You Probably Don't Need the BSL Fabric

Last week, we received our “Tier 4 Research” license from CostQuest to access the entire Broadband Serviceable Location (BSL) Fabric, a set of “very fancy points” that underlies the National Broadband Map. Credit where credit is due, CostQuest was super helpful and made the process quick and painless (thanks, Mark and Lisa).

I’d still argue that the entire licensing model around the BSL Fabric is far more restrictive than it should be for a dataset so fundamental to public programs as large as BEAD, but that ship has sailed. The research license has limitations that preclude publishing (for example) reports about the BSL Fabric at a spatial resolution beyond an H3 Level 7 grid (around 5 sq km) or a county level. And of course, you have to go through the process of obtaining a license in the first place — not difficult, but also not trivial.

One of the questions I had before we got this data was — do we really need it? The publicly-available National Broadband Map already provides the Census block and H3 Level 8 hex grid associated for each BSL Location that appears in the map. I think of this as the “Public Fabric” — there’s no license agreement required to download public data from the map, anyone can compute this mapping of BSL Locations to H3 grid and Census block, and as far as I can tell there aren’t restrictions on its use. Indeed, it provides better spatial resolution than the research license allows for publication.

The Public Fabric should include every location from the BSL Fabric, with two key exceptions:

1. Locations served by no service provider whatsoever — these just won’t appear in the National Broadband Map at all.
2. Locations that are classified as non-BSLs. These locations are assumed to not subscribe to “mass market” broadband service, and include many institutions such as schools, libraries, or large enterprises. Obviously, some set of these locations are inaccurately classified, and many of these probably do receive mass market service, but since they’re “non-BSLs”, they’re not going to show up in the National Broadband Map.

Of course, locations not represented in the BSL Fabric won’t show up in either, and the National Broadband Map doesn’t include most metadata about a location (e.g., the predicted unit count or address).

I should note that using the National Broadband Map to approximate the Fabric isn’t a particularly novel idea — anecdotally I know of a few others who have been doing this for a while — but I haven’t seen any empirical comparisons to the actual BSL Fabric. So, the question is, how closely does the Public Fabric approximate the real BSL Fabric, at the Census block or a given H3 grid level?

I can’t answer that question publicly without violating the terms of our Fabric license. I can, however, answer the related question of how the Public Fabric compares to the actual Fabric at the county level. We can do this by taking the most recent National Broadband Map release, computing the Public Fabric, and comparing it to the corresponding release of the Fabric (i.e., the July 30, 2022 edition of each; since the NBM corresponding to the most recent map isn’t yet released we can’t compare to the new one). In the ideal case, the Public Fabric would match the BSL Fabric exactly for any given spatial boundary.

If we only care about BSLs, this approach turns out to work really well. There’s a nearly perfect equivalence per county: only seven counties nationwide have a different number of BSLs in the Fabric than appear in the National Broadband Map. All of these are either in Pacific Ocean territories on rural municipalities of Alaska.

Now, what happens if we add in the non-BSLs?

Percent difference on a per-county basis between the total number of locations in the BSL Fabric compared to the Public Fabric (i.e., the locations described in the National Broadband Map).

Not as good… but it’s still not terrible. Beyond a handful of outliers, the median difference in locations is 0.5%, and 98.5% of counties have a difference of <1.5%.

So, do you need the real Fabric? You’ll have to take my word for it beyond this very high level comparison, but if you only care about BSLs and you don’t need higher spatial resolution than what’s available in the public National Broadband Map data (i.e., Census block or H3 level 8 grid) — consider just using the Public Fabric to start with.

Data sources. The snapshot of the National Broadband Map used for the Public Fabric described in this post is from April 12, 2023, reflecting the June 30, 2022 filing window. The one-and-only BSL Fabric is from CostQuest.