#228: shrink the HUD on a screen the canvas is short on

The layout is authored in CANVAS units. With stretch "canvas_items" and
aspect "expand" over a 1920x1080 base, a 720x1280 phone scales by 0.375 and
gets 1920x3413 canvas units - while ANY 16:9-or-wider screen lands on
1920x1080 units however many pixels it really has. The canvas is three times
shorter, so a 148-unit button went from 4.3% of the screen's height to
13.7%. Measured at 3.16x, and confirmed twice: once by hand off the scene
file, once by the test failing with the rule disabled.

Nothing collided. That was checked pair by pair on both canvases and the
same six pairs overlap on each, all of them same-slot alternatives. The
fault was only that everything was enormous.

HudScale captures each HUD control's authored offsets once, then moves it
toward the edge it hangs off and scales the NODE. Scaling the rect alone was
tried first and was wrong: half the HUD draws its own contents at fixed
sizes - ResourceReadout lays its rows out in code, Labels carry their own
font size - so shrunken boxes kept full-size content and the build button
slid up underneath the resource list. Scaling the node takes content and
children with it.

The cost is that position (parent space) and size (local space) now differ
by that scale, and a lot of code here assumed they did not. Five places:

  ResourceReadout.content_bottom() added a scaled position to an unscaled
  height, overstating itself by more than a button - so _seat_hud_columns
  pushed the whole column down to clear space nothing occupied. A live bug,
  not just a bad measurement.
  HUD_COLUMN_MARGIN would have been the one gap still phone-sized.
  _layout_context_buttons wrote its stride and pitch unscaled, landing the
  action row on the building-action row - which is the second of the two
  regressions test_hud_layout was written for, in a new disguise.
  test_hud_layout and test_hangar_management compared authored rects.

And HudScale reapplies on resize, which wiped the two layouts BuildController
writes at runtime - so dragging a desktop window left the context row where
it was. Both re-run on size_changed now, deferred, because which of two
siblings connected first is not a thing to depend on.

test_hud_layout ran at one window shape, which happened to be the desktop
canvas. It now runs every check at both, so this cannot hide at whichever
shape was not being exercised.

Rejected: scaling the UI CanvasLayer does not work at all - Control anchors
resolve against the viewport rect, so bottom-anchored elements land
mid-screen. A single scaled wrapper Control with the HUD reparented under it
is the genuinely clean shape and would need no scale-aware arithmetic
anywhere, but it changes every node path and would break every get_node in
the controllers and tests.

MIN_SCALE floors the shrink at 0.45 rather than the proportional 0.316: a
landscape phone gets the same short canvas as a desktop, and there the
sticks are still thumb targets. That leaves the HUD at 1.42x its phone share
on a PC rather than 1.0x.

177/177 at JOBS=3.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_019AJMdB7iuHuLFZM1aBu3GD

controllers/BuildController.gd

@@ -63,6 +63,7 @@ var context_actions_expanded: bool = false
 
 
 func _ready() -> void:
+	get_viewport().size_changed.connect(_on_viewport_resized)
 	_seat_hud_columns()
 	build_button.pressed.connect(_on_build_pressed)
 	action_button.pressed.connect(_on_context_action_pressed)
@@ -97,6 +98,22 @@ func _ready() -> void:
 	_layout_context_buttons()
 
 
+# HudScale reapplies each control's authored offsets whenever the window
+# changes, which wipes the two layouts this controller writes at runtime.
+# Re-running them after a resize is what keeps a desktop window that gets
+# dragged to another size from leaving the context row where it was.
+func _on_viewport_resized() -> void:
+	# Deferred: HudScale reapplies on the same signal, and which of two
+	# siblings connected first is not a thing to depend on. By the end of the
+	# frame its offsets are in, and these two can seat against them.
+	_relayout_hud.call_deferred()
+
+
+func _relayout_hud() -> void:
+	_seat_hud_columns()
+	_layout_context_buttons()
+
+
 func _seat_hud_columns() -> void:
 	var readout: ResourceReadout = get_node_or_null("../UI/ResourceReadout")
 	if readout == null:
@@ -109,7 +126,10 @@ func _seat_hud_columns() -> void:
 			highest = minf(highest, button.position.y)
 	if is_inf(highest):
 		return
-	var shift: float = readout.content_bottom() + HUD_COLUMN_MARGIN - highest
+	# The margin is authored in the same canvas units as everything else, so
+	# it scales with the HUD (#228) or it would be the one gap that stayed
+	# phone-sized on a desktop.
+	var shift: float = readout.content_bottom() + HUD_COLUMN_MARGIN * readout.scale.y - highest
 	if shift <= 0.0:
 		return
 	for button in buttons:
@@ -222,7 +242,14 @@ func _clear_detected_context() -> void:
 
 
 func _layout_context_buttons() -> void:
-	var row_top: float = build_button.position.y + build_button.size.y + HUD_ROW_GAP
+	# position is in the UI layer's space and size is in the button's own,
+	# and since #228 those differ by the HUD scale on a short canvas. Every
+	# gap and stride below is therefore scaled, while the RECTS keep their
+	# authored size - HudScale renders them down rather than shrinking them.
+	# Mixing the two is what this function got wrong twice before (see
+	# test_hud_layout), and a desktop canvas makes it wrong again.
+	var k: float = build_button.scale.y
+	var row_top: float = build_button.position.y + build_button.size.y * k + HUD_ROW_GAP * k
 	var row_height: float = build_button.size.y
 	action_button.offset_top = row_top
 	action_button.offset_bottom = row_top + row_height
@@ -234,8 +261,8 @@ func _layout_context_buttons() -> void:
 	for i in range(buttons.size()):
 		var button: Control = buttons[i]
 		button.visible = true
-		button.offset_left = 188 + i * 164
-		button.offset_right = button.offset_left + 148
+		button.offset_left = (188.0 + float(i) * 164.0) * k
+		button.offset_right = button.offset_left + 148.0
 		button.offset_top = row_top
 		button.offset_bottom = row_top + row_height
 

core/GameState.gd

@@ -20,6 +20,10 @@ const CATALOG_CHECK = preload("res://core/CatalogCheck.gd")
 # survive the trip has to live here instead.
 
 var has_return_position: bool = false
+# Set by the menu when a save has just been loaded, cleared by whoever acts
+# on it - the same one-shot handoff as has_return_position above, and
+# deliberately NOT saved: it describes this launch, not the run.
+var resumed_from_save: bool = false
 var surviving_ship_count: int = STARTING_SHIP_COUNT
 var surviving_vessel_types: PackedStringArray = PackedStringArray()
 # #106: a mirror of what is flying, pushed by Fleet._sync_composition_state()

planets/PlanetAtmosphereController.gd

@@ -39,6 +39,23 @@ func _ready() -> void:
 		camera.position = GameState.return_world_position
 		camera.zoom = Vector2(GameState.return_camera_zoom, GameState.return_camera_zoom)
 		GameState.has_return_position = false
+	elif GameState.resumed_from_save:
+		# #227: a resumed run used to fall through to the brand-new-game
+		# branch below, which put the STARTING fleet at the world origin -
+		# so loading a save stranded the player in empty space with the
+		# wrong ships while their base sat wherever they had left it.
+		#
+		# The mothership is where a wiped-out fleet is already rebuilt (see
+		# BuildController._check_fleet_wipe), so it is the run's home whether
+		# or not this is a resume. Falls back to the position the save was
+		# written at, and only then to the origin, because a run can have no
+		# mothership standing.
+		var home: Variant = GameState.get_mothership_position()
+		var at: Vector2 = home if home != null else GameState.fleet_position
+		fleet.spawn_ships(GameState.fleet_vessel_types.size(), at,
+			GameState.fleet_vessel_types)
+		camera.position = at
+		GameState.resumed_from_save = false
 	elif GameState.intro_active:
 		# #108: the prologue opens with a convoy, and crews it itself. This
 		# node still runs first, so spawning the ordinary opening fleet here

tests/test_hangar_management.gd

@@ -83,8 +83,12 @@ func _process(delta: float) -> bool:
 		controller.transfer_button.visible = controller._has_transfer_context()
 		controller._layout_context_buttons()
 		var buttons: Array = [controller.building_action_button, controller.hangar_button]
+		# Rendered width, not the authored rect: #228 shrinks the HUD on a
+		# short canvas by scaling the node and leaving its rect alone, so
+		# offset_right overstates where a button actually ends.
 		for i in range(buttons.size() - 1):
-			if buttons[i].offset_right > buttons[i + 1].offset_left:
+			var ends: float = buttons[i].offset_left + buttons[i].size.x * buttons[i].scale.x
+			if ends > buttons[i + 1].offset_left:
 				_fail("context buttons %d and %d overlap" % [i, i + 1])
 				return true
 		controller._collapse_context_actions()

tests/test_hud_layout.gd

@@ -31,6 +31,9 @@ var controller: Node
 var readout: Node
 var stage: int = 0
 var elapsed: float = 0.0
+# A desktop canvas (1080 units) and a phone one (3413), where the HUD
+# scale is 0.45 and 1.0 respectively.
+var shapes: Array = [Vector2i(1920, 1080), Vector2i(720, 1280)]
 
 # Deliberately share a row: a building either offers its own action or a
 # hangar, never both, so their rects coinciding is the design.
@@ -52,23 +55,40 @@ func _process(delta: float) -> bool:
 				return false
 			scene.get_node("AsteroidField").set_process(false)
 			scene.get_node("EnemySpawner").set_process(false)
-			if not _check("as laid out at load"):
+			# --- E. at BOTH shapes ---------------------------------------
+			# #228 shrinks the HUD on a short canvas, so every relationship
+			# here now has a scale in it. A desktop canvas is what the
+			# default viewport gives, and the phone is where that scale is
+			# 1.0 - the arithmetic has to hold at both or one of them is
+			# being checked by accident.
+			root.size = shapes.pop_front()
+			stage = 1
+			return false
+		1:
+			if elapsed < 0.2:
+				return false
+			elapsed = 0.0
+			var label: String = "%.0f-unit canvas" % readout.get_viewport_rect().size.y
+			if not _check("as laid out at load, %s" % label):
 				return true
 
 			# --- C. the path that broke it ------------------------------
 			controller.context_actions_expanded = false
 			controller._layout_context_buttons()
-			if not _check("after _layout_context_buttons() collapsed"):
+			if not _check("collapsed, %s" % label):
 				return true
 
 			controller.context_actions_expanded = true
 			controller._layout_context_buttons()
-			if not _check("after _layout_context_buttons() expanded"):
+			if not _check("expanded, %s" % label):
 				return true
 
-			print("TEST_HUD_LAYOUT: PASS")
-			quit(0)
-			return true
+			if shapes.is_empty():
+				print("TEST_HUD_LAYOUT: PASS")
+				quit(0)
+				return true
+			root.size = shapes.pop_front()
+			return false
 	return false
 
 
@@ -110,13 +130,21 @@ func _check(context: String) -> bool:
 			var b: Control = buttons[b_name]
 			if a == null or b == null:
 				continue
-			if a.get_rect().intersects(b.get_rect()):
+			if _rendered(a).intersects(_rendered(b)):
 				return _fail("%s %s overlaps %s %s (%s)" \
-					% [a_name, a.get_rect(), b_name, b.get_rect(), context])
+					% [a_name, _rendered(a), b_name, _rendered(b), context])
 	print("clear of the readout (reaches %.0f) and of each other - %s" % [readout_bottom, context])
 	return true
 
 
+# Where a control actually lands, which since #228 is not its layout rect:
+# the HUD shrinks on a short canvas by scaling the node and leaving the rect
+# at its authored size, so position is in scaled parent units while size is
+# in unscaled local ones.
+func _rendered(c: Control) -> Rect2:
+	return Rect2(c.position, c.size * c.scale)
+
+
 func _is_exclusive(a: String, b: String) -> bool:
 	for pair in EXCLUSIVE_PAIRS:
 		if (a == pair[0] and b == pair[1]) or (a == pair[1] and b == pair[0]):

tests/test_hud_scales_to_screen.gd

@@ -0,0 +1,118 @@
+extends SceneTree
+
+# Forgejo #228: the HUD must not be three times larger on a PC than on a
+# phone.
+#
+# The layout is authored in CANVAS units. With stretch "canvas_items" and
+# aspect "expand" over a 1920x1080 base, a 720x1280 phone scales by 0.375 and
+# gets 1920x3413 canvas units, while ANY 16:9-or-wider screen lands on
+# 1920x1080 units however many pixels it really has. The canvas is three
+# times shorter, so an unscaled 148-unit button goes from 4.3% of the
+# screen's height to 13.7% of it.
+#
+# Nothing collides - that was checked pair by pair on both canvases and the
+# same six pairs overlap on each, all of them same-slot alternatives. The
+# fault is purely that everything is enormous.
+#
+# Covers:
+#   A. on a phone the HUD is left exactly as authored
+#   B. on a desktop canvas a control takes a similar FRACTION of the screen
+#      rather than three times more
+#   C. the shrink is bounded, so a landscape phone keeps thumb-sized targets
+#   D. it follows a resize rather than only applying once at startup
+
+const PHONE: Vector2i = Vector2i(720, 1280)
+const DESKTOP: Vector2i = Vector2i(1920, 1080)
+const BIG_DESKTOP: Vector2i = Vector2i(2560, 1440)
+
+var scene: Node
+var hud: Node
+var button: Control
+var frames: int = 0
+var stage: int = 0
+var phone_fraction: float = 0.0
+
+
+func _initialize() -> void:
+	var gs: Node = root.get_node("GameState")
+	gs.buildings.clear()
+	gs.reset_new_game()
+	gs.intro_active = false
+	scene = load("res://world/Main.tscn").instantiate()
+	root.add_child(scene)
+	current_scene = scene
+	hud = scene.get_node("UI/HudScale")
+	button = scene.get_node("UI/MapButton")
+
+
+func _fraction() -> float:
+	# Height of a HUD control as a share of the screen's height - the number
+	# that actually differs between the two, and the one a player sees.
+	# Rendered height, not the layout rect: the rect keeps its authored size
+	# and the node is scaled, so scale is where the shrink actually lives.
+	return (button.size.y * button.scale.y) / button.get_viewport_rect().size.y
+
+
+func _process(_delta: float) -> bool:
+	frames += 1
+	if frames < 10:
+		return false
+	frames = 0
+	match stage:
+		0:
+			root.size = PHONE
+			stage = 1
+			return false
+		1:
+			# --- A. authored size on the shape it was authored for -------
+			var k: float = hud.hud_scale()
+			if not is_equal_approx(k, 1.0):
+				return _fail("the phone HUD is being scaled by %.3f, but it is the authored shape" % k)
+			phone_fraction = _fraction()
+			print("phone %s: canvas %.0f tall, scale %.2f, control is %.1f%% of the screen" % [
+				PHONE, button.get_viewport_rect().size.y, k, phone_fraction * 100.0])
+			root.size = DESKTOP
+			stage = 2
+			return false
+		2:
+			# --- B. a similar share of a desktop screen ------------------
+			var pc_fraction: float = _fraction()
+			var ratio: float = pc_fraction / maxf(phone_fraction, 0.0001)
+			print("desktop %s: canvas %.0f tall, scale %.2f, control is %.1f%% of the screen (%.2fx the phone)" % [
+				DESKTOP, button.get_viewport_rect().size.y, hud.hud_scale(),
+				pc_fraction * 100.0, ratio])
+			if ratio > 1.75:
+				return _fail("a HUD control is %.2fx its phone share on a desktop screen - that is the bug" % ratio)
+			# --- C. and not shrunk into uselessness ----------------------
+			if hud.hud_scale() < 0.4:
+				return _fail("the HUD shrank to %.2f, past anything pressable" % hud.hud_scale())
+			root.size = BIG_DESKTOP
+			stage = 3
+			return false
+		3:
+			# --- D. follows the resize -----------------------------------
+			var big: float = _fraction()
+			print("big desktop %s: control is %.1f%% of the screen" % [BIG_DESKTOP, big * 100.0])
+			if absf(big - phone_fraction) / maxf(phone_fraction, 0.0001) > 0.75:
+				return _fail("a bigger desktop screen did not keep the control proportional (%.1f%% vs %.1f%%)" \
+					% [big * 100.0, phone_fraction * 100.0])
+			# Back to the phone: the HUD must return, not stay shrunk.
+			root.size = PHONE
+			stage = 4
+			return false
+		4:
+			if not is_equal_approx(hud.hud_scale(), 1.0):
+				return _fail("returning to the phone shape left the HUD at %.3f" % hud.hud_scale())
+			if absf(_fraction() - phone_fraction) > 0.001:
+				return _fail("the HUD did not come back to its authored size on resize")
+			print("resizing back to the phone restored the authored size")
+			print("TEST_HUD_SCALES_TO_SCREEN: PASS")
+			quit(0)
+			return true
+	return false
+
+
+func _fail(message: String) -> bool:
+	print("TEST_HUD_SCALES_TO_SCREEN: FAIL (%s)" % message)
+	quit(1)
+	return true

tests/test_resume_spawn.gd

@@ -0,0 +1,150 @@
+extends SceneTree
+
+# Forgejo #227: a resumed run comes back at its own mothership.
+#
+# Loading a save used to fall through to the brand-new-game branch in
+# PlanetAtmosphereController, which spawns STARTING_SHIP_COUNT interceptors
+# at the world origin. So a resume stranded the player in empty space with
+# the wrong ships, while their base sat wherever they had left it - which on
+# a charted map is thousands of units away.
+#
+# Covers:
+#   A. resuming puts the fleet at the mothership, not at the origin
+#   B. and crews it with the ships the save actually had
+#   C. with no mothership standing, it falls back to where the save was made
+#   D. a NEW game is untouched - still the starting fleet at the origin
+
+const HOME: Vector2 = Vector2(7400.0, -3100.0)
+const AWAY: Vector2 = Vector2(-5200.0, 900.0)
+
+var game_state: Node
+var scene: Node
+var elapsed: float = 0.0
+var stage: int = 0
+var expected_crew: int = 0
+
+
+func _initialize() -> void:
+	game_state = root.get_node("GameState")
+	SaveGame.clear()
+
+
+func _fresh_state() -> void:
+	game_state.buildings.clear()
+	game_state.extractors.clear()
+	game_state.reset_progression()
+	game_state.reset_new_game()
+	game_state.reset_quests()
+	game_state.intro_active = false
+
+
+func _load_scene() -> void:
+	if scene != null and is_instance_valid(scene):
+		scene.free()
+	scene = load("res://world/Main.tscn").instantiate()
+	root.add_child(scene)
+	current_scene = scene
+
+
+func _process(delta: float) -> bool:
+	elapsed += delta
+	match stage:
+		0:
+			if elapsed < 0.3:
+				return false
+			# A run with a base at HOME and the fleet parked far from it.
+			_fresh_state()
+			game_state.add_building(GameState.BUILDING_MOTHERSHIP, HOME)
+			# Capacity for the crew below. Since #219 the fleet holds
+			# BASE_FLEET_SIZE (1) plus one berth per supplied fleet office,
+			# so a save with three ships flying is a save that had two
+			# offices standing - without them spawn_ships clamps to 1 and
+			# this would assert nothing.
+			game_state.add_building(GameState.BUILDING_FLEET_OFFICE, HOME + Vector2(400.0, 0.0))
+			game_state.add_building(GameState.BUILDING_FLEET_OFFICE, HOME + Vector2(800.0, 0.0))
+			game_state.fleet_vessel_types = PackedStringArray([
+				GameState.VESSEL_INTERCEPTOR, GameState.VESSEL_INTERCEPTOR,
+				GameState.VESSEL_INTERCEPTOR])
+			game_state.fleet_position = AWAY
+			expected_crew = game_state.fleet_vessel_types.size()
+			if game_state.max_fleet_size() < expected_crew:
+				return _fail("the fleet cannot hold %d, so a resumed crew would prove nothing (cap %d)" \
+					% [expected_crew, game_state.max_fleet_size()])
+			if not SaveGame.save(game_state):
+				return _fail("could not write the save this test resumes from")
+			_fresh_state()
+			var result: Dictionary = SaveGame.load_into(game_state)
+			if not bool(result.get("ok", false)):
+				return _fail("the save would not load: %s" % result.get("reason", ""))
+			game_state.resumed_from_save = true
+			_load_scene()
+			elapsed = 0.0
+			stage = 1
+			return false
+		1:
+			if elapsed < 0.6:
+				return false
+			var fleet: Node2D = scene.get_node("Fleet")
+			# --- A. at the mothership, not the origin -------------------
+			var at: Vector2 = fleet.get_centroid()
+			if at.distance_to(HOME) > 400.0:
+				return _fail("resumed at %s, but the mothership is at %s" % [at, HOME])
+			if at.distance_to(Vector2.ZERO) < 400.0:
+				return _fail("resumed at the world origin, which is the new-game behaviour")
+			print("resumed at the mothership (%.0f units from it)" % at.distance_to(HOME))
+
+			# --- B. with the crew the save had --------------------------
+			if fleet.ships.size() != expected_crew:
+				return _fail("resumed with %d ships, the save had %d" \
+					% [fleet.ships.size(), expected_crew])
+			print("resumed with the saved crew of %d" % fleet.ships.size())
+			if game_state.resumed_from_save:
+				return _fail("the resume flag was left raised, so the next scene would re-use it")
+
+			# --- C. no mothership: fall back to where the save was made --
+			_fresh_state()
+			game_state.fleet_vessel_types = PackedStringArray([GameState.VESSEL_INTERCEPTOR])
+			game_state.fleet_position = AWAY
+			game_state.resumed_from_save = true
+			_load_scene()
+			elapsed = 0.0
+			stage = 2
+			return false
+		2:
+			if elapsed < 0.6:
+				return false
+			var fleet2: Node2D = scene.get_node("Fleet")
+			var at2: Vector2 = fleet2.get_centroid()
+			if at2.distance_to(AWAY) > 400.0:
+				return _fail("with no mothership, resumed at %s rather than the saved %s" % [at2, AWAY])
+			print("with no mothership, resumed where the save was written")
+
+			# --- D. a new game is untouched -----------------------------
+			_fresh_state()
+			game_state.resumed_from_save = false
+			_load_scene()
+			elapsed = 0.0
+			stage = 3
+			return false
+		3:
+			if elapsed < 0.6:
+				return false
+			var fleet3: Node2D = scene.get_node("Fleet")
+			if fleet3.get_centroid().distance_to(Vector2.ZERO) > 400.0:
+				return _fail("a new game no longer starts at the origin (%s)" % fleet3.get_centroid())
+			if fleet3.ships.size() != GameState.STARTING_SHIP_COUNT:
+				return _fail("a new game starts with %d ships, expected %d" \
+					% [fleet3.ships.size(), GameState.STARTING_SHIP_COUNT])
+			print("a new game still starts with %d ships at the origin" % fleet3.ships.size())
+
+			SaveGame.clear()
+			print("TEST_RESUME_SPAWN: PASS")
+			quit(0)
+			return true
+	return false
+
+
+func _fail(message: String) -> bool:
+	print("TEST_RESUME_SPAWN: FAIL (%s)" % message)
+	quit(1)
+	return true

ui/HudScale.gd

@@ -0,0 +1,78 @@
+extends Node
+
+# Forgejo #228: the HUD is proportionally three times too big on a PC.
+#
+# The layout is authored in CANVAS units against a portrait phone. With
+# stretch "canvas_items"/"expand" and a 1920x1080 base, a 720x1280 phone
+# scales by 0.375 and so gets a canvas 1920x3413 units tall - while ANY
+# 16:9-or-wider screen lands on 1920x1080 units whatever its pixel count.
+# The canvas is three times shorter, so a 148-unit button goes from 4.3% of
+# the screen's height to 13.7% of it. Nothing collides; everything is just
+# enormous, and the thumb sticks in particular.
+#
+# Each HUD element's offsets are measured FROM ITS OWN ANCHOR, so scaling
+# them scales the element's size and its distance from the edge it hangs off
+# together. That is the whole trick: one multiplier per control, no layout
+# rules to re-derive, and anchoring keeps working untouched.
+
+# The canvas the HUD was drawn for. Not a magic number: 1280 / (720 / 1920).
+const AUTHORED_CANVAS_HEIGHT: float = 3413.0
+
+# Below this the controls stop being worth pressing. A landscape phone gets
+# the same short canvas as a desktop, and there the sticks are still thumb
+# targets - so the shrink is bounded rather than purely proportional.
+const MIN_SCALE: float = 0.45
+
+# Authored offsets and sizes, captured before anything is touched.
+# Re-reading them after a resize would measure a previous scaling and let the
+# HUD ratchet away to nothing - the same trap NeonPanel documents.
+var _authored: Dictionary = {}
+
+
+func _ready() -> void:
+	var layer: Node = get_parent()
+	for child in layer.get_children():
+		if child == self or not (child is Control):
+			continue
+		var c: Control = child
+		# Full-bleed things - the modal menus, the breach overlay - hang off
+		# every edge at once and have nothing to shrink.
+		if is_equal_approx(c.anchor_left, 0.0) and is_equal_approx(c.anchor_right, 1.0) \
+				and is_equal_approx(c.anchor_top, 0.0) and is_equal_approx(c.anchor_bottom, 1.0):
+			continue
+		_authored[c] = [c.offset_left, c.offset_top,
+			c.offset_right - c.offset_left, c.offset_bottom - c.offset_top]
+	get_viewport().size_changed.connect(_apply)
+	_apply()
+
+
+func hud_scale() -> float:
+	var height: float = get_viewport().get_visible_rect().size.y
+	if height <= 0.0:
+		return 1.0
+	# Only ever shrinks. A canvas TALLER than the phone's is not a reason to
+	# inflate the HUD over the game.
+	return clampf(height / AUTHORED_CANVAS_HEIGHT, MIN_SCALE, 1.0)
+
+
+func _apply() -> void:
+	var k: float = hud_scale()
+	for control in _authored:
+		if not is_instance_valid(control):
+			continue
+		var a: Array = _authored[control]
+		var c: Control = control
+		# Position moves toward the edge the element hangs off; the RECT keeps
+		# its authored size and the node is scaled instead.
+		#
+		# Scaling the rect alone was wrong: half the HUD draws its own
+		# contents at fixed canvas sizes - ResourceReadout lays its rows out
+		# in code - so a shrunken rect kept full-size content inside it, and
+		# the build button slid up underneath the resource list. Scaling the
+		# node takes the drawn content with it.
+		c.pivot_offset = Vector2.ZERO
+		c.offset_left = float(a[0]) * k
+		c.offset_top = float(a[1]) * k
+		c.offset_right = c.offset_left + float(a[2])
+		c.offset_bottom = c.offset_top + float(a[3])
+		c.scale = Vector2(k, k)

ui/MainMenu.gd

@@ -140,6 +140,7 @@ func _on_continue_pressed() -> void:
 	# A resumed run never replays the prologue, whatever it was doing when
 	# it was written.
 	GameState.intro_active = false
+	GameState.resumed_from_save = true
 	_start("res://world/Main.tscn")
 
 

ui/ResourceReadout.gd

@@ -79,7 +79,13 @@ func rows_height() -> float:
 # Bottom edge of the rows in this Control's parent space - the line anything
 # stacked below the readout has to clear.
 func content_bottom() -> float:
-	return position.y + rows_height()
+	# Rendered, not laid out. #228 scales the HUD down on a short canvas by
+	# scaling the NODE while leaving its rect at the authored size, so
+	# rows_height() is in unscaled local units while position is already in
+	# scaled parent ones. Mixing the two overstated this by more than a
+	# button's height and pushed the columns down to clear space that was
+	# not occupied.
+	return position.y + rows_height() * scale.y
 
 
 func _add_glyph(resource_type: String, at: Vector2) -> void:

world/Main.tscn

@@ -17,6 +17,7 @@
 
 
 [ext_resource type="Script" path="res://world/Starfield.gd" id="1"]
+[ext_resource type="Script" path="res://ui/HudScale.gd" id="hud_scale"]
 [ext_resource type="Script" path="res://vessels/Fleet.gd" id="2"]
 [ext_resource type="Script" path="res://controllers/GameController.gd" id="3"]
 [ext_resource type="Script" path="res://enemies/EnemySpawner.gd" id="4"]
@@ -151,6 +152,9 @@ script = ExtResource("12")
 [node name="UI" type="CanvasLayer" parent="."]
 layer = 1
 
+[node name="HudScale" type="Node" parent="UI"]
+script = ExtResource("hud_scale")
+
 [node name="ResourceReadout" type="Control" parent="UI"]
 mouse_filter = 2
 offset_left = 28