Draw the whole serve right-handed, and let the pull wind it up

The racket kept changing hands because the art disagreed with itself: p_idle,
p_bounce and p_toss held it in his LEFT hand while p_smash and p_swing used his
right, so every new frame copied whichever reference it was handed. The idle, the
bounce pair, the toss and the four wind-up frames are all regenerated with the
side named — drawn from behind, his right hand is the right-hand side of the
image, and only the racket HEAD crosses behind his back.

What separates the four wind-up stages is how far the racket has fallen, and that
needed the racket head pinned to a body landmark per frame — above his head,
beside his ear, between his shoulder blades, at the back of his thighs. Described
as "further into it each time" it drew four near-identical poses twice running.
They are wired late in the pull, Oliver's call: frame 1 owns the first half and
the three coiled frames are packed into the top half.

assets/_derive.py is deleted. Fixing generated art by mangling it afterwards is
always worse than regenerating; three guards in _regen.py turn the bad rolls that
used to install silently into automatic re-rolls — a portrait reference canvas
(the generator answered with a 2x2 grid), a non-magenta background (a green sheet
keyed away four racket heads), and a mirrored frame.

The bounce pair is one drawing from one sheet now, differing only in the free
hand, so BOUNCE_IDLE moved to his left side and was re-measured off the sprite.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

HANDOFF.md

@@ -409,41 +409,66 @@ Known generator quirks, all still true:
 - **Colour-match any regenerated frame** of an existing pair before shipping: fit a per-channel
   linear map onto the original's mean and sd, then verify with a mean-difference number.
 
-## Session log: the serve is the gesture (2026-07-31, after the art pass)
+## Session log: the serve is the gesture (2026-08-01)
 
 Oliver's notes on the art pass: the zombies are good, the runner "flickers very hard", and the
-player has three problems. All four fixed, all four measured first.
+player has three problems. The serve is now a drawn wind-up driven by the pull.
 
-- **The pull IS the wind-up.** Four new frames (`p_pull1..4`, `assets/_regen_batches.json` batch
-  `player_pull`) — racket above the head, behind the head, behind the back, fully coiled — chosen in
-  `drawChar` off `pullFrac(G.aim.mag)`, the same 0..1 that sets the shot's power. Drag further, coil
-  further; release fires `p_smash`. Verified per stage with drawImage instrumentation:
-  mag 10/50/90/130 -> pull1/pull2/pull3/pull4.
-- **The racket must stay in ONE hand across a whole sheet, and the generator will not do it unasked.**
-  Oliver, straight off the file browser: *"the racket changes the hand in 3 and 4"*. It did — frames
-  1-2 came back right-handed and 3-4 left-handed. Both mirrored into line (a back-view figure mirrors
-  for free) and the batch prompt now names the side explicitly. The convention the rest of the art
-  uses: **racket on the LEFT-HAND SIDE OF THE IMAGE, free arm on the RIGHT** — which is also the side
-  the game drops the bounced ball (`BOUNCE_IDLE.dx` > 0) and tosses the served one (`P.x + 0.35`).
-  `p_smash` is the deliberate exception: the racket comes over to the right to meet the ball it is
-  actually hitting.
-- **The bounce pair is now ONE drawing.** `p_bounce2` is `p_bounce` with the free arm sheared up 12
-  px (0.10 m) — everything else is pixel-identical, so the swap cannot flicker. The swap follows the
-  BALL now (`|sin| > 0.5`) instead of its rising phase, and `BOUNCE_IDLE` was re-measured off the
-  sprite (`apex` 0.55 -> 0.6, `dx` 0.5 -> 0.43) so the ball tops out under the raised palm instead of
-  passing beside it.
-- **`p_toss` had the tossed ball drawn into it.** Cut, without re-cropping the canvas: the game
-  scales him off `img.height`, so trimming the dead space would have shrunk him. Still 100x308.
-- **The runner.** Two measured defects, both in the PNGs, no code change. Its four frames were
-  bbox-cropped and drawn centred on the body's x, so hip centres sat -2.0/+7.1/-6.0/-0.1 px off
-  centre — a 13 px lateral jump at 12 frames a second, which is the flicker. And Oliver's own
-  diagnosis was right: all four frames leaned the same way and lifted the SAME leg, so the cycle
-  never used his other side. Frames 3 and 4 are mirrored (he is front-on, so a mirror costs nothing)
-  and all four are padded to one width with the hip on the centre line.
+- **The pull IS the wind-up.** Four new frames (`p_pull1..4`, batch `player_windup`), chosen in
+  `drawChar` off `pullFrac(G.aim.mag)`. What separates the four is HOW FAR THE RACKET HAS FALLEN, and
+  that only came out once the prompt pinned the racket head to a different BODY LANDMARK per frame —
+  above his head, beside his ear, between his shoulder blades, down at the back of his thighs.
+  Described as "further into the wind-up each time" it drew four near-identical poses twice running
+  (Oliver: *"pull3 and pull4 are now basically the same image"*), which is the generator quirk the
+  pipeline notes already warned about, in a place I did not expect it. The stages are NOT spread evenly: **frame 1 owns the whole
+  first half of the pull and the three coiled frames are packed into the top half**, Oliver's call,
+  so the part of the gesture that costs power is the part that visibly costs effort. Release fires
+  `p_smash`. Verified: 30/74/95/125 px of drag -> pull1/2/3/4.
+- **The whole player is RIGHT-HANDED now, and that was the real bug.** `p_idle`, `p_bounce` and
+  `p_toss` had been drawn holding the racket in his LEFT hand while `p_smash` and `p_swing` used his
+  right, so any new frame copied whichever reference it was given and the racket changed hands
+  mid-serve. Everything from the idle to the wind-up was regenerated with the hand named explicitly:
+  **drawn from behind, his right hand is the RIGHT-HAND SIDE OF THE IMAGE**. Only the racket HEAD
+  crosses behind his back in the deep frames — the arm never does.
+- **The bounce pair is one drawing from one sheet**, differing only in the free hand, so the swap
+  cannot flicker; it follows the BALL (`|sin| > 0.5`) rather than its rising phase. He pats the ball
+  with his LEFT hand now, so `BOUNCE_IDLE` moved to the other side and was re-measured off the sprite
+  (`dx` 0.36, `apex` 0.68). The served toss moved to `+0.12` — nearly over his head, because the
+  tossing hand is on the left and the racket meets it on the right, and a big offset either way is
+  wrong for one of those two frames.
+- **The runner** still carries a hand-mirrored fix from the previous commit (see OPEN) — its frames
+  need regenerating, not mangling.
 
-Not done, and the honest reason: the runner's cadence is still time-driven (12 fps) rather than
-stepped off distance covered like every other body. It did not read as wrong once the frames stopped
-jumping, and it is one line if a playtest says otherwise.
+**THE RULE, from Oliver, that this session existed to learn:** *"fixing generated art by mangling
+with it afterwards is always worse than regenerating. it's just a fact."* Mirroring a frame,
+shearing a limb, cutting a prop out — a wrong frame is wrong in more ways than the one you spotted
+(perspective, foreshortening, where the weight sits) and none of those survive a flip. There was an
+`assets/_derive.py` doing exactly that; it is deleted. Fix the PROMPT and re-run the batch.
+
+**Three pipeline guards were added to `_regen.py`, each after a bad sheet installed silently:**
+1. **The reference's ASPECT is the answer's aspect.** One tall reference figure gave a portrait
+   reference canvas and the generator answered with a 2x2 GRID, which `_fine.py` cannot cut (it
+   segments columns). `ref_sheet` now widens to a landscape strip with one slot per frame asked for.
+2. **A non-magenta background is a rejected roll.** "It sometimes ignores the magenta" was written
+   off as harmless because the key uses the corner colour — until it returned a GREEN sheet and keyed
+   away four grey-green racket heads with it, installing four men holding a grip and nothing else.
+3. **`arm_left` rejects a MIRRORED frame.** The generator flips the later frames of a serve about
+   half the time, which puts the racket in the other hand halfway through the animation. Oliver
+   caught it twice by eye before the check existed. **It is a net, not a proof**: it asks whether
+   anything reaches the top of the LEFT half, and a mirrored frame whose racket head hangs over there
+   passes. A tighter discriminator was tried (ink mass in the top third, left vs right) and does NOT
+   separate the two — measured 0.61/0.57/0.41 on a mirrored sheet against 0.54/0.56/0.55 on a good
+   one. Every wind-up sheet still has to be looked at.
+
+Still imperfect, and worth a look on the phone: the empty tossing arm WANDERS between the four
+wind-up frames (up, out, bent, up) instead of staying thrown up at the ball. The roll that held the
+arm still put the racket back in the wrong hand, and the racket is the load-bearing half of the
+gesture, so this is the set that shipped. One more roll may fix both.
+
+Also worth knowing: a batch of raised-arm poses has no honest scale to anchor on, so frame 0 is
+always a throwaway STANDING frame (`p_pull0`, deleted after installing). And a photo reference of a
+real serve is what finally got the poses right — Oliver supplied a 10-frame sequence, and the four
+stages are its frames 3, 4, 5 and 6, the last being maximum tension before release.
 
 ## Session log: the art-resolution pass (2026-07-31)
 
@@ -582,12 +607,19 @@ Nothing here is broken; these are known gaps and unverified risks.
 8. **A chapter can be skipped.** `G.bossDue` takes the max, so going a long stretch without ever
    clearing the court means you face the current chapter, not a backlog. Deliberate; noted in case it
    ever reads as a bug.
-9. **Step 6 stays PARKED** by Oliver's decision: the crowd turning into zombies (spawning from the
+9. **The runner's four frames were fixed BY HAND and must be regenerated.** `z_f1..4` are currently
+   hip-registered (padded to one width with the hip on the centre line) and frames 3 and 4 are
+   mirrored copies, because all four were drawn leaning the same way and lifting the same leg. That
+   is exactly the post-processing Oliver ruled out — it shipped before he stated the rule. The batch
+   to write asks for a four-frame front-on sprint whose stride and lean ALTERNATE, and the numbers to
+   check afterwards are the per-frame hip centroid (it was 13 px of lateral jump at 12 frames a
+   second) and which leg leads.
+10. **Step 6 stays PARKED** by Oliver's decision: the crowd turning into zombies (spawning from the
    stands — attacks from behind, high spectacle, `x_crowd` sheets already exist) and the ball-boy zombie
    that catches your shots. Recorded so the ideas are not lost, not as a to-do.
-10. **`ump_no_deal` does not exist.** Oliver asked for its removal; `ump_no_appeal` was removed as the
+11. **`ump_no_deal` does not exist.** Oliver asked for its removal; `ump_no_appeal` was removed as the
     closest match. Worth one confirmation that no other line is still unwanted.
-11. **Four sounds are missing and need RESEARCH first** (find/choose the source before implementing —
+12. **Four sounds are missing and need RESEARCH first** (find/choose the source before implementing —
     every existing sfx is either synthesised in `audio.js` or a wav in `assets/audio/`):
     - the **ball machine** while it moves — a squeaking wheel/castor, so the furniture announces
       itself. It already telegraphs its shot (`machWind` 0.9 s); this is the *approach* cue.