Sunday Misc: policy windows and paradigm shifts

Mar 09, 2025

"Misc” newsletters are my attempt to capture fragments/kernels of thoughts — and also try to continue writing in the midst of a general overload of life. (Though my 15-month-old is the most wonderful thing to ever happen to me.)

Policy windows and paradigm shifts

I am the odd fellow who has one foot in two fairly different worlds: public policy/government services on one hand, and technology/software on the other.

A virtue of that is I can sometimes see the strength of one world’s mental models when the other would be dismissive. And I can also see the occasional parallels.

One parallel that came to me recently was between two concepts:

In government, the notion of policy windows: that, due to the many interconnected factors that need to come together to get something big done in policy —say, broad public demand, a new Congress, a Presidential agenda including X — often good ideas just have to wait for their time.

The corollary is that the common shape of a thing in policy is: 20 years of policy and research papers being written, 6 months of frantic horsetrading, and a new law on an issue that won’t come up again for another 20 years.

In technology, there is the idea of a paradigm shift (a superset that includes the phenomenon of platform shifts.) The rough idea here is that in technology (especially software) there are these moments in time when things shift rather dramatically. Examples to me include the rise of cloud computing, or the iPhone, or social media.

Now what’s funny to me is that my experience is people in one of these worlds are generally skeptical of this concept from the other one.

Tech to policy: “Wait, so your plan is to write papers on the same topic for 20 years and hope something changes?”

Policy to tech: “Wait, so just because of this minor shift in the cost of some minor technical operation, you think the world is upended?”

But the funny thing is I see a lot of symmetry in these ideas! And I think the skeptical reactions come from neither side seeing the texture around each idea.

A half-baked idea from me: both policy windows and paradigm shifts are about the fact big changes in complex systems rely heavily (mostly?) on second- and third-order consequences.

For policy windows, the idea is that you must surf the second-order effects of things you don’t control (elections, particular leaders) when the time comes. And you must be ready and looking for that wave.

For paradigm shifts, the idea is that small changes in the shape of the technological frontier triggers large and cascading second- and third-order effects. Cloud — my quick/least-bad description being that we are now able to chain together a bunch of commodity computers and rent out the very reliable composite by the hour for cheap — made it so that software could be iterated on with much more direct user feedback much more rapidly. That was the thing that mattered. So in technology, being early to see how a small thing could play out into a big thing becomes a valuable skill.

You can sort of see how the concepts might sound less compelling in one world or the other. I think they’re closer in shape than they sound.

Evaluating AI models for SNAP

At work, I’ve been building “evaluations” (tests) of AI language models’ capabilities related to the SNAP program. I’ve written two public blogs so far describing some of that work (with an eye towards enabling folks who may not be programmers but are domain experts in some of these impactful areas to be able to do similar):

There’s more to come here, but part of doing this in public really is to reduce barriers. Feedback welcomed!

On the GetCalFresh (partial) sunset

GetCalFresh.org, the simplified SNAP application I and others built and scaled across California for a number of years (I left in 2019), is sunsetting in June of 2025.

More specifically, it’s something a bit more nuanced: some of the GCF services (outreach and client engagement/assistance like text/email/live chat support) will continue, but the simplified online application itself will sunset. While I’m glad some will live on, I’m understandably a bit nervous about this!

There’s some interesting pieces to put together here in looking at what might happen next:

SNAP participation rates for all states (what percent of eligible families are enrolled) was just released last month, and between FY 2020 and FY 2022 California’s participation rate rose from 67% to 81%. From a quick scan, that is a larger increase than most states, even though nationally there was improvement in participation. I do think it is a valid view that a streamlined “first step” played a significant role in this, though you can’t measure that directly. (But see point 3!)
After GCF sunsets, the sole way to apply will be BenefitsCal, the new county-run benefits portal built over the past 5 years. BenefitsCal is a big improvement on the prior portals that existed. A lot of credit is due to the CalSAWS (county consortium) team that built it. It is hard to do such things in such a… baroque governance environment with many competing priorities and objectives and big constraints. It also still has space for improvement. Much to the team’s credit (as well as credit to advocates who pushed for this) they report data on conversion success. They even report on screen-by-screen dropoff! This is a big deal and puts it in the top 1% of government IT projects in my mind. That said, only ~70% of people who start an application submit it successfully. So there are concrete outcomes to work towards with targeted user experience improvements here. (For example, the expenses screen has significant dropoff. Given a required interview where expenses will be covered, and given that expenses drive benefit amount more than eligibility [broadly speaking], this would be an area where more direct encouragement to submit then and there rather than quit would likely increase success.)
When GCF goes away in June, we will be able to see the measurable impact of a simplified online SNAP application in the total number of online applications, a data point reported monthly. (If you’re an econ PhD looking for a natural experiment on the impact of reduced administrative burden, you might want to look at this!) From there, I think the many decisionmakers engaged on public benefits in California will be in a good place — and honestly, one of the more clearly measured context — to decide on what’s next.

On a personal note: choosing to be more decisive is more possible than I thought

As mentioned, we have a now 15-month-old. I have traditionally been a massive overthinker of problems. (Did this newsletter give that away?)

At a certain point in the past year and change, though, I found it was all simply too much, objectively, to think through rigorously.

And it turns out… making the choice to be decisive — to make decisions immediately and not delay them at all — is much more possible than I’d imagined.

One of the more useful heuristics I learned: if you don’t know what your decision should be, ask yourself “what additional information do I actually need and that I can actually get cheaply to make this decision?”

A surprising amount of the time I’ve found the answer is… I just had a fuzzy sensation of needing more info. So just decide.

So far, I endorse this change. We’ll see how it continues going.