diff --git a/components/primitives/EscenaDialogo.tsx b/components/primitives/EscenaDialogo.tsx
index 36a2c54..7fbccb1 100644
--- a/components/primitives/EscenaDialogo.tsx
+++ b/components/primitives/EscenaDialogo.tsx
@@ -3,19 +3,23 @@
 import { motion, useTransform, type MotionValue } from 'framer-motion';
 import { Wordmark } from '@/components/primitives/Wordmark';
 
-// One beat of the pinned conversation, staged as two calling cards laid on a
-// table: the reader's, in white, and creeer's, in navy, handed back.
+// One beat of the pinned conversation. Not a card on a page — the whole frame
+// is the message: the reader's situation set large on the light ground, then
+// creeer's answer set large on azul-profundo, the section's own background
+// turning over with it.
+//
+// One thing per frame, never both. The two halves share a grid cell so the
+// answer takes the stage the question leaves instead of stacking under it, and
+// they arrive from opposite sides so the swap reads as a relay.
 //
 // The cards are driven by STATE, not by scroll position. Mapping opacity
 // straight onto scroll made every intermediate position a valid frame, so
-// stopping mid-swipe left creeer's card as a half-transparent grey rectangle —
-// it read as a rendering fault. Scroll now only decides which beat we are on;
-// the cards animate to that beat over time and always land somewhere finished,
-// wherever the reader lets go.
+// stopping mid-swipe left the answer as a half-transparent slab that read as a
+// rendering fault. Scroll now only picks the beat; the frame animates to it
+// over time and always lands somewhere finished, wherever the reader lets go.
 //
-// Every scene stays mounted and stacked in the same place. Mounting only the
-// active one would animate more cheaply and would strip five sixths of the
-// section out of the static export.
+// Every scene stays mounted. Rendering only the active one would animate more
+// cheaply and would strip five sixths of the section out of the static export.
 
 const EASE = [0.22, 1, 0.36, 1] as const;
 
@@ -44,66 +48,63 @@ export function EscenaDialogo({
   // The one thing still tied to scroll. A drifting numeral never looks broken
   // at any position it can be caught in, so it costs nothing and keeps the
   // stage from feeling frozen between beats.
-  const cifraY = useTransform(progreso, [i * paso - paso, (i + 1) * paso + paso], [60, -60]);
+  const cifraY = useTransform(progreso, [i * paso - paso, (i + 1) * paso + paso], [70, -70]);
 
   return (
     <motion.li
       data-activa={activa ? 'true' : 'false'}
-      animate={{ opacity: activa ? 1 : 0, y: activa ? 0 : 26 }}
-      transition={{ duration: 0.45, ease: EASE }}
-      className="absolute inset-x-0 top-1/2 -translate-y-1/2"
+      animate={{ opacity: activa ? 1 : 0 }}
+      transition={{ duration: 0.4, ease: EASE }}
+      className="absolute inset-0 flex items-center"
       style={{ pointerEvents: activa ? 'auto' : 'none' }}
     >
       <motion.span
         aria-hidden="true"
         style={{ y: cifraY }}
-        className="pointer-events-none absolute -top-32 right-0 select-none font-sans font-black leading-none text-azul-profundo/[0.045] text-[9rem] md:text-[17rem]"
+        animate={{ color: respondida ? 'rgba(255,255,255,0.06)' : 'rgba(10,29,61,0.05)' }}
+        transition={{ duration: 0.5, ease: EASE }}
+        className="pointer-events-none absolute right-0 top-8 select-none font-sans font-black leading-none text-[11rem] md:text-[22rem]"
       >
         {n}
       </motion.span>
 
-      {/* One card per frame — never both. Both live in the same grid cell so
-          the second can take the stage the first leaves, instead of stacking
-          under it: the question is asked, it goes, and then the answer is the
-          only thing on screen. They come in from opposite sides so the swap
-          reads as one card replacing another, not as one fading into the same
-          spot. */}
-      <div className="grid">
+      <div className="relative grid w-full">
+        {/* The reader's situation. */}
         <motion.div
-          animate={{
-            opacity: respondida ? 0 : 1,
-            y: respondida ? -52 : 0,
-            rotate: respondida ? -3.4 : -0.8,
-            scale: respondida ? 0.96 : 1,
-          }}
+          animate={{ opacity: respondida ? 0 : 1, y: respondida ? -60 : 0 }}
           transition={{ duration: 0.5, ease: EASE }}
           style={{ pointerEvents: respondida ? 'none' : 'auto' }}
-          className="col-start-1 row-start-1 self-center justify-self-start max-w-xl rounded-xl border border-azul-profundo/10 bg-white p-7 shadow-[0_28px_70px_-32px_rgba(10,29,61,0.5)] md:p-9"
+          className="col-start-1 row-start-1 max-w-5xl"
         >
-          <p className="text-[0.65rem] font-black uppercase tracking-[0.34em] text-dorado">
+          <p className="text-[0.7rem] font-black uppercase tracking-[0.34em] text-dorado md:text-xs">
             {n} · {name}
           </p>
-          <p className="mt-4 font-sans font-black leading-[1.12] text-azul-profundo text-2xl md:text-4xl">
+          <p className="mt-6 font-sans font-black leading-[0.98] text-azul-profundo text-4xl md:text-6xl lg:text-7xl">
             «{situacion}»
           </p>
         </motion.div>
 
-        {/* creeer's card, handed back: navy against the reader's white, tilted
-            the other way, and coming in from the opposite margin. */}
+        {/* creeer's answer, on the navy the whole frame turns to. The point
+            label repeats here on purpose: with the question gone from the
+            screen, it is the only thing tying the answer to what it answers. */}
         <motion.div
-          animate={{
-            opacity: respondida ? 1 : 0,
-            y: respondida ? 0 : 64,
-            rotate: respondida ? 1.6 : 5,
-            scale: respondida ? 1 : 0.95,
-          }}
+          animate={{ opacity: respondida ? 1 : 0, y: respondida ? 0 : 60 }}
           transition={{ duration: 0.55, ease: EASE }}
           style={{ pointerEvents: respondida ? 'auto' : 'none' }}
-          className="col-start-1 row-start-1 self-center justify-self-end max-w-xl rounded-xl bg-azul-profundo p-7 shadow-[0_34px_80px_-30px_rgba(10,29,61,0.75)] md:mr-6 md:p-9"
+          className="col-start-1 row-start-1 max-w-5xl"
         >
-          <span aria-hidden="true" className="absolute left-0 top-7 h-10 w-[3px] rounded-r bg-dorado md:top-9" />
-          <Wordmark size="sm" onDark />
-          <p className="mt-4 text-base leading-relaxed text-white/75 md:text-xl">{respuesta}</p>
+          <div className="flex items-center gap-4">
+            <span aria-hidden="true" className="h-[3px] w-10 rounded bg-dorado" />
+            <p className="text-[0.7rem] font-black uppercase tracking-[0.34em] text-dorado md:text-xs">
+              {n} · {name}
+            </p>
+          </div>
+          <p className="mt-6 font-sans font-bold leading-[1.06] text-white text-3xl md:text-5xl lg:text-[3.6rem]">
+            {respuesta}
+          </p>
+          <div className="mt-10">
+            <Wordmark size="sm" onDark />
+          </div>
         </motion.div>
       </div>
     </motion.li>
diff --git a/components/sections/PuntosInflexion.tsx b/components/sections/PuntosInflexion.tsx
index 31767b8..377cd75 100644
--- a/components/sections/PuntosInflexion.tsx
+++ b/components/sections/PuntosInflexion.tsx
@@ -111,11 +111,21 @@ function Escenificado() {
 
       {/* One viewport of stage, six viewports of scroll to move through it. */}
       <div ref={contenedor} className="relative" style={{ height: `${TOTAL * 100}vh` }}>
-        <div data-escena={activo} data-tiempo={tiempo} className="sticky top-0 h-screen overflow-hidden">
+        <div data-escena={activo} data-tiempo={tiempo} className="sticky top-0 h-screen overflow-hidden bg-gris-claro">
           <motion.div style={{ y: fondoY }} className="absolute inset-0">
             <FondoPuntos className="-bottom-10 -right-16 w-80 md:w-[34rem] text-azul-profundo/10" distance={32} />
           </motion.div>
 
+          {/* The frame itself turns over on the answer. The whole stage goes to
+              azul-profundo, not a panel inside it — the answer is not a card on
+              the reader's page, it is creeer's ground. */}
+          <motion.div
+            aria-hidden="true"
+            animate={{ opacity: respondida ? 1 : 0 }}
+            transition={{ duration: 0.55, ease: [0.22, 1, 0.36, 1] }}
+            className="absolute inset-0 bg-azul-profundo"
+          />
+
           <Container className="relative h-full">
             {/* left-6 / md:left-12, not left-0: an absolutely positioned child
                 resolves against Container's padding box, so `left-0` lands on
@@ -133,7 +143,13 @@ function Escenificado() {
                 <li key={item.name}>
                   <span
                     className={`block h-px origin-right transition-all duration-500 ${
-                      i === activo ? 'w-10 bg-dorado' : i < activo ? 'w-6 bg-dorado/40' : 'w-4 bg-azul-profundo/20'
+                      i === activo
+                        ? 'w-10 bg-dorado'
+                        : i < activo
+                          ? 'w-6 bg-dorado/40'
+                          : respondida
+                            ? 'w-4 bg-white/25'
+                            : 'w-4 bg-azul-profundo/20'
                     }`}
                   />
                 </li>
diff --git a/components/sections/__tests__/PuntosInflexion.test.tsx b/components/sections/__tests__/PuntosInflexion.test.tsx
index 4c1edfe..e3c5f82 100644
--- a/components/sections/__tests__/PuntosInflexion.test.tsx
+++ b/components/sections/__tests__/PuntosInflexion.test.tsx
@@ -9,12 +9,15 @@ describe('<PuntosInflexion />', () => {
     expect(screen.getByText(/llegan a un punto de inflexión/i)).toBeInTheDocument();
   });
 
-  it('renders the six inflection points from the dueña deck', () => {
+  it('labels both halves of every exchange with its inflection point', () => {
     render(<PuntosInflexion />);
-    // Each point now renders once, not twice: the mobile carousel is gone and
-    // the conversation stacks vertically at every breakpoint.
+    // Twice per point, once on each frame. The pinned sequence shows one thing
+    // at a time, so when creeer's answer is on screen the question that
+    // prompted it is gone — the repeated label is the only thing left tying the
+    // two together. Dropping it from the answer would leave six statements
+    // floating with nothing to answer.
     for (const name of ['Crecer', 'Consolidarse', 'Recuperar rentabilidad', 'Simplificar', 'Expandirse', 'Replantear el portafolio']) {
-      expect(screen.getByText(new RegExp(name))).toBeInTheDocument();
+      expect(screen.getAllByText(new RegExp(name))).toHaveLength(2);
     }
     expect(screen.getAllByRole('listitem')).toHaveLength(6);
   });