PHREEQC
Geochemical simulation software developed by the United States Geological Survey for simulating aqueous chemical equilibria and reaction paths, featuring deterministic output and traceability.
Timeline (last 90 days)
When the agent reads the PHREEQC simulator report via a section index rather than context, GPT-5.1's accuracy drops by 11.3 percentage points.
After re-running all 11,007 logged simulator calls, it was found that the two GPT models computed the key values about as frequently as before.
PHREEQC is a geochemical simulator used by the U.S. Geological Survey; its reports can be up to 846,800 characters long and are deterministic.
The agent used PHREEQC to answer 158 execution-verified multiple-choice questions, in order to precisely measure the impact of retrieval on accuracy.
All Facts (4)
When the agent reads the PHREEQC simulator report via a section index rather than context, GPT-5.1's accuracy drops by 11.3 percentage points.
50%UnverifiedAfter re-running all 11,007 logged simulator calls, it was found that the two GPT models computed the key values about as frequently as before.
50%UnverifiedPHREEQC is a geochemical simulator used by the U.S. Geological Survey; its reports can be up to 846,800 characters long and are deterministic.
50%UnverifiedThe agent used PHREEQC to answer 158 execution-verified multiple-choice questions, in order to precisely measure the impact of retrieval on accuracy.
50%