the proxy part on top of the owned part, and one baseline in `line-strong` under the seven columns. The mana values 0 to 6+ are labelled under it. The count above a column is owned plus proxy, in mono. From 1,000 it is abbreviated, one decimal under 10,000 and whole from there ("1.4k", "12k", "1.3m"), half up, never longer than four characters through 999m, so it fits a 30px column at a 320px viewport (spec 077 decisions 1 and 5; `MTG::DeckShape#short_count`). The exact column count is in the chart's `aria-label` only; the legend shows the curve's totals. Everything else is exact and delimited: the legend, the types' counts and suffixes, the caption, the zone heads and every `aria-label`.
 - The numbers are custom properties in a `style` attribute: `--owned`, `--proxy` and `--max` (the tallest column's owned plus proxy), all unitless. The CSS makes a part's height `96px × part ÷ max`. No value in a `style` attribute is a colour.
 - Owned is a solid `ink-muted` fill. Proxy is the hatch: lines at 45°, 1.5px wide in `line-strong` every 5px on `surface-raised`, with a 1px `ink-muted` outline. The two outlines share one line.
 - Under the chart sit the legend, with a swatch for each series ("Owned 30", "Proxy 6"), and "Average mana value 2.3" (and, when a nonland card has no mana value, the note as a second `c-curve__avg` paragraph, spec 069 decision 12). Counts, legend and average repeat what the chart shows, so no value is carried by fill or pattern alone.
@@ -66,15 +66,15 @@ The `details` copy, shown below 640px of the page container, holds the same thre
 **Card types** (`c-types`, spec 060 decision 2, AC-2.5).
 
 - Each type is one row: its name, its count (owned plus proxy, mono) at the end, and, when it has proxies, a muted "N proxy" (mono, 12px, `ink-muted`) after the count. Rows are divided by `line`.
-- A type with no proxies keeps an empty `c-types__proxy`. The element has a minimum width of 4.5rem, so the counts of every row line up whether or not a row has a suffix.
+- The list is a grid of three columns (`minmax(0, 1fr) auto auto`, a `space-2` column gap) and each row is a `subgrid` of it, so the count column is as wide as the widest count and the suffix column as wide as the widest suffix: every count ends at one x whatever a suffix says ("1,234 proxy" is wider than the old 4.5rem minimum). A type with no proxies keeps an empty `c-types__proxy`, which keeps a 4.5rem floor so a list with short suffixes looks as it did. The name cell has `min-width:0` and `overflow-wrap:anywhere`: with a five-digit count and suffix at a 320px viewport ("Planeswalkers 12,345 12,345 proxy") the name wraps inside its cell, even mid-word, and no row passes the list (spec 077 decision 3). Counts and suffix are exact.
 - With no card in the counted zones the part keeps its heading and says "No cards yet." in a `c-manacap` paragraph in place of the list (spec 069 decision 14).
 
-**The mana bar** (`c-manabar`, `c-manacap`, spec 060 decisions 3 and 4, AC-2.6, AC-2.7).
+**The mana bar** (`c-manabar`, `c-manacap`, `c-manacap__item`, spec 060 decisions 3 and 4, AC-2.6, AC-2.7).
 
 - One bar, 32px high, with a 1px `line-strong` outline and `radius-sm` corners, split into segments whose widths are proportional to their symbol counts (`--n`). Consecutive segments are divided by a 1px `line-strong` border.
-- Each segment fills with its `mana-*` token and its text is the matching `on-mana-*` token, the same in both themes. The segment's label is its letter and count ("U 44"), in mono.
+- Each segment fills with its `mana-*` token and its text is the matching `on-mana-*` token, the same in both themes. The segment's label is its letter and count ("U 44"), in mono. From 1,000 the count is abbreviated ("W 2.6k": six characters, about 43.5px on a Mac and 44.7px on Linux, which fits one line in the narrowest five-equal-segment bar at a 320px viewport); the caption and the `aria-label` keep the exact count (spec 077 decision 2).
 - A segment under 10% of the symbols adds `c-manabar__seg--narrow`, which hides its label: the server sets the class, with no JavaScript. The caption carries that segment's count.
-- The caption (`c-manacap`) names every colour present with its count and its percentage: "Blue 44 · 73% · Red 16 · 27%". In a five-colour deck whose white segment is under 10%, that segment shows no label and the caption still gives white's count and percentage.
+- The caption (`c-manacap`) names every colour present with its count and its percentage: "Blue 44 · 73% · Red 16 · 27%". In a five-colour deck whose white segment is under 10%, that segment shows no label and the caption still gives white's count and percentage. The caption is a run of `c-manacap__item` spans, "Blue 44 · 73% ·" and, last, "Red 16 · 27%", separated by one space, each `white-space:nowrap`: a line breaks only between colours, so no line begins with "·" and a name never parts from its count; the paragraph's text is the same string as before (spec 077 decision 4; `MTG::DeckShape#caption_items`).
 - The bar is `role="img"` and its `aria-label` reads the counts ("Coloured mana symbols in Main: blue 44, red 16").
 - With no coloured symbol the part has no bar and no caption and says "No coloured mana symbols." in a `c-manacap` paragraph (spec 069 decision 14).
 - The `mana-*` fills are used here and in `ManaCost`'s pips only (README Charts rule).
@@ -85,7 +85,7 @@ The `details` copy, shown below 640px of the page container, holds the same thre
 - The panel itself is `surface-raised` with a 1px `line` border, `radius-md` corners, `space-4` padding and `space-6` between its parts. The title is the sans family at 20px and weight 600; the scope line is mono and `ink-muted`.
 - Fills and outlines hold 3:1 on `surface`, `surface-raised` and `surface-sunken` in both themes (the tokens' contrast is guarded by spec 057 AC-2.3).
 
-**Width.** The panel works in 279px, the space inside its border and padding when it stacks in a 360px frame (313px wide, 279px inside); every count, the legend and the types fit with no horizontal overflow (spec 060 AC-2.4). Beside the zones it is 320px wide (spec 060 decision 6). Measured on the built page (spec 069, #256): a phone's 360px viewport gives a 278px interior (8px body margin, 16px page padding, 2px border and 32px panel padding), one pixel under the 279px the component was drawn for. With the app's classic scrollbar in a 360px browser frame the interior is 263px, and a four-digit count in the 6+ column ("1,357") overflows its column by about 1.85px. That is a known limit, filed as [#320](https://github.com/plainprogrammer/Collector/issues/320) with the other layout limits at extreme counts (crowded mana bar labels at four-digit counts, a proxy suffix wider than 4.5rem, a caption that can wrap with a leading dot), and spec 069 changes no CSS.
+**Width.** The panel works in 238px, the space inside its border and padding at a 320px viewport (a phone's 320px less 16px body margin, 32px page padding, 2px border and 32px panel padding), and every count, the legend, the types and the mana labels fit with no horizontal overflow at the worst-case counts of spec 069's fixture (spec 077 AC-1.3, AC-2.2, AC-3.1). It was drawn for 279px (spec 060 AC-2.4); the measured interiors are 238px (a 320px viewport), 263px (a 360px browser frame, which keeps a 15px scrollbar a phone doesn't), 278px (a 360px viewport) and 286px (the 320px aside; beside the zones the panel is 320px wide, spec 060 decision 6). **Known limits:** (1) a mana segment narrower than its label: a label of four characters ("U 44") needs about 14% of the bar, five ("W 999") about 17% and six ("W 2.6k") about 20% at a 320px viewport (measured to the half percent on a 236px bar), and the rule hides a label only under 10% (spec 060 decision 4), so a segment between 10% and those shares wraps its label and cuts it off; the caption carries the count; it exists at ordinary counts too, and is [#330](https://github.com/plainprogrammer/Collector/issues/330) (spec 077 decision 2, ruled option A). (2) A Card types row with a five-digit count and suffix at a 320px viewport wraps its name, even mid-word; counts of six digits are not specified.
 
 **The two copies** (spec 060 decision 7, AC-3.4). A closed `details` hides its content whatever the CSS says, and `::details-content` would put a browser floor under the page, so the deck page renders the panel twice and each copy hides the other with `display: none`.
 
diff --git a/docs/design-system/previews/DeckPage.html b/docs/design-system/previews/DeckPage.html
index 971d4381..2f14a518 100644
--- a/docs/design-system/previews/DeckPage.html
+++ b/docs/design-system/previews/DeckPage.html
@@ -55,7 +55,7 @@ <h1 class="c-item__title">Izzet Tempo</h1>
       <ul class="c-types"><li><span>Lands</span><span class="c-types__n">24</span><span class="c-types__proxy">2 proxy</span></li><li><span>Creatures</span><span class="c-types__n">12</span><span class="c-types__proxy">2 proxy</span></li><li><span>Instants</span><span class="c-types__n">16</span><span class="c-types__proxy">1 proxy</span></li><li><span>Sorceries</span><span class="c-types__n">4</span><span class="c-types__proxy">3 proxy</span></li><li><span>Artifacts</span><span class="c-types__n">4</span><span class="c-types__proxy"></span></li></ul></div>
     <div class="c-shape__part"><h3 class="c-shape__heading">Mana symbols</h3>
       <ul class="c-manabar" role="img" aria-label="Coloured mana symbols in Main: blue 44, red 16"><li class="c-manabar__seg c-manabar__seg--u" style="--n:44"><span>U 44</span></li><li class="c-manabar__seg c-manabar__seg--r" style="--n:16"><span>R 16</span></li></ul>
-        <p class="c-manacap">Blue 44 · 73% · Red 16 · 27%</p></div>
+        <p class="c-manacap"><span class="c-manacap__item">Blue 44 · 73% ·</span> <span class="c-manacap__item">Red 16 · 27%</span></p></div>
       </div>
     </details>
     <aside class="c-deck__shape">
@@ -82,7 +82,7 @@ <h1 class="c-item__title">Izzet Tempo</h1>
       <ul class="c-types"><li><span>Lands</span><span class="c-types__n">24</span><span class="c-types__proxy">2 proxy</span></li><li><span>Creatures</span><span class="c-types__n">12</span><span class="c-types__proxy">2 proxy</span></li><li><span>Instants</span><span class="c-types__n">16</span><span class="c-types__proxy">1 proxy</span></li><li><span>Sorceries</span><span class="c-types__n">4</span><span class="c-types__proxy">3 proxy</span></li><li><span>Artifacts</span><span class="c-types__n">4</span><span class="c-types__proxy"></span></li></ul></div>
     <div class="c-shape__part"><h3 class="c-shape__heading">Mana symbols</h3>
       <ul class="c-manabar" role="img" aria-label="Coloured mana symbols in Main: blue 44, red 16"><li class="c-manabar__seg c-manabar__seg--u" style="--n:44"><span>U 44</span></li><li class="c-manabar__seg c-manabar__seg--r" style="--n:16"><span>R 16</span></li></ul>
-        <p class="c-manacap">Blue 44 · 73% · Red 16 · 27%</p></div>
+        <p class="c-manacap"><span class="c-manacap__item">Blue 44 · 73% ·</span> <span class="c-manacap__item">Red 16 · 27%</span></p></div>
       </section>
     </aside>
     <div class="c-deck__zones">
diff --git a/docs/design-system/previews/DeckPagePhone.html b/docs/design-system/previews/DeckPagePhone.html
index 971d4381..2f14a518 100644
--- a/docs/design-system/previews/DeckPagePhone.html
+++ b/docs/design-system/previews/DeckPagePhone.html
@@ -55,7 +55,7 @@ <h1 class="c-item__title">Izzet Tempo</h1>
       <ul class="c-types"><li><span>Lands</span><span class="c-types__n">24</span><span class="c-types__proxy">2 proxy</span></li><li><span>Creatures</span><span class="c-types__n">12</span><span class="c-types__proxy">2 proxy</span></li><li><span>Instants</span><span class="c-types__n">16</span><span class="c-types__proxy">1 proxy</span></li><li><span>Sorceries</span><span class="c-types__n">4</span><span class="c-types__proxy">3 proxy</span></li><li><span>Artifacts</span><span class="c-types__n">4</span><span class="c-types__proxy"></span></li></ul></div>
     <div class="c-shape__part"><h3 class="c-shape__heading">Mana symbols</h3>
       <ul class="c-manabar" role="img" aria-label="Coloured mana symbols in Main: blue 44, red 16"><li class="c-manabar__seg c-manabar__seg--u" style="--n:44"><span>U 44</span></li><li class="c-manabar__seg c-manabar__seg--r" style="--n:16"><span>R 16</span></li></ul>
-        <p class="c-manacap">Blue 44 · 73% · Red 16 · 27%</p></div>
+        <p class="c-manacap"><span class="c-manacap__item">Blue 44 · 73% ·</span> <span class="c-manacap__item">Red 16 · 27%</span></p></div>
       </div>
     </details>
     <aside class="c-deck__shape">
@@ -82,7 +82,7 @@ <h1 class="c-item__title">Izzet Tempo</h1>
       <ul class="c-types"><li><span>Lands</span><span class="c-types__n">24</span><span class="c-types__proxy">2 proxy</span></li><li><span>Creatures</span><span class="c-types__n">12</span><span class="c-types__proxy">2 proxy</span></li><li><span>Instants</span><span class="c-types__n">16</span><span class="c-types__proxy">1 proxy</span></li><li><span>Sorceries</span><span class="c-types__n">4</span><span class="c-types__proxy">3 proxy</span></li><li><span>Artifacts</span><span class="c-types__n">4</span><span class="c-types__proxy"></span></li></ul></div>
     <div class="c-shape__part"><h3 class="c-shape__heading">Mana symbols</h3>
       <ul class="c-manabar" role="img" aria-label="Coloured mana symbols in Main: blue 44, red 16"><li class="c-manabar__seg c-manabar__seg--u" style="--n:44"><span>U 44</span></li><li class="c-manabar__seg c-manabar__seg--r" style="--n:16"><span>R 16</span></li></ul>
-        <p class="c-manacap">Blue 44 · 73% · Red 16 · 27%</p></div>
+        <p class="c-manacap"><span class="c-manacap__item">Blue 44 · 73% ·</span> <span class="c-manacap__item">Red 16 · 27%</span></p></div>
       </section>
     </aside>
     <div class="c-deck__zones">
diff --git a/docs/design-system/previews/DeckShape.html b/docs/design-system/previews/DeckShape.html
index 599a504c..eb2967e1 100644
--- a/docs/design-system/previews/DeckShape.html
+++ b/docs/design-system/previews/DeckShape.html
@@ -37,12 +37,12 @@
       <ul class="c-types"><li><span>Lands</span><span class="c-types__n">24</span><span class="c-types__proxy">2 proxy</span></li><li><span>Creatures</span><span class="c-types__n">12</span><span class="c-types__proxy">2 proxy</span></li><li><span>Instants</span><span class="c-types__n">16</span><span class="c-types__proxy">1 proxy</span></li><li><span>Sorceries</span><span class="c-types__n">4</span><span class="c-types__proxy">3 proxy</span></li><li><span>Artifacts</span><span class="c-types__n">4</span><span class="c-types__proxy"></span></li></ul></div>
     <div class="c-shape__part"><h3 class="c-shape__heading">Mana symbols</h3>
       <ul class="c-manabar" role="img" aria-label="Coloured mana symbols in Main: blue 44, red 16"><li class="c-manabar__seg c-manabar__seg--u" style="--n:44"><span>U 44</span></li><li class="c-manabar__seg c-manabar__seg--r" style="--n:16"><span>R 16</span></li></ul>
-        <p class="c-manacap">Blue 44 · 73% · Red 16 · 27%</p></div>
+        <p class="c-manacap"><span class="c-manacap__item">Blue 44 · 73% ·</span> <span class="c-manacap__item">Red 16 · 27%</span></p></div>
       </section>
     </aside>
     </div></div>
   <!-- The details copy is shown only below 640px of a page container (.c-page), so it sits in a 313px .c-page wrapper:
-       the page's own container rule shows it here, as it would on a 360px phone (313px wide, 279px inside). -->
+       the page's own container rule shows it here, as it would on a 360px phone (on a 360px phone the fold is 312px wide, 278px inside). -->
   <div class="pv__cell"><p class="pv__cap">details copy, closed, 313px (shown below 640px)</p>
     <div class="c-page" style="width:313px">
     <details class="c-shapefold">
@@ -69,7 +69,7 @@
       <ul class="c-types"><li><span>Lands</span><span class="c-types__n">24</span><span class="c-types__proxy">2 proxy</span></li><li><span>Creatures</span><span class="c-types__n">12</span><span class="c-types__proxy">2 proxy</span></li><li><span>Instants</span><span class="c-types__n">16</span><span class="c-types__proxy">1 proxy</span></li><li><span>Sorceries</span><span class="c-types__n">4</span><span class="c-types__proxy">3 proxy</span></li><li><span>Artifacts</span><span class="c-types__n">4</span><span class="c-types__proxy"></span></li></ul></div>
     <div class="c-shape__part"><h3 class="c-shape__heading">Mana symbols</h3>
       <ul class="c-manabar" role="img" aria-label="Coloured mana symbols in Main: blue 44, red 16"><li class="c-manabar__seg c-manabar__seg--u" style="--n:44"><span>U 44</span></li><li class="c-manabar__seg c-manabar__seg--r" style="--n:16"><span>R 16</span></li></ul>
-        <p class="c-manacap">Blue 44 · 73% · Red 16 · 27%</p></div>
+        <p class="c-manacap"><span class="c-manacap__item">Blue 44 · 73% ·</span> <span class="c-manacap__item">Red 16 · 27%</span></p></div>
       </div>
     </details>
     </div></div>
@@ -99,7 +99,7 @@
       <ul class="c-types"><li><span>Lands</span><span class="c-types__n">24</span><span class="c-types__proxy">2 proxy</span></li><li><span>Creatures</span><span class="c-types__n">12</span><span class="c-types__proxy">2 proxy</span></li><li><span>Instants</span><span class="c-types__n">16</span><span class="c-types__proxy">1 proxy</span></li><li><span>Sorceries</span><span class="c-types__n">4</span><span class="c-types__proxy">3 proxy</span></li><li><span>Artifacts</span><span class="c-types__n">4</span><span class="c-types__proxy"></span></li></ul></div>
     <div class="c-shape__part"><h3 class="c-shape__heading">Mana symbols</h3>
       <ul class="c-manabar" role="img" aria-label="Coloured mana symbols in Main: blue 44, red 16"><li class="c-manabar__seg c-manabar__seg--u" style="--n:44"><span>U 44</span></li><li class="c-manabar__seg c-manabar__seg--r" style="--n:16"><span>R 16</span></li></ul>
-        <p class="c-manacap">Blue 44 · 73% · Red 16 · 27%</p></div>
+        <p class="c-manacap"><span class="c-manacap__item">Blue 44 · 73% ·</span> <span class="c-manacap__item">Red 16 · 27%</span></p></div>
       </div>
     </details>
     </div></div>
@@ -109,7 +109,7 @@
 <div style="width:320px"><div class="c-shape">
   <div class="c-shape__part"><h3 class="c-shape__heading">Mana symbols</h3>
     <ul class="c-manabar" role="img" aria-label="Coloured mana symbols in Main: white 3, blue 20, black 4, red 18, green 15"><li class="c-manabar__seg c-manabar__seg--w c-manabar__seg--narrow" style="--n:3"><span>W 3</span></li><li class="c-manabar__seg c-manabar__seg--u" style="--n:20"><span>U 20</span></li><li class="c-manabar__seg c-manabar__seg--b c-manabar__seg--narrow" style="--n:4"><span>B 4</span></li><li class="c-manabar__seg c-manabar__seg--r" style="--n:18"><span>R 18</span></li><li class="c-manabar__seg c-manabar__seg--g" style="--n:15"><span>G 15</span></li></ul>
-        <p class="c-manacap">White 3 · 5% · Blue 20 · 33% · Black 4 · 7% · Red 18 · 30% · Green 15 · 25%</p></div>
+        <p class="c-manacap"><span class="c-manacap__item">White 3 · 5% ·</span> <span class="c-manacap__item">Blue 20 · 33% ·</span> <span class="c-manacap__item">Black 4 · 7% ·</span> <span class="c-manacap__item">Red 18 · 30% ·</span> <span class="c-manacap__item">Green 15 · 25%</span></p></div>
 </div></div>
 </div>
 <script>/* preview only: ?theme=dark */ if(new URLSearchParams(location.search).get("theme")==="dark")document.documentElement.dataset.theme="dark";</script>
diff --git a/docs/lessons/2026-11-11-build-a-measurement-table-cell-from-the-data.md b/docs/lessons/2026-11-11-build-a-measurement-table-cell-from-the-data.md
new file mode 100644
index 00000000..6fec1593
--- /dev/null
+++ b/docs/lessons/2026-11-11-build-a-measurement-table-cell-from-the-data.md
@@ -0,0 +1,25 @@
+---
+date: 2026-11-11
+spec: 077 (#320)
+tags: [specs, reviews, measurement, acceptance-criteria]
+---
+
+# Lesson: build a measurement table's cells from the data, and re-measure the width a variant depends on
+
+## Context
+
+Spec 077's problem statement has a table of what the four layout limits measure at four widths. I typed it from a JSON dump of one throwaway run.
+
+## What happened
+
+- The caption cell for the 320px viewport read "breaks inside 'Green / 2,591'" where the JSON showed the old caption breaking between colours, with no item split (the split was the 345px row). Two Opus spec reviews and Gate A carried the cell, and AC-4.3's failing variant was written for 320px.
+- It surfaced only when the plan needed a failing width for that variant and the supervisor re-read the numbers. The fix was a PATCH (spec 0.3.1) and the variant moved to 345px and 360px.
+
+## What to do next time
+
+Generate each cell of a measurement table from the measured value in the same step (a script that prints the table), or re-read every cell against the data before the spec is reviewed. When an acceptance criterion's failing variant depends on one cell ("an item splits at this width"), re-measure that width on its own before approving the criterion.
+
+## Signals to watch for
+
+- A table typed by hand from JSON or console output.
+- A failing variant whose width comes from a table row nobody re-measured.
diff --git a/docs/lessons/2026-11-11-sweep-the-widths-before-calling-a-wrapping-clause-unfailable.md b/docs/lessons/2026-11-11-sweep-the-widths-before-calling-a-wrapping-clause-unfailable.md
new file mode 100644
index 00000000..c1ad9f34
--- /dev/null
+++ b/docs/lessons/2026-11-11-sweep-the-widths-before-calling-a-wrapping-clause-unfailable.md
@@ -0,0 +1,25 @@
+---
+date: 2026-11-11
+spec: 077 (#320)
+tags: [testing, system-specs, narrow-frame, wrapping, acceptance-criteria]
+---
+
+# Lesson: sweep the widths before calling a wrapping clause unfailable
+
+## Context
+
+Spec 077 AC-4.3 says no line of the mana caption begins with "·". The spec's first draft said the clause "cannot fail on this fixture" because four measured widths showed dots only at the end of lines, and proposed proving it indirectly through the single-rect check on each item.
+
+## What happened
+
+- The plan's first task swept the old caption through 72 frame widths (255px to 468px in steps of 3px, one throwaway system spec reading the first character of each line with `Range` rects). A line began with "·" at 16 of them, including a 366px frame, a phone-width interior of 269px on a Mac and 272px on the Linux runner.
+- So the clause could fail after all: the failing variant now runs at that frame and expects a line to start with "·". Without the sweep the criterion would have been proved only by an indirect check.
+
+## What to do next time
+
+When a criterion is about where text wraps (a line that starts or ends with something, a name split from its count), measure a range of widths in one throwaway run before concluding that a clause cannot fail, and pick the failing variant's width from the result, checking that it holds for both the Mac's 15px and the runner's 12px scrollbar.
+
+## Signals to watch for
+
+- "No line begins with X", "never wraps inside Y" in a criterion.
+- A draft that says a clause "cannot fail on this fixture" after looking at three or four widths.
diff --git a/docs/specs/077-deck-shape-extreme-counts/plan.md b/docs/specs/077-deck-shape-extreme-counts/plan.md
new file mode 100644
index 00000000..ee650c8a
--- /dev/null
+++ b/docs/specs/077-deck-shape-extreme-counts/plan.md
@@ -0,0 +1,565 @@
+# Plan 077: The Deck Shape Panel Stays Tidy at Extreme Counts
+
+**Spec:** [spec.md](spec.md) 0.3.1 (approved, Gate A at 0.3.0; decision 2 ruled A; 0.3.1 is a PATCH from the measurements below) | **Issue:** [#320](https://github.com/plainprogrammer/Collector/issues/320) | **Branch:** `077-deck-shape-extreme-counts`
+
+## Global Constraints
+
+- Test first, one commit per task. Targeted `bin/rspec <file>` runs need no CI slot; a full `bin/ci`, or a push whose pre-push hook runs it, needs a slot from the coordinator.
+- Never run two test runs at once in this worktree (they share the test database). The untracked `spec/system/zz_077_*` throwaway files are the supervisor's; don't run, edit or commit them.
+- The app diff is exactly `app/models/mtg/deck_shape.rb`, `app/views/mtg/deck_shapes/_parts.html.erb` and `app/assets/stylesheets/collector/additions.css` (AC-6.1); the CSS edits stay inside the two DeckShape blocks (card types, mana bar) and use tokens (`var(--space-2)`, `var(--line)`…) wherever the block already does; nothing else in `additions.css` is touched, reordered or reformatted (four other workers edit it).
+- Commit trailer: the `Co-Authored-By` line the session's attribution reminder gives (earlier commits of this branch carry a different one; leave them), as a second message: `git commit -m "<subject>" -m "Co-Authored-By: …"`. One git operation per shell call (stage with `git add <paths>`, then `git commit`). `plan.md` itself is committed at Gate B, before batch A starts. Never `git add -A`: the untracked `spec/system/zz_077_*_spec.rb` throwaways must not be committed.
+- Off limits: `.kamal/secrets`, `config/master.key`, `.env*`, `storage/`, `*.sqlite3`; every `git grep` ends with `':!.kamal/secrets'`.
+- Run `bin/rubocop -a` on each Ruby file written (the project's hook does too); array literals use the omakase style `[ a, b ]`. New public code carries a YARD comment: a summary, `@param`, `@return`, and the why (`.claude/rules/code-documentation.md`); a `{…}` in a comment is YARD link syntax, so put braces in backticks.
+- Geometry examples size every viewport frame from the measured scrollbar (the viewport plus `window.innerWidth − documentElement.clientWidth`) and compare against `clientWidth`, never `innerWidth` (lessons of 2026-11-11 on phone widths and measurement traps); the 366px and 640px frames are fixed frame widths by the spec's words, and the aside is a 1280px window.
+- Every geometry example has a failing variant in the same file (spec Notes; lesson `an-example-that-cannot-fail-proves-nothing`): the old text or the old rule injected into the page, expecting the predicate to be false.
+
+## Goal
+
+`MTG::DeckShape` gets `short_count` and `caption_items`; the partial writes a curve column's count and a mana segment's label through `short_count` and the caption as unbreakable items; the card types list becomes a grid whose rows are subgrids; one nowrap rule is added for the items; the component document, three previews and the area doc say so (spec decisions 1 to 5, FR-1 to FR-4).
+
+## Pre-implementation gates
+
+- **Simplicity:** one model class gains two methods and two constants, one partial changes, two CSS blocks change. No new component, dependency, class or abstraction. Pass.
+- **Anti-abstraction:** `short_count` is a few lines of integer arithmetic; `number_to_human` is not used (spec decision 5: it keeps two significant digits, 123,456 would read "120k"). Pass.
+- **Integration-first:** the contract is the two public methods' return values (AC-1.1, AC-4.1) and the rendered markup (AC-1.2, AC-2.1, AC-4.2); Task 1 and Task 2 write them as examples before any code. Pass.
+
+## Measurements taken before this plan (2026-11-11, headless Firefox, the unchanged code; throwaway `zz_077_measure_spec.rb`)
+
+- **The label band** (a 320px viewport, a 236px bar): a mana label fits its segment (one line, within the segment's `clientWidth` less 2px) from a share of 20% for six characters, 17% for five and 14% for four, on a half-percent grid. The fixture's five equal segments are 19.99% to 20.02%, right at the six-character line (0.44px of room): the thin margin of AC-2.2.
+- **The leading dot:** the old caption starts a line with "·" at 16 of 72 frame widths from 255px to 468px (a fold open), among them frames of 285px, 321px, 363px, 366px, 369px, 402px and 438px; at a 366px frame the interior is 269px on a Mac and 272px on the Linux runner, both in the list, so the spec's 366px frame is robust to the scrollbar's width. At a 320px viewport (a 335px frame) the old caption breaks between colours, so the variant of AC-4.3 does not use it; at 345px and 360px it splits "Green / 2,591" and "Black / 2,591".
+- These go into `verification.md` and into `DeckShape.md`'s known limit (Task 5).
+
+## The existing examples this changes (from a grep, not from memory)
+
+Command (run 2026-11-11 from the repo root): `git grep -n -i -E "c-curve__count|c-manabar__seg|c-manacap|c-types|\.caption\b|countRoom|1,357|2,591|4\.5rem" -- spec docs/design-system docs/architecture app ':!.kamal/secrets'`, read hit by hit.
+
+| Hit | Disposition |
+|---|---|
+| `spec/system/deck_shape_panel_spec.rb:129-134` (the AC-5.3 comment: "a CSS question raised with the maintainer, not decided here") and `:102-106` (`open_fold_fit` measures `.c-curve__count`, `.c-types li > *`, `.c-manacap` for clipping) | the comment is rewritten (Task 2 Step 4); the example's assertions are unchanged and must stay green (AC-6.2) |
+| `spec/models/mtg/deck_shape_spec.rb:253, 293, 315` (`caption` strings, `be_nil`) | kept: `caption` returns the same string (AC-4.1); they stay green and guard it |
+| `spec/requests/deck_shape_panel_spec.rb:79-82, 93` (segment texts "U 6", "R 3", caption strings) and `:112` (`.c-curve__count` texts `%w[0] * 7`) | kept: below 1,000 nothing changes and the paragraph's `text` is the same string (AC-4.2) |
+| `spec/system/design_system_deck_components_spec.rb:82-111, 327-475` (builds its own `c-curve`, `c-types`, `c-manabar` markup; AC-3.3 guards the types' alignment) | kept; run after the CSS (Task 4) |
+| `spec/requests/deck_shape_panel_spec.rb:78, 103, 116`, `spec/system/deck_shape_panel_spec.rb:154` (the segments' classes and style, the paint example's `.c-manabar__seg` tokens) | kept: the markup of a segment and its tokens don't change |
+| `docs/design-system/README.md:138` (the App additions list of `DeckShape`'s classes) | gains `c-manacap__item` (Task 5 Step 3b) |
+| `docs/design-system/components/DeckShape.md:15-27, 55-56, 60, 64, 68-69, 72-79, 88` | rewritten (Task 5) |
+| `docs/design-system/previews/DeckShape.html:45, 40, 72, 102, 112`, `DeckPage.html:58, 85`, `DeckPagePhone.html:58, 85` (the caption paragraphs and one stale width comment) | rewritten (Task 5) |
+| `docs/architecture/collection-and-lots.md:137` (the panel's paragraph), `:244-246` (Named constants), `:312` (the Issues bullet naming #320) | rewritten or extended (Task 5) |
+| `docs/specs/**` (060, 069 and their `verification.md`) | history, not edited |
+
+## File structure
+
+| File | Responsibility |
+|---|---|
+| `app/models/mtg/deck_shape.rb` | `SHORT_COUNT_STEP`, `SHORT_COUNT_UNITS`, `#short_count`, `#caption_items`; `#caption` built from the items |
+| `app/views/mtg/deck_shapes/_parts.html.erb` | the column count and the segment label through `short_count`; the caption as items |
+| `app/assets/stylesheets/collector/additions.css` | the card types block (grid, subgrid rows) and the mana bar block (`.c-manacap__item`) |
+| `spec/models/mtg/deck_shape_spec.rb` | AC-1.1, AC-4.1 |
+| `spec/requests/deck_shape_panel_spec.rb` | AC-1.2, AC-2.1, AC-4.2 |
+| `spec/system/deck_shape_panel_spec.rb` | a nested `describe` with AC-1.3, AC-2.2, AC-3.1, AC-3.2, AC-3.1b, AC-4.3 and their failing variants (spec FR-3), and the stale comment (AC-6.2) |
+| `docs/design-system/components/DeckShape.md`, three previews, `docs/architecture/collection-and-lots.md`, `docs/specs/README.md` | Story 5 |
+| `docs/specs/077-deck-shape-extreme-counts/{screenshots/,verification.md}` | evidence, the grep, the review record |
+
+Batches for implementer subagents (`sonnet`, read `plan.md` by line range, one TDD commit per task, the supervisor checks each batch's diff): **A** = Tasks 1–2, **B** = Tasks 3–4, **C** = Task 5. Task 6 is the supervisor's.
+
+---
+
+## Phase 1: The figures and the examples, red first
+
+**Implements:** FR-1, FR-3 | **Satisfies:** AC-1.1, AC-4.1 (model); AC-1.2, AC-2.1, AC-4.2 (request); AC-1.3, AC-2.2, AC-3.1, AC-3.2, AC-3.1b, AC-4.3 (system) | **Files:** `app/models/mtg/deck_shape.rb`, three spec files | **Interfaces:** `MTG::DeckShape#short_count(number) → String`, `MTG::DeckShape#caption_items → Array<String>`
+
+### Task 1: `short_count` and `caption_items`
+
+**Step 1: write the failing model examples.** Append inside `RSpec.describe MTG::DeckShape` in `spec/models/mtg/deck_shape_spec.rb`, before the final `end`:
+
+```ruby
+  # Spec 077 decision 5: a count in a slot of fixed width is written with a unit from 1,000.
+  describe "#short_count" do
+    it "writes 0 to 999 as they are and 1,000 up with one decimal under 10,000, whole from there, half up (AC-1.1)", :aggregate_failures do
+      expected = { 0 => "0", 999 => "999", 1_000 => "1k", 1_049 => "1k", 1_050 => "1.1k", 1_350 => "1.4k", 1_357 => "1.4k", 9_949 => "9.9k",
+                   9_950 => "10k", 9_999 => "10k", 10_000 => "10k", 12_345 => "12k", 123_456 => "123k", 999_499 => "999k", 999_500 => "1m",
+                   1_250_000 => "1.3m", 12_000_000 => "12m" }
+
+      figures = shape
+      expected.each { |number, text| expect(figures.short_count(number)).to eq(text), "for #{number}" }
+      expect(expected.keys.map { figures.short_count(it).length }.max).to be <= 4
+    end
+
+    it "rounds 1,349 down to 1.3k and a count that reaches 1,000 of its unit moves to the next, 9,999,500 being 10m", :aggregate_failures do
+      expect(shape.short_count(1_349)).to eq("1.3k")
+      expect(shape.short_count(9_999_500)).to eq("10m")
+      expect(shape.short_count(2_000_000_000)).to eq("2b")
+    end
+  end
+
+  # Spec 077 decision 4: the caption as items a view can keep whole.
+  describe "#caption_items" do
+    # Spec 069's worst case: 123 owned five-colour creatures, 1,234 proxies of a hybrid card and 4 Bolts, red being 2,595.
+    def worst_case
+      own(card(name: "Worst", type_line: "Creature — Elf", cost: "{2}{W}{U}{B}{R}{G}", mana_value: 7.0), 123)
+      proxy(card(name: "Gleemax", cost: "{W/U}{U/B}{B/R}{R/G}{G/W}", mana_value: 1_000_000.0), 1_234)
+      own(card(name: "Bolt", cost: "{R}", mana_value: 1.0), 4)
+    end
+
+    it "gives one item per colour, each but the last ending in the separator, and `caption` is the items joined (AC-4.1)", :aggregate_failures do
+      worst_case
+      figures = shape
+
+      expect(figures.caption).to eq("White 2,591 · 20% · Blue 2,591 · 20% · Black 2,591 · 20% · Red 2,595 · 20% · Green 2,591 · 20%")
+      expect(figures.caption_items).to eq([ "White 2,591 · 20% ·", "Blue 2,591 · 20% ·", "Black 2,591 · 20% ·", "Red 2,595 · 20% ·", "Green 2,591 · 20%" ])
+      expect(figures.caption_items.join(" ")).to eq(figures.caption)
+    end
+
+    it "has one item with no trailing separator for one colour, and none for no colour (AC-4.1)", :aggregate_failures do
+      own(card(cost: "{1}"))
+      expect(shape.caption_items).to eq([])
+      expect(shape.caption).to be_nil
+
+      own(card(cost: "{R}"), 3)
+      expect(shape.caption_items).to eq([ "Red 3 · 100%" ])
+    end
+  end
+```
+
+**Step 2: run it and watch it fail.** `bin/rspec spec/models/mtg/deck_shape_spec.rb -e "short_count" -e "caption_items"` → `NoMethodError: undefined method 'short_count'` and `'caption_items'`. (The existing `caption` examples are untouched and green.)
+
+**Step 3: implement** in `app/models/mtg/deck_shape.rb`. Add the constants after `COLOURS` (after line 122):
+
+```ruby
+  # Counts below this are written in full; from it a count in a slot of fixed width takes a unit, and each unit is this many of the one
+  # before (spec 077 decision 5). The abbreviation is uniform, not width-driven: the server cannot know the width, and a count that
+  # fits at a 360px viewport still overflows at 320px.
+  SHORT_COUNT_STEP = 1_000
+  # The units of an abbreviated count: thousands, millions, billions.
+  SHORT_COUNT_UNITS = %w[k m b].freeze
+```
+
+Replace `#caption` (lines 138-142: from the comment `# "White 2 · 25% · Blue 4 · 50%": every colour present with its count and percentage.` through the `end` after `.join(" · ").presence`) with:
+
+```ruby
+  # "White 2 · 25% · Blue 4 · 50%": every colour present with its count and percentage, exact (the abbreviation is for a slot of
+  # fixed width and this is prose). Built from {#caption_items}.
+  # @return [String, nil] nil when there is no coloured symbol
+  def caption = caption_items.join(" ").presence
+
+  # The caption as one string per colour, "White 2 · 25% ·" and, last, "Blue 4 · 50%": every item but the last ends in the separator, so a
+  # view that sets each item as an unbreakable unit lets a line break only between items and never before a "·" (spec 077 decision 4,
+  # #320). Joined with one space they are {#caption}, character for character.
+  # @return [Array<String>] empty when there is no coloured symbol
+  def caption_items
+    items = symbols.map { |s| "#{COLOURS[s.colour].capitalize} #{delimited(s.count)} · #{s.percent}%" }
+    items.each_with_index.map { |item, index| index == items.size - 1 ? item : "#{item} ·" }
+  end
+
+  # A count written for a slot of fixed width, the curve's column count and a mana segment's label (spec 077 decisions 1, 2 and 5, #320).
+  # Below 1,000 it is the number as it is. From there it takes a unit: one decimal under 10 of the unit with a trailing ".0" dropped
+  # ("1.4k", "2k"), whole from 10 ("12k", "123k"), rounded half up on exact integers (1,350 is "1.4k", 1,349 "1.3k"). A value that
+  # rounds to 1,000 of its unit moves to the next unit (999,500 is "1m"), and one that rounds to 10.0 is written whole (9,999 is "10k"),
+  # so nothing is longer than four characters through "999m". It is not `number_to_human`, which keeps two significant digits (123,456
+  # would read "120k"). The exact number stays in every `aria-label`, the legend, the caption and the types.
+  #
+  # @example
+  #   short_count(999)     # => "999"
+  #   short_count(1_357)   # => "1.4k"
+  #   short_count(9_999)   # => "10k"
+  #   short_count(999_500) # => "1m"
+  # @param number [Integer] a count, 0 or more
+  # @return [String]
+  def short_count(number)
+    return delimited(number) if number < SHORT_COUNT_STEP
+
+    value = Rational(number, SHORT_COUNT_STEP)
+    unit = 0
+    while value.round >= SHORT_COUNT_STEP && unit < SHORT_COUNT_UNITS.size - 1
+      value /= SHORT_COUNT_STEP
+      unit += 1
+    end
+    tenths = (value * 10).round
+    body = if value < 10 && tenths < 100
+      (tenths % 10).zero? ? (tenths / 10).to_s : format("%d.%d", *tenths.divmod(10))
+    else
+      value.round.to_s
+    end
+    "#{body}#{SHORT_COUNT_UNITS[unit]}"
+  end
+```
+
+(`{#caption_items}` and `{#caption}` are YARD links to methods of the same class and resolve; if `bin/yard doc` warns, put them in backticks.)
+
+**Step 4: run** `bin/rspec spec/models/mtg/deck_shape_spec.rb` → all green (the 069 `caption` examples included), then `bin/rubocop -a app/models/mtg/deck_shape.rb spec/models/mtg/deck_shape_spec.rb`.
+
+**Step 5: commit** (two calls): `git add app/models/mtg/deck_shape.rb spec/models/mtg/deck_shape_spec.rb`, then `git commit -m "feat(deck-shape): short_count and caption_items on the figures (#320)"` with the trailer.
+
+### Task 2: the request and system examples, red
+
+**Step 1: the request examples.** Append inside `RSpec.describe "Deck shape panel", type: :request` in `spec/requests/deck_shape_panel_spec.rb`, before the final `end`:
+
+```ruby
+  # Spec 077: the worst case's numbers in a slot of fixed width are abbreviated, and everything that can carry the exact number does.
+  describe "counts in a slot of fixed width (spec 077, #320)" do
+    # Spec 069's worst case: a 6+ column of 1,357 (123 owned, 1,234 proxy), five single cards and Bolt in column 1, 20 lands, and a
+    # five-colour mana bar of about 2,591 symbols a colour (red 2,595: Bolt's {R} times 4).
+    before do
+      own(card(name: "Worst", type_line: "Creature — Elf", cost: "{2}{W}{U}{B}{R}{G}", mana_value: 7.0), 123)
+      proxy(card(name: "Gleemax", type_line: "Creature — Elf", cost: "{W/U}{U/B}{B/R}{R/G}{G/W}", mana_value: 1_000_000.0), 1_234)
+      own(card(name: "Bolt", cost: "{R}", mana_value: 1.0), 4)
+      5.times { own(card(type_line: "Sorcery", cost: "{1}", mana_value: 1.0)) }
+      own(card(name: "Island", type_line: "Basic Land — Island", cost: "", mana_value: 0.0), 20)
+      get deck_path(deck)
+    end
+
+    it "abbreviates a column's count from 1,000 and keeps the exact numbers in the label and the legend (AC-1.2)", :aggregate_failures do
+      expect(html.css("aside .c-curve__count").map(&:text)).to eq(%w[0 9 0 0 0 0 1.4k])
+      expect(html.at_css("aside .c-curve [role=img]")["aria-label"]).to include("6+, 1,357 cards (123 owned, 1,234 proxy)")
+      expect(html.css("aside .c-curve__legend li").map(&:text)).to eq([ "Owned 132", "Proxy 1,234" ])
+    end
+
+    it "abbreviates a segment's label and keeps the bar's label and the caption exact (AC-2.1)", :aggregate_failures do
+      expect(html.css("aside .c-manabar__seg span").map(&:text)).to eq([ "W 2.6k", "U 2.6k", "B 2.6k", "R 2.6k", "G 2.6k" ])
+      expect(html.at_css("aside .c-manabar")["aria-label"]).to eq("Coloured mana symbols in Main: white 2,591, blue 2,591, black 2,591, red 2,595, green 2,591")
+      expect(html.at_css("aside .c-manacap").text).to eq("White 2,591 · 20% · Blue 2,591 · 20% · Black 2,591 · 20% · Red 2,595 · 20% · Green 2,591 · 20%")
+    end
+
+    it "sets each caption item as a `c-manacap__item` span and the paragraph's text is the caption string (AC-4.2)", :aggregate_failures do
+      items = html.css("aside .c-manacap .c-manacap__item")
+
+      expect(items.map(&:text)).to eq([ "White 2,591 · 20% ·", "Blue 2,591 · 20% ·", "Black 2,591 · 20% ·", "Red 2,595 · 20% ·", "Green 2,591 · 20%" ])
+      expect(html.at_css("aside .c-manacap").text).to eq(items.map(&:text).join(" "))
+    end
+
+    it "keeps the types' counts and suffix exact (the rows have room)" do
+      expect(html.css("aside .c-types li").first.css("span").map(&:text)).to eq([ "Creatures", "1,357", "1,234 proxy" ])
+    end
+  end
+```
+
+**Step 2: the system examples.** Append a nested `describe` inside `RSpec.describe "Deck shape panel", type: :system do` in `spec/system/deck_shape_panel_spec.rb`, before that block's final `end` (spec FR-3 names this file; its top-level `before` already builds spec 069's worst-case fixture on `deck`: a 6+ column of 1,357, five mana segments of about 2,591 (red 2,595), a five-colour caption, and its `set_theme`, `card(name:, type_line:, cost:, mana_value:)` and `create_lot_in_zone` helpers are reused; nothing in the file above changes):
+
+```ruby
+  # Spec 077 (#320): the panel at extreme counts at a 320px, 345px and 360px viewport (a real 320px phone, a 360px browser frame, a
+  # 360px phone), a 366px frame (where the old caption leads a line with a dot), a 640px frame (the fold again) and the 320px aside at
+  # 1280px, in both themes. The viewport frames are sized from the measured scrollbar (the viewport plus `innerWidth - clientWidth`: 15px
+  # on a Mac, about 12px on the Linux runner) and read against `documentElement.clientWidth` (lessons of 2026-11-11 on phone widths and
+  # measurement traps); the 366px and 640px frames are fixed frame widths by the spec's words, not viewports. Each geometry example has a
+  # failing variant: the old text or the old rule injected into the page.
+  # The file's memoized helpers (user, account, deck) plus the two lets below are at the cop's limit of five: another deck an example needs is
+  # a local (`plain = create(:deck, …)`, as AC-3.2 does), not another `let`.
+  describe "extreme counts (spec 077, #320)" do
+    let(:rows_deck) { create(:deck, account:, name: "Rows") }
+    let(:wide_deck) { create(:deck, account:, name: "Wide rows") }
+
+    def plain_card(name:, type_line: "Creature — Elf") = card(name:, type_line:, cost: "{1}", mana_value: 2.0)
+    def own_in(in_deck, entry, quantity) = create_lot_in_zone(in_deck, "main", entry:, quantity:)
+    def proxy_in(in_deck, entry, quantity) = create(:deck_proxy, zone: in_deck.zone("main"), account:, entry:, quantity:)
+
+    # AC-3.1's rows: "Creatures" 1,357 with "1,234 proxy", "Instants" 9,999 with "9,999 proxy", "Artifacts" 3 and "Lands" 20 with none.
+    def build_rows
+      own_in(rows_deck, plain_card(name: "Rows elf"), 123)
+      proxy_in(rows_deck, plain_card(name: "Rows elf proxy"), 1_234)
+      proxy_in(rows_deck, plain_card(name: "Rows bolt", type_line: "Instant"), 9_999)
+      own_in(rows_deck, plain_card(name: "Rows ring", type_line: "Artifact"), 3)
+      own_in(rows_deck, card(name: "Rows island", type_line: "Basic Land — Island", cost: "", mana_value: 0.0), 20)
+    end
+
+    # AC-3.1b's rows: "Planeswalkers" 12,345 with "12,345 proxy" (a proxy takes at most 9,999, so two) and "Lands" 20, a second row so that
+    # "the counts end at one x" can fail.
+    def build_wide_row
+      proxy_in(wide_deck, plain_card(name: "Wide jace", type_line: "Legendary Planeswalker — Jace"), 9_999)
+      proxy_in(wide_deck, plain_card(name: "Wide ajani", type_line: "Legendary Planeswalker — Ajani"), 2_346)
+      own_in(wide_deck, card(name: "Wide island", type_line: "Basic Land — Island", cost: "", mana_value: 0.0), 20)
+    end
+
+    def widths = [ 320, 345, 360, :frame366, :frame640, :aside ]
+    def fold_panel = "details.c-shapefold .c-shape"
+    def open_fold = page.execute_script('document.querySelector("details.c-shapefold").open = true')
+    def inject_css(rules) = page.execute_script("const style = document.createElement('style'); style.textContent = #{rules.to_json}; document.head.append(style)")
+
+    # A frame whose viewport is `viewport` px wide with the fold open: the viewport plus the scrollbar the frame draws (measured, then
+    # the frame is reopened if it was not 15px). Returns the viewport it got, for the example to assert.
+    def open_fold_in_frame(path, viewport)
+      width = viewport + 15
+      2.times do
+        open_in_narrow_frame(path, width:, ready: "details.c-shapefold")
+        scrollbar = within_narrow_frame do
+          open_fold
+          page.evaluate_script("window.innerWidth - document.documentElement.clientWidth")
+        end
+        break if viewport + scrollbar == width
+
+        width = viewport + scrollbar
+      end
+      within_narrow_frame { page.evaluate_script("document.documentElement.clientWidth") }
+    end
+
+    # Opens the deck page as `width` says (a viewport in px, a fixed frame, or :aside at 1280px) and yields the panel's selector, inside
+    # the frame where there is one.
+    def at_width(width, path)
+      case width
+      when Integer
+        expect(open_fold_in_frame(path, width)).to eq(width)
+        within_narrow_frame { yield fold_panel }
+      when :frame366, :frame640
+        open_in_narrow_frame(path, width: { frame366: 366, frame640: 640 }.fetch(width), ready: "details.c-shapefold")
+        within_narrow_frame do
+          open_fold
+          yield fold_panel
+        end
+      when :aside
+        visit path
+        page.driver.browser.manage.window.resize_to(1280, 900)
+        expect(page).to have_css("aside.c-deck__shape")
+        yield "aside .c-shape"
+      end
+    end
+
+    # What the panel measures: per curve column its room (the column's width less its count's scrollWidth), what is drawn past the
+    # padding box, per visible mana label its line count, its text width, its segment's content width and whether the text's box is inside
+    # its segment's box, the Card types rows' right edges, and per caption item its line count and the first character of every caption line.
+    def measure(selector)
+      page.evaluate_script(<<~JS)
+        (() => { const panel = document.querySelector(#{selector.to_json}), edge = panel.getBoundingClientRect().right - parseFloat(getComputedStyle(panel).paddingRight)
+          const textRange = (el) => { const r = document.createRange(); r.selectNodeContents(el); return r }
+          const lines = (el) => new Set([...textRange(el).getClientRects()].map((q) => Math.round(q.top))).size
+          const cols = [...panel.querySelectorAll(".c-curve__col")], rows = [...panel.querySelectorAll(".c-types li")], list = panel.querySelector(".c-types")
+          const labels = [...panel.querySelectorAll(".c-manabar__seg span")].filter((s) => s.getClientRects().length)
+          const starts = new Map(), caption = panel.querySelector("p.c-manacap"), walker = document.createTreeWalker(caption, NodeFilter.SHOW_TEXT)
+          for (let node = walker.nextNode(); node; node = walker.nextNode()) { for (let i = 0; i < node.length; i++) { const r = document.createRange(); r.setStart(node, i); r.setEnd(node, i + 1)
+            const box = r.getClientRects()[0]; if (box && node.data[i] !== " " && !starts.has(Math.round(box.top))) starts.set(Math.round(box.top), node.data[i]) } }
+          return { viewport: document.documentElement.clientWidth,
+            scrolls: panel.scrollWidth > panel.clientWidth,
+            beyond: [...panel.querySelectorAll("*")].filter((el) => el.getClientRects().length && el.getBoundingClientRect().right > edge + 0.5).length,
+            clipped: [...panel.querySelectorAll(".c-curve__count, .c-types li > *, .c-curve__legend li, .c-curve__avg, .c-manacap")].filter((el) => el.scrollWidth > el.clientWidth + 1).length,
+            countRoom: cols.length ? Math.min(...cols.map((col) => col.getBoundingClientRect().width - col.querySelector(".c-curve__count").scrollWidth)) : 0,
+            labels: labels.map((s) => { const box = textRange(s).getBoundingClientRect(), seg = s.parentElement.getBoundingClientRect()
+              return { lines: lines(s), text: box.width, room: s.parentElement.clientWidth, inside: box.left >= seg.left - 0.5 && box.right <= seg.right + 0.5 } }),
+            rights: rows.map((li) => li.children[1].getBoundingClientRect().right),
+            suffixGap: rows.map((li) => Math.abs(li.getBoundingClientRect().right - li.children[2].getBoundingClientRect().right)),
+            suffixWidth: rows.map((li) => li.children[2].getBoundingClientRect().width),
+            rowOver: rows.filter((li) => li.getBoundingClientRect().right > list.getBoundingClientRect().right + 0.5 || li.scrollWidth > li.clientWidth + 1).length,
+            cut: rows.flatMap((li) => [...li.children]).filter((el) => el.scrollWidth > el.clientWidth + 1).length,
+            itemLines: [...panel.querySelectorAll(".c-manacap__item")].map(lines),
+            lineStarts: [...starts.values()] } })()
+      JS
+    end
+
+    # `clientWidth` is a whole number, so a 46.38px segment reads 46 and a label has 0.44px to spare at a 320px viewport (the bound is 2px).
+    def counts_fit?(found) = found["countRoom"] >= 0 && !found["scrolls"] && found["beyond"].zero? && found["clipped"].zero?
+    def labels_fit?(found) = found["labels"].any? && found["labels"].all? { |label| label["lines"] == 1 && label["inside"] && label["text"] <= label["room"] - 2 }
+    def counts_aligned?(found) = (found["rights"].max - found["rights"].min) <= 0.5
+    def types_tidy?(found) = counts_aligned?(found) && found["suffixGap"].all? { it <= 0.5 } && found["rowOver"].zero? && found["cut"].zero?
+    def wide_row_tidy?(found) = found["rowOver"].zero? && found["cut"].zero? && counts_aligned?(found) && found["rights"].size == 2
+    def floor_held?(found) = counts_aligned?(found) && found["suffixWidth"].all? { it >= 71.5 }
+    def caption_whole?(found) = found["itemLines"].any? && found["itemLines"].all?(1) && found["lineStarts"].none?("·")
+
+    # Measures in each theme (layout cannot depend on it, so there is no wait for the 0.12s colour transition here; the second reading is a
+    # guard) and expects the block to hold.
+    def expect_in_both_themes(width, selector)
+      %w[light dark].each do |theme|
+        set_theme(theme)
+        found = measure(selector)
+        expect(yield(found)).to be(true), "at #{width} in #{theme}: #{found.inspect}"
+      end
+    end
+
+    it "keeps every curve count inside its column and the panel inside its box at every width, in both themes (AC-1.3)", :aggregate_failures do
+      widths.each { |width| at_width(width, deck_path(deck)) { |panel| expect_in_both_themes(width, panel) { counts_fit?(it) } } }
+    end
+
+    it "can fail: the full count in the 6+ column overflows its column at a 320px and a 345px viewport (AC-1.3)", :aggregate_failures do
+      [ 320, 345 ].each do |viewport|
+        at_width(viewport, deck_path(deck)) do |panel|
+          page.execute_script('document.querySelector("details.c-shapefold .c-curve__col:last-child .c-curve__count").textContent = "1,357"')
+          expect(counts_fit?(measure(panel))).to be(false), "at #{viewport}"
+        end
+      end
+    end
+
+    it "keeps every mana label on one line inside its segment at every width, in both themes (AC-2.2)", :aggregate_failures do
+      widths.each { |width| at_width(width, deck_path(deck)) { |panel| expect_in_both_themes(width, panel) { labels_fit?(it) } } }
+    end
+
+    it "can fail: a label written in full wraps inside its segment at a 320px viewport (AC-2.2)" do
+      at_width(320, deck_path(deck)) do |panel|
+        page.execute_script('document.querySelectorAll("details.c-shapefold .c-manabar__seg span").forEach((span) => { span.textContent = span.textContent.replace("2.6k", "2,591") })')
+        expect(labels_fit?(measure(panel))).to be(false)
+      end
+    end
+
+    it "ends every Card types count at one x with the suffix at the row's edge, nothing cut off, at every width (AC-3.1)", :aggregate_failures do
+      build_rows
+      widths.each { |width| at_width(width, deck_path(rows_deck)) { |panel| expect_in_both_themes(width, panel) { types_tidy?(it) } } }
+    end
+
+    it "can fail: flex rows with a 4.5rem suffix put the 9,999 proxy row's count left of the others' (AC-3.1)" do
+      build_rows
+      at_width(320, deck_path(rows_deck)) do |panel|
+        inject_css(".c-types{display:block !important} .c-types li{display:flex !important;gap:var(--space-2)} .c-types__n{margin-left:auto !important}")
+        expect(types_tidy?(measure(panel))).to be(false)
+      end
+    end
+
+    it "keeps each empty suffix cell at its 4.5rem floor and the counts at one x when no row has a proxy (AC-3.2)", :aggregate_failures do
+      plain = create(:deck, account:, name: "Plain")
+      own_in(plain, plain_card(name: "Plain elf"), 5)
+      own_in(plain, plain_card(name: "Plain ring", type_line: "Artifact"), 3)
+      [ 360, :aside ].each { |width| at_width(width, deck_path(plain)) { |panel| expect_in_both_themes(width, panel) { floor_held?(it) } } }
+    end
+
+    it "can fail: without the floor the empty suffix cells collapse (AC-3.2)" do
+      plain = create(:deck, account:, name: "Plain")
+      own_in(plain, plain_card(name: "Plain elf"), 5)
+      at_width(360, deck_path(plain)) do |panel|
+        inject_css(".c-types__proxy{min-width:0 !important}")
+        expect(floor_held?(measure(panel))).to be(false)
+      end
+    end
+
+    it "keeps a five-digit row inside the list with the counts at one x at a 320px viewport, the name wrapping (AC-3.1b)" do
+      build_wide_row
+      at_width(320, deck_path(wide_deck)) { |panel| expect_in_both_themes(320, panel) { wide_row_tidy?(it) } }
+    end
+
+    it "can fail: without the name cell's wrap rule the name spills over its cell (AC-3.1b)" do
+      build_wide_row
+      at_width(320, deck_path(wide_deck)) do |panel|
+        inject_css(".c-types li > :first-child{overflow-wrap:normal !important}")
+        expect(wide_row_tidy?(measure(panel))).to be(false)
+      end
+    end
+
+    it "keeps every caption item on one line and no caption line starting with a dot at every width (AC-4.3)", :aggregate_failures do
+      widths.each { |width| at_width(width, deck_path(deck)) { |panel| expect_in_both_themes(width, panel) { caption_whole?(it) } } }
+    end
+
+    it "can fail: without the nowrap rule an item splits at a 345px and a 360px viewport and a line starts with a dot in a 366px frame (AC-4.3)", :aggregate_failures do
+      [ 345, 360, :frame366 ].each do |width|
+        at_width(width, deck_path(deck)) do |panel|
+          inject_css(".c-manacap__item{white-space:normal !important}")
+          found = measure(panel)
+          expect(caption_whole?(found)).to be(false), "at #{width}: #{found.inspect}"
+          expect(found["lineStarts"]).to include("·") if width == :frame366
+        end
+      end
+    end
+  end
+```
+
+**Step 3: run them and watch them fail.** `bin/rspec spec/requests/deck_shape_panel_spec.rb -e "spec 077"` → the four examples fail (the column count reads "1,357", the label "W 2,591", no `c-manacap__item`). `bin/rspec spec/system/deck_shape_panel_spec.rb -e "extreme counts"` → expected red: AC-1.3 (320 and 345: room −5.43 and −1.85), AC-2.2 (320 and 345: the label wraps or doesn't fit; at 345 the label is 50.8px in a 51.4px segment, over the 2px bound), AC-3.1 and AC-3.1b (the old flex rows misalign at every width; the five-digit row's name spills), AC-4.3 (no `c-manacap__item` yet, so `itemLines` is empty and `caption_whole?` is false). The "can fail" examples and the AC-3.2 floor example are already green (they inject the old text or rule, or guard what exists today), except the AC-4.3 variant: it is green for the wrong reason now (no item exists to split), so after Task 3 it must be re-run and must still pass, at all three widths, with the leading dot found at the 366px frame. Record the run's failures in the commit body. If an example is green here that the spec says should be red, stop and read it: it can't fail.
+
+**Step 4: the stale comment.** In `spec/system/deck_shape_panel_spec.rb`, replace the comment above the AC-5.3 example (lines 129-134, from "# A phone's 360px viewport" to "...and a text element cut off at its own box.") with:
+
+```ruby
+  # A phone's 360px viewport has no scrollbar and the frame keeps a classic one, so the frame is 360px plus the measured scrollbar
+  # (375px with the Mac's 15px). The chain is the viewport's 360px less the body's 16px (8px margin a side, a 344px body), the main's
+  # 32px padding (a 312px container), the details' 2px border (310px) and the panel's 32px padding (a 278px inside). A frame with the
+  # scrollbar inside a 360px width is 15px narrower (263px inside), where the fixture's count in full ("1,357") overflowed its column by
+  # about 1.85px; spec 077 (#320) abbreviates it ("1.4k"), and the "extreme counts" examples below measure that frame and a 320px
+  # viewport. Each check can fail: the panel scrolling sideways, an element drawn past the padding box's right edge, and a text element
+  # cut off at its own box.
+```
+
+**Step 5: commit** (two calls): `git add spec/requests/deck_shape_panel_spec.rb spec/system/deck_shape_panel_spec.rb`, then `git commit -m "test(deck-shape): extreme-count examples, red (#320)"` (body: which examples fail and why).
+
+---
+
+## Phase 2: The partial and the CSS
+
+**Implements:** FR-1, FR-2 | **Satisfies:** AC-1.2, AC-1.3, AC-2.1, AC-2.2, AC-3.1, AC-3.1b, AC-3.3, AC-4.2, AC-4.3, AC-6.1 | **Files:** `app/views/mtg/deck_shapes/_parts.html.erb`, `app/assets/stylesheets/collector/additions.css`
+
+### Task 3: the partial
+
+**Step 1: edit `app/views/mtg/deck_shapes/_parts.html.erb`.** Three changes, nothing else:
+
+Line 7, the column's count: `<%= number_with_delimiter(column.total) %>` becomes `<%= figures.short_count(column.total) %>`.
+
+Line 39, the segment's label: `<span><%= segment.colour %> <%= number_with_delimiter(segment.count) %></span>` becomes `<span><%= segment.colour %> <%= figures.short_count(segment.count) %></span>`.
+
+Line 42, the caption: `<p class="c-manacap"><%= figures.caption %></p>` becomes
+
+```erb
+    <p class="c-manacap"><%= safe_join(figures.caption_items.map { |item| tag.span(item, class: "c-manacap__item") }, " ") %></p>
+```
+
+The legend (lines 17-18), the types' counts and suffix (line 28) and the `aria-label`s stay `number_with_delimiter` / exact. Update the partial's header comment (line 1) to: `<%# The three parts of the deck shape panel for Magic (spec 069, DeckShape.md): a part of zero has no element. A curve column's count and a mana segment's label are in slots of fixed width, so they are abbreviated from 1,000 (spec 077, MTG::DeckShape#short_count); every other number is exact. %>`.
+
+**Step 2: run** `bin/rspec spec/requests/deck_shape_panel_spec.rb spec/models/mtg/deck_shape_spec.rb` → green. `bin/rspec spec/system/deck_shape_panel_spec.rb -e "extreme counts"` → AC-1.3 and AC-2.2 and both of their variants green; AC-3.1, AC-3.1b and AC-4.3 still red (no CSS). The AC-4.3 variant is now green for the right reason: an item exists, splits at 345px and 360px and a line starts with "·" at the 366px frame.
+
+**Step 3: commit** (two calls): `git add app/views/mtg/deck_shapes/_parts.html.erb`, then `git commit -m "feat(deck-shape): abbreviate the curve count and the segment label, caption as items (#320)"`.
+
+### Task 4: the CSS
+
+**Step 1: the card types block.** In `app/assets/stylesheets/collector/additions.css`, in the block headed `/* ---------- DeckShape card types (spec 060)`, replace the four rules `.c-types`, `.c-types li`, `.c-types__n`, `.c-types__proxy` (the lines after the heading, through the `/* reserved even when empty… */` comment) with:
+
+```css
+/* Spec 077 (#320): the list is a grid of three columns and every row takes them as its own (subgrid), so the count column is as
+   wide as the widest count and the suffix column as wide as the widest suffix, and every count ends at one x whatever a suffix
+   says. The suffix cell keeps its 4.5rem floor so a list of short suffixes looks as it did; the name cell may wrap, even mid-word,
+   so a five-digit count and suffix at a 320px viewport never push a row past the list. */
+.c-types { display:grid; grid-template-columns:minmax(0, 1fr) auto auto; column-gap:var(--space-2); margin:0; padding:0; list-style:none; border-top:1px solid var(--line); }
+.c-types li { display:grid; grid-template-columns:subgrid; grid-column:1 / -1; align-items:baseline; min-height:32px; padding:var(--space-1) 0; border-bottom:1px solid var(--line); font:400 14px/20px var(--font-sans); }
+.c-types li > :first-child { min-width:0; overflow-wrap:anywhere; }
+.c-types__n { margin-left:auto; font:400 14px/20px var(--font-mono); } /* in a grid cell an auto margin right-aligns it, as in the old flex row */
+.c-types__proxy { min-width:4.5rem; text-align:right; font:400 12px/16px var(--font-mono); color:var(--ink-muted); } /* reserved even when empty, so a list with short suffixes keeps its look */
+```
+
+**Step 2: the mana bar block.** After the line `.c-manacap { margin:0; font:400 13px/18px var(--font-sans); color:var(--ink-muted); }` add:
+
+```css
+.c-manacap__item { white-space:nowrap; } /* spec 077 (#320): a caption line breaks only between colours, so a name stays with its count and no line starts with a dot */
+```
+
+**Step 3: run** `bin/rspec spec/system/deck_shape_panel_spec.rb spec/system/design_system_deck_components_spec.rb spec/design_system_files_spec.rb spec/requests/deck_shape_panel_spec.rb spec/requests/query_counts_spec.rb spec/models/mtg/deck_shape_spec.rb spec/deck_core_spec.rb` → all green. If AC-2.2 fails at the 320px viewport by under 1px on a label (the thin 0.44px room, spec Notes), read the measured `text` and `room` in the failure message; do not move the bound: ask the supervisor (a spec update).
+
+**Step 4: commit** (two calls): `git add app/assets/stylesheets/collector/additions.css`, then `git commit -m "feat(deck-shape): card types rows share columns, caption items stay whole (#320)"`.
+
+---
+
+## Phase 3: The documents
+
+**Implements:** FR-4 | **Satisfies:** AC-5.1, AC-5.2, AC-5.3 | **Files:** `docs/design-system/components/DeckShape.md`, `docs/design-system/previews/{DeckShape,DeckPage,DeckPagePhone}.html`, `docs/architecture/collection-and-lots.md`, `docs/specs/README.md`
+
+### Task 5: the component document, the previews and the area doc
+
+**Step 1: `DeckShape.md`.** Make these edits (each a unique string):
+
+- In the markup example, the caption line `<p class="c-manacap">Blue 44 · 73% · Red 16 · 27%</p></div>` becomes `<p class="c-manacap"><span class="c-manacap__item">Blue 44 · 73% ·</span> <span class="c-manacap__item">Red 16 · 27%</span></p></div>`.
+- The curve bullet "The count above a column is owned plus proxy, in mono." gains: ` From 1,000 it is abbreviated, one decimal under 10,000 and whole from there ("1.4k", "12k", "1.3m"), half up, never longer than four characters, so it fits a 30px column at a 320px viewport (spec 077 decision 1, 5; \`MTG::DeckShape#short_count\`). The exact column count is in the chart's \`aria-label\` only; the legend shows the curve's totals. Everything else is exact and delimited: the legend, the types' counts and suffixes, the caption, the zone heads and every \`aria-label\`.`
+- The Card types bullet "A type with no proxies keeps an empty `c-types__proxy`…" is replaced by: `- The list is a grid of three columns (\`minmax(0, 1fr) auto auto\`, a \`space-2\` column gap) and each row is a \`subgrid\` of it, so the count column is as wide as the widest count and the suffix column as wide as the widest suffix: every count ends at one x whatever a suffix says ("1,234 proxy" is wider than the old 4.5rem minimum). A type with no proxies keeps an empty \`c-types__proxy\`, which keeps a 4.5rem floor so a list with short suffixes looks as it did. The name cell has \`min-width:0\` and \`overflow-wrap:anywhere\`: with a five-digit count and suffix at a 320px viewport ("Planeswalkers 12,345 12,345 proxy") the name wraps inside its cell, even mid-word, and no row passes the list (spec 077 decision 3). Counts and suffix are exact.`
+- The mana bar bullet "Each segment fills… The segment's label is its letter and count ("U 44"), in mono." gains: ` From 1,000 the count is abbreviated ("W 2.6k": six characters, about 43.5px, which fits one line in the narrowest five-equal-segment bar at a 320px viewport); the caption and the \`aria-label\` keep the exact count (spec 077 decision 2).`
+- The mana bar heading's class list "(`c-manabar`, `c-manacap`, spec 060 decisions 3 and 4, AC-2.6, AC-2.7)" gains `c-manacap__item`.
+- The caption bullet ends with: ` The caption is a run of \`c-manacap__item\` spans, "Blue 44 · 73% ·" and, last, "Red 16 · 27%", separated by one space, each \`white-space:nowrap\`: a line breaks only between colours, so no line begins with "·" and a name never parts from its count; the paragraph's text is the same string as before (spec 077 decision 4; \`MTG::DeckShape#caption_items\`).`
+- The **Width** paragraph is replaced by:
+
+  `**Width.** The panel works in 238px, the space inside its border and padding at a 320px viewport (a phone's 320px less 16px body margin, 32px page padding, 2px border and 32px panel padding), and every count, the legend, the types and the mana labels fit with no horizontal overflow at the worst-case counts of spec 069's fixture (spec 077 AC-1.3, AC-2.2, AC-3.1). It was drawn for 279px (spec 060 AC-2.4); the measured interiors are 238px (a 320px viewport), 263px (a 360px browser frame, which keeps a 15px scrollbar a phone doesn't), 278px (a 360px viewport) and 286px (the 320px aside; beside the zones the panel is 320px wide, spec 060 decision 6). **Known limits:** (1) a mana segment narrower than its label: a label of four characters ("U 44") needs about 14% of the bar, five ("W 999") about 17% and six ("W 2.6k") about 20% at a 320px viewport (measured to the half percent on a 236px bar), and the rule hides a label only under 10% (spec 060 decision 4), so a segment between 10% and those shares wraps its label and cuts it off; the caption carries the count; it exists at ordinary counts too, and is [#NNN](https://github.com/plainprogrammer/Collector/issues/NNN) (spec 077 decision 2, ruled option A). (2) A Card types row with a five-digit count and suffix at a 320px viewport wraps its name, even mid-word; counts of six digits are not specified.` (Replace `NNN` with the number the coordinator gives for the issue before this task's commit; if it is not known yet, write the sentence with the title instead and fix it in the supervisor's follow-up commit.)
+
+**Step 2: the previews.** In `docs/design-system/previews/DeckShape.html`, `DeckPage.html` and `DeckPagePhone.html`, replace every `<p class="c-manacap">Blue 44 · 73% · Red 16 · 27%</p>` (use `replace_all`) with `<p class="c-manacap"><span class="c-manacap__item">Blue 44 · 73% ·</span> <span class="c-manacap__item">Red 16 · 27%</span></p>`; in `DeckShape.html` the five-colour caption `White 3 · 5% · Blue 20 · 33% · Black 4 · 7% · Red 18 · 30% · Green 15 · 25%` becomes five spans (`White 3 · 5% ·`, `Blue 20 · 33% ·`, `Black 4 · 7% ·`, `Red 18 · 30% ·`, last `Green 15 · 25%`, a space between); in `DeckShape.html` line 45 the comment "(313px wide, 279px inside)" becomes "(on a 360px phone the fold is 312px wide, 278px inside)"; the wrappers' own widths (lines 44, 46–47, 76–77) stay: AC-5.2 asks for the comment only. Open each preview's markup only through these strings; no other line changes.
+
+**Step 3: the area doc, `docs/architecture/collection-and-lots.md`.**
+
+- Line 137, the panel's paragraph: after "…(decisions 9, 14, AC-4.6–AC-4.8)." insert: ` The visible number in a curve column's count and in a mana segment's label is abbreviated from 1,000 (\`MTG::DeckShape#short_count\`: 1,357 is "1.4k", 12,345 "12k", 1,250,000 "1.3m", half up on integers, at most four characters), because both sit in a slot of fixed width; every other number and every \`aria-label\` stays exact and delimited; the caption is a run of unbreakable items (\`caption_items\`, one per colour, so a line never starts with "·"); and the Card types list is a grid whose rows are subgrids, so the counts line up whatever the suffix (spec 077 decisions 1 to 5, #320; it supersedes, in part, spec 060 decision 2, AC-2.2, AC-2.4, AC-2.5 and AC-2.6 (the written-in-full numbers, the fixed 4.5rem minimum, the 279px) and spec 069 decision 14 for those two numbers). A mana segment narrower than its label (labels of four, five and six characters need about 14%, 17% and 20% of the bar at a 320px viewport; the 10% rule is spec 060 decision 4) is a known limit, [#NNN](https://github.com/plainprogrammer/Collector/issues/NNN) (spec 077 decision 2, ruled option A).`
+- Named constants (after the "narrow threshold" row, line 246): add two rows: `| \`MTG::DeckShape::SHORT_COUNT_STEP\` | 1,000 | \`app/models/mtg/deck_shape.rb\` | A count in a slot of fixed width is written in full below it and takes a unit from it, each unit 1,000 of the one before | spec 077 decision 5; #320 |` and `| \`MTG::DeckShape::SHORT_COUNT_UNITS\` | \`k\`, \`m\`, \`b\` | \`app/models/mtg/deck_shape.rb\` | The units of an abbreviated count | spec 077 decision 5; #320 |`.
+- Line 312, the Issues bullet: replace "Open on the panel: its layout at extreme counts, [#320](…) (a four-digit 6+ count in a 360px browser frame, crowded mana bar labels, a long proxy suffix in the Card types rows: spec 060's CSS, spec 069 verification "Findings")." with "Open on the panel: its mana labels in segments narrower than their label, [#NNN](https://github.com/plainprogrammer/Collector/issues/NNN) (spec 077 decision 2; the other limits of [#320](https://github.com/plainprogrammer/Collector/issues/320) are fixed by spec 077)."
+- Sources and the Distillation log are written by `distill-feature` at the end (Task 6), not here.
+
+**Step 3b: the design-system README.** In `docs/design-system/README.md` line 138 (the "App additions (spec 060…)" list), where `DeckShape`'s classes are named, add `c-manacap__item`.
+
+**Step 4: the index.** In `docs/specs/README.md` change the 077 row's Status from `Draft` to `Approved` and its Plan cell from `none` to `present`.
+
+**Step 5: verify by reading** each statement against the code (`short_count` examples, the CSS lines, the previews' markup); run `bin/yard doc` (0 warnings). **Commit** (two calls): `git add docs/design-system docs/architecture/collection-and-lots.md docs/specs/README.md`, then `git commit -m "docs: deck shape panel at extreme counts (spec 077, #320)"`.
+
+---
+
+## Phase 4: Evidence, review, finish (the supervisor's)
+
+### Task 6
+
+1. **The follow-up issue's number** (the coordinator files it from the Gate B message): replace every `#NNN` and `/issues/NNN` in `DeckShape.md` and `collection-and-lots.md`, cite it in `spec.md`'s Open Questions (a 0.3.2 changelog row: "issue number cited"), and commit. The measurements are in `verification.md` (from the section above).
+2. **Screenshots**: the throwaway `spec/system/zz_077_screenshots_spec.rb` with `PHASE=after`; "before" shots exist (captured from the old code at the plan's start); look at every shot in both themes; commit them under `docs/specs/077-deck-shape-extreme-counts/screenshots/` as JPEGs (`magick in.png -resize 1000x -quality 70 out.jpg`), delete the throwaways, list them in `verification.md`.
+3. **`verification.md`**: AC table (each AC and the example title that proves it), the supersession grep, the mutation checks (each "can fail" example proved by the injection), deviations, screenshots.
+4. **Mode B review** (read-only Opus, `sdd-review`), recorded under `## Implementation review` with the verdict first.
+5. **`session-wrap`, `distill-feature`** (area docs' Sources and Distillation log, the index row to Frozen with "retirement owed"), commit; ask for a CI slot, `bin/ci`, push, draft PR "Closes #320", CI green, `gh pr ready`, `gh pr edit --add-assignee @me`, `worker_done`.
+
+## Self-review
+
+- **Spec coverage:** FR-1 (Tasks 1, 3), FR-2 (Task 4), FR-3 (Tasks 1, 2: the model and request examples and the nested `describe` in `spec/system/deck_shape_panel_spec.rb`), FR-4 (Tasks 5, 6); AC-1.1 (T1), AC-1.2/2.1/4.2 (T2 request), AC-1.3/2.2/3.1/3.1b/4.3 (T2 system, each with a failing variant; AC-4.3's at 345px, 360px and the 366px frame), AC-3.2 (T2 system example and its variant), AC-3.3 (T4 Step 3), AC-4.1 (T1), AC-5.1 to AC-5.3 (T5), AC-6.1 (the diff check in T6), AC-6.2 (T2 Step 4, T4 Step 3, T6 `bin/ci`). **Gap check:** AC-2.3 (narrow segments keep the class) is spec 069's AC-4.6 example, `spec/requests/deck_shape_panel_spec.rb` "marks a segment under 10%…", kept green.
+- **Placeholder scan:** the only deliberate placeholder is the follow-up issue's number `#NNN`, which the coordinator supplies after Gate B (Task 5 says where it goes and what to do if it is unknown).
+- **Type consistency:** `short_count(number)` and `caption_items` are named identically in the model, the partial, the examples and the documents; `SHORT_COUNT_STEP` and `SHORT_COUNT_UNITS` in the model and the area doc rows.
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/after-phone320-dark.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-phone320-dark.jpg
new file mode 100644
index 00000000..74205701
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-phone320-dark.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/after-phone320-light.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-phone320-light.jpg
new file mode 100644
index 00000000..6844ed90
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-phone320-light.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/after-phone360-dark.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-phone360-dark.jpg
new file mode 100644
index 00000000..aa928da3
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-phone360-dark.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/after-phone360-light.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-phone360-light.jpg
new file mode 100644
index 00000000..ac9ed923
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-phone360-light.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/after-w640-dark.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-w640-dark.jpg
new file mode 100644
index 00000000..469ca001
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-w640-dark.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/after-w640-light.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-w640-light.jpg
new file mode 100644
index 00000000..9d2dd8b0
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-w640-light.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/after-wide1280-dark.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-wide1280-dark.jpg
new file mode 100644
index 00000000..2e9d596d
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-wide1280-dark.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/after-wide1280-light.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-wide1280-light.jpg
new file mode 100644
index 00000000..2256177e
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/after-wide1280-light.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/before-phone320-dark.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-phone320-dark.jpg
new file mode 100644
index 00000000..9d38a9cb
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-phone320-dark.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/before-phone320-light.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-phone320-light.jpg
new file mode 100644
index 00000000..e14805ce
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-phone320-light.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/before-phone360-dark.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-phone360-dark.jpg
new file mode 100644
index 00000000..9934920b
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-phone360-dark.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/before-phone360-light.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-phone360-light.jpg
new file mode 100644
index 00000000..cc61411e
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-phone360-light.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/before-w640-dark.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-w640-dark.jpg
new file mode 100644
index 00000000..8266a747
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-w640-dark.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/before-w640-light.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-w640-light.jpg
new file mode 100644
index 00000000..2c88a432
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-w640-light.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/before-wide1280-dark.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-wide1280-dark.jpg
new file mode 100644
index 00000000..552bc7b4
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-wide1280-dark.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/screenshots/before-wide1280-light.jpg b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-wide1280-light.jpg
new file mode 100644
index 00000000..1be7b1da
Binary files /dev/null and b/docs/specs/077-deck-shape-extreme-counts/screenshots/before-wide1280-light.jpg differ
diff --git a/docs/specs/077-deck-shape-extreme-counts/spec.md b/docs/specs/077-deck-shape-extreme-counts/spec.md
new file mode 100644
index 00000000..6d989fd0
--- /dev/null
+++ b/docs/specs/077-deck-shape-extreme-counts/spec.md
@@ -0,0 +1,230 @@
+# Feature 077: The Deck Shape Panel Stays Tidy at Extreme Counts
+
+**Status:** Approved by the coordinator under the maintainer's delegation (2026-11-11, Gate A) at 0.3.0, after two spec review rounds; decision 2 ruled A by the maintainer
+**Version:** 0.3.3
+**Created:** 2026-11-11
+**Last Updated:** 2026-11-11
+**Depends on:** [spec 060](../060-deck-components/spec.md) (the `DeckShape` component and its CSS), [spec 069](../069-deck-shape-panel/spec.md) (the panel's figures; its verification "Findings" 1 to 4 are this spec's input), both merged
+**Branch:** `077-deck-shape-extreme-counts`
+**Issue:** [#320](https://github.com/plainprogrammer/Collector/issues/320) (design-system lane, filed by the maintainer's ruling from spec 069's implementation review)
+**Frozen:** the current rules are in [`collection-and-lots.md`](../../architecture/collection-and-lots.md) (whole spec); the panel's markup is `DeckShape` in the [design system](../../design-system/README.md); see the [specs index](../README.md)
+
+---
+
+## Changelog
+
+| Version | Date | Change |
+|---------|------|--------|
+| 0.3.3 | 2026-11-11 | PATCH (`sdd-spec-update`, spec only, approved by the coordinator under the maintainer's delegation; behaviour unchanged): AC-2.2's label bound is the segment's `clientWidth` with the label on one line and inside its segment, no longer "less 2px", because the Linux runner's Firefox draws the six-character label 44.70px wide where the Mac draws 43.56px (a 46px content width; margins 1.30px and 2.44px); CI failed the 2px gap by 0.7px while the label fits. The shares of the label band (14%, 17%, 20%) were measured on a Mac with the 2px gap, so they state where a label has a pixel a side |
+| 0.3.2 | 2026-11-11 | PATCH (a pointer only): the follow-up issue for the mana-label band is #330 |
+| 0.3.1 | 2026-11-11 | PATCH (`sdd-spec-update`, spec only; found while writing the plan, before it was approved): the measurement table's caption column is corrected (at a 320px viewport the old caption breaks between colours, not inside one; at 345px it splits "Green / 2,591"), so AC-4.3's failing variant runs at 345px and 360px, not 320px; the sweep found the leading dot (16 of 72 frame widths), so AC-4.3 gains a 366px frame and its no-leading-dot clause can fail; the label shares are measured, 14%, 17% and 20%, not the arithmetic 13.5%, 16.5% and 19.5%; AC-3.1b adds "not cut off" and a variant for the name cell's wrap rule; the 0.84px room is 0.44px (an integer `clientWidth`). No behaviour changes |
+| 0.3.0 | 2026-11-11 | Spec review round 2 (Opus, read-only; REVISE: 0 blocking, 1 major, 4 minor, 3 nit), all taken: the fixture's red is 2,595 (AC-4.1's strings); the band's shares are 13.5%, 16.5% and 19.5% in both places; the Non-Goal and Goal 1 agree on five-digit Card types rows; AC-5.1 and AC-5.3 require the band and its follow-up issue to be recorded and the 060 AC-2.4 row is qualified; option B is costed; the method is named (`short_count`); AC-2.2's thin room is flagged for the CI runner; the preview comment's chain |
+| 0.2.0 | 2026-11-11 | Spec review round 1 (Opus, read-only; REVISE: 0 blocking, 4 major, 10 minor, 7 nit), all taken: AC-1.2's numbers follow spec 069's real fixture (column 1 reads "9", "Owned 132"); AC-2.2 measures against the segment's `clientWidth` less 2px and the label's width is the synthesized bold's (about 43.5px); Goal 1 is narrowed to what is fixed and the unfixed mana band is measured and named, with a narrow rule that follows the label offered as option B for Gate A; decision 1 states its trade-off (a column's exact count is in the `aria-label` only); decision 3's alternative corrected, the name cell's `overflow-wrap`, FR-2's CSS points and AC-3.1b for five-digit counts; AC-4.3's unfailable clause explained; AC-6.2's comment update; the `allow_browser` floor; spec 074's possible padding change; the preview's stale comment; nits |
+| 0.1.0 | 2026-11-11 | First draft from issue #320, spec 069's verification "Findings", `DeckShape.md`, the lessons of 2026-11-11 on phone widths, measurement traps and worst-case states, and a measurement of the four limits in headless Firefox (below). Routing: the issue asks for a ruling per item, and the rulings change rules of two shipped specs (060, 069), so it is a spec, not a direct fix |
+
+## Problem Statement
+
+Spec 069 drew the deck shape panel on spec 060's CSS and found four layout limits that show only at extreme counts, a deck or a collection with more than a thousand cards of one kind. It left them in its `verification.md` and `DeckShape.md` for a ruling. The worst-case fixture is spec 069's own: a Main zone with 123 owned and 1,234 proxy copies of one creature (a 6+ column of 1,357), a hybrid-cost proxy that puts about 2,591 coloured symbols on each of the five colours, and 20 lands.
+
+> **Terms.** A *viewport* is the width a phone reports (360px). A *frame* is the test's iframe, which keeps a 15px classic scrollbar (so a frame is the viewport plus the scrollbar; lesson `docs/lessons/2026-11-11-a-phone-width-criterion-needs-a-measured-chain.md`). The *interior* is the space inside the open panel's border and padding: viewport − 16 (body) − 32 (page padding) − 2 (border) − 32 (panel padding). *Abbreviated* means a count written with a unit: 1,357 is "1.4k".
+
+### Measurement (2026-11-11, headless Firefox, the fixture above, the fold open)
+
+| Where | Interior | 6+ count's room in its column | A mana label ("W 2,591") against its segment | Type counts' right edges (Creatures, Lands) | Caption lines |
+|---|---|---|---|---|---|
+| A 360px viewport (375px frame) | 278px | +0.28px | 50.8px of 54.4px, touching the borders | 231.8, 239 (7.2px apart) | breaks inside "Black / 2,591" and ends a line with "·" |
+| A 360px frame (a 345px viewport) | 263px | **−1.85px** | 50.8px of 51.4px | 216.8, 224 | breaks inside "Green / 2,591" |
+| A 320px viewport (335px frame; a 320px phone, the narrowest the page is meant for) | 238px | **−5.43px** | **wraps onto two lines** (36px) | 191.8, 199 | breaks between colours (no item splits) |
+| The aside at 1280px (a 320px panel) | 286px | +1.43px | 50.8px of 56px (5.2px to spare) | 1120.8, 1128 | breaks inside "Black / 2,591" |
+
+So the four limits are real, and two are worse than the issue says: the curve's count also overflows at a 320px viewport by 5.43px, and the mana labels wrap onto two lines there. The caption's leading dot is real: the two spaces around each "·" allow a break on either side of it, and a sweep of the old caption (a fold open in frames from 255px to 468px wide, in steps of 3px, on the unchanged code) found a line starting with "·" at 16 of 72 widths, among them a 321px frame (an interior of 224px), a 363px to 369px frame (266px to 272px, a phone's width) and a 402px frame (305px).
+
+## Goals
+
+- At any curve count, and at the worst-case fixture's mana bar and Card types rows (counts up to 9,999 per type, a suffix of "9,999 proxy"), on a 320px to 360px viewport and in the 320px aside, no count is cut off or overflows its column, and the panel does not scroll sideways; with five-digit type counts the rows stay inside the list (the name may wrap).
+- Each of the four limits has a ruling (decisions 1 to 4): all four are recommended as fixes, none as an accepted limit, and the narrower cases a fix does not reach are named, with their measured band (Error Scenarios).
+- What the abbreviation hides from a sighted reader is stated, not implied: a curve column's exact count is in the chart's `aria-label` only (decision 1); the caption, the legend's totals, the types' counts and the zone heads stay exact.
+
+## Non-Goals
+
+- The 10% rule that hides a segment's label (spec 060 decision 4, spec 069 decision 9) and the segments' proportional widths, unless the maintainer takes option B of decision 2 at Gate A.
+- Counts outside the panel (the zone heads, the Owned, Proxy and Total cells, the lists): they have room and stay exact.
+- A real-phone-only claim: the 360px frame's 263px interior is a test frame (#318), but the 320px viewport's 238px interior is a real phone, so the fixes are measured at both.
+- Any view of another page, any change to what the panel counts, to the export, to a query, a model other than the panel's figures, or to the tokens.
+- A type's name staying on one line when its count and suffix have five digits or more at a 320px viewport: the rows stay inside the list and the name may wrap (AC-3.1b); counts of six digits or more in a Card types row are not specified.
+
+## Decisions
+
+Each limit of the issue, with the options, my recommendation (the first option) and the reason. Gate A rules on these.
+
+### 1. The curve's count (limit 1): abbreviate it from 1,000
+
+- **Fix, recommended: abbreviate.** The count above a column reads "1.4k" from 1,000 on ("999" stays "999"). Four characters at most, so it fits a 238px interior's 30.57px column (about 29px of text). The rule is uniform, not width-driven, because the server cannot know the width and a count that fits at 360px still overflows at 320px.
+- **The trade-off:** a sighted reader loses the exact count of a column of 1,000 or more. It stays in the chart's `aria-label` (the whole column and its owned and proxy split), and the column's cards are in the zone tables; the legend shows the curve's totals, not a column's split. The curve is a shape, and "1.4k" is the shape's number.
+- Alternatives: let the count shrink or wrap (a 12px font is already the smallest the design system sets, and a wrapped number above a bar reads as two numbers); overflow into the gap (two adjacent four-digit columns would touch, and the last column spills into the panel's padding); accept the limit (it is real on a 320px phone, and costs nothing to fix).
+
+### 2. The mana bar's labels (limit 2): abbreviate the label's count the same way
+
+- **Fix, recommended: abbreviate.** A segment's label reads "W 2.6k" from 1,000 on: six characters, about 43.5px (the label is 600 weight and the page loads the mono face at 400 only, so the browser draws a synthesized bold of about 7.26px a character), which fits the 46.4px of the narrowest five-equal-segment case (a 238px interior) on one line. Spec 060's Charts rule keeps its letter and count; the caption beside the bar keeps the exact count and the percentage ("White 2,591 · 20%"), and the bar's `aria-label` the exact count.
+- **What this leaves (measured, 2026-11-11):** a segment narrower than its label. A label fits its segment (one line, inside its content width less 2px) from a share of the bar of 20% for six characters ("W 2.6k"), 17% for five ("W 999") and 14% for four ("U 44"), measured to the half percent on a 236px bar (a 238px interior); so segments between 10% (the narrow rule's line) and those shares have no room for their label, at ordinary counts too (a 12% segment of "U 44" at a 320px viewport). The labels wrap onto two lines and are cut off at the bar's 30px (the "before" shots of a deck whose red segment is large show it). The fix for the issue's case, five segments of about equal four-digit counts, does not reach it.
+- Alternatives:
+  - **B, a narrow rule that follows the label (the reviewer's suggestion; a larger change):** the server marks a segment narrow when its share of the bar is below `(label width + 2px) ÷ 232px`, using the label's length, so the label hides whenever it cannot fit at the narrowest interior, and the caption (always complete) carries the count. It closes the whole band above and is Ruby only. It also changes spec 060 decision 4 (ruled by the maintainer on 2026-11-10 as "a segment under 10% drops the label") and what the page shows at ordinary counts: a 12% segment loses the label it has on a 360px phone, where it fits. **Gate A question:** recommend A, with the band recorded in `DeckShape.md` and offered as its own issue; take B instead if the maintainer wants the band closed here.
+  - B's costs, for the maintainer: it replaces AC-2.3, adds 060 decision 4 and 069 decision 9 and AC-4.6 to the supersession lists and the index, drops the band's Error Scenarios row and widens Goal 1; and "Ruby only" means the model hard-codes a font measurement (about 7.26px a character for the synthesized bold, and the 232px bar), which drifts silently if the font, its weight or the page padding changes (spec 074), and it hides labels in the aside and on a 360px phone where they fit.
+  - Hide the label when the count has four digits (the Charts rule wants a count on every segment); lower the narrow rule's 10% (the same ruling as B, blunter); accept (the 320px phone wraps the label onto two lines).
+- Not changed under A: the 10% rule and the narrow class.
+
+### 3. A type row's count (limit 3): the rows share their columns
+
+- **Fix, recommended: a grid.** The list is a grid of three columns (name, count, proxy suffix) and each row takes the list's columns as its own (`subgrid`), so the count column is as wide as the widest count and the suffix column as wide as the widest suffix, whatever it is, and every row's count ends at the same x. The suffix cell keeps its 4.5rem minimum, so a list whose suffixes are short looks as before. The counts and the suffix stay exact ("1,234 proxy"). The name cell takes `min-width:0` and `overflow-wrap:anywhere`, so when a name, a count and a suffix together are wider than the row (five-digit counts at a 320px viewport) the name wraps inside its cell, even mid-word, and the row never runs past the list.
+- Alternatives: widen the minimum to 6rem (it fixes "1,234 proxy" and fails at "12,345 proxy", and costs 24px of every row's name column on a phone); abbreviate the suffix ("1.2k proxy" is 10 characters, 72.0px, exactly the 4.5rem minimum, so it would also fix the alignment with no CSS; it is not recommended because a type's exact proxy count is shown nowhere else and there is room for it); accept.
+
+### 4. The mana caption (limit 4): each colour is one unbreakable item
+
+- **Fix, recommended: items.** The caption is a run of items, "Blue 44 · 73% ·" and the last "Red 16 · 27%", each unbreakable (`white-space:nowrap`), separated by one ordinary space. A line can break only between items, so no line begins with "·", a colour's name never parts from its count or percentage, and a line may end with a trailing "·" (typographically harmless, and what the page does today). The caption's text is unchanged, character for character, so the strings of specs 060 and 069 and of the examples that read it still hold; the figures' `caption` method keeps returning the whole string.
+- Alternatives: a non-breaking space before every "·" (changes the string, which Nokogiri and the model examples compare exactly); a list of items with the dots drawn by CSS (the text the page reads changes, and a screen reader reads a caption of pseudo-elements unevenly); accept.
+
+### 5. One abbreviation, written once, for the two places that need it
+
+A count in a slot of fixed width, the curve's column and a mana segment, is abbreviated from 1,000; a count anywhere else is exact and delimited. The abbreviation is the figures object's (`MTG::DeckShape`), the extension's, so the deck's core files, which name no collectible, are untouched.
+
+- **From 1,000:** under 10,000 it has one decimal with a trailing ".0" dropped ("1.4k", "2k"); from 10,000 it is whole ("12k", "123k"). Millions use "m" the same way ("1.2m", "12m"), billions "b". It rounds half up on integers, exactly (1,350 is "1.4k"; 1,349 is "1.3k"). A value whose rounding reaches 1,000 of its unit moves to the next unit (999,500 is "1m"; 999,499 stays "999k"), and one that rounds to 10.0 under the one-decimal rule is written whole (9,950 and 9,999 are "10k"), so no abbreviation is longer than four characters through "999m". It is not `ActiveSupport::NumberHelper.number_to_human`, which keeps two significant digits (123,456 would read "120k").
+- **Below 1,000** the number is written as it is ("0", "123", "999").
+- The `aria-label`s, the legend, the types' counts and suffixes, the caption and the zone heads are not abbreviated.
+
+## Supersedes
+
+| Document | What it says | What changes |
+|---|---|---|
+| [Spec 060](../060-deck-components/spec.md) AC-2.2 ("the count above the column reads owned plus proxy"), AC-2.6 ("a segment's label reads its letter and count") | the number is written in full | from 1,000 the visible number is abbreviated (decision 5); below 1,000, and in every `aria-label`, nothing changes |
+| Spec 060 decision 2, AC-2.5, and `DeckShape.md` "Card types" ("the element has a minimum width of 4.5rem so the counts of every row line up") | a fixed minimum aligns the counts | the rows share the list's columns; the 4.5rem minimum stays as a floor (decision 3) |
+| Spec 060 AC-2.4 and `DeckShape.md` "Width" (the known limit and its mention of #320) | the panel fits in 279px | it fits in 238px (a 320px viewport) and in 278px, at four-digit counts, except the mana-label band of decision 2 (option A); the limit sentence about the 6+ count is replaced by the measured interiors |
+| [Spec 069](../069-deck-shape-panel/spec.md) decision 14 ("counts … use `number_with_delimiter`") and its Error Scenarios row "A four-digit count in the 6+ column in a 360px browser frame" | the curve's count and the segment's label are delimited in full; the overflow is a known limit | those two numbers are abbreviated from 1,000 (decision 5) and the limit is gone |
+| Spec 069 `verification.md` "Findings" 1 to 4 | recorded, not fixed, for the maintainer | fixed (this spec's verification names each) |
+
+Recorded in the specs index, in both directions. Specs 060 and 069 are frozen and are not edited.
+
+Existing examples this changes (a grep, not memory; the command is repeated in `verification.md`): `git grep -n -i -E "c-curve__count|c-manabar__seg|c-manacap|c-types|\.caption\b|countRoom|1,357|2,591|4\.5rem" -- spec ':!.kamal/secrets'`. Hits that change: `spec/system/deck_shape_panel_spec.rb` (the AC-5.3 example's comment says "a CSS question raised with the maintainer, not decided here", and its fixture's 1,357 now reads "1.4k"), `spec/requests/deck_shape_panel_spec.rb` (the segments' texts "U 6" and the caption lines hold; a new example for the items), `spec/models/mtg/deck_shape_spec.rb` (new examples for the abbreviation and the items; the `caption` examples stay). Hits left alone: `spec/system/design_system_deck_components_spec.rb` (builds its own markup with delimited-free numbers and reads the types' alignment, which the grid keeps) and the previews' text.
+
+## User Stories
+
+### Story 1: The curve's counts fit their columns
+
+**As a** collector with a very large deck on a phone
+**I want** the number above each curve column to fit it
+**So that** no number is cut off or runs into the next column
+
+- [ ] **AC-1.1** Given the abbreviation When it is given 0, 999, 1,000, 1,049, 1,050, 1,350, 1,357, 9,949, 9,950, 9,999, 10,000, 12,345, 123,456, 999,499, 999,500, 1,250,000 and 12,000,000 Then it returns "0", "999", "1k", "1k", "1.1k", "1.4k", "1.4k", "9.9k", "10k", "10k", "10k", "12k", "123k", "999k", "1m", "1.3m" and "12m", and no result is longer than four characters.
+- [ ] **AC-1.2** Given spec 069's system fixture (`spec/system/deck_shape_panel_spec.rb`: Worst at mana value 7, 123 owned; Gleemax, mana value 1,000,000, 1,234 proxy; Bolt 4 owned and five single cards at mana value 1; 20 lands) When the curve renders Then the 6+ column's count reads "1.4k", column 1 reads "9", the chart's `aria-label` reads "6+, 1,357 cards (123 owned, 1,234 proxy)" and the legend reads "Owned 132" and "Proxy 1,234".
+- [ ] **AC-1.3** Given the same fixture in the open fold at a 320px viewport (interior 238px), a 345px viewport (a 360px frame; 263px), a 360px viewport (278px) and a 640px frame (a 545px container: the fold again), and in the aside at 1280px (286px), in both themes When each count's room in its column (the column's width less the count's `scrollWidth`) is read Then it is at least 0 for all seven columns, no text is cut off at its own box, no element is drawn past the panel's padding box's right edge and the panel does not scroll sideways. A variant that sets the 6+ count back to "1,357" fails at the 320px and 345px viewports (so the example can fail).
+
+### Story 2: The mana bar's labels have room
+
+**As a** collector with a deck of many symbols
+**I want** each segment's label to sit inside its segment, on one line
+**So that** the bar reads as a bar, not as crowded text
+
+- [ ] **AC-2.1** Given a segment of 2,591 symbols When the bar renders Then its label reads "W 2.6k" (its letter, a space and the abbreviation), a segment of 44 reads "U 44", the caption reads "White 2,591 · 20% · …" and the bar's `aria-label` reads "white 2,591, …", all exact.
+- [ ] **AC-2.2** Given the fixture's five segments of about 2,591 at the viewports and the aside of AC-1.3, in both themes When each label's text is measured with a `Range` Then it is on one line (its height is at most one line, 16px), its width is at most its segment's `clientWidth` (the content box, the divider's border excluded; the first segment has no divider) and the text's box is inside its segment's box (so no label overlaps another). A variant that writes the count in full fails at the 320px viewport (the label wraps onto two lines). Measured margins at a 320px viewport (a 46px content width): 2.44px on a Mac (a 43.56px label) and 1.30px on the Linux runner (44.70px).
+- [ ] **AC-2.3** Given a segment under 10% When the bar renders Then it still carries `c-manabar__seg--narrow` and no visible label (spec 069 AC-4.6 unchanged).
+
+### Story 3: The types' counts line up whatever the suffix
+
+**As a** collector
+**I want** every type's count to end at the same place
+**So that** I can compare them down the list
+
+- [ ] **AC-3.1** Given Card types rows "Creatures" 1,357 with "1,234 proxy", "Instants" 9,999 with "9,999 proxy", "Artifacts" 3 and "Lands" 20 with no suffix When the list renders at the viewports and the aside of AC-1.3 Then every row's count ends at the same x (within 0.5px), every suffix ends at the row's right edge, and no row's name, count or suffix is cut off (`scrollWidth` at most `clientWidth`) or runs past its row. A variant of the previous CSS (flex rows, a 4.5rem minimum) fails: the "Instants" and "Creatures" counts are about 7px left of the others'.
+- [ ] **AC-3.1b** Given a row "Planeswalkers" 12,345 with "12,345 proxy" at a 320px viewport (the name, the count and the suffix need about 247px of the 238px interior) When the list renders Then no row runs past the list's right edge the counts still end at one x, and no name, count or suffix is cut off; the name may wrap inside its cell, mid-word. A variant without the name cell's `overflow-wrap` rule fails (the name, about 95px, spills over its roughly 85px cell).
+- [ ] **AC-3.2** Given Card types rows none of which has a proxy When the list renders Then the counts line up and each row has its empty suffix cell of at least 4.5rem as before (nothing moves for an ordinary deck).
+- [ ] **AC-3.3** Given the same rows in the `design_system_deck_components_spec` markup (`c-types` with its three children per row) When it is measured Then it still passes: name at the row's left edge, suffix at the right edge, counts aligned.
+
+### Story 4: The caption never breaks inside an item or leads with a dot
+
+**As a** collector
+**I want** each colour's count and percentage to stay together
+**So that** the caption is not misread
+
+- [ ] **AC-4.1** Given the figures of spec 069's fixture (red is 2,595: Bolt's `{R}` adds 4 to the 2,591 of the other four) When `caption` is read Then it returns the string the code builds today ("White 2,591 · 20% · Blue 2,591 · 20% · Black 2,591 · 20% · Red 2,595 · 20% · Green 2,591 · 20%"), and `caption_items` returns five strings, "White 2,591 · 20% ·", "Blue 2,591 · 20% ·", "Black 2,591 · 20% ·", "Red 2,595 · 20% ·" and "Green 2,591 · 20%"; a deck with one coloured symbol has one item with no trailing "·"; a deck with no coloured symbol has `caption_items` of `[]` and `caption` nil (as before).
+- [ ] **AC-4.2** Given the rendered caption When its text is read Then it equals `caption` exactly (the items are separated by one space, so `have_text` and Nokogiri's `text` read the same string as before).
+- [ ] **AC-4.3** Given the fixture's caption at the viewports and the aside of AC-1.3, in both themes When each item's text is measured with a `Range` Then every item is on one line (a single client rect), and no line of the caption begins with "·". The widths also include a 366px frame (an interior of 269px on a Mac, 272px on the Linux runner, both inside the range where the old caption leads a line with "·"). A variant without the nowrap rule fails at the 345px and 360px viewports (an item splits: "Green / 2,591" and "Black / 2,591"; at a 320px viewport the old caption happens to break between colours, so it is not used) and at the 366px frame (a line starts with "·").
+
+### Story 5: The documents
+
+**As a** developer, or an agent, drawing the panel
+**I want** the component document and the area doc to say what is shown and what is exact
+**So that** the next count in a fixed slot follows the same rule
+
+- [ ] **AC-5.1** `docs/design-system/components/DeckShape.md` states decisions 1 to 5: the abbreviation and where it applies (and that the legend, the types, the caption, the zone heads and the `aria-label`s are exact), the grid and the 4.5rem floor, the caption's items and that its text is unchanged, the markup example's caption in items, the "Width" paragraph rewritten with the measured interiors (238px, 263px, 278px, 286px), the 6+ known limit and its #320 mention removed, and the unfixed mana-label band (decision 2, option A) and the five-digit Card types rows recorded as limits, with their shares (about 14%, 17% and 20% of the bar for labels of four, five and six characters) and the issue that offers the band. (Verified by reading.)
+- [ ] **AC-5.2** The three previews that draw the caption (`DeckShape.html`, `DeckPage.html`, `DeckPagePhone.html`) wrap each caption item in a `c-manacap__item`, so the previews match the markup, and `DeckShape.html`'s comment "313px wide, 279px inside" is corrected to the measured chain (the fold 312px wide, 278px inside). (Verified by reading, and by the previews' existing examples, if any, passing.)
+- [ ] **AC-5.3** `docs/architecture/collection-and-lots.md` states the abbreviation in the panel's paragraph (which numbers are abbreviated, from 1,000), adds a Named constants row for the 1,000 threshold, and its Issues bullet no longer lists #320 as open and, under option A, names the follow-up issue for the mana-label band; `docs/specs/README.md` has this spec's row with 060 and 069 in "Supersedes" and 077 in their "Superseded by". (Verified by reading.)
+
+### Story 6: Nothing else moves
+
+**As a** maintainer
+**I want** the change to be small and to touch only the panel
+**So that** it merges cleanly beside the deck-page work of spec 074
+
+- [ ] **AC-6.1** The branch's diff to `app/` is `app/models/mtg/deck_shape.rb`, `app/views/mtg/deck_shapes/_parts.html.erb` and `app/assets/stylesheets/collector/additions.css` only; the CSS change is inside the DeckShape blocks (the card types block and the mana bar block) and adds no colour, font family or spacing literal that the tokens could give (`spec/design_system_files_spec.rb` passes unchanged), and no other stylesheet changes. (Verified by the diff.)
+- [ ] **AC-6.2** `spec/deck_core_spec.rb` is unchanged and passes (the deck core files name no collectible), the page's statement counts are unchanged (`spec/requests/query_counts_spec.rb` passes unchanged) and spec 069's AC-5.1 to AC-5.3 examples pass with their assertions unchanged (the AC-5.3 example's comment, which calls the 360px frame's overflow "a CSS question raised with the maintainer, not decided here", is updated in the same commit: a stale comment is a bug).
+
+## Functional Requirements
+
+- **FR-1.** `MTG::DeckShape` gains the abbreviation as the instance method `short_count(number)` (decision 5; AC-1.1's model examples and the partial call it) and `caption_items` (decision 4); `caption` and every method that builds an `aria-label` are unchanged. The partial `mtg/deck_shapes/_parts` writes the column count and the segment's label through the abbreviation and the caption through the items.
+- **FR-2.** `additions.css` changes inside the DeckShape card types block (the list a grid of three columns, `minmax(0,1fr) auto auto`, with the column gap on the list; each row `grid-column:1/-1` on `subgrid` columns, its `min-height`, baseline alignment and borders as now; the count keeps `margin-left:auto` so it stays right-aligned in its column; the suffix cell keeps its 4.5rem minimum; the name cell `min-width:0; overflow-wrap:anywhere`) and the mana bar block (one rule, `.c-manacap__item { white-space:nowrap }`), each with a comment saying why, citing spec 077, and no value outside the tokens beyond the literals the blocks already use.
+- **FR-3.** Examples first, each red before its fix: the abbreviation and the items in `spec/models/mtg/deck_shape_spec.rb`; the rendered markup in `spec/requests/deck_shape_panel_spec.rb`; the layout in `spec/system/deck_shape_panel_spec.rb` (the viewports and the aside, both themes), sizing each frame from the measured scrollbar as `open_phone_frame` does and comparing against `documentElement.clientWidth`, not `innerWidth` (lesson on measurement traps).
+- **FR-4.** The documents of Story 5 and the screenshots of the non-functional requirements.
+
+## Non-Functional Requirements
+
+- **Compatibility:** no migration, no schema change, the export unchanged at version 4, upgrade-safe from the previous release.
+- **Hosted-door cost per account (north star rule 4):** none; a few string operations per page view and a stylesheet.
+- **Accessibility:** unchanged names and labels: the `aria-label`s keep the exact numbers, the abbreviated visible text is in `aria-hidden` columns and an `img`-role bar, so a screen reader reads the exact figures once; the caption stays a paragraph with the same text.
+- **Performance:** the abbreviation is integer arithmetic on at most 12 numbers a copy (the panel renders twice); the statement counts are unchanged.
+- **Visual change:** yes, only at counts of 1,000 or more and in the Card types list's alignment (none at ordinary counts). Before and after screenshots of the worst-case panel at 375px (a 360px viewport), 640px and 1280px, in both themes, in the pull request: "before" is the old rules injected as a `<style>` (`.claude/memory/pr-screenshots-workflow.md`), and the old abbreviation-free text is drawn by the same fixture with the figures' abbreviation stubbed to the delimited number.
+
+## Error Scenarios
+
+| Scenario | Behaviour |
+|---|---|
+| A count of exactly 1,000, or 999 | "1k" and "999" (AC-1.1) |
+| A count that rounds up into the next unit (9,950, 999,500) | "10k" and "1m", never "10.0k" or "1000k" |
+| A column or segment of 0 | "0" (a curve column; a segment of zero has no element) |
+| A deck whose curve counts all fit (every count under 1,000) | Nothing changes: the numbers are written in full |
+| A column of 1,000 or more in the aside, where the number would fit | Still abbreviated: the rule is one rule for every width; the exact count is in the chart's `aria-label` (decision 1's trade-off) |
+| A type's count or suffix of 10,000 or more ("12,345 proxy") at a 320px viewport (a name such as "Planeswalkers", a five-digit count and a five-digit suffix need about 247px of the 238px interior) | The rows still share their columns, the counts line up and no row runs past the list; the name wraps inside its cell, even mid-word (AC-3.1b). Recorded in `DeckShape.md` |
+| A mana segment between 10% and about 14% (a label of four characters, "U 44"), 17% (five) or 20% (six, "W 2.6k") of the bar, whose label does not fit at a 320px viewport | Not changed under decision 2's A: the label wraps and is cut off; the caption carries the count. Recorded in `DeckShape.md` as a limit with its band, and offered as its own issue (or taken here as B) |
+| A caption with one colour | One item, no trailing "·" |
+| A caption item wider than the panel (not possible at the counts above: the longest item is well under 200px) | It would overflow like any unbreakable word; AC-4.3 fails first |
+| The browser lacks `subgrid` | Not reachable: `ApplicationController` allows modern browsers only (`allow_browser versions: :modern`: Safari 17.2, Chrome 120, Firefox 121 and later), and all have `subgrid` |
+
+## Open Questions
+
+None. Gate A (2026-11-11) approved decisions 1 to 5 as written, and the maintainer ruled decision 2 **A**: spec 060's 10% rule stays, and the mana-label band is recorded in `DeckShape.md` as a documented limit with the shares above and becomes its own issue, [#330](https://github.com/plainprogrammer/Collector/issues/330) (filed by the coordinator; cited in `DeckShape.md`, the area doc's Issues bullet and here). Option B is not taken.
+
+## Notes for the plan
+
+- The label band and the leading-dot sweep were measured before the plan (`verification.md`); the AC-1.3, AC-2.2, AC-3.1 and AC-4.3 examples and the model examples are written red first.
+- The interiors (238, 263, 278, 286px) are derived from the page's own chain (viewport, body margin, page padding, border, panel padding) and the examples assert the viewport, as `open_phone_frame` does. Spec 074 (#326, branch `074-deck-pages-phone-width`) may change the deck page's padding: if it merges first, re-measure the interiors and update the numbers here and in `DeckShape.md`.
+- AC-2.2's room at a 320px viewport is thin (`clientWidth` is a whole number, so a 46.38px segment reads 46 and the bound is 44px against a 43.56px label: 0.44px): the first green run showed the runner's mono face 1.1px wider (44.70px), which failed the 2px gap the spec had drawn from the Mac, and the gap was dropped in 0.3.3 (the criterion is about wrapping and overflow).
+- Each geometry example measures both themes in one frame (set `data-theme`, wait about 0.6s, measure again) so the cost is one frame per width; theme cannot change layout, so the second measurement is a guard, not a second example.
+- The frame sizes: a viewport of 320px needs a frame of 320 + the measured scrollbar (335px with the Mac's 15px, about 332px on the Linux runner), a 345px viewport a frame of 360px; size every frame from the measured scrollbar as `open_phone_frame` does, and assert the viewport. The aside is measured at a 1280px window (a 320px panel, 286px inside).
+- Each system example needs its failing variant (the lesson `an-example-that-cannot-fail-proves-nothing`): inject the old rule (`.c-types li { display:flex }` and the old `.c-types__proxy`) or remove the nowrap rule with a `<style>` and expect the example to fail; say in the example's comment which it is.
+- The caption's items are measured by `Range` client rects of each item's text (a single rect is one line); a line's first character is read from the character rects, as the throwaway measurement did.
+- `app/assets/stylesheets/collector/additions.css` is shared with the other workers of this run: keep the edit inside the two DeckShape blocks and don't reorder or reformat anything else. `spec/design_system_files_spec.rb` counts the strings "spec 019", "spec 021" and "spec 023" across the whole file; "spec 077" in the comments is fine.
+- `bin/rspec spec/system/design_system_deck_components_spec.rb` after the CSS edit (it builds its own `c-types` and `c-manabar` markup, and AC-3.3 keeps it green).
+- Run `bin/yard doc` for the new public methods' comments (a summary, `@param` and `@return`, and the "why" of the abbreviation's rounding).
+
+## Out of Scope (Future Considerations)
+
+- Abbreviating counts elsewhere on the deck page (zone heads, table cells), which have room.
+- A width-driven rule (container queries that choose an abbreviation or a label): the server-side rule is uniform and enough.
+- The curve's blank band when it is empty (spec 069 Findings 5), a separate matter.
+- Real-phone checks on a physical 320px device.
+
+## Documents that change
+
+- `app/models/mtg/deck_shape.rb`, `app/views/mtg/deck_shapes/_parts.html.erb`, `app/assets/stylesheets/collector/additions.css` (the card types block and the mana bar block); `spec/models/mtg/deck_shape_spec.rb`, `spec/requests/deck_shape_panel_spec.rb`, `spec/system/deck_shape_panel_spec.rb` (new and changed examples).
+- `docs/design-system/components/DeckShape.md`; `docs/design-system/previews/DeckShape.html`, `DeckPage.html`, `DeckPagePhone.html` (the caption's items).
+- `docs/architecture/collection-and-lots.md` (the panel's paragraph, a Named constants row, the Issues bullet, Sources and the Distillation log); `docs/specs/README.md` (this spec's row; 060's and 069's "Superseded by").
+- No ADR: the rulings are this spec's decisions; no new area doc: the design system has none and the panel's rules are in `collection-and-lots.md`.
+- Left as history, not edited: specs 060 and 069 and their `verification.md` files.
diff --git a/docs/specs/077-deck-shape-extreme-counts/verification.md b/docs/specs/077-deck-shape-extreme-counts/verification.md
new file mode 100644
index 00000000..683be3e9
--- /dev/null
+++ b/docs/specs/077-deck-shape-extreme-counts/verification.md
@@ -0,0 +1,84 @@
+# Verification: spec 077, the deck shape panel at extreme counts
+
+Spec 0.3.3, plan `plan.md` (commit `8bee6a51`). Issue [#320](https://github.com/plainprogrammer/Collector/issues/320); the follow-up for the mana-label band is [#330](https://github.com/plainprogrammer/Collector/issues/330).
+
+## Acceptance criteria
+
+Examples are in `spec/models/mtg/deck_shape_spec.rb` (model), `spec/requests/deck_shape_panel_spec.rb` (request) and the nested `describe "extreme counts (spec 077, #320)"` of `spec/system/deck_shape_panel_spec.rb` (system, headless Firefox).
+
+| AC | Proved by |
+|---|---|
+| AC-1.1 | model "writes 0 to 999 as they are and 1,000 up with one decimal under 10,000, whole from there, half up (AC-1.1)" (all 17 values, none longer than four characters) and "rounds 1,349 down to 1.3k and a count that reaches 1,000 of its unit moves to the next, 9,999,500 being 10m" |
+| AC-1.2 | request "abbreviates a column's count from 1,000 and keeps the exact numbers in the label and the legend (AC-1.2)" |
+| AC-1.3 | system "keeps every curve count inside its column and the panel inside its box at every width, in both themes (AC-1.3)" (320, 345, 360, a 366px and a 640px frame, the 1280px aside); can fail: "the full count in the 6+ column overflows its column at a 320px and a 345px viewport (AC-1.3)" |
+| AC-2.1 | request "abbreviates a segment's label and keeps the bar's label and the caption exact (AC-2.1)"; a segment of 44 reading "U 44" is covered by spec 069's existing request example ("U 6", "R 3": below 1,000 nothing changes) |
+| AC-2.2 | system "keeps every mana label on one line inside its segment at every width, in both themes (AC-2.2)"; can fail: "a label written in full wraps inside its segment at a 320px viewport (AC-2.2)" (it fails through the `lines == 1` clause: every label wraps onto two lines, text 36px, `inside` true, measured under the 0.3.3 bound) |
+| AC-2.3 | spec 069's request example "marks a segment under 10% of the symbols narrow in the page (AC-4.6)", unchanged and green |
+| AC-3.1 | system "ends every Card types count at one x with the suffix at the row's edge, nothing cut off, at every width (AC-3.1)"; can fail: "flex rows with a 4.5rem suffix put the 9,999 proxy row's count left of the others' (AC-3.1)" |
+| AC-3.1b | system "keeps a five-digit row inside the list with the counts at one x at a 320px viewport, the name wrapping (AC-3.1b)"; can fail: "without the name cell's wrap rule the name spills over its cell (AC-3.1b)" |
+| AC-3.2 | system "keeps each empty suffix cell at its 4.5rem floor and the counts at one x when no row has a proxy (AC-3.2)"; can fail: "without the floor the empty suffix cells collapse (AC-3.2)" |
+| AC-3.3 | `spec/system/design_system_deck_components_spec.rb`, "DeckShape card types (AC-2.5)", unchanged and green after the CSS |
+| AC-4.1 | model "gives one item per colour, each but the last ending in the separator, and `caption` is the items joined (AC-4.1)" and "has one item with no trailing separator for one colour, and none for no colour (AC-4.1)"; the 069 `caption` examples (lines 253, 293, 315) unchanged and green |
+| AC-4.2 | request "sets each caption item as a `c-manacap__item` span and the paragraph's text is the caption string (AC-4.2)" |
+| AC-4.3 | system "keeps every caption item on one line and no caption line starting with a dot at every width (AC-4.3)"; can fail: "without the nowrap rule an item splits at a 345px and a 360px viewport and a line starts with a dot in a 366px frame (AC-4.3)" |
+| AC-5.1 | `docs/design-system/components/DeckShape.md`, read against the code and the CSS (batch C's report and the supervisor's reading) |
+| AC-5.2 | the three previews' caption paragraphs and `DeckShape.html`'s width comment, read |
+| AC-5.3 | `docs/architecture/collection-and-lots.md` (the panel's paragraph, two Named constants rows, the Issues bullet) and `docs/specs/README.md` (the 077 row, 060's and 069's "Superseded by"), read |
+| AC-6.1 | `git diff --stat origin/main...HEAD -- app/` names `app/models/mtg/deck_shape.rb`, `app/views/mtg/deck_shapes/_parts.html.erb`, `app/assets/stylesheets/collector/additions.css` only; the CSS diff is inside the card types and mana bar blocks (10 insertions, 4 deletions) |
+| AC-6.2 | `spec/deck_core_spec.rb`, `spec/requests/query_counts_spec.rb` and spec 069's AC-5.1 to AC-5.3 examples green and unchanged (the AC-5.3 example's comment is rewritten, its assertions are not) |
+
+## The failing variants (each proved able to fail)
+
+Each "can fail" example injects the old text or the old rule into the page and expects its predicate to be false; all six pass, so each predicate can fail. Batch B also ran a scratch copy of the AC-3.1b variant with the measurement printed: with `overflow-wrap:normal` on the name cell at a 320px viewport it reads `rowOver 0, cut 1, rights [184.6, 184.6]`, so the predicate goes false through `cut` (a name spilling over its cell), not through count misalignment: the variant is the proof that the wrap rule is tested. The red run before the code (commit `9c1d80a2`): the request examples failed 3 of 4 (the fourth, "keeps the types' counts and suffix exact", guards what exists and can't be red; the plan's "the four examples fail" was imprecise), the system examples failed 5 of 12 groups (AC-1.3 at 320, 345 and the 366px frame; AC-2.2 at 320 and 345; AC-3.1; AC-3.1b on count misalignment; AC-4.3 for lack of items), the six variants and AC-3.2 green.
+
+## Measurements taken before the plan (2026-11-11, headless Firefox, the unchanged code)
+
+- **Four limits, spec 069's fixture** (a table is in the spec): the 6+ count's room in its column was +0.28px at a 360px viewport, −1.85px at a 360px frame, **−5.43px** at a 320px viewport, +1.43px in the aside; the mana labels wrapped onto two lines at 320px; the Card types counts of the "1,234 proxy" row sat 7.2px left of the others'; the caption broke inside "Black / 2,591" and ended lines with "·".
+- **The label band:** at a 320px viewport (a 236px bar) a mana label fits its segment from 14% of the bar for four characters ("U 44"), 17% for five ("W 999") and 20% for six ("W 2.6k"), on a half-percent grid (`#330` offers closing it).
+- **The leading dot:** a sweep of the old caption in frames of 255px to 468px (steps of 3px) found a line starting with "·" at 16 of 72 widths, among them frames of 285px, 321px, 363px, 366px, 369px, 402px and 438px; a 366px frame is an interior of 269px on a Mac and 272px on the Linux runner, both in the list. At a 320px viewport the old caption breaks between colours (no item split); at 345px and 360px it splits "Green / 2,591" and "Black / 2,591".
+
+## The supersession grep (plan, "The existing examples this changes")
+
+`git grep -n -i -E "c-curve__count|c-manabar__seg|c-manacap|c-types|\.caption\b|countRoom|1,357|2,591|4\.5rem" -- spec docs/design-system docs/architecture app ':!.kamal/secrets'`, read hit by hit; the dispositions are the plan's table. Edited: the AC-5.3 comment of `spec/system/deck_shape_panel_spec.rb` (stale), `DeckShape.md`, three previews, `docs/design-system/README.md` (the class list), the area doc. Kept: the `caption` examples of the model and request specs (the string is unchanged), `design_system_deck_components_spec.rb` (builds its own markup, run green after the CSS).
+
+## Deviations
+
+1. The request spec's fourth example is a guard and green before the code (plan Task 2 Step 3 said all four fail; recorded above).
+2. Spec 0.3.1 (a PATCH, spec only) was made while planning: the table's caption column corrected, AC-4.3's variant widths changed and a 366px frame added, the label shares measured (14%, 17%, 20%), AC-3.1b's "not cut off" and its variant (the spec's changelog row). The plan, reviewed twice, follows it.
+3. The follow-up issue #330 was filed by the coordinator (decision 2 ruled option A by the maintainer, 2026-11-11).
+4. A 0.3.2 row of the spec cites #330 (a pointer only).
+5. The spec's Notes say to wait about 0.6s after a theme switch; the geometry examples don't (`testing.md` forbids sleeps, and the second reading in the other theme is a guard, since layout cannot depend on the theme; the helper's comment says so).
+6. The non-functional requirements describe the "before" shots as made by injecting the old rules and stubbing the abbreviation; they were taken from the unchanged code at the start of the branch (the same fixture, run before any app change), which is the real "before", not an emulation.
+7. `docs/architecture/collection-and-lots.md`'s Sources and Distillation log are written by `distill-feature` at the end (the plan leaves them there), not by Task 5.
+8. **GitHub CI failed AC-2.2 on the Linux runner at a 320px viewport** (PR #331, run 43153405100, shard 2/4; the other thin check, the 366px frame's leading dot, passed). The six-character label measured 44.70px there against 43.56px on the Mac (the runner's mono face is wider), in a 46px content width: 1.30px of margin on the runner, 2.44px on the Mac, and the spec's bound (text at most `clientWidth` less 2px) was 44px. The label was on one line and inside its segment, so no code or CSS was wrong: the 2px gap was a number the spec had drawn from a Mac measurement. Ruled (the coordinator, under the maintainer's delegation): spec PATCH 0.3.3 drops the gap, so a label fits when it is on one line, its text is within the segment's `clientWidth` and its box is inside its segment's box; `labels_fit?` follows. The failing variant (the unabbreviated label at 320px) still fails under the new bound, through its `lines` clause alone (two lines per label); the `text` and `inside` clauses pass there. The band shares of the spec (14%, 17%, 20%) were measured on a Mac with the old 2px gap and stay as the shares at which a label has a pixel a side.
+
+## Screenshots
+
+In [`screenshots/`](screenshots/): factory data, placeholder card images; each panel in `-light` and `-dark`, "before" being the unchanged code, "after" this branch. The fixture is spec 069's worst case (a 6+ column of 1,357, five equal mana segments of about 2,591, 20 lands) plus a colourless Instant proxy of 9,999 that gives a 10,002 column and a "9,999 proxy" Card types row.
+
+- `before-phone360-*`, `after-phone360-*`: a 375px frame (a 360px viewport), the fold open.
+- `before-phone320-*`, `after-phone320-*`: a 335px frame (a 320px viewport).
+- `before-w640-*`, `after-w640-*`: a 640px frame (the fold again).
+- `before-wide1280-*`, `after-wide1280-*`: the 320px aside at 1280px.
+
+Every shot was looked at. Before: the 6+ count "1,357" crowds its column and the label "W 2,591" touches its borders (at 320px it wraps onto two lines and is cut off), the "Creatures" and "Instants" counts sit left of the "Artifacts" and "Lands" counts, and the caption breaks inside a colour. After: the counts read "10k" and "1.4k" over their columns, every label reads "W 2.6k" on one line inside its segment, every Card types count ends at one x, and the caption breaks only between colours; both themes read alike.
+
+## Implementation review
+
+SPEC-ALIGNED WITH MINOR FINDINGS (round 1, `sdd-review` Mode B, read-only Opus review of `46568b3d..86e54034` at spec 0.3.2, with the supervisor's fresh run beside it: 253 examples, 0 failures over the model, request and system deck-shape specs, `design_system_deck_components_spec`, `deck_page_spec`, `design_system_files_spec`, `deck_core_spec`, `query_counts_spec` and `deck_pages_spec`; rubocop clean on the four changed Ruby files; the comment ratchet "3 added or changed public objects, all with a comment")
+
+- **Checked, no finding:** `short_count` by hand (1,049, 1,050, 9,949, 9,950, 999,499, 999,500, 1,250,000, 9,999,500, 2,000,000,000), `caption` unchanged, the YARD comments, what the partial abbreviates and what stays exact, the CSS (only the two DeckShape blocks, tokens only), AC-6.1 (the app diff is exactly three files), the unchanged guards (`deck_core_spec`, `query_counts_spec`, specs 060 and 069), the documents against the code, 20 example titles cited here (each exists once), and the screenshot pairs.
+- **No blocking or major findings.** Minor findings and their resolution:
+  1. The area doc's Sources and Distillation log had no spec 077 entry: left to `distill-feature` (deviation 7), done in the distillation commit.
+  2. Three geometry predicates would pass if the panel were not drawn, and "all seven columns" was not counted: fixed in `430295a3` (`measure` returns the column count and the narrowest column, `counts_fit?` requires seven columns of positive width, `counts_aligned?` requires positive right edges).
+  3. Two "can fail" examples proved only that the predicate could be false: fixed in `430295a3` (the AC-3.1 variant expects the count edges to spread over 5px, the AC-4.3 variant at 345px and 360px expects an item on two lines).
+  - Nits: the version and unrecorded deviations (5, 6, 7 above, fixed here); the literal "U 44" of AC-2.1 had no example of its own (a request example "writes counts under 1,000 in full in a slot of fixed width (AC-2.1)" was added in `430295a3`); "never longer than four characters" overstated past 999m (the area doc and `DeckShape.md` now say "through 999m", `366ae43c`).
+- **Owed to CI, not defects:** two checks have little margin and were proved on a Mac only: AC-2.2's 0.44px at a 320px viewport, and the 366px frame's leading dot (an interior of 272px on the Linux runner, inside the measured range). If either fails on the runner, the spec is updated through `sdd-spec-update` before a bound moves.
+
+## Distillation (`distill-feature`, spec 027)
+
+- `docs/architecture/collection-and-lots.md`: the panel's paragraph (written with Task 5) and the two Named constants rows and the Issues bullet naming #330, plus, here, the Sources bullet for spec 077 (#320, #330) and a Distillation log row ("spec 077's pull request" until the pull request has a number). The design system has no area doc or log: `DeckShape.md` was written in Task 5.
+- `docs/specs/README.md`: the 077 row is Frozen, by that area doc, Plan "retirement owed"; the supersession of specs 060 and 069 is in both directions (written with the spec).
+- No ADR proposed: the rulings are the spec's decisions and decision 2 is the maintainer's (#330).
+- No `CLAUDE.md` change: it restates no rule the feature changed (its decks sentence names the panel and `Catalog::DeckShape`, a pointer).
+- The plan's retirement follows the merge (state 1).
diff --git a/docs/specs/README.md b/docs/specs/README.md
index e32635b3..76c6472b 100644
--- a/docs/specs/README.md
+++ b/docs/specs/README.md
@@ -75,7 +75,7 @@ A plan is deleted from the tree in a commit of its own after its feature's pull
 | [057 Design System: Deck Rules, Navigation and Pickers](057-deck-design-rules/spec.md) | Frozen, by [design system](../design-system/README.md) (the rules, previews and CSS), [catalog core](../architecture/catalog-core.md) and [collection and lots](../architecture/collection-and-lots.md) (decision 12's wording) | [Design system](../design-system/README.md) (navigation, colour and charts, the Decks vocabulary, grouped pickers, the card page's stats and "In decks", Locations); the area docs [catalog core](../architecture/catalog-core.md) and [collection and lots](../architecture/collection-and-lots.md) for the re-export wording only | retired in `20bb951812466842686dc4333a8bf504b22306ba`; `git show 20bb951812466842686dc4333a8bf504b22306ba^:docs/specs/057-deck-design-rules/plan.md` | 029 decision 8 (the header's links, Decks after Collection); 037 AC-6.4's 220px and its option list in part; 004 FR-1's "updated by re-exporting" guidance, 019 AC-4.2, the re-export parts of its decision 1 and FR-3, its FR-4 and its re-export Error Scenarios row, 021 decision 8, Story 3, the Non-Goal on `ItemTile.md`, FR-4 and the Open Questions entry (decision 12) | |
 | [058 Store Each Card's Mana Value from the Catalog](058-mtg-mana-value/spec.md) | Frozen, by [catalog core](../architecture/catalog-core.md) and [catalog operations](../architecture/catalog-operations.md) | [Catalog core](../architecture/catalog-core.md) (the card extension), [catalog operations](../architecture/catalog-operations.md) (the refresh's `#reapply?`); the [upgrade notes](../releasing.md) | retired in `37f348f106fdf43b3660ace6290d013786bddc09`; `git show 37f348f106fdf43b3660ace6290d013786bddc09^:docs/specs/058-mtg-mana-value/plan.md` | none (adds a second reason to 011 AC-2.3's `#reapply?` hook; changes no rule) | |
 | [059 Open a Location's Lots in Bulk Mode from Its Refusal Page](059-bulk-mode-from-location-refusal/spec.md) | Frozen, by [collection and lots](../architecture/collection-and-lots.md) | [Collection and lots](../architecture/collection-and-lots.md) (locations, bulk mode's selection); [design system](../design-system/README.md) (Locations) | retired in `8d9f59223e75ea8357fb151d1c7077cd049717f0`; `git show 8d9f59223e75ea8357fb151d1c7077cd049717f0^:docs/specs/059-bulk-mode-from-location-refusal/plan.md` | 041 decision 5, its Non-Goal and Out of Scope entries for the refusal page's shortcut. 006 FR-4 (the submissions that start, change or clear a selection) and Story 4 (Edit many as the only way in). 032 AC-5.2 and AC-5.4, where they list the refusal page's controls | 067: Out of Scope entry "decks as locations (#49)" as it touches the refusal page's shortcut (a deck has its own; the place shortcut and its route are unchanged) |
-| [060 Design System: Deck Components and Previews](060-deck-components/spec.md) | Frozen, by [design system](../design-system/README.md) (the components, their previews and the CSS in `additions.css`) | [Design system](../design-system/README.md) (the deck components: `DeckShape`, `ProxyTag`, `DeckPage`, `DecksIndex`, `AddToDeck`, `UseMyCopy`) | retired in `7e9c5850773b19319ce1cda1bf6b9fe0aed127cb`; `git show 7e9c5850773b19319ce1cda1bf6b9fe0aed127cb^:docs/specs/060-deck-components/plan.md` | none of a shipped spec (draws what spec 057 left to #249). It changes three statements of the issue design, which is not a spec: decision 10's 640px stacking (992px; a closed `details` on a phone), decision 13's percentages-only caption and decision 7's single choice page (a quantity on every lot) | 062 (decision 11 and AC-4.1 to AC-4.3 for the Decks page until #254: row actions, no Proxy column; the name link, which 062 made plain text until #253, is restored by 066); 066 (decisions 4, 7 and 9, until #254: decision 17 and AC-3.10 (the row menu), the menu parts of decision 25, AC-3.5 and AC-3.11, decision 10's proxy counts and "Add a proxy", the "Add a proxy" and proxy-count parts of AC-3.7 and AC-3.9); 071 (decision 14 and AC-5.1 in part: a printing with several finishes has one proxy option per finish, not "up to two options") |
+| [060 Design System: Deck Components and Previews](060-deck-components/spec.md) | Frozen, by [design system](../design-system/README.md) (the components, their previews and the CSS in `additions.css`) | [Design system](../design-system/README.md) (the deck components: `DeckShape`, `ProxyTag`, `DeckPage`, `DecksIndex`, `AddToDeck`, `UseMyCopy`) | retired in `7e9c5850773b19319ce1cda1bf6b9fe0aed127cb`; `git show 7e9c5850773b19319ce1cda1bf6b9fe0aed127cb^:docs/specs/060-deck-components/plan.md` | none of a shipped spec (draws what spec 057 left to #249). It changes three statements of the issue design, which is not a spec: decision 10's 640px stacking (992px; a closed `details` on a phone), decision 13's percentages-only caption and decision 7's single choice page (a quantity on every lot) | 062 (decision 11 and AC-4.1 to AC-4.3 for the Decks page until #254: row actions, no Proxy column; the name link, which 062 made plain text until #253, is restored by 066); 066 (decisions 4, 7 and 9, until #254: decision 17 and AC-3.10 (the row menu), the menu parts of decision 25, AC-3.5 and AC-3.11, decision 10's proxy counts and "Add a proxy", the "Add a proxy" and proxy-count parts of AC-3.7 and AC-3.9); 071 (decision 14 and AC-5.1 in part: a printing with several finishes has one proxy option per finish, not "up to two options"); 077 (in part: decision 2, AC-2.2, AC-2.4, AC-2.5 and AC-2.6's visible numbers and the card types' fixed 4.5rem minimum, for the extreme counts of [#320](https://github.com/plainprogrammer/Collector/issues/320)) |
 | [061 The Phone Line's Wrap Is Stated at a Phone's Width](061-place-line-wrap-at-phone-width/spec.md) | Frozen, by [collection and lots](../architecture/collection-and-lots.md) | [Collection and lots](../architecture/collection-and-lots.md) (the table's place line) | retired in `1c0d5ee074242ad292e522b94df96b291f82954d`; `git show 1c0d5ee074242ad292e522b94df96b291f82954d^:docs/specs/061-place-line-wrap-at-phone-width/plan.md` | 048 AC-1.8 (a five-character set code wraps the phone line at 360px: true only with a classic scrollbar), the Error Scenarios row for an account with no location, and the sentences of decision 4 that say so; its ruling and AC-1.7 stand ([#288](https://github.com/plainprogrammer/Collector/issues/288), [#271](https://github.com/plainprogrammer/Collector/issues/271)) | |
 | [062 Create, Rename and Remove a Deck](062-create-rename-remove-a-deck/spec.md) | Frozen, by [collection and lots](../architecture/collection-and-lots.md), [data portability](../architecture/data-portability.md) and [tenancy and accounts](../architecture/tenancy-and-accounts.md) (the design system's Decks pages are in [`DecksIndex.md`](../design-system/components/DecksIndex.md)) | [Collection and lots](../architecture/collection-and-lots.md), [data portability](../architecture/data-portability.md), [tenancy and accounts](../architecture/tenancy-and-accounts.md), [design system](../design-system/README.md) | retired in `624c513a993a037534b13f7572525d65e6c8238c`; `git show 624c513a993a037534b13f7572525d65e6c8238c^:docs/specs/062-create-rename-remove-a-deck/plan.md` | 029 decisions 4, 7, 9 and AC-1.13 (a location is a name only; the limit; name uniqueness per account, for decks and zones), 032 FR-3 (the sixth lot-identity value's form, from version 4), 060 decision 11 and AC-4.1 to AC-4.3 for the Decks page until #253 and #254 (name as text, row actions, no Proxy column). Scope notes for the issues it touches: from [#252](https://github.com/plainprogrammer/Collector/issues/252) it takes only the label "Deck › Zone" wherever a lot's place is shown and the safe Edit copy of a lot in a zone (decision 6; #252's grouped pickers, "Deck (all zones)" in the collection filter and the tiles' design stay open); for [#251](https://github.com/plainprogrammer/Collector/issues/251), an import that changes a deck's format applies the file as the truth, with no move to Considering (#251 moves cards when the collector changes the format in the app) | 064 (Non-Goal "changing a deck's format ... in the app a deck keeps the format it was created with": the app changes it; AC-8.2's import rule stands); 065: AC-5.1, AC-5.2, AC-5.5, AC-5.6, AC-5.7a, AC-5.8a, decision 1's places-only surface list, decision 6's read-only field; 066: decision 5's plain-text deck name (the name is a link to the deck page); 067: decision 7's clause "no shortcut into bulk mode", decision 5's "no bulk-mode shortcut until #252" and AC-3.3's "no 'Move them in bulk mode' button" (AC-3.8 stands); 068: AC-4.4's equal Cards and Owned (the Decks page counts proxies in Cards) |
 | [063 Small Icons Keep Their Weight (Stroke Follows Size)](063-small-icon-stroke/spec.md) | Frozen, by [design system](../design-system/README.md) (Iconography, the App additions line, `additions.css`) | [Design system](../design-system/README.md) (Iconography; `additions.css`) | retired in `375c33e05e502eec749f84a1cafa9b0d79b00ba1`; `git show 375c33e05e502eec749f84a1cafa9b0d79b00ba1^:docs/specs/063-small-icon-stroke/plan.md` | The README's Iconography line ("1.5px stroke, 20px"): 1.5 at 20px and above, 2 at 14px to 16px, 2.5 at 12px; spec 060 decision 16 is extended, not superseded (the previews' attributes stay 1.5) ([#294](https://github.com/plainprogrammer/Collector/issues/294)) | |
@@ -91,3 +91,4 @@ A plan is deleted from the tree in a commit of its own after its feature's pull
 | [074 Deck Pages at Phone Width (Row Menus Clear of the Tab Bar, the Tab Bar's Missing Shots, the Stale Deck Page Preview)](074-deck-pages-phone-width/spec.md) | Frozen, by [design system](../design-system/README.md) (`Menu`, `TableActions`, `TabBar`, `DeckPage`, the Navigation section; `additions.css`; `menu_controller.js`) | [Design system](../design-system/README.md) (`Menu`, `TableActions`, `TabBar`, `DeckPage`; `additions.css`; the `menu` controller) | retirement owed | Nothing; it restores [060](060-deck-components/spec.md) AC-3.11 (the zone tables' aligned columns) after 068's empty menu cell (decision 5, AC-2.6), changing no criterion of either ([#326](https://github.com/plainprogrammer/Collector/issues/326)) | |
 | [076 Direct fix of #317, no spec: a select's label keeps a gap to its select](076-select-label-gap/verification.md) | Approved (a direct fix of [#317](https://github.com/plainprogrammer/Collector/issues/317), no spec and no plan; its evidence is `verification.md`) | [Design system](../design-system/README.md) (`FilterSelect`, `NumberEntry`; `additions.css`) | none (no plan) | none | |
 | [073 "Free to use" and "In decks" on the Card Page](073-card-page-deck-stats/spec.md) | Frozen, by [collection and lots](../architecture/collection-and-lots.md) and [catalog core](../architecture/catalog-core.md) (the card page's stats and "In decks" section; the visitor rule and the card page's order); the card page's markup is `ItemPage` in the [design system](../design-system/README.md) | [Collection and lots](../architecture/collection-and-lots.md), [catalog core](../architecture/catalog-core.md), [design system](../design-system/README.md) (`ItemPage`, `ItemPagePhone`, `SingleStat`) | retirement owed | In part 004 FR-9 (the single stat; no "In decks" or "Free to use"), its Non-Goal and Out of Scope entries and AC-8.1's order; 071 AC-1.4 and its "where #50 sits" note (fulfilled); fulfils the Non-Goals that name #50 (029, 032, 033, 057, 070) and 057 decisions 7 and 8 (the strip and the section) | |
+| [077 The Deck Shape Panel Stays Tidy at Extreme Counts](077-deck-shape-extreme-counts/spec.md) | Frozen, by [collection and lots](../architecture/collection-and-lots.md) (the panel's paragraph and Named constants); the panel's markup is `DeckShape` in the [design system](../design-system/README.md) | [Design system](../design-system/README.md) (`DeckShape`), [collection and lots](../architecture/collection-and-lots.md) (the panel's paragraph, Named constants) | retirement owed | In part 060 decision 2, AC-2.2, AC-2.4, AC-2.5 and AC-2.6 (the visible numbers, the types' alignment) and 069 decision 14, its four-digit-count Error Scenarios row and Findings 1 to 4 ([#320](https://github.com/plainprogrammer/Collector/issues/320)) | |
diff --git a/spec/models/mtg/deck_shape_spec.rb b/spec/models/mtg/deck_shape_spec.rb
index 813b7c8f..abe97bed 100644
--- a/spec/models/mtg/deck_shape_spec.rb
+++ b/spec/models/mtg/deck_shape_spec.rb
@@ -362,4 +362,51 @@ def counts(figures = shape) = figures.symbols.to_h { [ it.colour, it.count ] }
       expect(shape(krenko).symbols_label).to eq("Coloured mana symbols in Commander + Main: red 1")
     end
   end
+
+  # Spec 077 decision 5: a count in a slot of fixed width is written with a unit from 1,000.
+  describe "#short_count" do
+    it "writes 0 to 999 as they are and 1,000 up with one decimal under 10,000, whole from there, half up (AC-1.1)", :aggregate_failures do
+      expected = { 0 => "0", 999 => "999", 1_000 => "1k", 1_049 => "1k", 1_050 => "1.1k", 1_350 => "1.4k", 1_357 => "1.4k", 9_949 => "9.9k",
+                   9_950 => "10k", 9_999 => "10k", 10_000 => "10k", 12_345 => "12k", 123_456 => "123k", 999_499 => "999k", 999_500 => "1m",
+                   1_250_000 => "1.3m", 12_000_000 => "12m" }
+
+      figures = shape
+      expected.each { |number, text| expect(figures.short_count(number)).to eq(text), "for #{number}" }
+      expect(expected.keys.map { figures.short_count(it).length }.max).to be <= 4
+    end
+
+    it "rounds 1,349 down to 1.3k and a count that reaches 1,000 of its unit moves to the next, 9,999,500 being 10m", :aggregate_failures do
+      expect(shape.short_count(1_349)).to eq("1.3k")
+      expect(shape.short_count(9_999_500)).to eq("10m")
+      expect(shape.short_count(2_000_000_000)).to eq("2b")
+    end
+  end
+
+  # Spec 077 decision 4: the caption as items a view can keep whole.
+  describe "#caption_items" do
+    # Spec 069's worst case: 123 owned five-colour creatures, 1,234 proxies of a hybrid card and 4 Bolts, red being 2,595.
+    def worst_case
+      own(card(name: "Worst", type_line: "Creature — Elf", cost: "{2}{W}{U}{B}{R}{G}", mana_value: 7.0), 123)
+      proxy(card(name: "Gleemax", cost: "{W/U}{U/B}{B/R}{R/G}{G/W}", mana_value: 1_000_000.0), 1_234)
+      own(card(name: "Bolt", cost: "{R}", mana_value: 1.0), 4)
+    end
+
+    it "gives one item per colour, each but the last ending in the separator, and `caption` is the items joined (AC-4.1)", :aggregate_failures do
+      worst_case
+      figures = shape
+
+      expect(figures.caption).to eq("White 2,591 · 20% · Blue 2,591 · 20% · Black 2,591 · 20% · Red 2,595 · 20% · Green 2,591 · 20%")
+      expect(figures.caption_items).to eq([ "White 2,591 · 20% ·", "Blue 2,591 · 20% ·", "Black 2,591 · 20% ·", "Red 2,595 · 20% ·", "Green 2,591 · 20%" ])
+      expect(figures.caption_items.join(" ")).to eq(figures.caption)
+    end
+
+    it "has one item with no trailing separator for one colour, and none for no colour (AC-4.1)", :aggregate_failures do
+      own(card(cost: "{1}"))
+      expect(shape.caption_items).to eq([])
+      expect(shape.caption).to be_nil
+
+      own(card(cost: "{R}"), 3)
+      expect(shape.caption_items).to eq([ "Red 3 · 100%" ])
+    end
+  end
 end
diff --git a/spec/requests/deck_shape_panel_spec.rb b/spec/requests/deck_shape_panel_spec.rb
index 417237f0..c5f148fd 100644
--- a/spec/requests/deck_shape_panel_spec.rb
+++ b/spec/requests/deck_shape_panel_spec.rb
@@ -169,4 +169,54 @@ def proxy(entry, quantity = 1, zone: "main") = create(:deck_proxy, zone: deck.zo
 
     expect(response).to have_http_status(:not_found)
   end
+
+  # Spec 077: the worst case's numbers in a slot of fixed width are abbreviated, and everything that can carry the exact number does.
+  describe "counts in a slot of fixed width (spec 077, #320)" do
+    # Spec 069's worst case: a 6+ column of 1,357 (123 owned, 1,234 proxy), five single cards and Bolt in column 1, 20 lands, and a
+    # five-colour mana bar of about 2,591 symbols a colour (red 2,595: Bolt's {R} times 4).
+    before do
+      own(card(name: "Worst", type_line: "Creature — Elf", cost: "{2}{W}{U}{B}{R}{G}", mana_value: 7.0), 123)
+      proxy(card(name: "Gleemax", type_line: "Creature — Elf", cost: "{W/U}{U/B}{B/R}{R/G}{G/W}", mana_value: 1_000_000.0), 1_234)
+      own(card(name: "Bolt", cost: "{R}", mana_value: 1.0), 4)
+      5.times { own(card(type_line: "Sorcery", cost: "{1}", mana_value: 1.0)) }
+      own(card(name: "Island", type_line: "Basic Land — Island", cost: "", mana_value: 0.0), 20)
+      get deck_path(deck)
+    end
+
+    it "abbreviates a column's count from 1,000 and keeps the exact numbers in the label and the legend (AC-1.2)", :aggregate_failures do
+      expect(html.css("aside .c-curve__count").map(&:text)).to eq(%w[0 9 0 0 0 0 1.4k])
+      expect(html.at_css("aside .c-curve [role=img]")["aria-label"]).to include("6+, 1,357 cards (123 owned, 1,234 proxy)")
+      expect(html.css("aside .c-curve__legend li").map(&:text)).to eq([ "Owned 132", "Proxy 1,234" ])
+    end
+
+    it "abbreviates a segment's label and keeps the bar's label and the caption exact (AC-2.1)", :aggregate_failures do
+      expect(html.css("aside .c-manabar__seg span").map(&:text)).to eq([ "W 2.6k", "U 2.6k", "B 2.6k", "R 2.6k", "G 2.6k" ])
+      expect(html.at_css("aside .c-manabar")["aria-label"]).to eq("Coloured mana symbols in Main: white 2,591, blue 2,591, black 2,591, red 2,595, green 2,591")
+      expect(html.at_css("aside .c-manacap").text).to eq("White 2,591 · 20% · Blue 2,591 · 20% · Black 2,591 · 20% · Red 2,595 · 20% · Green 2,591 · 20%")
+    end
+
+    it "sets each caption item as a `c-manacap__item` span and the paragraph's text is the caption string (AC-4.2)", :aggregate_failures do
+      items = html.css("aside .c-manacap .c-manacap__item")
+
+      expect(items.map(&:text)).to eq([ "White 2,591 · 20% ·", "Blue 2,591 · 20% ·", "Black 2,591 · 20% ·", "Red 2,595 · 20% ·", "Green 2,591 · 20%" ])
+      expect(html.at_css("aside .c-manacap").text).to eq(items.map(&:text).join(" "))
+    end
+
+    it "keeps the types' counts and suffix exact (the rows have room)" do
+      expect(html.css("aside .c-types li").first.css("span").map(&:text)).to eq([ "Creatures", "1,357", "1,234 proxy" ])
+    end
+  end
+
+  describe "counts under 1,000 in a slot of fixed width (spec 077, #320)" do
+    before do
+      own(card(cost: "{U}"), 44)
+      own(card(cost: "{R}"), 56)
+      get deck_path(deck)
+    end
+
+    it "writes counts under 1,000 in full in a slot of fixed width (AC-2.1)", :aggregate_failures do
+      expect(html.css("aside .c-manabar__seg span").map(&:text)).to eq([ "U 44", "R 56" ])
+      expect(html.css("aside .c-curve__count").map(&:text)[1]).to eq("100")
+    end
+  end
 end
diff --git a/spec/system/deck_shape_panel_spec.rb b/spec/system/deck_shape_panel_spec.rb
index 2bb35735..4fe6022b 100644
--- a/spec/system/deck_shape_panel_spec.rb
+++ b/spec/system/deck_shape_panel_spec.rb
@@ -129,9 +129,10 @@ def open_phone_frame
   # A phone's 360px viewport has no scrollbar and the frame keeps a classic one, so the frame is 360px plus the measured scrollbar
   # (375px with the Mac's 15px). The chain is the viewport's 360px less the body's 16px (8px margin a side, a 344px body), the main's
   # 32px padding (a 312px container), the details' 2px border (310px) and the panel's 32px padding (a 278px inside). A frame with the
-  # scrollbar inside a 360px width is 15px narrower (263px inside), where the fixture's 1,357 count overflows its column by about
-  # 1.85px: a CSS question raised with the maintainer, not decided here. Each check can fail: the panel scrolling sideways, an element
-  # drawn past the padding box's right edge, and a text element cut off at its own box.
+  # scrollbar inside a 360px width is 15px narrower (263px inside), where the fixture's count in full ("1,357") overflowed its column by
+  # about 1.85px; spec 077 (#320) abbreviates it ("1.4k"), and the "extreme counts" examples below measure that frame and a 320px
+  # viewport. Each check can fail: the panel scrolling sideways, an element drawn past the padding box's right edge, and a text element
+  # cut off at its own box.
   it "fits the open fold's parts in the panel's interior at a real phone's 360px viewport, with nothing cut off and each count with room in its column (AC-5.3)", :aggregate_failures do
     expect(open_phone_frame).to eq(360)
     within_narrow_frame do
@@ -160,4 +161,220 @@ def open_phone_frame
       expect(paint).to eq({ "owned" => true, "proxyHatch" => true, "noBrand" => true, "segments" => 5, "segmentsRight" => true })
     end
   end
+
+  # Spec 077 (#320): the panel at extreme counts at a 320px, 345px and 360px viewport (a real 320px phone, a 360px browser frame, a
+  # 360px phone), a 366px frame (where the old caption leads a line with a dot), a 640px frame (the fold again) and the 320px aside at
+  # 1280px, in both themes. The viewport frames are sized from the measured scrollbar (the viewport plus `innerWidth - clientWidth`: 15px
+  # on a Mac, about 12px on the Linux runner) and read against `documentElement.clientWidth` (lessons of 2026-11-11 on phone widths and
+  # measurement traps); the 366px and 640px frames are fixed frame widths by the spec's words, not viewports. Each geometry example has a
+  # failing variant: the old text or the old rule injected into the page.
+  # The file's memoized helpers (user, account, deck) plus the two lets below are at the cop's limit of five: another deck an example needs is
+  # a local (`plain = create(:deck, …)`, as AC-3.2 does), not another `let`.
+  describe "extreme counts (spec 077, #320)" do
+    let(:rows_deck) { create(:deck, account:, name: "Rows") }
+    let(:wide_deck) { create(:deck, account:, name: "Wide rows") }
+
+    def plain_card(name:, type_line: "Creature — Elf") = card(name:, type_line:, cost: "{1}", mana_value: 2.0)
+    def own_in(in_deck, entry, quantity) = create_lot_in_zone(in_deck, "main", entry:, quantity:)
+    def proxy_in(in_deck, entry, quantity) = create(:deck_proxy, zone: in_deck.zone("main"), account:, entry:, quantity:)
+
+    # AC-3.1's rows: "Creatures" 1,357 with "1,234 proxy", "Instants" 9,999 with "9,999 proxy", "Artifacts" 3 and "Lands" 20 with none.
+    def build_rows
+      own_in(rows_deck, plain_card(name: "Rows elf"), 123)
+      proxy_in(rows_deck, plain_card(name: "Rows elf proxy"), 1_234)
+      proxy_in(rows_deck, plain_card(name: "Rows bolt", type_line: "Instant"), 9_999)
+      own_in(rows_deck, plain_card(name: "Rows ring", type_line: "Artifact"), 3)
+      own_in(rows_deck, card(name: "Rows island", type_line: "Basic Land — Island", cost: "", mana_value: 0.0), 20)
+    end
+
+    # AC-3.1b's rows: "Planeswalkers" 12,345 with "12,345 proxy" (a proxy takes at most 9,999, so two) and "Lands" 20, a second row so that
+    # "the counts end at one x" can fail.
+    def build_wide_row
+      proxy_in(wide_deck, plain_card(name: "Wide jace", type_line: "Legendary Planeswalker — Jace"), 9_999)
+      proxy_in(wide_deck, plain_card(name: "Wide ajani", type_line: "Legendary Planeswalker — Ajani"), 2_346)
+      own_in(wide_deck, card(name: "Wide island", type_line: "Basic Land — Island", cost: "", mana_value: 0.0), 20)
+    end
+
+    def widths = [ 320, 345, 360, :frame366, :frame640, :aside ]
+    def fold_panel = "details.c-shapefold .c-shape"
+    def open_fold = page.execute_script('document.querySelector("details.c-shapefold").open = true')
+    def inject_css(rules) = page.execute_script("const style = document.createElement('style'); style.textContent = #{rules.to_json}; document.head.append(style)")
+
+    # A frame whose viewport is `viewport` px wide with the fold open: the viewport plus the scrollbar the frame draws (measured, then
+    # the frame is reopened if it was not 15px). Returns the viewport it got, for the example to assert.
+    def open_fold_in_frame(path, viewport)
+      width = viewport + 15
+      2.times do
+        open_in_narrow_frame(path, width:, ready: "details.c-shapefold")
+        scrollbar = within_narrow_frame do
+          open_fold
+          page.evaluate_script("window.innerWidth - document.documentElement.clientWidth")
+        end
+        break if viewport + scrollbar == width
+
+        width = viewport + scrollbar
+      end
+      within_narrow_frame { page.evaluate_script("document.documentElement.clientWidth") }
+    end
+
+    # Opens the deck page as `width` says (a viewport in px, a fixed frame, or :aside at 1280px) and yields the panel's selector, inside
+    # the frame where there is one.
+    def at_width(width, path)
+      case width
+      when Integer
+        expect(open_fold_in_frame(path, width)).to eq(width)
+        within_narrow_frame { yield fold_panel }
+      when :frame366, :frame640
+        open_in_narrow_frame(path, width: { frame366: 366, frame640: 640 }.fetch(width), ready: "details.c-shapefold")
+        within_narrow_frame do
+          open_fold
+          yield fold_panel
+        end
+      when :aside
+        visit path
+        page.driver.browser.manage.window.resize_to(1280, 900)
+        expect(page).to have_css("aside.c-deck__shape")
+        yield "aside .c-shape"
+      end
+    end
+
+    # What the panel measures: per curve column its room (the column's width less its count's scrollWidth), what is drawn past the
+    # padding box, per visible mana label its line count, its text width, its segment's content width and whether the text's box is inside
+    # its segment's box, the Card types rows' right edges, and per caption item its line count and the first character of every caption line.
+    def measure(selector)
+      page.evaluate_script(<<~JS)
+        (() => { const panel = document.querySelector(#{selector.to_json}), edge = panel.getBoundingClientRect().right - parseFloat(getComputedStyle(panel).paddingRight)
+          const textRange = (el) => { const r = document.createRange(); r.selectNodeContents(el); return r }
+          const lines = (el) => new Set([...textRange(el).getClientRects()].map((q) => Math.round(q.top))).size
+          const cols = [...panel.querySelectorAll(".c-curve__col")], rows = [...panel.querySelectorAll(".c-types li")], list = panel.querySelector(".c-types")
+          const labels = [...panel.querySelectorAll(".c-manabar__seg span")].filter((s) => s.getClientRects().length)
+          const starts = new Map(), caption = panel.querySelector("p.c-manacap"), walker = document.createTreeWalker(caption, NodeFilter.SHOW_TEXT)
+          for (let node = walker.nextNode(); node; node = walker.nextNode()) { for (let i = 0; i < node.length; i++) { const r = document.createRange(); r.setStart(node, i); r.setEnd(node, i + 1)
+            const box = r.getClientRects()[0]; if (box && node.data[i] !== " " && !starts.has(Math.round(box.top))) starts.set(Math.round(box.top), node.data[i]) } }
+          return { viewport: document.documentElement.clientWidth,
+            columns: cols.length,
+            narrowest: cols.length ? Math.min(...cols.map((col) => col.getBoundingClientRect().width)) : 0,
+            scrolls: panel.scrollWidth > panel.clientWidth,
+            beyond: [...panel.querySelectorAll("*")].filter((el) => el.getClientRects().length && el.getBoundingClientRect().right > edge + 0.5).length,
+            clipped: [...panel.querySelectorAll(".c-curve__count, .c-types li > *, .c-curve__legend li, .c-curve__avg, .c-manacap")].filter((el) => el.scrollWidth > el.clientWidth + 1).length,
+            countRoom: cols.length ? Math.min(...cols.map((col) => col.getBoundingClientRect().width - col.querySelector(".c-curve__count").scrollWidth)) : 0,
+            labels: labels.map((s) => { const box = textRange(s).getBoundingClientRect(), seg = s.parentElement.getBoundingClientRect()
+              return { lines: lines(s), text: box.width, room: s.parentElement.clientWidth, inside: box.left >= seg.left - 0.5 && box.right <= seg.right + 0.5 } }),
+            rights: rows.map((li) => li.children[1].getBoundingClientRect().right),
+            suffixGap: rows.map((li) => Math.abs(li.getBoundingClientRect().right - li.children[2].getBoundingClientRect().right)),
+            suffixWidth: rows.map((li) => li.children[2].getBoundingClientRect().width),
+            rowOver: rows.filter((li) => li.getBoundingClientRect().right > list.getBoundingClientRect().right + 0.5 || li.scrollWidth > li.clientWidth + 1).length,
+            cut: rows.flatMap((li) => [...li.children]).filter((el) => el.scrollWidth > el.clientWidth + 1).length,
+            itemLines: [...panel.querySelectorAll(".c-manacap__item")].map(lines),
+            lineStarts: [...starts.values()] } })()
+      JS
+    end
+
+    # A label fits when it is on one line and its text is within its segment's content width (spec AC-2.2). Margins at a 320px viewport,
+    # where `clientWidth` is a whole 46px: 2.44px on a Mac (a 43.56px label) and 1.30px on the Linux runner (44.70px: its mono face is wider).
+    # A page whose panel is not drawn measures nothing, so the seven columns and a column with a width are part of "fits".
+    def counts_fit?(found)
+      found["countRoom"] >= 0 && !found["scrolls"] && found["beyond"].zero? && found["clipped"].zero? && found["columns"] == 7 && found["narrowest"] > 0
+    end
+    def labels_fit?(found) = found["labels"].any? && found["labels"].all? { |label| label["lines"] == 1 && label["inside"] && label["text"] <= label["room"] }
+    def counts_aligned?(found) = found["rights"].any? && found["rights"].min > 0 && (found["rights"].max - found["rights"].min) <= 0.5
+    def types_tidy?(found) = counts_aligned?(found) && found["suffixGap"].all? { it <= 0.5 } && found["rowOver"].zero? && found["cut"].zero?
+    def wide_row_tidy?(found) = found["rowOver"].zero? && found["cut"].zero? && counts_aligned?(found) && found["rights"].size == 2
+    def floor_held?(found) = counts_aligned?(found) && found["suffixWidth"].all? { it >= 71.5 }
+    def caption_whole?(found) = found["itemLines"].any? && found["itemLines"].all?(1) && found["lineStarts"].none?("·")
+
+    # Measures in each theme (layout cannot depend on it, so there is no wait for the 0.12s colour transition here; the second reading is a
+    # guard) and expects the block to hold.
+    def expect_in_both_themes(width, selector)
+      %w[light dark].each do |theme|
+        set_theme(theme)
+        found = measure(selector)
+        expect(yield(found)).to be(true), "at #{width} in #{theme}: #{found.inspect}"
+      end
+    end
+
+    it "keeps every curve count inside its column and the panel inside its box at every width, in both themes (AC-1.3)", :aggregate_failures do
+      widths.each { |width| at_width(width, deck_path(deck)) { |panel| expect_in_both_themes(width, panel) { counts_fit?(it) } } }
+    end
+
+    it "can fail: the full count in the 6+ column overflows its column at a 320px and a 345px viewport (AC-1.3)", :aggregate_failures do
+      [ 320, 345 ].each do |viewport|
+        at_width(viewport, deck_path(deck)) do |panel|
+          page.execute_script('document.querySelector("details.c-shapefold .c-curve__col:last-child .c-curve__count").textContent = "1,357"')
+          expect(counts_fit?(measure(panel))).to be(false), "at #{viewport}"
+        end
+      end
+    end
+
+    it "keeps every mana label on one line inside its segment at every width, in both themes (AC-2.2)", :aggregate_failures do
+      widths.each { |width| at_width(width, deck_path(deck)) { |panel| expect_in_both_themes(width, panel) { labels_fit?(it) } } }
+    end
+
+    it "can fail: a label written in full wraps inside its segment at a 320px viewport (AC-2.2)" do
+      at_width(320, deck_path(deck)) do |panel|
+        page.execute_script('document.querySelectorAll("details.c-shapefold .c-manabar__seg span").forEach((span) => { span.textContent = span.textContent.replace("2.6k", "2,591") })')
+        expect(labels_fit?(measure(panel))).to be(false)
+      end
+    end
+
+    it "ends every Card types count at one x with the suffix at the row's edge, nothing cut off, at every width (AC-3.1)", :aggregate_failures do
+      build_rows
+      widths.each { |width| at_width(width, deck_path(rows_deck)) { |panel| expect_in_both_themes(width, panel) { types_tidy?(it) } } }
+    end
+
+    it "can fail: flex rows with a 4.5rem suffix put the 9,999 proxy row's count left of the others' (AC-3.1)", :aggregate_failures do
+      build_rows
+      at_width(320, deck_path(rows_deck)) do |panel|
+        inject_css(".c-types{display:block !important} .c-types li{display:flex !important;gap:var(--space-2)} .c-types__n{margin-left:auto !important}")
+        found = measure(panel)
+        expect(types_tidy?(found)).to be(false)
+        expect(found["rights"].max - found["rights"].min).to be > 5
+      end
+    end
+
+    it "keeps each empty suffix cell at its 4.5rem floor and the counts at one x when no row has a proxy (AC-3.2)", :aggregate_failures do
+      plain = create(:deck, account:, name: "Plain")
+      own_in(plain, plain_card(name: "Plain elf"), 5)
+      own_in(plain, plain_card(name: "Plain ring", type_line: "Artifact"), 3)
+      [ 360, :aside ].each { |width| at_width(width, deck_path(plain)) { |panel| expect_in_both_themes(width, panel) { floor_held?(it) } } }
+    end
+
+    it "can fail: without the floor the empty suffix cells collapse (AC-3.2)" do
+      plain = create(:deck, account:, name: "Plain")
+      own_in(plain, plain_card(name: "Plain elf"), 5)
+      at_width(360, deck_path(plain)) do |panel|
+        inject_css(".c-types__proxy{min-width:0 !important}")
+        expect(floor_held?(measure(panel))).to be(false)
+      end
+    end
+
+    it "keeps a five-digit row inside the list with the counts at one x at a 320px viewport, the name wrapping (AC-3.1b)" do
+      build_wide_row
+      at_width(320, deck_path(wide_deck)) { |panel| expect_in_both_themes(320, panel) { wide_row_tidy?(it) } }
+    end
+
+    it "can fail: without the name cell's wrap rule the name spills over its cell (AC-3.1b)" do
+      build_wide_row
+      at_width(320, deck_path(wide_deck)) do |panel|
+        inject_css(".c-types li > :first-child{overflow-wrap:normal !important}")
+        expect(wide_row_tidy?(measure(panel))).to be(false)
+      end
+    end
+
+    it "keeps every caption item on one line and no caption line starting with a dot at every width (AC-4.3)", :aggregate_failures do
+      widths.each { |width| at_width(width, deck_path(deck)) { |panel| expect_in_both_themes(width, panel) { caption_whole?(it) } } }
+    end
+
+    it "can fail: without the nowrap rule an item splits at a 345px and a 360px viewport and a line starts with a dot in a 366px frame (AC-4.3)", :aggregate_failures do
+      [ 345, 360, :frame366 ].each do |width|
+        at_width(width, deck_path(deck)) do |panel|
+          inject_css(".c-manacap__item{white-space:normal !important}")
+          found = measure(panel)
+          expect(caption_whole?(found)).to be(false), "at #{width}: #{found.inspect}"
+          expect(found["lineStarts"]).to include("·") if width == :frame366
+          expect(found["itemLines"].any? { it > 1 }).to be(true), "at #{width}: no item spans two lines: #{found.inspect}" if width.is_a?(Integer)
+        end
+      end
+    end
+  end
 end