NLSS
A conversation you can check
The Natural Language Statistics Suite (NLSS) connects a researcher's question with statistical procedures in R. You work through a local agent: describe your study, discuss the analysis and ask what the findings mean. NLSS supplies reusable procedures and preserves their inputs, settings and results.
The researcher remains responsible for the design, methodological choices and conclusions. The agent helps make those choices explicit, invokes NLSS, reads the output and writes a response or report. Statistical calculations follow the selected R routine; the agent's explanation is freely written and can vary.
A result with a trail behind it
An analysis can leave you with a table, a figure and a readable report. It also leaves a saved calculation record: which data were used, which options were selected and what R returned. The automatic protocol and an authored report serve different purposes. The protocol records numerical evidence; the report explains its meaning in the study's context.
See both for the learning study: read the result section, then inspect its analysis protocol. Can you follow a reported number back to the calculation?
For supported runs in a complete current project, exact replay checks the preserved input and recorded execution environment before repeating the calculation. Keep the complete project when moving or sharing it. Reproduction depends on the data, software versions, settings and random seeds where relevant; it is not a promise that every future environment or agent will produce identical prose. See the run contract and replay guide.
This gives conversational work a reusable statistical foundation. Menu-driven statistical software and hand-written R can also support reproducible analysis. NLSS makes fixed procedures and saved records part of an agent-assisted workflow, reducing the need to generate a new analysis implementation for each conversation. The researcher still has to judge whether that procedure fits the question.
Local calculation, an explicit model choice
R runs in your local working environment. That does not by itself mean every part of the conversation stays on your computer: an agent may use a remote model, depending on your setup. Choose the agent and model arrangement appropriate to your data. This website's teaching studies are synthetic and can be explored without sharing real participant records.
The examples demonstrate defined workflows with independently checked numerical results. They do not establish the validity of an instrument, the effectiveness of an intervention or universal correctness across all supported analyses. Explore the examples or find a method.
Project and citation
NLSS is developed by Mike Hammes and distributed as open-source software under Apache-2.0. The analyses on this site use version 2.0.1, released 17 September 2026. The release and citation file identify the software behind these examples.
To cite the version used here:
Hammes, M. (2026). docmh/nlss: NLSS (v2.0.1) [Computer software]. Zenodo. https://doi.org/10.5281/zenodo.22808408
Cite the version you actually use. The concept DOI identifies the software across versions; it is distinct from this version's DOI.
Original teaching data, text and figures are shared under CC BY 4.0; original generator, verification and website code under Apache-2.0. Attribute Mike Hammes / NLSS, identify the material version and indicate changes. Third-party software and assets retain their own terms. The material license does not grant general rights to the NLSS name or mark.