A Vahtian product
An AI-assisted analysis can look finished while the reasoning behind it is scattered across prompts, generated code, and reruns. StudyVahti Vault keeps the whole observational study in one local workspace — question to claim — and checks whether it still holds together. A study you can defend, on your own machine.
Get StudyVahti Vault · €49Launch price · regular €99 · one payment, all 1.x updates included · 14-day refund if the demo study does not run end to end on your machine
The final analysis looks coherent, but the path to it is not. Which code the model wrote, why an exclusion happened, whether Table 1 still comes from the current script — the answers live in a chat you can no longer find. Traditional reproducibility asks whether the analysis can be rerun. AI-assisted work has to answer a harder question: can it be reconstructed and defended? StudyVahti Vault is built for that second question.
| The coherence chain, checked | research question, dictionary, plan, decisions, outputs, and claims — linked, with 15 checks that flag drift between them |
| Analysis from your own notes | the engine fits the model (risk ratio, odds ratio, hazard ratio, mean difference), every sensitivity analysis, Table 1, participant-flow and missingness maps — no numbers typed by hand |
| A causal diagram that checks itself | draw the DAG; the vault flags a mediator or collider that slipped into the adjustment set |
| A provenance graph + audit trail | the visual, tamper-evident record of how the study holds together, from data to claim |
| A study assistant built in | drafts methods, proposes sensitivity analyses, checks consistency — on your local model or in Claude Code; it suggests, you decide |
| A complete demo study | a fabricated cohort worked end to end, so you learn the whole workflow by watching before you start |
Models suggest structure. You decide the analysis. The assistant drafts, proposes, and checks — but every output lands as a suggestion for your review, and it flags what it cannot trace rather than inventing it. It records support for a study's defensibility, never whether its conclusions are true.
Estimands are named before models. Verdicts are three-way — ok / flag / not checkable, and robust / not robust / not resolvable at this n — because a threshold on a noisy estimate should not be forced into pass or fail. Copy stays with associated with, not causes, unless the design earns it — and causal wording is flagged for your review. Your data never enters the vault: the Data Room keeps a fingerprint, never the rows.
Turn a study idea into a protocol-ready plan: PICOTS, estimands, reporting standards, feasibility.
Run and document the study: computed analysis, decisions, sensitivity analyses, and 15 checks.
Plan with the free Planner, run and document here, check the trail with the free AuditLite. The Vault is the middle rung.
A complete, fabricated cohort study ships inside — worse performance status and one-year mortality — worked from the research question all the way to the manuscript claims. You can see the computed forest plot, the reconciled participant flow, the provenance graph, and every check passing. Ten minutes of reading and you know what each stage of your own study should look like.
auditlite packageStudyVahti Vault is built by Heidi Andersén, MD PhD, a thoracic-oncology clinician-scientist in Finland who runs observational studies herself. The workflow comes out of an active research practice, not a template shop. ORCID · About Vahtian
No. Your rows never enter the vault at all — the Data Room records only where the data lives and its SHA-256 fingerprint. The only network address in the scripts is localhost, and the assistant runs on a model on your own machine.
You need to know your study. The Vault structures the analysis and computes the common models (risk ratio, odds ratio, hazard ratio, mean difference) and the sensitivity analyses from your own notes. It does not choose your science, and it does not replace a statistician — it makes the analysis you run easier to check and defend.
No. You still run your models in R, Stata, or SAS if you prefer. The Vault is the workspace that keeps the question, the decisions, the outputs, and the claims coherent and traceable — and it can compute and check the common models for you.
Covered. The scripts are open source (Apache-2.0), and the vault builds STROBE coverage and methods and results drafts from your own notes, each sentence traceable to a note.
QualiVahti Local is the same local-first, human-first workspace idea for qualitative interview studies. StudyVahti Vault is for quantitative observational studies — the unit of work is the variable and the analytic decision, not the transcript and the code.
Plan the study for free with StudyVahti Planner. When you are running and writing it up — and want the analysis computed, the decisions logged, and the whole thing checked for drift — StudyVahti Vault is the workspace.
Get StudyVahti Vault · €49Secure checkout by Polar (merchant of record, handles EU VAT). Instant download. Lab and doctoral-school licenses: hello@vahtian.com