Constellate: An Experiment and Retrospective

JSTOR Labs -

When we set out in 2019 to build Constellate, our goal was to create a sustainable platform that would provide text analysis access to content in JSTOR and Portico. By the time we sunset in 2025, that had transformed into making the power of computational analysis accessible to any curious mind, whether they were a first‑year undergrad or a seasoned historian. While we succeeded in building a vibrant learning community and advancing computational literacy, we were unable to achieve the scale and sustainability needed for a viable long-term service. Six years later, as we close this chapter, we can reflect on both what we accomplished and what we learned about the challenges of this space.

A community that scaled beyond the platform

  • 11,954 people created Constellate accounts—enough to fill a small town. Those users launched 23,152 JupyterLab sessions, poked at data, broke things, fixed them, and learned by doing.
  • Nearly 4,000 people tuned into our Constellate Slack community at one point or another. 
  • The fall 2024 Skill Build program drew 400+ registrants for a ten week session, culminating in ten group projects that proved synchronous learning can, in fact, scale.
  • Our JupyterNotebooks were forked 216 times (meaning people took them to adapt and adopt).

Each of these metrics tells the same story: the demand for hands‑on, code‑curious learning is real, and welcoming design beats steep prerequisites every time.

What we learned about sustainability

Building a sustainable service in this space proved more challenging than anticipated. Despite meaningful engagement from our community, we wrestled with: 

  • Converting individual enthusiasm into institutional adoption at the scale needed 
  • Competing with free alternatives, existing institutional tools, and other products providing more content-focused text analysis services 
  • Navigating long sales cycles 
  • Supporting delivery of content and of learning was complicated
  • Balancing the diverse needs of beginners and advanced users within a single platform

These challenges don't diminish what we built together, but they're important lessons for anyone working in this space.

What lives on

The sunsetting of a service isn’t the end of its ideas. In our case, the assets—and the people—are already finding new homes:

  • Educational DNA in Ithaka S+R. Nathan Kelber and Zhuo Chen are leading new AI cohorts in ITHAKA S+R, carrying forward Constellate’s workshop playbook.
  • Permanent open content. The notebooks in our GitHub repo remain freely available, and the Constellate YouTube channel continues to rack up views.
  • JSTOR’s next leap. JSTOR has committed to keeping its corpus available for text analysis and is even contemplating native visualizations—an idea road‑tested in Constellate’s builder.
  • A community infrastructure recycled. Our Slack community is being re‑purposed to support S+R’s research cohorts, so the conversation never really stops.
  • Our people stay on. All Constellate engineers, educators, and UX staff remain employed at ITHAKA.
  • Skills redeployed. Our staff briefly supported Seeklight (JSTOR’s new AI metadata creation tool that is an element of the new JSTOR Digital Stewardship Services) and the JSTOR AI Research Tool, and now channels its expertise into JSTOR Access in Prison. One engineer has joined the Platform Orchestration & Workflow (POW) team, bringing Constellate’s DevOps lessons to the wider platform.

Lessons we learned

  1. The pipeline is everything. A 12–18‑month SaaS sales cycle in B2B is the rule, not the exception. If you don’t feed the funnel one year, you’ll feel it for years to come.
  2. Find your advocate. Early‑career faculty turned out to be our most ardent champions.
  3. Content is nice; skills are priceless. The true differentiator for Constellate wasn’t exclusive datasets but the ability to help people wield whatever data they had.
  4. Experiment relentlessly, but measure ruthlessly. Rapid prototypes were great, yet the team’s best work happened when we paired bold ideas with clear metrics.

An experiment with mixed outcomes 

Constellate was both a product and an experiment—and while the product couldn't achieve sustainability at this time, the experiment taught us valuable lessons about building a new SaaS service for and serving the computational literacy needs of academic communities. Thousands of learners now know their way around Python, R, and distant reading because of this project, and that impact will outlast the platform itself.  

We're grateful to everyone who believed in this vision, invested their time in learning with us, and helped build a community around accessible computational analysis. Your engagement made this work meaningful.

—Amy Kirchhoff & the Constellate Team