Wide viewports (>= 960 px, leftover margins fitting a mini) place
opponents alternating left and right of the field, stacking extra
rows upward. Narrow screens keep the existing row above the field.
Tim's 'daalsnelheid' referred only to how flying clusters float back
down after a jump, including the apex ease (scaled with the halved
terminal speed). Restore spawn fall, unsupported/released singles and
dropped cargo to the original fallSpeed. Manual jumps also descend at
full speed. Cargo that is part of a descending cluster still rides
with it. Planet ratios and upward launch are unchanged.
Document the progression schedule, bot skill, +50% minimum jump,
halved descent speeds, additive +25% landing-chain bonus, visual-only
swap tween, and the 10/13 grounded-row red pulse (flying blocks do
not count).
Add the Offline Strijd single-player mode: local bots with the same
shot/miss mechanics as multiplayer, a 1-then-N-for-(N-1) opponent
schedule, skill from 5% rising 1% per round, and a HUD plus final
screen. Bot AI rolls every 0.5s and may make a vertical match-making
move. Raise the minimum jump height by 50% and halve every descent
speed, keeping planet ratios. Landing a cluster into a match adds a
stacking +25% jump bonus. Block swaps tween visually in 100 ms with
easeOutCubic. Columns with at least 10 grounded rows pulse red on the
player field and on opponent/bot minis.
Unit tests for t1111 (push column, attach at bottom, successive shots,
two launched blocks). A reproduction of the video column (repeated
launches under a flying cluster plus falling blocks). A randomized
soak of thousands of ticks across several seeds, asserting no overlap,
no illegal colour change and no vanish without a score. Falling-below-
a-gap and hanging-chain tests for the t1112 stackTop and
releaseUnsupported fixes. makeGame enables strictInvariants by default.
A flying cluster is one rigid body with a single anchorY/vy and integer
cell offsets, so members stay grid-aligned. A block shot into a cluster
pushes the hit column up by one cell and attaches at the bottom, and
the same rule applies to successive shots (t1111). Thrown blocks collide
with each other the same way. Collisions are swept between every pair
(launched vs cluster, launched vs launched, falling vs flying, cluster
vs cluster).
stackTop no longer ignores gaps, so falling blocks cannot teleport up
onto a hanging still. releaseUnsupported repeats until a hanging chain
is stable. An invariant checker (no overlap, no duplicate cells) runs
at the end of every tick in tests and under ?debug.
Adds the poisoned-pointer plus correct-touch regression that failed
before the t1106 fix, unit and Playwright coverage for under-flyer and
landed-cluster drags, thrown-falling merges, cell-corner grabs (mouse
and touch), mixed landing matches, and a 3x flame size check with a
screenshot.
Still stacks under a flying cluster are draggable; the flyer itself is
not. Landed clusters can be rearranged, and swapping in or out updates
membership (t1107). A thrown block collides with falling blocks the same
way two flying clusters do: they merge with velocity averaged by count
and cannot pass through each other (t1108). Grabbing uses the grid cell
under the pointer, with a half-tile fallback for mid-animation blocks
(t1109). Coloured cargo that lands into a mixed row with resting blocks
launches immediately; greys wait until they recolor (t1110).
On a real phone, PointerEvent coordinates can be wrong while TouchEvent
changedTouches stay correct, so a one-tile drag walked the whole column.
While a touchstart is live, Touch Events lead and pointer move/up/cancel
are ignored. Document-level non-passive touch listeners prevent scrolling,
setPointerCapture is gone, cancel ends without launching, and layout
freezes mid-drag. ?debug=drag shows an overlay.
Rocket flames are 3x as large in width and length, extending downward
under the block (t1110).