Cepeda lost in every runoff scenario tested
En la víspera del silencio electoral colombiano, una encuesta de la firma brasileña Atlas Intel colocó al candidato de izquierda Iván Cepeda y al radical de derecha Abelardo de la Espriella en empate técnico, contradiciendo mediciones realizadas apenas horas antes. La distancia de 1.4 puntos entre ambos no solo reconfiguró el mapa de la primera vuelta, sino que reveló una tendencia más profunda: en todos los escenarios de segunda vuelta, la izquierda pierde. Lo que está en juego no es solo una elección, sino la pregunta de si las encuestas reflejan la voluntad ciudadana o la construyen, pues Atlas Intel enfrenta una investigación del CNE por posibles irregularidades metodológicas que habrían beneficiado sistemáticamente a los candidatos de derecha.
- Una sola encuesta borró de golpe la ventaja de 44% que Invamer le había dado a Cepeda el día anterior, sembrando incertidumbre a horas del cierre de campaña.
- De la Espriella ganó casi cinco puntos respecto a la medición anterior de Atlas Intel, consolidando un impulso que la estrategia del 'voto útil' de Paloma Valencia no ha podido frenar.
- En todos los duelos de segunda vuelta, Cepeda pierde: 8.7 puntos ante De la Espriella, 3.1 ante Valencia, y un empate ante Fajardo, quien aparece muy rezagado en los agregadores.
- La credibilidad de los propios datos está en entredicho: el CNE investiga a Atlas Intel por ajustes metodológicos que, según el análisis de La Silla Vacía, inflaron el apoyo a la derecha justo antes de las primarias de marzo.
- El país entra al silencio electoral con dos relatos contradictorios sobre quién lidera, y sin certeza de si las encuestas miden la realidad o la distorsionan.
Atlas Intel, la firma brasileña contratada por Revista Semana, publicó su última medición antes del silencio electoral con una conclusión que sacudió el ambiente político: Iván Cepeda, con 37.7%, y Abelardo de la Espriella, con 36.3%, quedaban en empate técnico. El contraste con la encuesta de Invamer del día anterior —que le daba a Cepeda un cómodo 44%— era imposible de ignorar.
El avance de De la Espriella fue notable: casi cinco puntos ganados desde la medición anterior de la misma firma, y una ventaja de 22.4 puntos sobre Paloma Valencia, cuya apuesta por el voto útil no logró despegar. Geográficamente, Cepeda mantuvo el Pacífico y Bogotá, pero la costa Caribe, la Amazonía y los Llanos Orientales se inclinaron hacia su rival.
El panorama de segunda vuelta resultó aún más adverso para la izquierda. En todos los escenarios probados, Cepeda perdía: ante De la Espriella por 8.7 puntos, ante Valencia por 3.1, y empataba con Fajardo. Esto coincidía con proyecciones anteriores de Guarumo, aunque divergía de las cifras más equilibradas de Invamer.
Sobre todo este cuadro pesaba una sombra metodológica. La Silla Vacía analizó los microdatos de Atlas Intel y encontró que la firma había modificado su universo muestral de maneras que ninguna otra encuestadora replicaba. Esos ajustes, que se intensificaron justo antes de las primarias de marzo, tuvieron el efecto de reducir el apoyo a Cepeda y elevar el de los candidatos de derecha. El CNE ya investiga posibles irregularidades técnicas. La pregunta que queda flotando al inicio del silencio electoral es si esas encuestas estaban midiendo la opinión pública o, de alguna manera, contribuyendo a moldearla.
Atlas Intel, a Brazilian polling firm hired by Revista Semana, released its final measurement before Colombia's electoral silence took hold, and the numbers told a starkly different story than surveys conducted just hours earlier. On the left, Iván Cepeda stood at 37.7 percent. On the radical right, Abelardo de la Espriella had climbed to 36.3 percent. The gap between them—1.4 points—amounted to a technical tie, a dramatic reversal from Invamer's survey the day before, which had shown Cepeda commanding 44 percent and a far more comfortable position heading into a potential runoff.
The rightward shift was particularly striking when viewed across the broader field. De la Espriella had gained nearly five points since the previous Atlas Intel measurement, and he now held a commanding 22.4-point lead over Paloma Valencia, the centrist candidate betting on strategic voting. Valencia's "useful vote" strategy appeared to be failing her; she could not gain traction against De la Espriella's momentum, and the gap between them remained consistent across Atlas Intel's three most recent polls.
But the real alarm for Cepeda lay not in the first round but in the runoff scenarios. Across every second-round matchup the poll tested, the leftist candidate lost. De la Espriella would beat him by 8.7 points. Valencia would defeat him by 3.1 points. Even Sergio Fajardo, who appeared far behind in La Silla Vacía's poll aggregator, would tie him. This represented a sharp departure from Invamer's more balanced runoff projections released the previous day, though it aligned with earlier modeling from Guarumo that had already suggested this troubling pattern for the left.
Geographically, the picture was uneven. The Pacific coast remained Cepeda's stronghold, and Bogotá kept him competitive in the capital. But the Caribbean coast, the Amazon region, and the Orinoco plains all tilted toward De la Espriella. These regional shifts mattered in a country where geography often determined electoral outcomes.
President Gustavo Petro's approval rating, meanwhile, had stabilized at 41 percent, though his disapproval had climbed to 56.7 percent—a concerning sign for a leftist candidate hoping to benefit from the sitting president's support. The political environment appeared to be hardening against the left as the campaign entered its final phase.
Yet Atlas Intel's credibility was itself under scrutiny. The firm faced an investigation by Colombia's National Electoral Council (CNE) over possible technical irregularities. When La Silla Vacía analyzed the firm's microdata—the underlying sample composition and weighting decisions—it found something unusual: Atlas Intel had been adjusting its survey universe in ways no other polling organization did. These methodological shifts, which intensified just before March's primary elections at a crucial moment in the campaign, had the effect of boosting right-wing candidates while simultaneously reducing support for Cepeda. The pattern raised questions about whether the polls were measuring public opinion or shaping it, and whether the firm's technical choices reflected genuine shifts in voter sentiment or something else entirely. The CNE's investigation would need to determine whether these adjustments were defensible statistical practice or something more troubling.
Notable Quotes
Atlas Intel adjusted its survey universe in ways no other polling organization did, with shifts that boosted right-wing candidates while reducing support for Cepeda— La Silla Vacía analysis of CNE microdata
The Hearth Conversation Another angle on the story
Why does a single poll matter so much if other surveys showed something different?
Because this one came last—right before the electoral silence began. It's the final word voters hear before they can't hear anything else. And it contradicted what Invamer had just shown, which makes people wonder which one is real.
The gap between Cepeda and De la Espriella is only 1.4 points. That's basically nothing.
Exactly. But it's a nothing that changes everything in a runoff. If Cepeda can't win the first round decisively, he loses in every second-round scenario. That's not a polling quirk—that's a structural problem.
You mentioned Atlas Intel is under investigation. Does that mean we should ignore this poll?
Not ignore it. But hold it carefully. The firm adjusted its methodology in ways other pollsters don't. Those adjustments happened to benefit the right. Whether that's innocent statistical practice or something else—that's what the CNE is trying to figure out.
What about the regional breakdown? Does that tell us anything real?
It tells us De la Espriella is consolidating outside the cities. Cepeda has Bogotá and the coast, but De la Espriella has the Caribbean, the Amazon, the plains. In a close race, that geography matters.
Petro's approval is 41 percent. Is that good or bad for Cepeda?
It's a problem. Cepeda is the left's candidate, and the president's disapproval is climbing toward 57 percent. That's a headwind, not a tailwind.