There are a lot of dockets out there. Halcyon catalogs hundreds of new ones every week. All are important, but some are more important than others. And right now, Halcyon has been examining one of the hottest dockets around: California’s ongoing proceeding entitled Order Instituting Rulemaking to Oversee the Resource Adequacy Program, Consider Program Reforms and Refinements, and Establish Forward Resource Adequacy Procurement Obligations. You can check it out here (free authentication required to view).

There’s a simpler way to say this title: it is California’s examination of utility reserve margins. The California Public Utility Commission is inquiring into the amount of idle generation capacity which the state needs to keep available to ensure the lights stay on during periods of high demand or forced outages or both. Reserve margins are a tradeoff between cost and reliability, and the nuances of that trade are hotly debated by the biggest players in any given state — utilities, advocates, suppliers, and customers.

It’s also a text-heavy undertaking. The CPUC proposed two options for statewide resource adequacy; it has received more than two dozen counter-proposals. Not only that, but those counter-proposals and supporting comments (almost 250 of them to date) arrive in a chunky fashion. Some months see a dozen comments; some months see that many comments every week…and some months zero comments at all. A human can read these all — but doing so is a slog, and it’s a hurry-up-and-wait slog at that. It also requires constant referencing back (“what did San Diego Gas & Electric say in November 2023 again? Or August 2024?”) and a clear taxonomy (“is this respondent one I care about?”). It’s critically important, and it’s also a lot to ask of human capabilities.

Enter the LLM, via the inbox 

Large language models can be good at generating text. But perhaps more importantly for this purpose, they are also quite good at reading it. That ability to read is the foundation of Halcyon Alerts, which reads each day’s new uploads, identifies relevant parties, and summarizes the contents in context. 

In our case, our data science team, which has set up daily alerts for California Public Utilities Commission, received an email summarizing two California utility company responses to the state’s proposal (one for, and one against). That same docket, the team noticed, received 21 updates in one day. That’s a lot for a person to read, even with summarization, and it’s also an indication that something significant is up for discussion in the proceeding. 

Importantly, this is an entirely passive undertaking on the team’s part — the alerts were set months ago, and their instructions run in the background. When new information enters, the alerts pop up, and they also do so in a live information framework, including a timeline, key players, and of course, summarization.

Going deeper with Halcyon Helpdesk

There is a next step to take with a docket such as California’s resource adequacy proceeding: Halcyon Helpdesk. Where alerts provide universal updates on what is new, Helpdesk is a way to go deeper: specifically designed summarization, a focus on custom arguments, and near real-time tracking of evolving regulatory positions — all transformed into structured outputs.  

For this proceeding, for instance, Halcyon set up a comprehensive analysis that creates a summary of summaries, essentially — tying every argument about resource adequacy into one large structured output. We think of it as being breadth-first: capturing everything in one place, but in a like-for-like way that allows for a holistic understanding of a proceeding. This is a (relatively) time-consuming first step, but it is also the framework on which all other actions depend. 

Breadth-first then allows us to go into depth with a customer. For instance, we can look at specific pro and con arguments, which entities are propounding them, and why, and then continue to update that analysis as more comments are made in the proceeding. This second step, going deeper, is also faster and easily iterated. We find that the marginal time needed to update Helpdesk analysis of a proceeding decreases significantly, while the comprehensiveness of information stays the same.  

Building an analysis with Halcyon Helpdesk combines an initial fairly intensive blueprinting of the information and arguments in a proceeding, followed by an iterative and rapid re-running or updating as often as needed. And as with Halcyon’s alerts, Helpdesk processes are always at the ready: when there is another big drop of comments, utility plans, or a Commission decision, a Helpdesk inquiry can be completed in a day or two. Follow-on analysis can take hours, or in some cases, even minutes. 

Bring us a challenge

Sound interesting to you? Bring us a challenge like California’s ongoing resource adequacy proceeding, and let’s see what we can do with you, and for you. 

Comments or questions? We’d love to hear from you - sayhi@halcyon.eco, or find us on LinkedIn and Twitter