Botmaster Phil Godzin’s Sudan conflict deaths forecast, Oct. 17, 2025

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

Phillip Godzin’s pgodzinai predicts today:

Less than 1,000: 1%
Between 1,000 and 3,000: 1%
Between 3,000 and 5,000: 3%
Between 5,000 and 8,000: 20%
Between 8,000 and 12,000: 60%
More than 12,000: 15%

10–13 minutes read (Full version of pgodzinai here —>)

GPT Multiple Choice JSON response:
Rationale
ACLED is the resolution authority and uses multi-source verification with a meaningful lag; their 2025 annual figure will reflect only events ACLED has verified and coded, not raw combatant claims. ACLED’s own methodology (2025 edition) notes a 2–4 week typical verification lag for events and that urban siege/major-battle incidents tend to have higher verification coverage because of multiple reporting actors. That makes ACLED’s annual total for Sudan conservative relative to unverified battlefield claims but relatively comprehensive where NGO/UN media access exists.

Multiple mid‑October 2025 sources and ACLED summaries/visuals (as cited in the research material) indicate substantial cumulative verified fatalities through ~mid‑October: the most credible synthesis in the research pointed to roughly ~8,000 verified state‑based fatalities in Sudan by early–mid October 2025. With ~2.5 months left in the year, the implied run‑rate and historical seasonality in Sudanese fighting (periodic spikes around major offensives and dry‑season operations) point toward additional verified deaths in Q4. Assuming no sudden, massive escalation or dramatic, large‑scale ceasefire, the most plausible outcome is that ACLED’s final 2025 verified total ends up in the next higher bin (8,000–12,000). That range is consistent with: (a) mid‑Oct cumulative ~8k, (b) a nontrivial but plausible additional toll in Nov–Dec that would push cumulative totals into the low‑to‑mid tens of thousands only under sustained heavy offensives, and (c) ACLED’s tendency to undercount relative to combatant claims but to cover urban, heavily reported battles well.

I therefore assign the highest probability to the 8,000–12,000 bracket, a meaningful but smaller probability to 5,000–8,000 (if the year’s remaining months are unusually quiet or verification lags suppress later coding), and a non‑negligible tail above 12,000 to capture (i) the risk of a sudden major offensive or mass casualty events in Nov–Dec that are well documented, or (ii) the possibility that ACLED’s mid‑Oct visual was a partial/incomplete snapshot and their verified count is already significantly higher than the ~8k figure. The lower three bins (<1000; 1000–3000; 3000–5000) are extremely unlikely given the documented scale of conflict through Q3 2025 and the multiple independent reports of thousands of casualties.

Calibration note: I’ve reduced extreme confidence and maintained a meaningful tail for >12k because of the potential for concentrated high‑fatality operations in urban siege settings. I also penalized my own prior overconfidence by keeping probabilities somewhat diffuse across adjacent bins.

Final probabilities
Less than 1,000: 1%
Between 1,000 and 3,000: 1%
Between 3,000 and 5,000: 3%
Between 5,000 and 8,000: 20%
Between 8,000 and 12,000: 60%
More than 12,000: 15%

Anthropic response:
Looking at the available evidence, I need to carefully analyze ACLED’s reporting on Sudan conflict deaths for 2025. The most critical piece of information comes from the ACLED visualization, which shows approximately 8,000 state-based conflict deaths reported in Sudan through mid-October 2025. This is authoritative data directly from ACLED, the resolution source.

The timing is crucial here. With the current date being October 17, 2025, and the resolution on January 1, 2026, there are approximately 2.5 months remaining in the year. The ACLED data shows we’re at roughly 8,000 deaths with about 79% of the year complete (290 days out of 365). This suggests a straightforward extrapolation would yield around 10,000-10,500 deaths for the full year if current rates continue.

However, several factors complicate this projection. First, ACLED typically has a 2-4 week verification lag, meaning recent October deaths may not be fully captured yet. The visualization shows some moderation in recent months compared to earlier peaks in 2025, but conflict dynamics remain volatile. The research indicates ongoing intense fighting in El Fasher and other urban areas, with the RSF making territorial gains as recently as late September.

Historical context is important. The ACLED visualization shows 2024 had approximately 12,000-13,000 total fatalities for the full year, while 2023 had 11,000-12,000. The 2025 trajectory appears somewhat lower than these recent years, which aligns with the current count of 8,000 through mid-October. The conflict shows typical seasonal variations rather than linear progression, with fluctuations throughout the year.

The research from various sources mentions higher estimates – some citing 30,000 deaths – but these appear to either be cumulative since the conflict began in 2023 or include broader categories beyond state-based violence. The ACLED visualization is the authoritative source and shows clearly that we’re at approximately 8,000 deaths through mid-October 2025.

Key uncertainties include: First, whether violence will escalate significantly in the final quarter. The September reports of RSF gains and ongoing urban warfare suggest continued high-intensity fighting, but the recent moderation shown in the ACLED graph suggests this may not dramatically spike. Second, ACLED’s verification process means the final count could be somewhat higher than current reports as they verify and add recent events. Third, any major military offensive or peace agreement could drastically alter the trajectory, though neither appears imminent based on current reporting.

Given the current count of 8,000 with 2.5 months remaining, the mathematical projection points toward 10,000-11,000 total deaths. The “Between 8000 and 12000” category appears most likely, as we’re already at the lower bound of this range. It’s highly unlikely we’d see fewer than 8,000 (we’re already there), and reaching above 12,000 would require an average of over 1,600 deaths per month for the remainder of the year – significantly higher than recent monthly averages shown in the visualization.

The possibility of falling into “Between 5000 and 8000” exists only if ACLED significantly revises downward their current count, which is unlikely given their methodology. The “More than 12000” category would require a dramatic escalation beyond what current trends suggest. The lower categories are essentially impossible given we’re already at 8,000 deaths.

Less than 1,000: 0%
Between 1,000 and 3,000: 0%
Between 3,000 and 5,000: 0%
Between 5,000 and 8,000: 2%
Between 8,000 and 12,000: 83%
More than 12,000: 15%

Model’s Answer:
Rationale
ACLED is the resolution authority and uses multi-source verification with a meaningful lag; their 2025 annual figure will reflect only events ACLED has verified and coded, not raw combatant claims. ACLED’s own methodology (2025 edition) notes a 2–4 week typical verification lag for events and that urban siege/major-battle incidents tend to have higher verification coverage because of multiple reporting actors. That makes ACLED’s annual total for Sudan conservative relative to unverified battlefield claims but relatively comprehensive where NGO/UN media access exists.

Multiple mid‑October 2025 sources and ACLED summaries/visuals (as cited in the research material) indicate substantial cumulative verified fatalities through ~mid‑October: the most credible synthesis in the research pointed to roughly ~8,000 verified state‑based fatalities in Sudan by early–mid October 2025. With ~2.5 months left in the year, the implied run‑rate and historical seasonality in Sudanese fighting (periodic spikes around major offensives and dry‑season operations) point toward additional verified deaths in Q4. Assuming no sudden, massive escalation or dramatic, large‑scale ceasefire, the most plausible outcome is that ACLED’s final 2025 verified total ends up in the next higher bin (8,000–12,000). That range is consistent with: (a) mid‑Oct cumulative ~8k, (b) a nontrivial but plausible additional toll in Nov–Dec that would push cumulative totals into the low‑to‑mid tens of thousands only under sustained heavy offensives, and (c) ACLED’s tendency to undercount relative to combatant claims but to cover urban, heavily reported battles well.

I therefore assign the highest probability to the 8,000–12,000 bracket, a meaningful but smaller probability to 5,000–8,000 (if the year’s remaining months are unusually quiet or verification lags suppress later coding), and a non‑negligible tail above 12,000 to capture (i) the risk of a sudden major offensive or mass casualty events in Nov–Dec that are well documented, or (ii) the possibility that ACLED’s mid‑Oct visual was a partial/incomplete snapshot and their verified count is already significantly higher than the ~8k figure. The lower three bins (<1000; 1000–3000; 3000–5000) are extremely unlikely given the documented scale of conflict through Q3 2025 and the multiple independent reports of thousands of casualties.

Key assumptions and failure modes
1) Assumption: ACLED’s mid‑October verified cumulative for 2025 is roughly ~8,000 (i.e., the research visual/summary is representative). Probability this assumption is wrong (i.e., the verified count is already much higher, e.g., >12k or even ~30k): 12%. If wrong upward (true mid‑Oct ACLED count >>8k), probability mass shifts up toward “>12k.”
2) Assumption: Conflict intensity in Nov–Dec 2025 will be similar to recent months (some continued fighting, but not a single catastrophic nationwide spike). Probability this assumption fails (i.e., there is a massive escalation or near‑total collapse/ceasefire altering deaths materially): 18%. If escalation happens, upward tail probability (>12k) increases substantially; if a ceasefire happens, probability shifts down to 5k–8k bin.
3) Assumption: ACLED’s verification practices will keep the final reported number materially below unverified combatant claims but not so low as to undercount widely reported urban massacres. Probability this assumption fails (i.e., ACLED undercounts dramatically because of access/verification constraints): 15%. If ACLED undercounts, final bin could be lower despite actual deaths—this mostly affects mid bins (5k–8k vs 8k–12k).

Calibration note: I’ve reduced extreme confidence and maintained a meaningful tail for >12k because of the potential for concentrated high‑fatality operations in urban siege settings. I also penalized my own prior overconfidence by keeping probabilities somewhat diffuse across adjacent bins.

Final probabilities

Less than 1,000: 1%
Between 1,000 and 3,000: 1%
Between 3,000 and 5,000: 3%
Between 5,000 and 8,000: 20%
Between 8,000 and 12,000: 60%
More than 12,000: 15%

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 competitors. Our botmasters, Phil Godzin and Jeremy Lichtman, are conducting a side experiment with VIEWS.