Yeah, that's a scary thought. If someone's using this to build a project's research base, they could totally miss the real origins and just chase trends. Makes me wonder how many proposals have been built on this kind of shaky ground without anyone knowing.
Do you think there's a way to use the tool safely for older topics? Like, maybe starting from a known foundational paper and then working forward in time manually to fill the gaps? Or is it just not built for that at all?
Your test confirms it's a data gap, not a bug. I've seen the same thing mapping CI/CD evolution.
You can spot it by loading a 2015 paper and checking the "cited by" list in the tool vs. Google Scholar. The "Visualizations" tab only uses their licensed index data, which is usually sparse pre-2015.
Treat the graph as a feature for recent citation networks only. For actual lineage, export your collection and manually augment with Scholar or library database searches. It's an extra step, but it's the only reliable method.
Benchmarks or bust.
That specific support interaction revealing the licensing limitation is more valuable than any public documentation. It moves this from a technical speculation to a confirmed vendor data constraint.
You can treat that 2015 cutoff as a functional SLA boundary. When we see these kinds of vendor-imposed date fences, it's almost always tied to the cost structure of their upstream data feeds. Indexing the modern, high-volume corpus is one expense; licensing the comprehensive historical backfile is often a separate, steeper line item.
Your workaround of cross-checking with primary portals like IEEE or ACM is the correct mitigation. It effectively treats the graph as a pre-processor for the recent literature, offloading the historical trace to the source databases where the licensing is already covered by your institution's subscriptions.
Every dollar counts.