diff --git a/PROJECT.md b/PROJECT.md
index 180f210..872c290 100644
--- a/PROJECT.md
+++ b/PROJECT.md
@@ -552,12 +552,33 @@ The threshold is not a flip: the colour is already gold before you reach it, so
something you watch coming. **It is OFF as of its first playtest** (*"hide the indicator for now"*)
— `TUNE.aimCloud` gates the whole block and the tuning panel can flip it back live. Kept rather than
deleted: it is the only thing that shows the gesture's strength before the ball leaves, and the
-lesson it teaches has no other teacher. Only inside the window where a release would still BE a flick (`FLICK_MS`/`TUNE.flickMin`, the same
-gate `endTouch` uses) — past that the gesture is a run and a target hanging off it would be a lie.
-Rally modes only (`rallyFeel`); gore's flick free-aims at a body, not at a lane. The placement ring
+lesson it teaches has no other teacher. Shown from `TUNE.flickMin` for as long as the drag lives —
+since the crossing-fired flick (below) the vector stays live until contact, so there is no release
+deadline for it to lie past. Rally modes only (`rallyFeel`); gore's flick free-aims at a body, not
+at a lane. The placement ring
turns `UI.bad` when the shot really is out, so the two rings together ARE the lesson: what you asked
for, what your timing gave you, and the difference between them.
+**...and the flick fires the moment it IS a flick** (2026-08-20, Oliver: *"trying to move the
+approaching ball with a flick to the left is almost impossible ... standing closer to the grundlinie
+and more right from the center eases this restriction"*). The aim math was symmetric to the
+centimetre — the defect was LATENCY. The swing waited for the RELEASE, so the whole gesture ran as
+dead time on top of the 60 ms windup, and ~120 ms of lateness drops contact quality 1.0 -> 0.08,
+where `flickReach` halves the width and `aimError` drowns what is left. A LEFT swipe is the slowest
+direction a right thumb has (~250-350 ms vs ~150 for right), the service line leaves only ~700-870 ms
+of budget against even the slowest rung (~1.1-1.3 s at the baseline — measured), and any swipe over
+`FLICK_MS` (300 ms) was silently DEAD: no swing, no feedback, ball passes. Left + service line was
+exactly the slowest gesture meeting the shortest budget. Two changes, rally modes only
+(`rallyFeel`), input.js + one line in `trySwing`: **the swing fires when the drag crosses
+`flickMin`** (in `G.phase === 'rally'`, once per gesture) rather than at release, and **the aim
+stays LIVE** — `trySwing` reads `G.swipe` at contact, so the ball takes the vector under the thumb
+NOW; the released `flickAim` only carries it when the ball arrives after the finger lifted. The
+FLICK_MS deadline is gone with it where the crossing can fire — it survives only where it cannot
+(gore's free-aim flick, the serve-phase strike), where a late release must stay a rescue drag. The
+serve gesture, the duel and gore are untouched; keys unchanged. Verified with real touch events:
+held-swipe contact aims exactly `swayOf`, a 360 ms swipe (previously dead) swings and aims, and a
+release-before-arrival still aims through `takeFlickAim`.
+
**The umpire and the boy.** `aside()` refused during any interlude and the match runs on a permanent
one, so he was silently muted for the entire mode — the one mode that is literally a tennis match he
is sitting over (`G.interlude && !G.match` now, the same shape as `sendBoy`). Six dry lines: two on
diff --git a/index.html b/index.html
index fc3cbb9..06d6b67 100644
--- a/index.html
+++ b/index.html
@@ -659,6 +659,6 @@ if (location.hostname !== 'splatter-tennis.wuwi.one') {
-
+