Run Binding refers to linking a specific AI run to a defined, fixed data state. It is the connection that makes each execution explainable in terms of the data conditions it ran under.
When a run finishes, only the answer remains. Which schema, which field definitions, which applied policy produced it disappears unless it is bound at the time. What an audit asks for is not the result but the conditions behind it, and without that link the result cannot be defended.
This differs in purpose from experiment tracking. Experiment tracking records hyperparameters and metrics so training runs can be compared. Run binding ties a production execution to a data state so it can be returned to under the same conditions. A bound run can be re-run on the same saved data state, and if the data has changed since, a comparison shows what changed. In Syntitan, this is handled as a run record: each run records the data version, settings and evaluation conditions it used, so you can re-run on the saved version and compare it with the previous version.