Knowledge work · 18 Aug 2026
The frontier should remember
A research system becomes more useful when it can retain the evidence, decisions and corrections that changed earlier conclusions.
A second research assignment should start with the useful work from the first. That requires more than loading yesterday’s answer into context. A later reader needs to know which source supported a conclusion, when that observation was made and which correction changed it.
For financial work, the date matters as much as the number. Feeding a later price or an amended filing into an earlier decision changes the question under test. Keeping dated evidence alongside the conclusion lets an agent reopen its work without quietly rewriting what was known at the time.