Versioned prompts & schemes
Available versions let you inspect changes to prompt templates and coding schemes alongside associated study records.
Cross-cutting · Reproducibility
Keep LLM study methods and trial records inspectable
Export prompts, parameters, coding and results so another researcher can assess what was run.
Record what the study used
Model responses can vary, provider behavior changes, and model aliases may resolve differently over time. Recording protocol details helps a reader inspect the study and understand what was done; it cannot guarantee exact repetition of an earlier response.
Review recorded provenance such as the design, resolved prompts, template and coding versions, model identifiers, parameters, seed status where available, and analysis inputs. Confirm that the records and exports match the materials you intend to report.
Version changes and recorded provenance
Full trial metadata
Trial records can include resolved prompts, model identifiers, sampling parameters, provider seed status where available, usage, cost, and validity details.
Method Specification Prompt
A generated method summary can help communicate design, prompts, parameters, and coding. Review it for omissions and inaccuracies before sharing.
Deterministic condition resolution
Inspect how design levels map to saved snippet versions and resolved text; verify that the saved materials represent the intended manipulation.
Provider drift and limits
Model identifiers can change
A model alias may resolve differently over time, and a provider may update or retire a model. Note the model and provider information available for each run.
Stochastic responses and seeds
A later response can differ even with the same prompt; a provider may not honor a requested seed. Report those limits rather than promising an exact rerun.
Review the export inventory
Reproducibility package
Enterprise subscribers can export the available study materials, trial data, coding, analyses, and method summary for review. Check package contents before treating it as a complete record.
Data export
Export raw or coded data in supported formats and inspect the files before re-use or sharing.
When reporting, check model identifiers, settings, replications, exclusions, and coding against the actual run. Review privacy, licensing, and institutional requirements before sharing prompts or response data; not every artifact will necessarily be shareable.
See the prompt-sensitivity walkthrough and review the reproducibility FAQ. For help drafting a report, see write-up assistance. The same documentation capability extends to project records in R&D documentation.
Study LLM behavior with a guided research workflow.
Design controlled text studies, collect model responses, and explore what changes across conditions. You decide what the evidence means.