diff --git a/PROJECT.md b/PROJECT.md
index 872c290..6b2c281 100644
--- a/PROJECT.md
+++ b/PROJECT.md
@@ -579,6 +579,24 @@ serve gesture, the duel and gore are untouched; keys unchanged. Verified with re
 held-swipe contact aims exactly `swayOf`, a 360 ms swipe (previously dead) swings and aims, and a
 release-before-arrival still aims through `takeFlickAim`.
 
+**...and an aimed shot keeps its DIRECTION** (2026-08-20, same report, second round — the crossing
+fire alone "did not feel much different", and his sharpened observation was that the flick only
+seemed to work when the opponent already stood where he aimed). Measured with the opponent
+teleported at contact: the aim never tracked the body — no magnet exists — but at rushed quality
+the SYMMETRIC scatter (±aimError·(1-q)) outweighed the quality-shrunk aim, so a mistimed left flick
+landed anywhere in a ~4 m blob around "slightly left": usually centre, sometimes right, and the eye
+anchors that on wherever the opponent stands. The opponent's position was a proxy for preparation
+time (a body parked wide = its return flew a long slow diagonal = time = quality). The dial:
+**`TUNE.aimTrust` — an OUTWARD scatter roll is untouched** (a committed mistimed swipe still sails
+at the exact tuned rate; `flickReach` stays the only out dial) **but an inward roll keeps only
+`aimTrust` (0.3) of itself**, so bad timing costs width, never the direction that was asked for.
+Player lane shots only: the opponent's symmetric scatter is its width mechanic, and gore's
+free-aim never used lanes. Verified in real rallies at worst-case q≈0.08: 14/14 rushed left flicks
+landed left (−0.8..−3.5), 9/9 right flicks right — previously the same condition was a coin toss.
+The suite pins the direction guarantee and the knob relationship (`aimTrust·aimError` must stay
+under a committed swipe's worst-case aim). Whether placement should also BEAT an unwounded body
+more often (`aiRead` on wide balls — "dial 2") is open, awaiting the phone verdict on this one.
+
 **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 06d6b67..738a0e0 100644
--- a/index.html
+++ b/index.html
@@ -659,6 +659,6 @@ if (location.hostname !== 'splatter-tennis.wuwi.one') {
     </div>
   </div>
 </div>
-<script type="module" src="src/main.js?v=290"></script>
+<script type="module" src="src/main.js?v=291"></script>
 </body>
 </html>
diff --git a/src/config.js b/src/config.js
index c08011a..82555a4 100644
--- a/src/config.js
+++ b/src/config.js
@@ -64,6 +64,14 @@ export const TUNE = {
                            // send the ball longer — solveShot still lands it on the same spot — it
                            // gets there flatter and sooner, which is what a smash is.
   aimError: 2.4,           // m of scatter at quality 0
+  aimTrust: 0.3,           // an aimed rally shot TRUSTS its flick: the scatter may still push the
+                           // ball further OUT (the chosen out-risk keeps its exact shape) but only
+                           // this fraction of an INWARD roll survives, so a mistimed left flick
+                           // lands SHORTER LEFT rather than center-or-right "wherever the dice
+                           // said". Player's lane shots only — the opponent's symmetric scatter is
+                           // its width mechanic, and the gore free-aim never used lanes. Must stay
+                           // under 0.5*flickWidth*flickReach/aimError (~0.43) or a committed
+                           // mistimed swipe can cross back past centre (pinned by the suite).
   // NOT `aimReach` — that name is taken further down by the slingshot's aim-ray LENGTH (22 m), and
   // the duplicate key silently won: every aimed shot reached 22x its width until a test caught it.
   flickWidth: 3.95,        // m off centre a PERFECTLY timed flick reaches. The singles line is 4.115,
diff --git a/src/gameplay.js b/src/gameplay.js
index 47fea10..7c02991 100644
--- a/src/gameplay.js
+++ b/src/gameplay.js
@@ -178,7 +178,18 @@ function doHit(ch, quality, aim) {
     // it just stops being a coin toss you are punished for calling.
     const reach = ch.side === 'P' ? TUNE.flickReach + (1 - TUNE.flickReach) * quality : 1;
     sway = swayOf(aim);   // -1..1, continuous on touch — see swayOf
-    tx = sway * (toFar ? TUNE.flickWidth : 2.7) * reach + err();
+    // AN AIMED SHOT KEEPS ITS DIRECTION. The scatter used to be symmetric, and at rushed quality
+    // it was BIGGER than the quality-shrunk aim — a mistimed left flick landed anywhere in a 4 m
+    // blob around "slightly left", i.e. usually centre and sometimes right. It read as the ball
+    // landing wherever the opponent stood (2026-08-20, "trying to move the approaching ball with
+    // a flick to the left is almost impossible"). An OUTWARD roll is untouched — the committed
+    // swipe's chance of sailing wide is a chosen risk with a tuned shape (flickReach) and this
+    // must not soften it — but an inward roll now keeps only TUNE.aimTrust of itself, so bad
+    // timing costs WIDTH, never the direction that was asked for. Player only: the opponent's
+    // symmetric scatter is its width mechanic (matchWide clamps it, not this).
+    let e = err();
+    if (ch.side === 'P' && sway && e * Math.sign(sway) < 0) e *= TUNE.aimTrust;
+    tx = sway * (toFar ? TUNE.flickWidth : 2.7) * reach + e;
     const tzBase = aim.az < 0 ? SHOT_DEPTH.deep : aim.az > 0 ? SHOT_DEPTH.short : SHOT_DEPTH.mid;
     tz = (toFar ? tzBase : COURT.L - tzBase) + err() * 0.8;
     // A FORWARD FLICK IS A SMASH. Up-screen is the far court, so flicking INTO the court asked for
diff --git a/test.js b/test.js
index 292e0d6..cbab5f7 100644
--- a/test.js
+++ b/test.js
@@ -852,6 +852,30 @@ test('a gentle flick cannot be played out — only a committed one can', () => {
   assert.ok(safe > 0.4 && safe < 1, 'the ring turns hot inside the swipe range, not at either end');
 });
 
+test('an aimed shot keeps its DIRECTION: scatter may push it wider, never back past centre', () => {
+  // Dial 1 (2026-08-20). At rushed quality the symmetric scatter outweighed the quality-shrunk
+  // aim, so a mistimed left flick landed "wherever" — usually centre, read as the ball following
+  // the opponent. The model mirrors doHit's lane path: outward rolls untouched (the chosen
+  // out-risk keeps its exact tuned shape), inward rolls keep only aimTrust of themselves.
+  const land = (s, q, u) => {   // u in [-1,1]: the raw scatter roll
+    const reach = TUNE.flickReach + (1 - TUNE.flickReach) * q;
+    let e = u * TUNE.aimError * (1 - q);
+    if (s && e * Math.sign(s) < 0) e *= TUNE.aimTrust;
+    return s * TUNE.flickWidth * reach + e;
+  };
+  const line = COURT.W / 2;
+  for (let q = 0; q <= 1; q += 0.1) {
+    assert.ok(land(0.5, q, -1) > 0, 'a committed right swipe stays right at q=' + q.toFixed(1)
+      + ': ' + land(0.5, q, -1).toFixed(2));
+    assert.ok(land(-0.5, q, 1) < 0, '...and a committed left swipe stays left, mirrored');
+  }
+  assert.ok(land(1, 0, 1) > line, 'the outward tail is untouched: a mistimed full swipe can still sail');
+  assert.ok(land(1, 1, 1) < line, 'and a perfectly timed one still never does');
+  // the knob relationship the direction guarantee rests on — retuning must keep it
+  assert.ok(TUNE.aimTrust * TUNE.aimError < 0.5 * TUNE.flickWidth * TUNE.flickReach,
+    'aimTrust*aimError must stay under a committed swipe\'s worst-case aim');
+});
+
 test('the flick aims from the LIVE swipe: contact reads G.swipe, no release needed', () => {
   // The swing fires the moment the drag crosses flickMin now (input.js), so at contact the finger
   // is usually still down and no released flickAim exists yet. The aim has to come from the live