Also: internet penetration makes the model worse, not better. The Gulf states had the highest connectivity and the lowest instability.
Full paper, data, and replication code are open access. Fourth in the FICSS Computational Macrohistory series.
The most valuable part of the analysis isn't the correct classifications — it's the three failures.
Jordan had the highest stress score and didn't collapse. Syria had a low score and descended into civil war. Egypt sat exactly at the threshold.
Each failure tells you something the model is missing.
New paper: "Structural Stress and Political Instability in the MENA Region" — testing whether a formal index of structural conditions could have distinguished Arab Spring outcomes across 11 countries.
Result: 8/11 correct. Youth unemployment and regime type are the critical variables.
👇
You can't predict the path. But you can map the landscape.
5 strategies for seeing beyond the Lyapunov Wall — from separating fast & slow variables to identifying the "attractors" that constrain where a system can go.
In 1961, Edward Lorenz discovered the butterfly effect by accident — a rounding error of 0.000127 produced completely different weather patterns.
The same mathematics applies to societies. It's why CMH is honest about what it can and can't predict.
Mohamed Bouazizi's self-immolation didn't cause the Tunisian revolution.
It was the spark. The fuel had been accumulating for a decade — in unemployment data, inequality indices, and the fragility of a regime too repressive to be legitimate.
New paper quantifying the fuel:
doi.org/10.5281/zeno...
New from FICSS: "The Mirror Paradox" — what happens when a science must account for the fact that its own predictions change the thing it predicts.
From a bank run triggered by a BBC broadcast to the mathematics of self-fulfilling prophecies.
galenfontaise.substack.com/p/the-mirror...
The country with the BIGGEST youth bulge in 2011 among Tunisia, Egypt, and Saudi Arabia?
Saudi Arabia.
The country that had a revolution because of its youth? Not Saudi Arabia.
Demographics alone don't predict revolution. New paper on what actually does:
doi.org/10.5281/zeno...
Hot take: "we predicted the Arab Spring with math" 🚫
What we actually did: built a stress index, tested it on 3 countries, got correct classifications, and wrote 15 pages of limitations explaining why this proves almost nothing yet.
Science is slow. That's the point.
doi.org/10.5281/zeno...
If you had 18 months of warning before the next humanitarian crisis —
what would you do with that time?
That's the question CMH is trying to answer.
New from FICSS: "The Lyapunov Wall" — why mathematics says we can't predict the future beyond ~5 years, and what we can do instead.
Starts with Asimov. Ends with something more useful than science fiction.
galenfontaise.substack.com/p/the-lyapun...
Full guide: www.ficss.institute/working-pape...
What if Saudi Arabia had been a semi-authoritarian regime instead of a full autocracy in 2011?
Our model says: it would have ranked as the MOST revolutionary country in the region.
One variable. +0.58 shift. Completely different prediction.
That's the anocracy effect.
doi.org/10.5281/zeno...
Saudi Arabia had HIGHER youth unemployment than Egypt in 2010.
Higher internet penetration than Tunisia.
And yet: no revolution.
Why? Because math says regime type isn't just another variable — it's the multiplier that determines whether stress becomes uprising.
New paper 👇
doi.org/10.5281/zeno...
The Arab Spring warning signals were in public databases.
World Bank. ILO. Freedom House.
What was missing wasn't data.
It was the framework to read it — and the will to act.
With A8, the axiomatic foundation is complete.
A1-A8 define what Computational Macrohistory IS:
A science that acknowledges limits, quantifies uncertainty, and submits to empirical test.
The theory is set. The empirical work begins.
galenfontaise.substack.com/p/the-test-o...
5/5
The backtesting protocol:
1. Train on past (e.g., 1950-2000)
2. Predict held-out period (e.g., 2001-2020)
3. Compare predictions vs outcomes
4. Report ALL results
No cherry-picking. No hindsight adjustments.
Transparent track records build credibility.
4/5
The metrics:
- Brier Score—are probabilities calibrated?
- AUC-ROC—can we distinguish cases?
- Calibration plots—do 70% predictions happen 70% of the time?
And the hardest part: PUBLISHING FAILURES.
Not just successes. Everything.
3/5
The requirement:
Every model must generate testable predictions:
Pₘ(X(t+Δt) | X(t))
Not "instability is possible" but "P(revolution) = 45% ± 10% within 3 years."
Specific. Temporal. Observable. Independent.
If it can't fail, it's not science.
2/5
🧵 What separates science from storytelling?
Both explain the world. Both can be compelling.
But only science can be WRONG.
Axiom A8—Computational Falsifiability—completes the CMH foundation.
1/5
Libya and Tunisia both ousted their dictators in 2011.
One built a democracy. One became a civil war.
The difference? The invisible structural scaffolding underneath — institutions, state capacity, elite consensus.
CMH measures it.
Very interesting!
The hard limit isn't computing power. It's chaos theory.
Small errors compound exponentially. After ~10 years, even perfect models tell you nothing.
Calibrated probability in the 1-5 year window. That's enough to matter.
This is why CMH requires:
- Multiple predictions
- Track records over time
- Calibration analysis (do 60% predictions come true 60% of the time?)
One prediction proves nothing. A pattern of predictions reveals everything.
#Probability #Forecasting #Statistics #Methodology
tinyurl.com/4s5dxur4
Both outcomes were in the forecast. We assigned probabilities to each.
Judging probabilistic predictions by single outcomes is like judging a poker player by one hand. You need many hands to see if they're skilled.
"There's a 60% chance of instability" doesn't mean "instability will probably happen."
It means: in 100 similar situations, roughly 60 would see instability and 40 wouldn't.
If instability occurs, the prediction wasn't "right."
If stability persists, the prediction wasn't "wrong."
This is it.
The moment we've been waiting for.
Go Canada, GO! 🇨🇦🏒
#TeamCanada #GoCanadaGo #Olympics2026 #Hockey #WinterOlympics2026
Overconfidence: claiming precision at long horizons.
Underconfidence: refusing to predict when short-term signal exists.
Calibration: matching confidence to horizon.
That's the discipline.
galenfontaise.substack.com/p/the-horizo...
#Forecasting #Prediction #DataScience #Methodology
5/5
What survives at long horizons?
✓ Demographics (decades of inertia)
✓ Physical constraints (geography)
✓ Cyclical patterns
✓ Distributions, not points
✓ Directions, not magnitudes
Structure persists. Specifics don't.
4/5
Realistic horizons for social prediction:
1-2 years → quantitative (70-80%)
3-5 years → scenarios (60-70%)
5-10 years → trends only (50-60%)
>20 years → speculation
The trade-off is fundamental: PRECISION vs HORIZON.
Pick one.
3/5