diff --git a/.github/workflows/action-format.yml b/.github/workflows/action-format.yml
index adc2396e45..ad68eaffc7 100644
--- a/.github/workflows/action-format.yml
+++ b/.github/workflows/action-format.yml
@@ -83,7 +83,7 @@ jobs:
         run: yarn install --pure-lockfile
 
       - name: Format JSON files
-        run: yarn format-json
+        run: yarn format-json && ruby bin/format-array.rb
 
       - name: 'Commit formatted code'
         run: |
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index a20d8ff07d..494e396366 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -153,4 +153,10 @@ jobs:
         run: yarn install
 
       - name: Verify that json files are formatted correctly
-        run: yarn test-json-formatting
+        run: |
+          yarn format-json && ruby bin/format-array.rb
+          if ! git diff --quiet --exit-code; then
+            echo "please format the files with prettier and bin/format-array.rb, or apply the following changes:"
+            git diff
+            exit 1
+          fi
diff --git a/.prettierignore b/.prettierignore
deleted file mode 100644
index 2d1a836a16..0000000000
--- a/.prettierignore
+++ /dev/null
@@ -1,18 +0,0 @@
-exercises/book-store/canonical-data.json
-exercises/bowling/canonical-data.json
-exercises/change/canonical-data.json
-exercises/connect/canonical-data.json
-exercises/diamond/canonical-data.json
-exercises/dominoes/canonical-data.json
-exercises/flatten-array/canonical-data.json
-exercises/forth/canonical-data.json
-exercises/go-counting/canonical-data.json
-exercises/minesweeper/canonical-data.json
-exercises/ocr-numbers/canonical-data.json
-exercises/rectangles/canonical-data.json
-exercises/saddle-points/canonical-data.json
-exercises/scale-generator/canonical-data.json
-exercises/sieve/canonical-data.json
-exercises/transpose/canonical-data.json
-exercises/variable-length-quantity/canonical-data.json
-exercises/word-search/canonical-data.json
diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index 300a498cd3..86b3fe317f 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -304,12 +304,12 @@ We are maintaining this section, since many open issues link to it.
 
 ## Formatting
 
-This repository uses [prettier][prettier] to automatically format its JSON files.
+This repository uses [prettier][prettier] and a [custom array formatter][format-array] to automatically format its JSON files.
 If you've added or modified a JSON file, you can format it using:
 
 ```shell
 yarn install
-yarn format-json
+yarn format-json && ruby bin/format-array.rb
 ```
 
 Note: if you use VS Code as your editor, you can install the [prettier plugin][prettier-vs-code] to automatically handle formatting for you.
@@ -336,4 +336,5 @@ Note: if you use VS Code as your editor, you can install the [prettier plugin][p
 [improve-exercise-metadata]: https://github.com/exercism/legacy-docs/blob/main/you-can-help/improve-exercise-metadata.md
 [legacy-docs]: https://github.com/exercism/legacy-docs
 [prettier]: https://prettier.io/
+[format-array]: https://github.com/exercism/problem-specifications/blob/main/bin/format-array.rb
 [prettier-vs-code]: https://marketplace.visualstudio.com/items?itemName=esbenp.prettier-vscode
diff --git a/bin/format-array.rb b/bin/format-array.rb
new file mode 100644
index 0000000000..24b5fdbf59
--- /dev/null
+++ b/bin/format-array.rb
@@ -0,0 +1,178 @@
+require 'json'
+
+# format-array expresses (and enacts) our per-file preferences for array formatting,
+# in cases where the preference differs from prettier.
+#
+# It has only one mode of operation:
+# Run it without any arguments to format the files.
+
+# format-array was written because we want to keep files consistently formatted,
+# while also optimising for human readability.
+#
+# As part of optimising for human readability,
+# some arrays should be formatted such that each element is on its own line,
+# whereas some other arrays should be formatted such that the entire array is on one line.
+#
+# We could not find an existing tool that allows us to specify array formatting how we wanted.
+
+# The configuration of array formatting preferences.
+#
+# format can be the following choices:
+#
+# * single_line (the entire array and all of its elements should be on a single line)
+# * multi_line (each element of the array is on its own line)
+# * multi_line_unless_single (like multi_line, except arrays with one element remain on a single line)
+# * multi_line_deep (multi_line, even applied to arrays within that array)
+# * padded (14 elements per line padded to 3 characters per element)
+#     (TODO: padding amount could be configurable, but we haven't needed it)
+formats = {
+  'book-store' => {
+    'basket' => :single_line,
+  },
+  'bowling' => {
+    'previousRolls' => :single_line,
+  },
+  'change' => {
+    'expected' => :single_line,
+  },
+  'connect' => {
+    'board' => :multi_line,
+  },
+  'diamond' => {
+    'expected' => :multi_line,
+  },
+  'dominoes' => {
+    'dominoes' => :single_line,
+  },
+  'flatten-array' => {
+    'array' => :multi_line_deep,
+    'expected' => :multi_line,
+  },
+  'forth' => {
+    'instructions' => :multi_line_unless_single,
+  },
+  'go-counting' => {
+    'board' => :multi_line,
+  },
+  'minesweeper' => {
+    'minefield' => :multi_line_unless_single,
+    'expected' => :multi_line_unless_single,
+  },
+  'ocr-numbers' => {
+    'rows' => :multi_line,
+  },
+  'rectangles' => {
+    'strings' => :multi_line_unless_single,
+  },
+  'saddle-points' => {
+    'matrix' => :multi_line,
+  },
+  'scale-generator' => {
+    'expected' => :single_line,
+  },
+  'sieve' => {
+    'expected' => :padded,
+  },
+  'transpose' => {
+    'lines' => :multi_line,
+    'expected' => :multi_line,
+  },
+  'variable-length-quantity' => {
+    'integers' => :single_line,
+    'expected' => :single_line,
+  },
+  'word-search' => {
+    'grid' => :multi_line,
+    'wordsToSearchFor' => :multi_line,
+  },
+}.each_value(&:freeze).freeze
+
+def single_line_arrays(contents, key)
+  # matches things of the form "key": [1, 2, 3]
+  # because this is not a multi-line regex,
+  # . does NOT match newlines.
+  contents.scan(/^( +)"#{key}": (\[.*\],?$)/)
+end
+
+def multi_line_arrays(contents, key)
+  # matches things of the form
+  # "key": [
+  #   1,
+  #   2,
+  #   3
+  # ]
+  # because this IS a multi-line regex (note the /m),
+  # . DOES match newlines.
+  #
+  # To find which closing bracket matches the opening bracket,
+  # we find the *first* closing bracket that is both:
+  # - the only thing on its line (except for maybe a comma afterward)
+  # - at the same indentation level as the key
+  #
+  # to find the first of these, make the match non-greedy (*? instead of *)
+  contents.scan(/^( +)"#{key}": (\[$.*?^\1\],?$)/m)
+end
+
+formats.each { |exercise, exercise_format|
+  filename = "#{__dir__}/../exercises/#{exercise}/canonical-data.json"
+  contents = File.read(filename)
+  exercise_format.each { |key, format_type|
+    replace = ->(old, new) {
+      # Include the key in both the search and the replacement,
+      # to avoid accidentally replacing something we didn't mean to.
+      # This has been observed to be important for transpose,
+      # where ["A1"] is both an input and an output.
+      contents.sub!(%Q("#{key}": #{old}), %Q("#{key}": #{new}))
+    }
+
+    case format_type
+    when :single_line
+      multi_line_arrays(contents, key).each { |indent, arr|
+        arr_lines = arr.lines
+        raise "impossible, array doesn't start with [ but instead #{arr_lines[0]}" if arr_lines[0] != "[\n"
+        raise "impossible, array doesn't end with ] but instead #{arr_lines[-1]}" unless arr_lines[-1].match?(/^ *\],?$/)
+        replace[arr, "[#{arr_lines[1...-1].map(&:strip).join(' ')}]#{',' if arr[-1] == ','}"]
+      }
+    when :multi_line, :multi_line_unless_single
+      single_line_arrays(contents, key).each { |indent, arr|
+        elements = JSON.parse(arr.delete_suffix(','))
+        next if elements.empty?
+        next if elements.size == 1 && format_type == :multi_line_unless_single
+        indented_elements = elements.map { |el|
+          js = JSON.generate(el)
+          # JSON.generate will output [1,2,3] but we want [1, 2, 3]
+          "#{indent}  #{el.is_a?(Array) ? js.gsub(',', ', ') : js}"
+        }
+        # all lines but the last need a trailing comma
+        replace[arr, "[\n#{indented_elements.join(",\n")}\n#{indent}]#{',' if arr[-1] == ','}"]
+      }
+    when :multi_line_deep
+      single_line_arrays(contents, key).each { |indent, arr|
+        # pretty_generate will render an empty array as:
+        # [
+        #
+        # ]
+        # whereas we just want []
+        pretty_lines = JSON.pretty_generate(JSON.parse(arr)).sub(/\[\s+\]/, '[]').lines
+        if pretty_lines == ['[]']
+          replace[arr, '[]']
+        else
+          raise "impossible, array doesn't start with [ but instead #{pretty_lines[0]}" if pretty_lines[0] != "[\n"
+          raise "impossible, array doesn't end with ] but instead #{pretty_lines[-1]}" if pretty_lines[-1] != "]"
+          replace[arr, "[\n" + pretty_lines[1...-1].map { |l| "#{indent}#{l}" }.join + "#{indent}]#{',' if arr[-1] == ','}"]
+        end
+      }
+    when :padded
+      multi_line_arrays(contents, key).each { |indent, arr|
+        elements = JSON.parse(arr)
+        padded_elements = elements.map { |el| '%3d' % el }
+        new_rows = padded_elements.each_slice(14).map { |row| indent + '  ' + row.join(', ') }
+        # all lines but the last need a trailing comma
+        replace[arr, "[\n#{new_rows.join(",\n")}\n#{indent}]"]
+      }
+    else
+      raise "unknown #{exercise} #{key} #{format_type}"
+    end
+  }
+  File.write(filename, contents)
+}
diff --git a/exercises/bowling/canonical-data.json b/exercises/bowling/canonical-data.json
index c8e8d13150..3c4a9f5afe 100644
--- a/exercises/bowling/canonical-data.json
+++ b/exercises/bowling/canonical-data.json
@@ -133,9 +133,7 @@
       "description": "last two strikes followed by only last bonus with non strike points",
       "property": "score",
       "input": {
-        "previousRolls": [
-          0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 10, 10, 0, 1
-        ]
+        "previousRolls": [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 10, 10, 0, 1]
       },
       "expected": 31
     },
diff --git a/exercises/transpose/canonical-data.json b/exercises/transpose/canonical-data.json
index 8cb13cdd03..14a9dae38b 100644
--- a/exercises/transpose/canonical-data.json
+++ b/exercises/transpose/canonical-data.json
@@ -64,7 +64,9 @@
       "description": "single line",
       "property": "transpose",
       "input": {
-        "lines": ["Single line."]
+        "lines": [
+          "Single line."
+        ]
       },
       "expected": [
         "S",
diff --git a/package.json b/package.json
index ed217f9067..f6049cb0f6 100644
--- a/package.json
+++ b/package.json
@@ -13,7 +13,6 @@
   "scripts": {
     "test": "ajv -s canonical-data.schema.json -d \"exercises/*/canonical-data.json\"",
     "test-one": "ajv -s canonical-data.schema.json -d",
-    "test-json-formatting": "prettier --check **/*.json",
     "format-json": "prettier --write **/*.json"
   },
   "resolutions": {