John C. MenziesSOCIETAL DYNAMICS: John C. Menzies · 28 May 2026 · Founder, Societal Dynamics
Three days before Colombia votes, the polls have Cepeda winning and the conventional reading has narrowed the race to two candidates. My framework reads it differently. The signal that matters most for Sunday’s outcome is something the polls cannot measure — and it has been visible since the eighth of March.
Where the race stands
Colombia votes on Sunday, 31 May. The country’s first leftist president, Gustavo Petro, is barred from a second consecutive term, and the race to succeed him has narrowed to three names. Iván Cepeda, the senator carrying the governing Pacto Histórico coalition, leads the polls. Abelardo De la Espriella, a hardline outsider modelling himself on the Salvadoran president Bukele and Argentina’s Milei, has surged in the final weeks. Paloma Valencia, the Centro Democrático senator and Álvaro Uribe’s political protégée, has collapsed in the same window after what most analysts now describe as a strategically poor campaign.
The most recent Invamer survey, published on 22 May from a sample of 3,800 voters, has Cepeda at 44.6%, De la Espriella at 31.6%, and Valencia at 14%. The institutional analysis from the Library of Congress notes that a weighted average of recent polls suggests Cepeda could narrowly lose to Valencia in a runoff while defeating or tying De la Espriella, but the polling has shifted enough in the final week that Valencia may no longer survive Round One at all. The dominant reading is that this is a two-horse race between Cepeda and De la Espriella, with Valencia effectively eliminated.
I think this reading is wrong. Not by a small amount. By a lot. Below I’ll explain why.
How I get to a different answer
I will not go into the framework itself in any detail in this post. The underlying mathematics is the subject of work that is not yet publicly released, and the specifics will appear in due course in a more formal venue. But the general approach is worth describing at a high level so the reasoning that follows is not opaque.
The framework I use is mechanistic rather than statistical. Most election forecasting works by taking polls and adjacent signals, fundamentals, sentiment data and prediction-market prices and then weighting them according to their historical predictive value. The results are aggregated into a probability over outcomes. That tradition does useful work when the underlying signal sits in the polls. It struggles when the signal sits in structural variables that polls do not measure.
What I do is build a system of ordinary differential equations — ODEs in the jargon — that track several dimensions of the political situation at once. Each equation describes how one variable changes over time as a function of the others. The system is solved forward from a calibrated starting state to produce a trajectory for the whole political environment from now until the date of resolution. The coupling between variables is specified by what is called a structured Jacobian — meaning the relationships are written down from theory and prior calibration, rather than fitted to data after the fact. This is the discipline that makes the framework predictive rather than descriptive, and that makes its calls falsifiable.
When two political mobilisations are competing, as in Colombia, where the anti-Petro right and the pro-Petro left are pitted against each other, the framework treats them as two coupled cascade systems and analyses the stability of the resulting two-cascade configuration. This is Verhulst-Lotka-Volterra (VLV) stability analysis, which is the topology of the coupling between the cascades and their respective state variables. The output is a regime classification — balance, contestation, or dominance — that tells me how much of the discharge energy in the system is going to end up flowing to which side once an electoral event releases it.
I will stop there with the technical description. The point of describing it at all is to make clear that what follows is not vibes-based reasoning. It is the output of a specific calibrated model that has been validated against earlier electoral cases.
The signal the polls cannot see
The single most important fact for understanding what is about to happen in Colombia did not happen in a poll. It happened on the eighth of March, the same day as Colombia’s legislative elections, when the three major coalitions held open primaries to select their presidential candidates.
5.84 million Colombians voted in the anti-Petro right-wing primary, the Gran Consulta. 589,000 voted in the Pacto Histórico’s Frente por la Vida primary on the left. Ten to one. That is the ratio of anti-Petro to pro-Petro voters who showed up to vote in their respective primaries. It is the largest right-versus-left primary turnout asymmetry in Colombian history, by a wide margin.
This is an empirical fact, not a poll. Those 5.84 million voters exist. They queued up at polling stations on a Sunday in March, cast a ballot for their preferred candidate within the right-wing field, and then went home. They are not going to disappear by Sunday the 31st of May. When they come back to vote in the actual first round of the presidential election, today, they show up in the count.
The polls cannot read this signal. Pollsters sample a slice of all eligible voters and weight to demographic targets. They cannot measure who actually got off the couch in March. The polling models do not incorporate primary turnout because primary turnout is not reliably predictive of general election turnout in the systems where most pollsters were trained, the United States, Britain and Western Europe. But in Colombia in 2026, after three years of deteriorating security, the assassination of an opposition presidential candidate, and the visible collapse of the Petro government’s flagship Total Peace policy, primary turnout is the single best signal of who is actually going to show up on Sunday.
The structural conditions reinforce the signal
The framework decomposes the structural pressure operating against the incumbent coalition into three components. I will describe each in Colombian terms rather than abstractly.
The first is material — the actual lived conditions of Colombian life under the Petro government. 40,663 homicides under his presidency to date, a 7.6% increase on the previous government. Fiscal expansion equivalent to 9% of government spending. 369,000 of 650,000 new jobs were created in the public sector. Real wage erosion. Cost of living which is particulalry acute for the poor.
The second is the dignity dimension — the cumulative sense that the Colombian state has failed to perform its most basic protective functions. The collapse of the Total Peace policy. Armed groups now operating in more than 700 municipalities. The China BRI deal that many Colombians read as a sovereignty erosion. 48 massacres and 249 deaths in the first four months of 2026 alone, per the Indepaz monitoring data — the highest pre-electoral body count in a decade.
The third is memorial — the specific traumatic events that crystallise political identity in the present moment. The assassination of Senator Miguel Uribe Turbay last year, the strongest single political memorial event in Colombia in more than thirty years, comparable in weight to the murder of Luis Carlos Galán in 1989. The lineage of right-wing victims it sits alongside in Colombian political memory.
Each of these three components is scored independently on a zero-to-one scale against the empirical evidence, on each side of the political divide. The anti-Petro side scores high on all three in a way the pro-Petro side cannot match, particularly on the dignity and memorial dimensions, where the asymmetry is most stark. The composite differential is large by the standards of the cross-case archive the framework has been calibrated against. In every prior case where this magnitude of structural differential was visible, the electoral discharge against the incumbent coalition was decisive.
What the framework predicts
Running the system forward from these inputs — the structural differential, the primary turnout asymmetry, the Uribe Turbay anchoring event, the high polarisation of the current Colombian polity — produces the following prediction.
In Round One on 31 May, Cepeda comes first but well short of the 50% needed to avoid a runoff. The framework places him in the 33 to 39% range, materially below his 44.6% polling figure. Valencia comes second in the 20 to 26% range despite her recent polling collapse. De la Espriella comes third in the 17 to 22% range, despite his current polling lead over Valencia. The combined right vote sits comfortably above Cepeda’s share.
In the Runoff on 21 June, the binary frame collapses the Round One field. Centrist and abstention voters who did not show up in Round One get drawn back in by the stark choice between Pacto Histórico continuity and the anti-Petro alternative. The framework’s consolidation mechanism, calibrated against numerous prior and predominantly western elections, produces a Runoff result in the vicinity of 55 to 45 in favour of Valencia — a margin of approximately 10% points.
The headline prediction: anti-Petro wins the presidency on 21 June with approximately 90% probability under the pure-structural read. Within that, Paloma Valencia is the most likely runoff winner at roughly 65%. Abelardo De la Espriella is the secondary anti-Petro path at about 25%. Cepeda holds the presidency in approximately 10% of outcomes.
Why I’m less certain about Valencia specifically
There is one place where the framework’s reading is genuinely uncertain, and I want to be transparent about it. Valencia carries a strong structural advange: her party apparatus, her Uribista lineage, the discipline of her organising operation. This should have translated into a clear consolidation of the anti-Petro vote around her in the final weeks. It hasn’t. Over the past five weeks she has lost roughly 8% points in the polling while De la Espriella has gained roughly 10%. Her campaign has failed to attack De la Espriella, her messaging has been muddled, and the qualitative narrative — what Colombians who follow politics are saying to each other — has shifted decisively in his direction.
When I run the framework with the late-May polling reality as the starting point rather than the pre-campaign structural inputs, I get a tighter result. Anti-Petro still wins, but by under 5% rather than 10%, with De la Espriella rather than Valencia as the survivor, and with the overall anti-Petro probability dropping from 90% to about 75%.
Both readings use the same framework. The difference is which set of initial conditions you trust. The structural read says polls systematically understate anti-Petro mobilisation and the count will reveal what the primary turnout already showed. The polling read accepts that the campaign-period developments have actually shifted the underlying structure, particularly through Valencia’s strategic failure to capitalise on her very strong structural advantage.
The truth is probably somewhere between the two reads. The Round One count on Sunday resolves which is closer.
What Sunday will resolve
Round One on 31 May is a genuinely informative test, and the specific predictions above are filed before the count precisely so that the resolution can be unambiguous. The most important single number is Cepeda’s actual vote share. If he lands at 33 to 39%, the framework’s structural read is vindicated and the Runoff becomes essentially a foregone anti-Petro victory. If he lands at 44% or higher, the polling read is closer to the truth and the Runoff is genuinely competitive.
The second test is the order between Valencia and De la Espriella for second place. If Valencia recovers and finishes second despite her recent polling collapse, that is an extraordinary structural-signal vindication and the Runoff becomes a near-certain Valencia win. If De la Espriella finishes second by the polling-implied margin, the polling read is closer and the runoff against Cepeda is a more uncertain contest where the better-organised political machine on one side plays against the louder media presence on the other.
Either outcome is informative for the framework. The verdict gets recorded against the specific version of the model that produced it. That is the discipline that protects the integrity of the work, and it is the discipline I am operating under here.
Why I’m posting this publicly
There are two reasons I am putting this on the public record three days before the vote rather than keeping it private. The first is the obvious one: a forecast that is not publicly timestamped before the event is much weaker evidence about the underlying framework’s accuracy than one that is. If the Colombian count vindicates this call, I want there to be no ambiguity about when the prediction was made and what the framework was saying.
The second is that I think Colombia is the first case where the framework’s read is dramatically different from the conventional reading of the race in a way that is visible to a general audience. A framework that agrees with the polls is uninteresting. A framework that disagrees sharply, and is then either vindicated or refuted by the count, is informative. The disagreement here is sharp enough to be diagnostic.
If the framework holds up here, the next cases on the calendar, Brazil October 2026, Hungary 2026, the US midterms November 2026, will provide further tests and allow for ongoing model development. Each one will be filed publicly in advance, with named predictions and explicit version-tagging, so that the cumulative track record is visible and verifiable over time. That said, public elections are a tough test for the Cascade Model and a necessary test.
Where this leaves us
Three days before Colombia votes, the polls and the conventional reading are pricing one outcome, and my framework is pricing a very different one. The framework says anti-Petro wins the presidency on 21 June with somewhere between 75 and 90% confidence, depending on how heavily you weight the structural signal against the campaign-period polling reality. The specific runoff survivor is genuinely uncertain — most likely Valencia under the structural read, most likely De la Espriella under the polling read. Cepeda holds the presidency in only a small minority of outcomes under either reading.
Whether Sunday’s count vindicates this prediction or refutes it, we will know more about the framework’s reach on Monday than we know today. I will post the verdict, honestly, against the prediction filed above, regardless of which way it goes. Subscribe if you want to see that resolution and the follow-up analysis.
Disclaimer and limitation of liability
This post is published by John C. Menzies under the Societal Dynamics imprint for general informational and educational purposes only. It does not constitute financial, investment, tax, legal, or professional advice and should not be relied upon as the basis for any decision. The forecasting framework described in this post is proprietary and has not been subject to independent peer review at the time of writing. Prior out-of-sample accuracy on earlier electoral cases, where described, has not been independently verified. Past forecasting accuracy does not guarantee future results.
To the maximum extent permitted by law, the author and Societal Dynamics exclude all liability for any direct, indirect, consequential, incidental, or special loss or damage arising from any use of, reliance on, or inability to use the content of this post. The reader's sole remedy in respect of any dissatisfaction with this content is to cease reading it. No advisory, fiduciary, or other professional relationship is created between the author and any reader of this post.
The forecasting framework discussed in this post is the intellectual property of John C. Menzies. Excerpts of this post may be quoted with attribution; the framework architecture, parameter values, scoring rubrics, and calibration archive are proprietary and may not be reproduced or reverse-engineered without written permission. © John C. Menzies / Societal Dynamics 2026. All rights reserved.