~90 min
My first qualitative study
- Choose the project
- Set your rigour
- Transcribe locally
- Build the codebook
- Code with review
- Decide it’s enough
Next step: QualiVahti Local →
Vahtian Learn
What are you stuck on right now?
Most people land here mid-panic: the sample your supervisor just questioned, the references that broke the night before the deadline, the reviewer comment you can’t decode. Pick the moment. Get one clear next step, and the sentence you can show your supervisor.
Browse by stage
The research journey
Find where you are stuck. Each stage is one decision and the guides that settle it.
I can’t turn my idea into a question
Turn a gap into a question you can actually finish.
My supervisor says my sample is too small
Build the sample-size argument you can put in front of them.
- How do I justify my qualitative sample size?A four-part sample-size argument you can paste into your methods, with a worked example.
- Registered Reports: reviewed before you run the studyGet the design reviewed while it is still cheap to change, and use the right checklist for the stage.
- The three questions of rigourWhat counts as knowledge, who interprets, and how it is made credible.
- Deductive, inductive, abductiveThree kinds of reasoning in plain terms, and why mixing them up weakens your argument.
- Code saturation and meaning saturationTwo kinds of “enough”, and how to decide when to stop.
My data won’t load, or isn’t clean enough to analyse
You cannot run a single test until the data is clean, structured, and de-identified. Do this first.
- Clean and prepare your data, in R and PythonThe whole workflow on one page: missing values, types, validation, duplicates, merges, and de-identifying before you share.
- Build a data dictionary that validatesDefine variables, types, allowed values, and validation before you collect.
- My survey export has two header rowsCombine the headers and reshape a SurveyMonkey or Qualtrics export for R and Python.
- My numbers turned to text in the CSVDecimal commas, semicolons, and locale, and how to import cleanly.
I don’t understand my own statistics
Read the few things that change your conclusion, in plain words.
This review is drowning me — when do I stop?
Set a defensible stopping rule, and screen a slice at a time.
Can I collect this without breaking the rules?
Keep sensitive data on your machine, and anonymise it properly.
I’ve coded everything — now what are my themes?
Get from codes to themes, with a worked example.
- I have coded my data. How do I find the themes?Codes are not themes: the descriptive-to-analytic move, with a worked example and a trail.
- You coded everything and still have no theoryThe step after themes: challenging, seeing, articulating, and the conceptual leap note.
- Local AI coding, with human reviewThe model suggests first-cycle codes; you review every one and record the decision.
- Writing discipline, and the detectorsThe literature's vocabulary moved in two years. Detectors flag careful multilingual prose at 61%.
- What LLMs can do, and where they stopWhere models help, five ways their output is over-trusted, and a workable division of labour.
- Canvas as an evidence mapEvery box is a live note, not a drawing. Trace a claim back to its source.
- A minimal model for qualitative textA small, interpretable classifier for coded text, and how to check it holds up.
I’m staring at a wall and can’t write
Beat the blank page with structure, not willpower.
- Agile writing. Four decisions that finish a paperA stalled paper is usually stuck on one decision. Find yours with a free self-check.
- Write from the results outFigures and methods first, introduction near the end, abstract last.
- My figures are “too low-resolution”. What do they want?Formats, DPI, font sizes, and colour-blind-safe colour, with a pre-submission checklist.
- Can every reader tell my figure colours apart?Build a palette from one colour and test it under simulated colour-blindness, in Python or R.
- My bibliography is broken. How do I recover it?Recover a corrupted Zotero or EndNote document, then stop it recurring.
- Zotero for researchersCollect, organise, annotate, and cite, plus FullVahti and Better BibTeX.
- Make your analysis code citableGive the version you actually ran a release, a DOI, and a code-availability statement.
The reviewers came back and I don’t know what they want
Triage the comments, decode the feedback, keep it polite.
- Triage a major-revisions letterProcess the hit, sort the comments, and handle two reviewers who contradict.
- Respond without losing a commentGive every request a stance, response, real change, and location.
- What does ‘more depth’ actually mean?The five vague supervisor comments, translated into concrete fixes.
- Every research project needs a CHANGELOGAnswer “why did you change that?” from a dated entry instead of memory.
I submitted and now they want a huge fee
Check the red flags, and know your options now.
Using AI without getting into trouble
Get AI’s help without sending your data away or letting it decide.
- ML, LLM, RAG, GraphRAG, DAGs and agentsOne map of the AI ideas behind research tools, and what each does not establish.
- Before interview data enters an AI modelThree risks, four checks, and when a local workflow removes the data transfer.
- GenAI without breaking your methodWhy a chatbot invents quotes, and how to use a model as a reading partner instead.
- Write an AI-use disclosureName the tool, version, task, timing, and human review. “AI was used” is not a method.
- What should I say about AI use in my paper?Four example statements — brainstorm, draft, edit, analysis — to copy and adapt.
- Document AI-assisted analysisRecord provenance, decisions, runs, and claims before memory becomes the method.
- Can a local model check your citations?Two laptop LLMs on 120 claim-vs-source pairs, and why agreement is not accuracy.
- Local models, agents & orchestrationRunning models locally and coordinating them, while the human keeps deciding.
- Agent memory & the pruning research gardenFour ways agents remember, where each leaks, and who should hold the pruning shears.
When you’re the reviewer
Judge a paper fairly, and defensibly.
Or follow a path
Learning paths
A guided route through one project, start to finish. Under two hours of reading each.
~70 min
Writing my first paper
- Write from the results out
- Make the code citable
- Disclose AI use
- Check the reference list
- Answer the reviewers
Next step: Manuscript Kit →
~60 min
Using AI without getting into trouble
Next step: AI Disclosure Kit →
Observational
Design a study you can defend
- Pin the question and the estimand
- Build the variable dictionary
- Lock the analysis plan before data
- Keep every decision in the audit trail
Next step: StudyVahti Vault →
~40 min
Organise your notes by what you know
- The epistemic notes method
- Keep a research changelog
- Set up the Obsidian vault
- Map the argument on a canvas
Next step: PhD Thesis Vault →
Try it in your browser
Small checks, no installation
Nothing uploads. Each runs on your own machine.
Pre-Submission CheckRun the whole chain in order before you submit: statements, reporting-guideline items, references, claims. It lists what to fix.
Check your reference listPaste references; check each DOI against Crossref and flag retractions, preprints, and duplicates.
Citation Integrity Quick CheckPaste a manuscript paragraph; flag citation-risk wording.
Qualitative Claim Quick CheckPaste findings; flag ungrounded themes and over-reach.
Codebook Builder & CheckBuild a thematic-analysis codebook and check its structure.
Sample-size explorerMove the design parameters; watch N and its stability respond.
SynthVahtiPool agreement across studies; the joint bivariate model runs in R.
Manuscript readiness checkRate 16 statements; see which of four areas is keeping your paper stuck, and one thing to do next.
Artikkelin lähtötilannetestiThe same check in Finnish: neljä akselia, tulos lasketaan selaimessasi.