SciFlow
Research is months of intertwined, nonlinear work — but we only ever study it in simple, linear settings: controlled tasks, retrospective interviews. SciFlow captures the full complexity: tracking research trajectories in real time. The goal: capture both the full trajectory of a research project and the intention behind each action — what researchers did, why they did it, and how it shaped what came next.
A research project, replayed
An example research trajectory — reconstructed from recordings, code sessions, messages, and document edits — replayed as a single timeline.
Interactive demo · synthetic data Two views of the same activity: left replays it as parallel streams you can scrub through; right auto-scrolls it as a single chronological feed.
The tool — SciFlowCollection
SciFlowCollection is a desktop app that acts as both a recording tool and a central hub for your research. Connect your existing workspace and see your edits, messages, commits, and revisions in one place. It connects only to the items you explicitly approve and captures your research activity in the background while you work as usual.
Currently supported integrations
Documents and notes in the folders you approve — proposals, meeting notes, planning docs.
Approved folders onlyLab and project discussion — questions, feedback, and decisions as they happen.
Approved channels onlyCommits, branches, issues, and pull requests — how the code evolves over the project.
Approved repos onlyThe paper as it is written — revision history of your approved LaTeX projects.
Approved projects onlyAI-assisted work sessions — what you asked for, what was produced, what you kept.
Session logs, opt-inPeriodic recordings of the screens where you do research work.
Approved screens onlyInteractive demo · synthetic data Explore the app below — click through the sidebar. Everything shown is demo data, not from a real participant.
- analysis-plan editedDemo Participant · Jul 16 at 07:49 AM
- Recorded 3 min (Entire Screen)Jul 16 at 06:24 AM
- message in #demo-projectDemo Advisor · Jul 16 at 02:00 AM
- push to demo-research-pipelinedemo-participant · Jul 15 at 04:12 PM
- edited main.tex on OverleafDemo Participant · Jul 15 at 11:24 AM
- Notes — Weekly Check-In editedDemo Participant · Jul 14 at 09:53 PM
- Recorded 47 min (2 windows)Jul 14 at 09:15 AM
- opened PR #28: validation checksdemo-participant · Jul 13 at 03:40 PM
- Drive● connected ↗
- GitHub● connected ↗
- Slack● connected ↗
- Overleaf● connected ↗
SW: analysis sprint — Jul 17, 7:45 AM (from Google Calendar)
You choose exactly which folders, channels, repositories, projects, and screens are shared. Anything you don't approve is never collected — and you can revoke access to any stream at any time.
Get involved — recruitment
Help us understand how research really unfolds. You keep doing your normal research while SciFlowCollection records your approved streams; we meet for three reflection interviews over the course of your participation, roughly every 4 months, to review your recorded activity together. You receive $150 at each reflection interview, up to $450 total — and you can pause or stop at any time without penalty.
If you're actively working on a research project: AI research assistants are an exciting direction, and a first step is understanding how real research workflows evolve over months.
Set up SciFlowCollection and choose which folders, channels, repos, and screens to approve.
Do your normally planned research; the tool records your approved streams in the background.
Review your recorded activity with us. $150 at each of the three interviews.
- Fill out the interest form (below).
- Book an onboarding session (below).
- Confirm participation with your PI.
Get started
Frequently asked questions
No existing dataset captures a research project at this granularity — the full arc from first idea through publication, across every tool where the work happens, paired with the researcher's own account of what they were doing and why. SciFlow fills that gap, producing the empirical foundation for understanding how research actually unfolds and for building AI assistants grounded in real workflows.
Only streams you approve: documents in approved Google Drive folders, research-related Gmail, approved Slack channels, approved GitHub repositories, approved Overleaf projects, Claude Code session logs, and recordings of approved screens. We also record the three reflection interviews where we review your activity together.
Your normal research is the data — there are no extra day-to-day tasks. Beyond a short onboarding session to set up the SciFlowCollection tool, participation is three reflection interviews over the course of the study, roughly every 4 months, where we review your recorded activity together.
You receive $150 at each reflection interview, for a maximum of $450 over the course of your participation.
Yes. You can pause or stop your participation at any time without penalty — just let us know. You can also revoke access to any individual stream (a folder, channel, repository, or screen) at any time while continuing to participate with the rest.
Your data is co-owned by you, PI Dongyeop Kang, and your Contributor PI. No data is publicly released without consent from you. You may participate without consenting to public release of your data.
We do not publicly release personally identifiable information. We ask that you avoid recording it whenever possible — the tool only touches the folders, channels, repos, and screens you approve. If any appears in collected materials, it is removed or redacted before any broader sharing or release.
The current purpose is to study how research workflows evolve and to evaluate AI research assistants — not to train models. If a future use case involves model training, PI Kang will contact you and obtain explicit agreement before that use. Until then, the data is not used for model training.
Team
Reach out to the study coordinator about participating, or to the PI for general inquiries and cross-project collaboration.
