A task is also a social object. Someone defines its semantics, someone operates it, someone accepts its quality thresholds, and someone decides how exceptions are handled. These responsibilities should be attached to the logical task rather than inferred from repository ownership.

Let responsibility be a mapping

where $K$ is a set of responsibility kinds and $A$ a set of accountable actors or roles. Different kinds include semantic approval, code maintenance, operational response, quality acceptance, and policy ownership.

This prevents the platform team from becoming the accidental owner of every concept simply because it executes the code. It also prevents domain teams from declaring ownership without accepting incident or change obligations.

The philosophy of action is relevant. An action is intelligible not only through its physical movement but through intention, context, and consequence. Likewise, a data task is not adequately described by CPU time and row count. It is an institutional action that transforms the evidential state of the organisation.

Conclusion

The task should be designed as the smallest unit that can be executed, explained, owned, tested, replayed, and audited without reconstructing hidden intent. This requires more than an orchestrator node. It requires contracts, preconditions, postconditions, typed dependencies, controlled effects, quality assertions, provenance, and recovery semantics.

When tasks are too coarse, errors become mysterious and change becomes expensive. When they are too fine, operations fragment into noise. The right granularity balances computational efficiency with epistemic clarity.

A disciplined Data Platform is therefore built from accountable actions. Its DAG is not merely a schedule. It is a formal narrative of how representations become governed claims. The platform becomes intelligible when its atomic unit of execution is also its atomic unit of responsibility.

Conceptual references

Hoare, C. A. R. “An Axiomatic Basis for Computer Programming” (1969). Dijkstra, E. W. A Discipline of Programming (1976). Milner, R. Communication and Concurrency (1989). Mac Lane, S. Categories for the Working Mathematician (1971). Anscombe, G. E. M. Intention (1957).