diff --git a/components/primitives/ParDialogo.tsx b/components/primitives/ParDialogo.tsx
index a3b2d21..21d068a 100644
--- a/components/primitives/ParDialogo.tsx
+++ b/components/primitives/ParDialogo.tsx
@@ -1,11 +1,14 @@
 'use client';
 
+import { useEffect, useRef } from 'react';
 import { motion } from 'framer-motion';
 import { Wordmark } from '@/components/primitives/Wordmark';
 import { fadeInUp } from '@/lib/motion';
 
 // One exchange of the page's conversation: the reader's situation, then what
-// creeer does about it.
+// creeer does about it. The answer arrives as you scroll into the pair — its
+// gold rule draws down and the block comes up to full contrast — so the section
+// reads as being answered rather than as a list that happens to have answers.
 //
 // The sides never swap from one pair to the next. Each voice keeps its own
 // margin, the way a conversation reads on a page; what alternates going down is
@@ -13,39 +16,114 @@ import { fadeInUp } from '@/lib/motion';
 // the reader the one cue that tells them who is speaking.
 //
 // Three cues carry that distinction, never colour alone: the speaker label, the
-// column, and the rule hanging off each block. The answer also drops half a
-// step below the situation on wide screens, so the eye reads it as a reply
-// rather than as a second column of the same thought.
+// column, and the rule hanging off each block.
+//
+// The answer text is in the DOM at all times, at every breakpoint — opacity,
+// never `display: none` or `hidden`, so screen readers still announce it and
+// the static export still ships it to search engines.
+//
+// `activo` latches: the section turns it on when the reader reaches the pair
+// and never turns it off again. An answer that faded back out once you scrolled
+// past would leave the page emptier the further you read, and would re-create
+// the failure this project has already hit twice — content stranded at opacity
+// zero with every test green, because jsdom has no layout and no scrolling.
 //
 // No motion.li here on purpose: Framer propagates variants through React
 // context, so the two motion.div children still inherit the stagger from the
-// <ul> the section wraps them in, even across this plain <li>. Making the item
-// itself a motion.li would add a third beat nobody asked for.
+// <ul> the section wraps them in, even across this plain <li>.
 
 export function ParDialogo({
   n,
   name,
   situacion,
   respuesta,
+  activo,
+  reducido,
+  onActivar,
 }: {
   n: number;
   name: string;
   situacion: string;
   respuesta: string;
+  activo: boolean;
+  reducido: boolean;
+  onActivar: (n: number) => void;
 }) {
+  const ref = useRef<HTMLLIElement>(null);
+
+  // The pair crossing the middle of the viewport is the one being answered.
+  // The negative margins collapse the root to a band across the centre, so
+  // "in view" means "where the reader is actually looking", not "anywhere on
+  // screen" — which with six pairs would fire all of them at once.
+  useEffect(() => {
+    const el = ref.current;
+    if (!el || reducido) return;
+    const observer = new IntersectionObserver(
+      ([entrada]) => {
+        if (entrada.isIntersecting) onActivar(n);
+      },
+      { rootMargin: '-45% 0px -45% 0px' },
+    );
+    observer.observe(el);
+    return () => observer.disconnect();
+  }, [n, reducido, onActivar]);
+
+  // Reduced motion gets the resolved state outright: everything at full
+  // contrast, every rule drawn. Never the dimmed half of an animation that
+  // will not run.
+  const resuelto = reducido || activo;
+
   return (
-    <li className="grid gap-5 md:grid-cols-2 md:gap-12">
-      <motion.div variants={fadeInUp} className="border-l-2 border-azul-profundo/25 pl-5">
-        <p className="text-xs font-black uppercase tracking-[0.2em] text-dorado">
-          {String(n).padStart(2, '0')} · {name}
-        </p>
-        <p className="mt-3 font-sans font-bold text-xl md:text-2xl text-azul-profundo leading-snug">
-          «{situacion}»
-        </p>
+    <li
+      ref={ref}
+      data-activo={resuelto ? 'true' : 'false'}
+      onMouseEnter={() => onActivar(n)}
+      onFocus={() => onActivar(n)}
+      className="grid gap-5 md:grid-cols-2 md:gap-12"
+    >
+      {/* Two nested layers on purpose, and they must stay nested. Framer writes
+          the resolved variant into the inline `style` — `opacity: 1` — and an
+          inline style beats a Tailwind class. Putting the reveal on the same
+          element as `variants` looked right in the markup and did nothing in the
+          browser: the answers were readable from the moment the section entered,
+          and only the gold rule, a transform Framer never touches, animated.
+          So: motion.div owns the entrance, the plain child owns the reveal. */}
+      <motion.div variants={fadeInUp}>
+        <div
+          className={`border-l-2 border-azul-profundo/25 pl-5 transition-opacity duration-500 motion-reduce:transition-none ${
+            resuelto ? 'opacity-100' : 'opacity-55'
+          }`}
+        >
+          <p className="text-xs font-black uppercase tracking-[0.2em] text-dorado">
+            {String(n).padStart(2, '0')} · {name}
+          </p>
+          <p className="mt-3 font-sans font-bold text-xl md:text-2xl text-azul-profundo leading-snug">
+            «{situacion}»
+          </p>
+        </div>
       </motion.div>
-      <motion.div variants={fadeInUp} className="border-l-2 border-dorado pl-5 md:mt-10">
-        <Wordmark size="sm" />
-        <p className="mt-3 text-gris-medio leading-relaxed">{respuesta}</p>
+
+      <motion.div variants={fadeInUp} className="md:mt-10">
+        <div
+          tabIndex={0}
+          aria-label={`Respuesta de creeer a: ${name}`}
+          className={`relative pl-5 transition-opacity duration-500 motion-reduce:transition-none ${
+            resuelto ? 'opacity-100' : 'opacity-0'
+          }`}
+        >
+          {/* The rule draws itself down as the answer arrives. A transform, not
+              a stroke: a dash pattern on a non-uniformly scaled SVG is exactly
+              what broke the hilo conductor, and this has to survive any
+              container it lands in. */}
+          <span
+            aria-hidden="true"
+            className={`absolute left-0 top-0 h-full w-0.5 bg-dorado origin-top transition-transform duration-700 ease-out motion-reduce:transition-none ${
+              resuelto ? 'scale-y-100' : 'scale-y-0'
+            }`}
+          />
+          <Wordmark size="sm" />
+          <p className="mt-3 text-gris-medio leading-relaxed">{respuesta}</p>
+        </div>
       </motion.div>
     </li>
   );
diff --git a/components/primitives/__tests__/ParDialogo.test.tsx b/components/primitives/__tests__/ParDialogo.test.tsx
index aa01985..d4ca31d 100644
--- a/components/primitives/__tests__/ParDialogo.test.tsx
+++ b/components/primitives/__tests__/ParDialogo.test.tsx
@@ -1,5 +1,5 @@
 import { render, screen } from '@testing-library/react';
-import { describe, it, expect } from 'vitest';
+import { describe, it, expect, vi } from 'vitest';
 import { ParDialogo } from '../ParDialogo';
 
 const par = {
@@ -9,15 +9,22 @@ const par = {
   respuesta: 'Me sumerjo hasta entender tu empresa desde dentro.',
 };
 
+const montar = (extra: Partial<Parameters<typeof ParDialogo>[0]> = {}) =>
+  render(
+    <ul>
+      <ParDialogo {...par} activo reducido={false} onActivar={() => {}} {...extra} />
+    </ul>,
+  );
+
 describe('<ParDialogo />', () => {
   it('renders both voices of the exchange', () => {
-    render(<ul><ParDialogo {...par} /></ul>);
+    montar();
     expect(screen.getByText(/El siguiente salto no tiene mapa/)).toBeInTheDocument();
     expect(screen.getByText(/Me sumerjo hasta entender tu empresa desde dentro/)).toBeInTheDocument();
   });
 
   it('attributes the answer to creeer, not to the reader', () => {
-    render(<ul><ParDialogo {...par} /></ul>);
+    montar();
     // Without a speaker label the answer reads as more description of the
     // problem. The wordmark is that label — and <Wordmark /> is the only thing
     // allowed to draw cre≡er, never hand-typed in the JSX.
@@ -25,22 +32,80 @@ describe('<ParDialogo />', () => {
   });
 
   it('labels the situation with its inflection point', () => {
-    render(<ul><ParDialogo {...par} /></ul>);
+    montar();
     expect(screen.getByText(/01 · Crecer/)).toBeInTheDocument();
   });
 
   it('tells the two voices apart by more than colour', () => {
-    const { container } = render(<ul><ParDialogo {...par} /></ul>);
+    const { container } = montar();
     // A reader who cannot separate gold from navy still has the speaker label,
-    // the column, and the rule hanging off each block. Colour alone would fail
-    // the spec's accessibility criterion.
-    const bloques = container.querySelectorAll('li > div');
-    expect(bloques).toHaveLength(2);
-    bloques.forEach((b) => expect(b.className).toMatch(/border-l-2/));
+    // the column, and a rule down the side of each block. Colour alone would
+    // fail the spec's accessibility criterion.
+    expect(container.querySelector('.border-l-2')).toBeInTheDocument();
+    expect(container.querySelector('span[aria-hidden="true"]')?.className).toMatch(/bg-dorado/);
   });
 
   it('is a list item, so the section can mount six of them as a list', () => {
-    render(<ul><ParDialogo {...par} /></ul>);
+    montar();
     expect(screen.getAllByRole('listitem')).toHaveLength(1);
   });
+
+  it('keeps the answer in the DOM before the reader reaches it', () => {
+    montar({ activo: false });
+    // The answer waits its turn by going transparent, never by leaving the
+    // document. `hidden` or `display: none` would take it away from screen
+    // readers and from the static export the crawler sees — half the section's
+    // message, gone, to buy an animation.
+    const respuesta = screen.getByText(/Me sumerjo hasta entender tu empresa desde dentro/);
+    expect(respuesta).toBeInTheDocument();
+    expect(respuesta.closest('[hidden]')).toBeNull();
+    expect(screen.getByLabelText(/Respuesta de creeer a: Crecer/).className).toMatch(/opacity-0\b/);
+  });
+
+  it('keeps the reveal off any element Framer writes inline styles onto', () => {
+    montar({ activo: false });
+    // Framer resolves `fadeInUp` into the inline style — `opacity: 1` — and an
+    // inline style beats a Tailwind class. When the reveal lived on the same
+    // element as `variants`, the markup read correctly and the browser ignored
+    // it: every answer was legible from the moment the section scrolled in,
+    // and only the gold rule animated. The reveal must sit on a plain child
+    // that Framer never writes to.
+    const bloque = screen.getByLabelText(/Respuesta de creeer a: Crecer/);
+    expect(bloque.getAttribute('style') ?? '').not.toMatch(/opacity/);
+  });
+
+  it('arrives already answered under reduced motion, whatever the scroll did', () => {
+    montar({ activo: false, reducido: true });
+    // These readers get no scroll reveal at all. If the resolved state depended
+    // on the animation having run, the six answers would sit at opacity zero
+    // for them — the exact failure this project has already shipped twice.
+    const bloque = screen.getByLabelText(/Respuesta de creeer a: Crecer/);
+    expect(bloque.className).toMatch(/opacity-100/);
+    expect(bloque.querySelector('span[aria-hidden="true"]')?.className).toMatch(/scale-y-100/);
+  });
+
+  it('lets the reader drive the conversation, not only the scroll', () => {
+    const activar = vi.fn();
+    const { container } = montar({ onActivar: activar });
+    // Hover and keyboard focus reach the same state scroll does, so the pair
+    // is not the exclusive property of whoever scrolls with a mouse wheel.
+    const item = container.querySelector('li')!;
+    item.dispatchEvent(new MouseEvent('mouseover', { bubbles: true }));
+    expect(activar).toHaveBeenCalledWith(1);
+  });
+
+  it('does not observe the viewport at all under reduced motion', () => {
+    const observar = vi.fn();
+    const original = window.IntersectionObserver;
+    window.IntersectionObserver = vi.fn(() => ({
+      observe: observar,
+      unobserve: () => {},
+      disconnect: () => {},
+    })) as unknown as typeof window.IntersectionObserver;
+
+    montar({ reducido: true });
+    expect(observar).not.toHaveBeenCalled();
+
+    window.IntersectionObserver = original;
+  });
 });
diff --git a/components/sections/PuntosInflexion.tsx b/components/sections/PuntosInflexion.tsx
index 139f68e..5fe0326 100644
--- a/components/sections/PuntosInflexion.tsx
+++ b/components/sections/PuntosInflexion.tsx
@@ -1,5 +1,6 @@
 'use client';
 
+import { useCallback, useState } from 'react';
 import { motion } from 'framer-motion';
 import { Container } from '@/components/primitives/Container';
 import { FondoPuntos } from '@/components/primitives/FondoPuntos';
@@ -10,6 +11,15 @@ import { fadeInUp, staggerContainer, viewportOnce, usePrefersReducedMotion, moti
 
 export function PuntosInflexion() {
   const reduced = usePrefersReducedMotion();
+
+  // How far down the conversation the reader has got. It only ever moves
+  // forward: an answer that faded back out on the way past would leave the page
+  // emptier the further you read. Scroll drives it, and so do hover and
+  // keyboard focus, so the reader can also run the conversation themselves
+  // instead of only being carried through it.
+  const [alcanzado, setAlcanzado] = useState(0);
+  const activar = useCallback((n: number) => setAlcanzado((previo) => Math.max(previo, n)), []);
+
   return (
     <section className="bg-gris-claro relative overflow-hidden">
       <FondoPuntos className="-bottom-6 -right-12 w-80 md:w-[32rem] text-azul-profundo/10" distance={32} />
@@ -32,13 +42,21 @@ export function PuntosInflexion() {
 
         {/* The conversation. One <li> per exchange, read top to bottom — no
             carousel at any breakpoint, because "te bajas" is the whole shape
-            the dueña asked for. The stagger reaches ParDialogo's two halves
-            through React context, so each pair arrives in two beats: the
-            situation, then the answer. */}
+            the dueña asked for. Each answer arrives when the reader reaches its
+            pair, so the section answers as you go down instead of sitting there
+            already answered. The stagger reaches ParDialogo's two halves
+            through React context. */}
         <motion.ul variants={staggerContainer} initial="hidden" whileInView="visible" viewport={viewportOnce} {...motionProps(reduced)}
-          className="mt-8 list-none p-0 flex flex-col gap-12 md:gap-16">
+          className="mt-8 list-none p-0 flex flex-col gap-16 md:gap-24">
           {puntosInflexion.items.map((item, i) => (
-            <ParDialogo key={item.name} n={i + 1} {...item} />
+            <ParDialogo
+              key={item.name}
+              n={i + 1}
+              {...item}
+              activo={i + 1 <= alcanzado}
+              reducido={reduced}
+              onActivar={activar}
+            />
           ))}
         </motion.ul>
       </Container>