Combine scientific tools in one study

A study can use different instruments for different parts of a question. A fluid simulation may identify a region for a kinetic follow-up; a radiation calculation may need a material table from another package. The worker can choose those tools from the advertised catalogue rather than receive a fixed sequence of commands. The resulting claim still needs a traceable chain of inputs and numerical checks.

Expose instruments and give the physical question

Commission the required execution paths and give the study access to the relevant capability descriptors. Describe the question, domain and resource budget in the brief. State an instrument requirement when it matters scientifically, but leave the experimental choices to the agent where possible.

The minimal service’s required_capability_prefixes is an any-of allowance: one matching family satisfies that particular instrument check. Listing FLASH and WarpX there does not mechanically require evidence from both. Put a two-solver requirement in the immutable study brief and require the reviewer to inspect both sets of receipts and their data relationship.

Installed tools and their model validity are separate facts. For example, an EOS interpolator cannot generate missing atomic physics, and an opacity model is not an EOS. A missing dependency should lead to an explicit tool/model decision, not an invented table or an unrelated substitute simulation.

Preserve the handoff

Record the upstream experiment, actual output time, source file hash and the code that creates the downstream input. Each submission has its own workspace. Declare the transferred data and its metadata in inputs; files in another experiment are not automatically available in the next one.

Handoff	Information needed to interpret it
Fluid state → kinetic patch	Coordinates, units, physical scale, species, temperature partition, field/current/charge consistency, interpolation and boundary assumptions
Material generator → radiation solver	Composition, thermodynamic variables and units, model/data provenance, coverage, interpolation, conversion, consistency checks and actual table consumption
Simulation → diagnostic	Source state/time, control volume, normalization, estimator, output cadence and numerical uncertainty

Treat conversion and interpolation as numerical methods. Test them on a case with independently known quantities, then inspect the realized downstream state. A file that loads successfully may still have swapped coordinates or the wrong energy normalization. The island audit shows a concrete failure: mixing cell-centre and boundary origins produced false 15%-threshold rejections. A known analytic potential exposed the error that the worker and its reviewer had missed.

Review the combined conclusion

Use exploration and lab.analyze while preparing a method, then submit the relevant source, runtime, input identities and validation receipts with lab.method. Changed source or inputs require the applicable new review. Fresh evidence must match the reviewed method and any committed repair.

The reviewer should distinguish what each model can establish. A collisionless kinetic patch initialized from a resistive-MHD snapshot is a local follow-up; its rate is not automatically a corrected global MHD rate. An EOS evaluation alone does not establish radiative transport. Compare controls that can separate handoff artifacts from the physical mechanism being claimed.

The recorded FLASH-to-WarpX investigation provides source-controlled transfer helpers, completed kinetic controls and an audit of its unresolved outcome. The separate radiation tool-selection proposal is a prospective design, not a recorded result.

See guided commissioning, the minimal research API and the evidence model.