A model’s answer is probabilistic, and it can differ from one run to the next. Research needs work that stays the same when someone checks it. These five steps keep each finding checkable: what produced it, whether it still holds, and where it disagrees with the rest of the project.
The five steps
Record
Each finding keeps its source: the session, the files it read and the tool calls behind it.
Re-run
Recorded tool calls run again, with no AI in the loop, to show whether the result still holds. Terminal commands are kept on record, not replayed.
Replicate
In a planned research task, a separate agent redoes the work and compares the numbers within set limits.
Flag
Numbers with no source, and findings that disagree, are marked for you to see.
You decide
You see the evidence for each result, then decide what to trust and what to build on.
How Athar applies it
Athar is a research workspace that runs on your own computer, and it applies this protocol to your project. It keeps the record behind each finding, re-runs recorded steps, replicates claims in planned research tasks, and shows you where findings disagree. You work with AI agents and choose the model for each step.