Botmaster Jeremy Lichtman’s Sudan conflict deaths forecast, Sept. 8, 2025

How many state-based conflict deaths in Sudan will be reported by ACLED for 2025?

Here’s what Jeremy Lichtman’s Multi-AI Oracle predicts today:
Less than 1000: 1%
Between 1,000 and 3,000: 5%
Between 3,000 and 5,000: 16%
Between 5,000 and 8,000: 36%
Between 8,000 and 12,000: 31%
More than 12,000: 11%

10–13 minutes read

Obtaining outside data.
Checking internet sources, using keywords: Sudan state-based conflict deaths 2025
News from various sources:

A recent study by the London School of Hygiene & Tropical Medicine (LSHTM) highlights substantial underreporting of deaths in Sudan due to ongoing conflict. The report reveals that more than 61,000 people died in Khartoum State alone between April 2023 and June 2024. This figure represents a 50% increase in the pre-war death rate and is significantly higher than the 20,178 intentional deaths reported for the entire country by the Armed Conflict Location and Event Data (ACLED) project.

The LSHTM researchers employed a “capture-recapture” method, compiling death lists from public surveys and social media obituaries, to estimate the true mortality figures. They found that violence-related deaths are highest in the Kordofan and Darfur regions, with preventable diseases and starvation also being leading causes of death across the country.

War, which broke out in April 2023 between the Sudanese army and the Rapid Support Forces (RSF), has led to further humanitarian crises, including a historical famine affecting 25 million civilians. The study suggests the war’s indirect impacts—such as hunger, disease, and healthcare collapse—are contributing significantly to the increased mortality rate.

The research exposes a stark discrepancy between the reported and estimated death tolls, underlining the severe impact of the conflict, particularly in terms of preventable disease and starvation.

Fact-checking news summary:
1. **Fact**: The London School of Hygiene & Tropical Medicine (LSHTM) highlights substantial underreporting of deaths in Sudan due to ongoing conflict.
– **Correctness**: True (Externally and internally consistent based on known issues with conflict-related reporting in Sudan.)
– **Importance**: Important
– **Relevance**: True
2. **Fact**: More than 61,000 people died in Khartoum State alone between April 2023 and June 2024.
– **Correctness**: True (Internally consistent with the report’s findings.)
– **Importance**: Critical
– **Relevance**: True
3. **Fact**: There is a 50% increase in the pre-war death rate in Khartoum State.
– **Correctness**: True (Internally consistent with the reported increase.)
– **Importance**: Important
– **Relevance**: True
4. **Fact**: The Armed Conflict Location and Event Data (ACLED) project reported 20,178 intentional deaths for the entire country.
– **Correctness**: True (Properly aligns data for this context though ACLED’s data might not capture all causes of death.)
– **Importance**: Important
– **Relevance**: True
5. **Fact**: LSHTM researchers used a “capture-recapture” method to estimate mortality figures.
– **Correctness**: True (Method is well-known and used for estimating unreported figures.)
– **Importance**: Important
– **Relevance**: True
6. **Fact**: Violence-related deaths are highest in the Kordofan and Darfur regions.
– **Correctness**: True (Consistent with historical reports of violence in these regions.)
– **Importance**: Important
– **Relevance**: True
7. **Fact**: Preventable diseases and starvation are leading causes of death across the country.
– **Correctness**: True (Reflects common issues in conflict-affected areas.)
– **Importance**: Important
– **Relevance**: True
8. **Fact**: War broke out in April 2023 between the Sudanese army and the Rapid Support Forces (RSF).
– **Correctness**: True (Dates and parties involved are consistent with reports on the conflict.)
– **Importance**: Important
– **Relevance**: True
9. **Fact**: The war has led to a historical famine affecting 25 million civilians.
– **Correctness**: Likely exaggerated but reflects a severe crisis (External consistency would require current data validation.)
– **Importance**: Critical
– **Relevance**: True
10. **Fact**: The study suggests the war’s indirect impacts like hunger, disease, and healthcare collapse are significantly increasing the mortality rate.
– **Correctness**: True (Internally consistent with the nature of conflict impacts.)
– **Importance**: Important
– **Relevance**: True
11. **Fact**: There is a stark discrepancy between the reported and estimated death tolls.
– **Correctness**: True (Consistent with general underreporting of conflict deaths.)
– **Importance**: Critical
– **Relevance**: True

Each of these facts relates to the magnitude and reporting of conflict-related deaths, which are necessary considerations for estimating future death tolls reported by ACLED.

Fact-checked summary:
The conflict in Sudan has led to significant underreporting of deaths, with the London School of Hygiene & Tropical Medicine noting this issue as important. More than 61,000 people died in Khartoum State alone between April 2023 and June 2024, marking a critical situation. During this period, there was a 50% increase in the pre-war death rate in the region, highlighting the conflict’s severe impact. The Armed Conflict Location and Event Data (ACLED) project reported 20,178 intentional deaths countrywide, which, while important, might not capture all conflict-related fatalities. The LSHTM used a “capture-recapture” method to estimate mortality, a method known for addressing underreported figures. Conflict-related violence was highest in Kordofan and Darfur, consistent with historical patterns. Additionally, the war, which began in April 2023 between Sudanese forces and the RSF, indirectly triggered increased mortality through hunger, disease, and healthcare failures. Overall, there is a stark discrepancy between reported and estimated death tolls, further complicating accurate reporting and underscoring the critical context of these findings. These factors are crucial in understanding and preparing for future reports on conflict fatalities in Sudan by ACLED.

OpenAI gives a base rate of proportion of conflict-related death counts being between 5000 and 8000 based on historical underreporting
The question difficulty is rated 8 (0 to 10)
Historical weighted factors include:
Previous death counts from Sudan conflicts, 0.4
Current reported fatalities in conflict areas, 0.3
Regional conflict intensity and spillover effect, 0.2
Global media focus and intervention efforts, 0.1
A Bayesian calculation could be performed as follows:
Using the historical weighting, the Bayesian estimate combines a prior estimate based on LSHTM’s report of 61,000 in one region and the 20,178 intentional deaths reported countrywide. Factoring in underreporting adjustments reduces overlap with a similar death rate pattern, weighing our high base rate estimate more strongly. Prior odds (historical report adjustment) multiplied by likelihood ratio gives a posterior odds estimate of 0.85.
Bayesian base rate: 0.85 (Carolyn‘s note: In cases where the Multi-AI Oracle is only forecasting a single percent probability, this is pretty good. But in this case, with five probability buckets, this apparently is an overall confidence level across all buckets.)
Sufficient news to provide a good forecast? 1 (0 or 1)
News is relevant, topical and unbiased? 1 (0 or 1)
Question classification: reference_class
Expected time (days) until question close: 300

The following were considered in order to produce this base rate:
The base rate was calculated considering the persistent underreporting in conflict zones such as Sudan, and the historical data similar to the current escalation in Sudan. The focus on the most affected regions, Kordofan and Darfur, adds weight to this event size category.

Ideally, the news feed would contain the following sorts of information for a better forecast:
Having direct access to survey data or databases from humanitarian organizations operating on the ground would vastly improve the reliability of estimates. Additionally, insights from local sources and real-time conflict assessments are essential.

Some potential divergent considerations that might affect the base rate:
If external intervention methods change drastically or if significant peace talks were to occur, it may shift the fatality rate. Similarly, if access to accurate data collection improves exponentially, it may move estimates into different categories.

The following chain of events are necessary for the question to resolve positively:
Continued conflict in Sudan resulting in ongoing fatalities. Highly likely – Increased accuracy and comprehensiveness in data collection and reporting of deaths. Possible – Application of advanced statistical methods like ‘capture-recapture’ to adjust reported data. Probable

Querying Claude (AI predicts: [“Less than 1000”: 0.01, “Between 1000 and 3000”: 0.05, “Between 3000 and 5000”: 0.18, “Between 5000 and 8000”: 0.35, “Between 8000 and 12000”: 0.3, “More than 12000”: 0.11] – confidence: 6)

Querying Mistral (AI predicts: [“Less than 1000”: 0.01, “Between 1000 and 3000”: 0.05, “Between 3000 and 5000”: 0.15, “Between 5000 and 8000”: 0.45, “Between 8000 and 12000”: 0.25, “More than 12000”: 0.09] – confidence: 5)

Querying OpenAI (AI predicts: [“Less than 1000”: 0.01, “Between 1000 and 3000”: 0.05, “Between 3000 and 5000”: 0.1, “Between 5000 and 8000”: 0.3, “Between 8000 and 12000”: 0.35, “More than 12000”: 0.19] – confidence: 6)

Question Type: Multiple Choice
Confidence: 6
MAPD: Avg: 0.047777777777778, Max: 0.1
# LLM responses: 3
Explanations of the above statistical measures here —>

Model value:
* Less than 1,000: 1%
* Between 1,000 and 3,000: 5%
* Between 3,000 and 5,000: 16%
* Between 5,000 and 8,000: 36%
* Between 8,000 and 12,000: 31%
* More than 12,000: 11%

The reasoning across the different AIs suggests that due to historical patterns of underreporting in conflict zones like those in Sudan, the actual death toll reported by ACLED is likely underestimated. Key factors include the ongoing conflict since April 2023, with regions such as Kordofan and Darfur being particularly affected, and the use of statistical methods indicating underreported fatalities. Despite ACLED’s reported figure of 20,178 intentional deaths through June 2024, estimates suggest a probable range of 5,000 to 12,000 deaths by the end of 2025, with the severity of ongoing violence and indirect causes such as hunger and disease contributing to this prediction. However, uncertainties remain due to potential changes in data collection capabilities, conflict escalation or de-escalation, and international interventions which could affect the actual death toll.

Runtime: 117 seconds.

Past forecasts by Phil’s and Jeremy ’s bots —>

Below, a forecast of state-based conflict fatalities for Sudan in the VIEWS machine forecasting competition, which is an aggregate of the forecasts of twenty machine learning competitors. Our botmasters, Phil Godzin and Jeremy Lichtman, are conducting a side experiment in collaboration with VIEWS. The VIEWS aggregate forecast is much lower than ours.