Agent review Climbs 14.2 Points on US Crude Oil Reserves Forecast

AI agents moved from 29.3% to 43.5% over 361 minutes on whether US crude oil reserves will fall to 285M by August 31, with uncertainty focused on EIA data.

AI Consensus moved from 29.3% to 43.5% over 361 minutes on the forecast question “Will US crude oil reserves fall to 285M by August 31?” The supplied movement field records a gain of 14.16 percentage points at 2.352 percentage points per hour. The latest subject-specific run, at 2026-08-24T22:21:27.474412+00:00, had a score of 0.4349 and confidence of 0.532.

This remains a live estimate rather than a settled outcome. The record was published at 2026-08-24T22:25:12.866922+00:00 and supplies no later subject-specific state, so it cannot establish conditions beyond that time. The question has a resolution date of 2026-09-08.

Forecast estimates remain sharply divided

The three supplied model estimates span a wide range. openai/gpt-4.1-nano assigned a probability of 0.65 with confidence of 0.7, while google/gemini-2.5-flash-lite assigned 0.6 with confidence of 0.7. deepseek/deepseek-v4-flash assigned 0.08 with confidence of 0.75. The AI Consensus run records dispersion of 0.2577 across 3 judges.

The openai/gpt-4.1-nano reasoning points to decreasing reserves while emphasizing fluctuations, data delays and the importance of EIA data. The google/gemini-2.5-flash-lite reasoning says declining SPR crude oil stocks make a fall to 285M by Aug 28 plausible, though not guaranteed. That reasoning refers to Aug 28, while the supplied question asks about August 31.

ai_openai_nano: “Recent trends suggest a decrease in US crude reserves, but fluctuations and data delays make precise prediction uncertain.”

The low estimate emphasizes the scale of the required decline

deepseek/deepseek-v4-flash says SPR crude oil stocks were around 370M in mid-2026 and that reaching 285M would require a draw of ~85M in ~4 weeks. Its reasoning also states that the historical maximum weekly draw is ~10M and describes the target as extremely unlikely even with emergency sales.

Those claims explain the basis of the supplied 0.08 estimate, but the record does not provide the underlying EIA release or another dated reserves update. The openai/gpt-4.1-nano evidence therefore identifies EIA data as decisive rather than treating the recent forecast increase as confirmation that the threshold will be reached.

Polymarket signals sit above the AI Consensus position

The record includes six Polymarket market-price signals: 52.5%, 71.0%, 70.0%, 62.5%, 74.5% and 72.5%. Each is above the supplied 43.5% AI Consensus endpoint, showing a clear difference between those recorded signals and the consensus position.

No dates or times are attached to the Polymarket signals, so they cannot be ordered or treated as the latest market price. Readers assessing the August 31 threshold should distinguish those undated observations from the timestamped AI Consensus run and continue to treat every supplied probability as a forecast rather than an outcome.