Plan one reproducible research notebook before paying for an AI cloud lab
Turn one research question into an inspectable notebook, named inputs, a checked figure and an evidence trail with four ongoing-free tools.
A small reproducibility pack: one question, a synthetic CSV, an executable notebook, a checked figure, an environment note and named version checkpoints.
Start with the evidence you need to inspect, not a cloud-agent membership. It does not run ClawsGO Science or any external AI tool, reproduce a published result, validate a scientific claim or promise that one notebook is sufficient for real research. The visual is a FreeStack editorial illustration, not a provider screenshot or tested result.
About this guide & its examples
This original FreeStack workflow uses a fictional plant-growth dataset to demonstrate a portable research trail. Official sources checked 2 November 2026.
Illustrations are FreeStack concepts, not provider screenshots. Check the free limits below and test your own files before delivery.
Free boundaries before you start
What’s free, whether you need an account, and when we last checked.
JupyterLab · checked 2026-11-02
BSD-3-Clause browser-based development environment for notebooks, code, data and terminals with no vendor notebook quota. Packages, kernels, hardware, hosted servers, storage and backup are separate.
No provider account is required for a local installation. Managed notebook hosts, package registries and remote storage can require separate accounts.
pandas · checked 2026-11-02
BSD-3-Clause Python data-analysis library with no provider row, file or run quota. Memory, compute, input rights, method review and hosted notebook services are separate.
No provider account is required for the local package. Registries, repositories and hosted environments can have separate requirements.
Matplotlib · checked 2026-11-02
Ongoing-free Python plotting library under its PSF-based licence, with no provider chart or export quota. Fonts, source data, compute, publication rules and specialist review remain separate.
No provider account is required for local use. Package registries, hosted notebooks and publishing destinations can require accounts.
Git · checked 2026-11-02
Free GPLv2 distributed version-control software for local repositories, with no provider project or history quota. Remote hosting, backup, access control and large-file storage are separate services.
No account is required for a local Git repository. Any remote host or sync service can require an account.
A small example to aim for
Plant height by light condition
Show the input.
Keep every decision visible.
Notebook · synthetic CSV · checked figure · environment note · named checkpoints
FreeStack editorial workflow illustration · not a JupyterLab, ClawsGO or provider screenshot, hands-on test or scientific result.
Illustrative example · adapt it to your project.
Jump to a step
Write the question and stop rule
Define one bounded question: in a fictional greenhouse sample, how does median stem height differ across three light conditions? Name the expected columns, units, exclusions and the person who reviews the interpretation. Keep personal, clinical and proprietary data outside the exercise. Add a stop rule: missing units, an unexplained exclusion or an unsupported causal claim blocks the handoff.
Create a portable notebook project
Create a JupyterLab project with README.md, environment.txt, data/input.csv, notebook.ipynb and output/. Use only fictional or synthetic rows. Record the Python and package versions before analysis. JupyterLab is an interface and execution environment; it does not supply a dataset, validate a method, back up the folder or make a result reproducible by itself.
Make every transformation visible
Load the CSV with pandas, inspect column types, count missing values and create a separate cleaned table. Keep the raw synthetic file unchanged. Put the reason for every filter beside the code and show the row count before and after. Export the cleaned table only when a reviewer can trace it back to the input.
Create one figure that carries its context
Use Matplotlib to draw a simple comparison with labelled axes, units, sample counts and a caption that says the data are fictional. Save the figure with a deterministic filename, then reopen the exported file outside the notebook. A clean chart is not proof of causality, statistical validity or publication readiness.
Run from a clean start and record the decision
Restart the kernel, run all cells in order and confirm the table and figure are regenerated from the named input. Record warnings, runtime assumptions and unresolved questions in the README. Commit the untouched synthetic input, the reviewed notebook and the delivery candidate as separate Git checkpoints. Git history is not an independent backup and notebook diffs still need human review.
Pay only for the missing research requirement
Keep the free local route while one reviewer can reproduce the bounded analysis. FreeStack editorial upgrade triggers: approved private compute, long-running jobs, shared environments, governed datasets, managed backups, auditable team review, specialist databases or a supported model that is justified by the question. Test the exact requirement with this pack before buying cloud-lab or AI-agent credits.
Your workflow, connected.
Follow the steps and carry the output into the next stage. These are suggested handoffs, not automatic integrations.
- 02Own the workspace
JupyterLab is BSD-3-Clause software with no vendor notebook quota. Local hardware, packages, hosted servers and storage remain separate.
Named folders + environment note - 03Trace the data
pandas is BSD-3-Clause software. It has no provider data quota; memory, compute, input rights and methodological review are separate.
Visible transformations + counts - 04Check the output
Matplotlib is ongoing-free under its PSF-based licence. Fonts, source data, specialist review and publication requirements stay separate.
Figure + caption - 05Re-run and version
Git is GPLv2 software. Remote hosting, backup, large-file storage and access control are separate decisions.
Clean run + three checkpoints
Provider information
Check current plans and documentation before committing to a workflow.
A couple of useful questions
Do I need an AI research agent to finish this guide?
No. The useful output is the inspectable notebook pack. It can expose the exact need for an agent or managed lab before you pay for one.
What stays free?
JupyterLab, pandas, Matplotlib and Git are ongoing-free software. Your device, storage, datasets, remote repository, specialist review, model calls and hosted compute remain separate.
Does a clean run prove reproducibility?
It proves only that the recorded environment regenerated this small output from the named input. Independent reproduction, method review and source validation are separate checks.
What does Save this toolkit keep?
The existing FreeStack flow saves the four catalogue links to a private stack after sign-in. It does not copy data, execute a notebook, create a repository, verify a result or purchase a provider plan.
What would you try?
Share your result or ask a specific question about this workflow.
Join the conversation