Make a digital citation format for Free Law Project and assign it to all case law

[mlissner]

This idea comes up over and over again and I've resisted for years because who wants another digital citation?

My resistance ends now, and I am ready to cast my lot with the advocates of these citations. Here's why I think it's time:

1. We're just bigger now.

   When we were smaller, it looked like hubris. Who is this weird organization that thinks it gets to add citations to things? Well, we're CourtListener, referenced by the library of congress and server to half a million people each week.

2. We have more of the data.

   It's easier now not to miss a case and to have to give it an out of sequence citation when we add it years later.

3. Peter Martin, God of Citations, supports it.

   He sent me an old paper he wrote, which I've only had time to skim so far. The gist though is that, as he puts it:

   rather than continuing merely to exhort courts to implement non-print-based non-proprietary citations, some non-profit entity should do it to their decisions. Rereading the memo I was struck by the thought that Courtlistener might in a position to do it.

4. Others keep bringing it up.

   When CL was new, this was a wild idea and I dismissed it. Now that others are asking for it, I think it's time.

---

I actually think this is pretty easy. Here's what I think we need to do:

1. Make an atomic counter for each year, starting at the number 1. Use it to generate citations for everything we import in the order we import it.

2. Make citations for all existing case law we have using atomic counters.

3. Make our citations work in the citation lookup, eyecite, and elsewhere.

4. Document the concept in a blog post, on CourtListener, on the neutral citation page, and elsewhere.

The one question is which citation format to use:

• CL — people might readily connect this with CourtListener, which is good.
• FREE — bold, gets the point across, and is educational about our mission and the purpose of the citations themselves. Talk about branding. In our URL.
• FLP — gets people back to the org itself and maybe that'll drive support for us.
• FREELAW — long but clearer than FLP and has some of the advocacy angle of FREE. It's also our URL, which is sort of like how Java namespaces thing (though they're weirdos).

I lean towards FREE. It's the kind of on-brand thing only we can do, and it'll stand out like hell. It could help adoption, and, if nothing else, it makes our point very loudly.

My second vote is for FLP. It's more serious, which lawyers like, but 99% of people, even those that use CourtListener, don't know what Free Law Project is, much less FLP.

FREELAW is fun, but it's kind of long and it's a bit preachy. Some might not go for it, but everybody likes freedom.

CL could work, but doesn't impose any advocacy. Missed opportunity.

[mlissner]

Some jerk named Freeman made a reporter in Mississippi in the 1840's, but I am undeterred:

Happily, nobody abbreviates it "Free" or "FREE".

[mlissner]

A friend suggests OPEN for the token. So....what's best? FLP, FREE, OPEN, CL, FREELAW or something else?

[grossir]

They would look like this, right?

• 2025 OPEN 11120
• 2025 FREE 888722
• 2025 FREELAW 98523
• 2025 FLP 74240
• 2025 CL 33122

I don't like how OPEN and FREE look when wrapped around numbers... Feels like an advertisement ?

If going for a conceptual reporter name, I like FREELAW best

CL feels nice too. I think it evokes WL, which is an already known reporter

[mlissner]

CL definitely evokes WL. Totally agree with that. I agree FREE feels a bit like an ad, and maybe that'd make people hesitate. OPEN feels more like something a coalition could do though. Or maybe it'd be Open, e.g.:

2025 Open 1238

That feels like the right spirit.

Maybe we target April 1 for the launch and make it very clear it's not a joke. We could ride the publicity of the holiday though.

[brianwc]

I think we're asking the wrong question: what should be the name of yet-another-proprietary-citation-system? If we name everything uniformly, irrespective of its jurisdiction, we're just duplicating LEXIS and WL. Instead, the beauty of 2025-NMSC-012 is that, in just about the shortest amount of characters possible, I only need to look at it to know what year the opinion is (2025) and what court it is (New Mexico Supreme Court). They're one of the smart courts that has solved this problem. They don't need yet another citation and we shouldn't give them one. Instead, the right thing to do is to make every other court follow their example, and the best way to do that, is to just start using the format that the court could adopt themselves if they wanted to do so. Alabama hasn't adopted neutral citations yet, but that doesn't mean we can't call the first opinion their Supreme Court releases in 2025 something like "2025-ALSC-1". We just start calling it that and eventually they adopt it too. This means we need a unique identifier for every court that hasn't already adopted one. But guess what?! We (basically) already have that. We had to come up with unique identifiers for each of our scrapers in Juriscraper and we used them in the filenames. So, the Alabama Supreme Court scraper is called "ala.py" which gives us the name of its neutral citation: "2025 ALA 1". We could certainly review some of the decisions we made on these and shorten some of them, but, in general, this approach preserves the main beauty of a neutral citation: you can tell just by looking at it what court you're talking about. 2025 CL 1 or 2025 FREE 1 will never do that.

[grossir]

I agree, the reporter abbreviation would be directly useful to the users.

What happens if, after we implement and publish these citations for the courts that do not have them, one of those courts creates its own neutral citation?

If the new official neutral citation has the same abbreviation as we do, which may happen if we use regular state abbreviations, how could we (or other people) tell them apart?

If they only assign citations from a give date forward, we could differentiate ours and theirs using that date; but users may not understand the difference

Maybe we should add an acronym to make a difference?

• "2025 ALA 1" -> "2025 ALA FLP 1"?

[NamedIdentity]

I think missing from the discussion is, which system is optimized for AI?

I think the days of humans using citations to track knowledge-objects are limited. Natural Language retrieval using AI systems is going to become the defacto standard.

Case law is essentially a stored computation result; the legal system is a system of applying rules and rational analysis; a computation. It has an input, and an output. It's subject to errors that get introduced by imperfect humans. It is a flawed corpus, but one we have invested our computational resources in producing so that we can optimize larger systems (society as a whole; rule of law).

AI will shake the legal system to it's foundations by providing a more consistent, less prejudicial application of rational analysis.

I expect the utility of case law to the application of law is going to become challenged by the capacity of AI to apply rational analysis to large corpus's of case data.

I think what people are missing, not comprehending, is AI will be expert in multiple domains. It won't just replace the lawyer, but the doctor, the engineer, the programmer. It's judgment will surpass that of expert witnesses and legal professionals.

We will likely see case law get overturned because of expert AI analysis demonstrating how applying the law in the manner it had been applied in the past, causes a serious harm to society.

So, I believe it's important that we should be asking what will best serve the AI systems that we will rely upon. What gets implemented shouldn't be down to media exposure, or which one we prefer, or other subjective measures. We should use rational analysis to model what the future is likely to be, and build to prepare for it.

[NamedIdentity]

With this tab opened, I asked Gemini 3 Pro Preview in Chrome:

"As an AI agent, what do you think? What would you advise?"

It Answered:

Based on the discussion thread and the specific context regarding AI optimization raised by NamedIdentity, here is my analysis and advice from an AI perspective.

The Core Analysis

The comment by NamedIdentity correctly identifies a shift in how legal data is consumed: shifting from manual human lookups to AI-mediated retrieval (RAG - Retrieval-Augmented Generation). However, the conclusion that citations are purely "human" artifacts that might become obsolete is only partially correct.

For an AI, a citation is not just a lookup tool; it is a unique identifier (UID) for ground truth.

1. Hallucination Mitigation: AI models struggle with ambiguity. If a case is referenced vaguely, a model may hallucinate details. A strict, logical, and machine-readable citation format acts as a "hard anchor," allowing the model to verify that "Case A" is definitively not "Case B," even if the facts are similar.
2. Tokenization & Search: The proposal to use the token FREE or FREELAW is suboptimal for AI.

• Common Tokens: Words like "Free" and "Open" appear frequently in general text. This creates "noise" in vector embeddings and keyword searches, making it harder for an AI to distinguish the citation from the concept of freedom or openness.
• Distinct Tokens: FLP, CL, or specific court codes (like ALSC) are distinct tokens. They are statistically unlikely to appear in natural prose, making them high-value signals for retrieval systems.

3. Jurisdiction is Data: The counter-proposal by brianwc (using court-specific identifiers like 2025-ALSC-1) is technically superior for AI analysis.

• Hierarchical Weighting: An AI analyzing legal arguments needs to know the court to determine authority (binding vs. persuasive). Embedding the court ID (e.g., NMSC vs ALSC) directly into the citation allows the model to instantly weight the authority of the text without needing a secondary metadata lookup.

My Advice

If the goal is to build a system optimized for the future of AI legal analysis, do not prioritize branding (FREE) over utility.

1. Adopt the "Juriscraper" Style (Structured Neutral Citations)
Support brianwc's approach of [Year] [Court_Code] [Sequential_Number].

• Why: This follows the "Semantic Web" principle. The citation itself carries metadata (Year + Jurisdiction). An AI reading 2025 ALSC 14 immediately knows the temporal and jurisdictional context. 2025 FREE 14 requires an external database to know if the case is from Alabama or Zimbabwe.
• Recommendation: Use the existing Juriscraper abbreviations. They are already standardized in code, making them easily ingestible by machines.

2. Avoid Generic Keywords
Advise strictly against FREE, OPEN, or FREELAW.

• Why: These are "stop words" or high-frequency tokens in natural language. Using them as unique identifiers introduces unnecessary collision risks and complicates Named Entity Recognition (NER) tasks for future models. FLP is acceptable; CL is acceptable; FREE is technically poor.

3. The "AI-Ready" Format
If the Free Law Project wants to "assign" citations where courts haven't, they should use a format that mimics the best-in-class official neutral citations, perhaps with a namespace suffix to indicate it is an assigned (unofficial) ID until adopted.

• Format: [Year] [Court] [FLP-ID]
• Example: 2025 ALSC 402 (if relying on brianwc's logic) or 2025 TNApp FLP 12.