Iran Regime Change Forecast (10/04/2026)

April 10, 2026 · View on GitHub

Question: What is the probability of forced regime change in Iran -- defined as the IRGC being removed from power and replaced by a non-autocratic form of government that formally recognises Israel and renounces the intention to eliminate it -- within 1 week, 1 month, and 1 year?

Headline Forecast

Chairman Forecast

Horizon	Probability	Confidence
+1 week (17 Apr)	<1%	High
+1 month (10 May)	1-3%	High
+1 year (10 Apr 2027)	4-9%	Moderate

Forced regime change in Iran remains highly unlikely across all time horizons. Despite unprecedented military degradation (85% of defense industrial base destroyed), 92% public dissatisfaction, and active diplomatic/covert pressure, the IRGC maintains institutional cohesion and internal control. The most probable pathway to regime change involves cascading failures across military humiliation, economic collapse, and elite fracture, requiring 9-18 months to materialise.

Stack

This forecast was generated by the Geopol Forecaster pipeline, a two-stage system inspired by:

Snowglobe (IQTLabs) -- Stage A actor simulation pattern (control/player loop with referee)
LLM Council (Karpathy) -- Stage B multi-lens council protocol (parallel query, blind peer review, chairman synthesis)

Orchestration via LangGraph with SQLite checkpointing. Grounding via Tavily search + RSS/ISW feeds. Vector memory via Pinecone (integrated index). All LLM calls routed through OpenRouter.

Experiment Design

Stage A -- Actor Simulation

Actors: 10 core (Khamenei, Netanyahu, Trump, IRGC, Hezbollah, CENTCOM, Mossad, IDF, Russia, MBS)
Timesteps: 3 (+1 week, +1 month, +1 year)
MC runs: 1
Actor temperature: 0.8 (exploratory)
Referee temperature: 0.4 (consistent)
Information hygiene: Actors see only referee-authored world state; no raw news, no other actors' private reasoning

Stage B -- Council Protocol

Lenses: 6 (Neutral, Pessimistic, Optimistic, Blindsides, Probabilistic, Historical)
Protocol: 3-stage (parallel query at 0.6 temp -> blind peer review at 0.4 -> chairman synthesis at 0.5)
Chairman max tokens: 12,000

Grounding

Tavily: 6 expanded queries, advanced depth, top 5 results each
RSS: Times of Israel, Jerusalem Post, ISW, CTP (48h window)

Model & Runtime

Model: Claude Sonnet 4.5 via OpenRouter (all roles)
Date: 10/04/2026

Simulation-Council Convergence

Convergence Chart

The simulation (Stage A) independently estimated a 12% probability of regime change success at the 1-year horizon -- higher than the council's 4-9% range. This divergence is informative: the simulation models mechanisms (covert support, economic pressure, military operations) but may overestimate their effectiveness absent internal IRGC fracture, which fresh data experts assess has not materialised.

Per-Lens Probability Table

Lens	+1 week	+1 month	+1 year
Neutral	<1%	~2%	~8%
Pessimistic	<1%	~1.5%	~5%
Optimistic	~1%	~5%	~15%
Blindsides	~2%	~8%	~20%
Probabilistic	<0.5%	~1.5%	~6%
Historical	<1%	~2%	~8%
Chairman	<1%	1-3%	4-9%
Simulation	~0.5%	~3%	~12%

Lens Spread at 1 Year

The Blindsides and Optimistic lenses are outliers at +1 year (15-20%), modelling low-probability cascading scenarios. The chairman's 4-9% range weights toward the consensus cluster (Neutral, Probabilistic, Historical) while acknowledging the tail risk.

Key Predictions

Key Findings

IRGC institutional resilience exceeds external military degradation. Despite 85% of the defense industrial base being destroyed, the IRGC maintains control through economic entrenchment (40%+ of economy), ideological selection, and internal security apparatus.
Opposition fragmentation is the binding constraint. 92% public dissatisfaction does not equal organised capacity for regime-threatening action. No parallel governance structures, unified leadership, or military coordination are visible.
Nuclear escalation continues regardless. 60-75% probability of Iran achieving 90% uranium enrichment at Fordow within a month, using multi-site hedging to survive Israeli interdiction.
The ceasefire is fragile and likely temporary. 40-55% probability of collapse within a month. Pakistan talks create a pause, not a resolution.
The definition bar is extremely high. Not just IRGC removal, but replacement with a government recognising Israel and renouncing eliminationism -- this requires a generational political transformation, not merely a coup.
Historical precedent is unfavourable. The IRGC has survived the Iran-Iraq War (1980-88), the 2009 Green Movement, the 2019 protests, and the 2022 Mahsa Amini uprising. External military pressure historically consolidates authoritarian regimes rather than toppling them.

Repository Contents

Iran-Regime-Change-Forecast-1004/
├── README.md
├── report/
│   ├── chairman_report.md          # Final synthesised forecast
│   ├── stage1_answers.md           # 6 individual lens answers
│   ├── stage2_reviews.md           # Blind peer reviews
│   ├── fresh_data.json             # Frozen Tavily + RSS news seed
│   └── simulation.json             # Full Stage A transcript
├── code_snapshot/                   # Frozen pipeline code at run time
│   ├── graph.py                    # LangGraph pipeline orchestrator
│   ├── pipeline.py                 # Backward-compat shim
│   ├── config.py                   # Model & simulation parameters
│   ├── llm.py                      # OpenRouter async client
│   ├── schemas.py                  # Pydantic schemas
│   ├── base_context.py             # Static conflict background
│   ├── pinecone_store.py           # Pinecone RAG integration
│   ├── actors/                     # Actor personas & roster
│   ├── simulation/                 # Stage A engine
│   ├── council/                    # Stage B protocol & lenses
│   ├── news/                       # Tavily + RSS grounding
│   ├── render/                     # Typst -> PDF rendering
│   └── pyproject.toml
├── graphics/
│   ├── convergence_all_lenses.png
│   ├── chairman_forecast.png
│   ├── key_predictions.png
│   └── 1yr_lens_spread.png
└── generate_graphics.py

Note: code_snapshot/ is a frozen snapshot of the pipeline at run time. For the latest code, see the parent repository.