Python authoring and execution

Python authoring and fixed-core execution

1. Edit and Save. The embedded Monaco editor edits Python text. The server validates syntax and automatically queues semantic validation of the exact immutable source, entry point, parameters and approved environment. Structured errors return to the browser or MCP client. Invalid source cannot run.
2. Isolated build. The trusted launcher starts the approved builder under runsc. Its private capability broker supplies authorized, revision-bound metadata. Authored Python composes a typed plan; the fixed server core validates that plan. The builder has no database credentials or publication authority.
3. Test. A valid inactive rule can compute private impacts. Numeric batches, ordered numeric sequences and scoped numeric transforms use the production arithmetic kernels against a consistent fact snapshot plus earlier proposed effects. No business values or revisions are published. Unsupported units are rejected before a result is described as complete.
4. Activate. An authorized action permits the exact validated version to change data. Test success alone does not activate it. Editing produces another inactive version.
5. Run. A separate Plan build and current execution authority support the active source version. The server checks authority and that exact active version before publication. Each unit has its own commit boundary; earlier committed units remain if a later unit fails. Validation and Test artifacts cannot run directly.

Metadata and result identity

Dimension names resolve against the selected application's metadata. Names do not introduce hidden business behavior. The saved source and its digest remain visible; the internal typed plan is execution evidence. Test results identify the source, plan, parameter values and baseline, with full coordinates, before/after values, deltas and per-unit outcomes. Results are private to their authorized caller and expire after 24 hours. Expiry removes the private payload while keeping its task identity for retry reconciliation.

See the MCP Python guide and Python environment administration.