With Connectors, Claude Fable Goes Toe-to-Toe with CoCounsel
On June 9, 2026, Anthropic released a 10-trillion parameter large language model named Fable. It comes from the "Mythos" family, a technology said to be so advanced that Anthropic has so far refused to release its more powerful members for fear they could be used to build bioweapons or mount calamitous cyberattacks. But how does Fable perform on the more benign task of legal analysis? Based on an early experiment, extremely well. The twist is that Fable may become unaffordable for people — law professors and law students among them — who have no client to charge; at best, they will reserve it for special occasions. That's because after June 22, 2026 you won't be able to use Fable as part of a conventional monthly Anthropic account. If you want to enjoy the wonders of 10 trillion parameters, you'll pay a la carte for the tokens.
This post shows you what Fable produced and how it stacks up against three rivals: Westlaw's CoCounsel standing alone, CoCounsel paired with Claude Opus, and ChatGPT 5.5 paired with CoCounsel. The grades, in the order the experiments unfold below: Claude Fable with CourtListener earns an A, CoCounsel by itself a B (after a small rescue operation), Claude Opus with CoCounsel a B+, and ChatGPT 5.5 with CoCounsel a strong A-. Now the economics. Had we paid by the token, the Fable effort would have run about $43. A comparable effort by a human associate would likely run well upwards of $12,000. Fable, in other words, produces work on par with a strong biglaw associate at perhaps 1% of the cost.
Fable, as it turns out, isn't the only model earning its keep. The experiments below show that a strong research tool such as CoCounsel can lift a no-longer-frontier model to partner-review quality on complex legal issues, at a flat monthly price with no marginal charge for the curious student or the professor without a client. I confess that is not the post I sat down to write based on some spectacular early work with Fable. I was prepared to extol Fable as flatly better than the alternatives across domains. The experiments said otherwise: Fable is in fact terrific, but the best choice depends on the pairing of writing model and research tool — and on whether someone else is paying for the tokens.
The legal issue
To evaluate my little AI experiment, you need to understand the legal issues involved. Hence an excursion from the world of AI with which many of my readers are familiar into the world of federal court jurisdiction, with which some may be rusty. The task at hand involves a case the United States Supreme Court will hear next term: Suncor Energy (U.S.A.), Inc. v. County Commissioners of Boulder County. It's a climate change case that has been bouncing around the American judicial system since 2018. Boulder County decided to sue the American subsidiary of an integrated Canadian energy company for increasing greenhouse gases and thus, it alleged, causing heat waves, wildfires, droughts, floods and other climate harms. Suncor USA, however, has no operations within Boulder County. To make a very long procedural story short, the case last ended up in the Colorado Supreme Court, which exercised its own original jurisdiction to hold 5-2 that neither the Constitution nor any federal statute preempted Boulder's state law claims. It did not, however, rule on the merits of Boulder's underlying tort theories. It just allowed the case to proceed in a trial level state court.
And it's from that Colorado Supreme Court decision that Suncor has sought United States Supreme Court review. Its original cert petition appears to have stated one issue: does federal law preempt Boulder's state law claims. That is clearly an extremely important matter. But when SCOTUS reviewed and granted the cert petition, it added another issue: "The parties are directed to brief and argue the following question: whether this court has statutory and Article III jurisdiction to hear this case." (!)
To the untrained eye, this addition is mysterious. Federal preemption is involved. And since the enactment of section 25 of the Judiciary Act of 1789 and decisions such as Fairfax's Devisee v. Hunter's Lessee (1812) and its more famous sequel Martin v. Hunter's Lessee (1816), it's clear that, in general, the Supreme Court can hear cases originating in state court in which a federal preemption defense is asserted even when, under the well-pleaded complaint rule, such cases could not be brought originally in federal court or removed thereto. But ... what if the plaintiff hasn't suffered the concrete and particularized injury generally necessary to bring cases in federal court? Maybe a diffuse interest in climate change is good enough to get you into a Colorado court, but in general it's not enough to get you into a federal court (except, perhaps, if you are Massachusetts and your state boundaries are being eaten up by rising sea levels). Federal courts, it is said, only decide current and non-speculative disputes. And Congress didn't give the Supreme Court roving jurisdiction to hear just any case in which it found a preemption dispute to exist. No. Congress in what is now 28 U.S.C. 1257 gave the Supreme Court jurisdiction to hear cases involving "Final judgments or decrees rendered by the highest Court of a State in which a decision could be had." Is a dispute that lets a case go forward to trial really "final"?
So that is in fact, I suspect, why the Supreme Court added the question. I think it wants an opportunity to determine whether state subdivisions can bring these claims at all and whether fossil fuel companies are going to have to endure multiple bet-the-company lawsuits brought by state entities in passionately pro-environmental venues and bond the gazillion dollar judgments that will perhaps ensue, before having their preemption argument vindicated in a federal forum more sympathetic to the importance of federal law and (O Canada) foreign relations. Procedure matters!
Claude Fable
And now we're ready to see how I asked AI for help. Here's what I asked Claude Fable. Although I used my own expertise in constitutional law to develop the prompt you will notice, as it turns out, that even that proved insufficient to spot all the issues that Claude Fable perceptively uncovered.
Use connectors in this order to help you analyze the legal issues that will be coming before the Supreme Court: (1) Lexis Protege; (2) midpage; (3) CourtListener. if one is not available, go to the next connector in order. The standing issue highlighted by the court is a bit opaque but I believe it is whether the county has a concrete and particularized injury here and whether the special solicitude doctrine of Massachusetts v. EPA is implicated here. There is a tricky issue. Those rules don't apply in state court necessarily, so it doesn[t matter there. But if the federal court enforces its own standing rules when the party asserting a federal interest loses, the Supreme Court loses the ability to review potentially bad preemption decisions. I believe this is the ASARCO exception. I am looking for about a 2000-word legal memo on the two issues involved. Tone should be professional, neutral and analytic with citations to case law and statutes.
Note that I asked it to use Lexis Protégé because I had seen Fable attempt to do so on prior runs. I was surprised by that effort, because I did not think Claude could connect to Protégé yet. Still, I thought the attempt was worth a try in my prompt. As it turned out, my doubt was justified: at least with a law school Lexis account, Claude does not yet connect to Lexis Protégé or to Westlaw's CoCounsel. I hope both of these research giants will liberalize their plans and let frontier models from Anthropic, OpenAI, and others connect. In my opinion, whichever legal research service provides the best connectivity to frontier models is the service likely to win in the long run.
Round 1A
I am not going to go through Fable's original production. Suffice it to say it was very good. I then began my parallel request to ChatGPT. I noticed, however, that ChatGPT's response appeared to be referencing documents pulled from the SCOTUS docket, something I had not given to Claude. Kudos to ChatGPT for initiative. We will come back to its performance in a moment. But I decided my initial prompt to Claude Fable had neglected its capabilities to connect with and pull down Supreme Court docket materials. So, after confirming with Fable that it had not yet consulted the filings before the Supreme Court and after it graciously volunteered to do so, I revised the Fable prompt.
Yes. Dig into the docket of the US Supreme Court if possible via CourtListener or Midpage. I believe CourtListener should have the materials. I don't know about Midpage.
Round 1B
And here is the 3300-word memo Fable produced in about 10 minutes as a Microsoft Word .docx file. (I didn't time it so that is just an approximation)
Let's walk though it. Although the memo is too short to have warranted a table of contents, for purposes of this blog entry, I fabricated one so you can see how well organized the document is. Honestly, I think it would be hard for even a top-flight human attorney to do any better.
Here's the first part. Cosmetically it looks just fine. And its introduction is clear and well written.
Fable then goes on to state the jurisdictional issues clearly.
We now see the first arguable problem, however. I didn't say Fable was perfect. The citations are correct in both form and substance – so far as they go.
What's missing from the response above is the knowledge that subsequent Supreme Court cases have cast doubt on the theory of Massachusetts v. EPA and a sense that theoretical problems with drainage infrastructure might be materially different than the alleged shrinkage in the size of the state of Massachusetts threatened by rising sea levels. So, I used human-in-the-loop to give Fable a second chance (feeling that perhaps I should have given it a better prompt but also grateful that human expertise is not entirely irrelevant as of June 2026). Here's a slightly cleaned up version of my prompt.
It is, to be sure, possible that Fable recognized that Massachusetts v. EPA was beside the point in this circumstance—because the Article III case or controversy actually before the Supreme Court rested on the injury to Suncor from having to endure state climate-change litigation, not on any injury to the State of Massachusetts. But if that was Fable's reasoning, it should have made the point clearly rather than dwelling on Massachusetts v. EPA in the first place.
You wrote that Massachusetts v. EPA supports standing for Boulder. But you neglected that there are factual distinctions between reduction in the physical size of a state and what is alleged here. You also missed that the special solicitude theory on which Massachusetts v. EPA relied has had doubt cast on it. The Supreme Court tends not to cite the case much any more. Do research using Consensus connectors and Midpage and CourtListener to examine this criticism. Give me a proposal on how you would revise the memo. Then create a revised memo. Don't overwrite the original. Propagate any changes throughout the document.
Here is version 2.0 of the memo from Fable:
It first revised the introduction as follows:
And then it rewrote the section on state standing. I provide two screen captures below. The first deals with whether there is really injury in fact.
The second screen capture shows Fable's discussion of the "Erosion" (did it intend that pun, given the Massachusetts coastline issues?) of Special Solicitude. You can judge for yourself. I think the writing quality is superb.
The Fable analysis then moves into the issue of how Suncor can gain federal court review over the state court's determination to reject federal preemption issues when (a) the plaintiff brought the case in state court and (b) the defendant can not remove it to federal court. And here it gracefully turns to the ASARCO issue. The writing is clear and parsimonious.
The standing question just sketched may never matter, because of where this case has been litigated. .“[T]he constraints of Article III do not apply to state courts, and accordingly the state courts are not bound by the limitations of a case or controversy or other federal rules of justiciability even when they address issues of federal law.” ASARCO, 490 U.S. at 617. .... The Supreme Court, by contrast, sits as a federal court even when reviewing state judgments. ... The apparent tension—state courts may hear what federal courts may not, yet the Court reviews state judgments only within Article III—is the seam the Court’s added question probes. ... ASARCO resolved the tension by asking whose standing matters. Standing must be shown by “any person invoking the power of a federal court,” and “standing in federal court is a question of federal law, not state law.” Hollingsworth, 570 U.S. at 704, 715.
And the Fable memo then explains the rationale.
The structural rationale is the one the Court’s grant implicates here. If plaintiff-side Article III defects barred review, a defendant who loses a federal defense—preemption above all—in state court could never obtain Supreme Court review of that federal ruling. State supreme courts would render effectively final judgments on the preemptive scope of federal law whenever the plaintiff could not have sued in federal court, and plaintiffs could insulate favorable preemption rulings simply by choosing state court.
But Fable spots a complication. The ASARCO rule governing Article III standing may not extend to interlocutory appeals, since the Supreme Court could still hear the case at a later stage. And Congress has not necessarily given the Court the full sweep of jurisdiction that Article III would theoretically permit.
Two cases address this wrinkle. In Nike, Inc. v. Kasky, 539 U.S. 654 (2003) (per curiam), the Court dismissed certiorari as improvidently granted, and Justice Stevens, concurring, suggested that ASARCO cannot supply a basis for Supreme Court jurisdiction over an interlocutory appeal. The Court might adopt that view. Alternatively, it might rely on the exception to the final-judgment requirement recognized in Cox Broadcasting Corp. v. Cohn, 420 U.S. 469, 477–83 (1975), which permits immediate review where reversal on the federal issue would foreclose further litigation and delaying review "might seriously erode federal policy."
This is all excellent work. But Fable then takes it up another notch with a very sophisticated explanation of how the multiple jurisdictional issues interact. Perhaps I am being unduly pessimistic, but I suspect few but the very sharpest attorneys would either think of this point or explain it as clearly.
There's another section of the brief in which Fable examines the position of the parties. And it notes that the prompt with which I initiated the discussion neglected another key axis of jurisdictional debate: whether the Colorado Supreme Court decision, which might appear interlocutory to the inexpert could in fact be recharacterized as final and thus comfortably within the scope of Supreme Court jurisdiction under 28 U.S.C. 1257. Here the key case is Atlantic Richfield Co. v. Christian, 590 U.S. 1 (2020). Fable clarifies the positions of the parties:
Petitioner
[P]etitioners’ lead argument is not Cox but Atlantic Richfield: because a C.A.R. 21 proceeding is, under Colorado law, a “self-contained original proceeding” in the state’s court of last resort, the decision discharging the order to show cause “terminated” that proceeding and is final under § 1257(a) regardless of further trial-court proceedings—just as the Montana supervisory-writ decision in Atlantic Richfield was final because “a supervisory writ proceeding is a self-contained case, not an interlocutory appeal,” with finality turning on “the nature of the [state-court] proceeding, not the issues the state court reviewed.
Respondent
Boulder answers that Colorado is different from Montana: the supervisory power exercised here flows from article VI, § 2 of the Colorado Constitution—captioned “Appellate jurisdiction”—not the § 3 grant of original jurisdiction, and the Colorado Supreme Court itself has called Rule 21 proceedings “interlocutory appeals” in other contexts.
Fable concludes with a summary about the jurisdictional question presented added by the Supreme Court. Its statement is both brief and sophisticated.
The added question is best read as Nike redux—now refracted through Atlantic Richfield. The live disputes are (1) whether a C.A.R. 21 proceeding is a self-contained original proceeding whose termination is “final” under § 1257(a) (or, failing that, final under Cox Broadcasting’s fourth category), and (2) whether ASARCO’s judgment-injury rationale extends to a ruling that merely allows the case to proceed. Boulder’s own standing—and the inapplicability of Massachusetts v. EPA’s special solicitude to political subdivisions—is analytically antecedent only on a view of certiorari review that the Court’s precedents reject. If the Court finds jurisdiction, expect it to reaffirm and perhaps extend ASARCO; if it dismisses, expect a separate writing on why state-court adjudication of federal preemption defenses should not evade review indefinitely. And if any opinion reaches the County’s standing, expect it to sound in the post-2023 register of United States v. Texas and Murthy rather than in Massachusetts v. EPA, whose special solicitude the Court has left unused for nearly two decades. Either disposition will shape every pending climate case filed in state court.
I hope to have persuaded you that Fable is a very strong legal model. Yes, I had to intervene and remind it that Massachusetts v. EPA and its special solicitude doctrine was skating on thin ice, but it recovered well from that correction and produced an outstanding memo. But, as I mentioned, if I were being charged by the token as will be the case in a few weeks, the product would have cost over $40. Obviously, in a world in which associate time for sophisticated work on behalf of large corporate clients bills out at $600+, however, and in which a memo like this would likely consume a minimum of 20 hours, it's hard to see how using Fable cuts the initial cost of legal work by anything less than 99%.
A key economic issue in using AI for legal work is how much of those savings are going to be eaten up by a now-separate process of validation. Even grounded AI hallucinates on occasion, and humans seem particularly intolerant of mistakes made by different intelligences. Already, though, we see AI tools emerging that greatly reduce the cost of basic cite checking, and I suspect those tools will keep evolving to the point where for a very manageable cost AI-generated content will contain fewer pure errors than does human-generated content.
CoCounsel
Perhaps I have successfully suggested that Claude Fable is an extraordinarily useful model for legal reasoning. But I have not yet proven that other large language models can't do equivalent work. Fable isn't special if others can do the same, particularly if they can do so at a price that invites casual use by impecunious law students or medium-pecunious law professors. Moreover, for the moment, Claude Fable does not use Zero Data Retention: Anthropic is currently storing Fable dialogs for 30 days. It doesn't train on them and the purpose appears to be to ensure that Fable is not, in fact, plotting new recipes for ricin or new ways of infiltrating nuclear power plant safety. Still, so long as this data retention policy remains in force, Fable may be inappropriate for general legal use.
My first effort was with Westlaw's AI product, CoCounsel.
Round 1
I found the Colorado Supreme Court decision and then tried out this initial prompt:
The Supreme Court of the United States has accepted certiorari in this case and has asked the parties to brief whether there is standing under Article III and statutory standing. What is the standing issue. Is ASARCO implicated here? Massachusetts v. EPA.
CoCounsel fell to pieces. Given that Colorado had understandably not addressed the federal standing issue, CoCounsel had nothing to go on and lacked the "imagination" to consider the matter.
The context provided does not explicitly address the standing issue under Article III or statutory standing in this case. Therefore, the specific standing issue that the Supreme Court of the United States is interested in is not detailed in the provided content.
Regarding the cases you mentioned, ASARCO Inc. v. Kadish and Massachusetts v. EPA, the context does not reference these cases directly or discuss their implications in relation to the standing issue in this case. Therefore, it is not clear from the provided content whether these cases are implicated or how they might relate to the standing issue.
Round 2
I thus did a little extra work. I basically kickstarted CoCounsel by telling it what the issues were in the abstract and then asking it to provide an answer:
Imagine a case in which a state supreme court such as Colorado has used original jurisdiction to determine whether a federal preemption defense is available in a case brought under state law. The state supreme court finds that there is no federal preemption, which allows the state case to proceed in a lower state court. Does the Supreme Court have jurisdiction under Article III and 28 USC 1257 to hear the case and address federal preemption in a petition brought by the defendant who lost on federal preemption in state court. What might jurisdiction depend on?
This time, CoCounsel did not give up. After about six minutes, it provided 20-pages worth of material, including secondary authority that Claude Fable had not been able to find. Having access to what may be the world’s largest compendium of legal materials is undoubtedly valuable for comprehensive research. Here's the report.
A touted feature of CoCounsel is its ability to verify its research to provide an audit trail: I took advantage of this by pressing the "Verify" button that came at the end of its initial report. Another seven or eight minutes of work from CoCounsel produced a 263-page document that contained the original 20-page report supplemented by a partitioning of the prose into 76 separate assertions each of which purported to be sourced in at least one case and one West headnote. Here's a screen capture of its verification of Assertion #75.
Lovely. But few people—other than the most overburdened junior associate billing endless hours for a client with unlimited resources—will actually read and use a 263-page version. Indeed, the Ghost platform that hosts this blog won't even let me upload it! At best, CoCounsel's 263-page report can serve as a defensive record to satisfy a court or sanctioning body if a CoCounsel-generated research summary later proves flawed. Or, I suppose it would be useful if, unlike the case here, it were unable to verify one of its citations.
CoCounsel + Claude Opus
Although inartfully written, the 20-page research report contains a lot of valuable information. One would hope, for example, that it has found better materials than the free CourtListener connector did or the far cheaper Midpage connector can. I thus fed the 20-page initial variant of the CoCounsel report to a large language model somewhat less powerful than Claude Fable to see if CoCounsel in combination with a general purpose AI could compete with Claude Fable augmented with plugins that accessed less comprehensive legal databases. I chose Claude Opus, which, until this week was the state of the art Anthropic model and will remain available to monthly Anthropic subscribers past June 22 without having to add on per token charges.
The memo that resulted was quite good, but not as good as that produced by Fable. I'm putting the memo here for those who want to judge for themselves.
Here's the lede:
And here's an analytic passage:
What's wrong with it? Two things. It too missed the fact that Massachusetts v. EPA is hanging by a jurisprudential thread. It thought that the case might not matter much because, after all, at the United States Supreme Court level it was Suncor that needed to establish injury. That's not a ridiculous perspective, but I don't think that is all that needs to be said. The other problem is subjective: style. Opus 4.8 was unable to escape the clunky compendium style of the CoCounsel report with which it had been provided.
To check my intuition that CoCounsel + Opus just wasn't as good as Fable with weaker connectors, I gave both memos to Gemini, 3.1 Pro Extended, which I assumed did not have a dog in the fight. Here's what it reported. When it refers to the "second memo," it is referring to the Fable memo, not the CoCounsel + Opus memo.
The second memo is the superior document because it bridges the gap between academic jurisdictional theory and the reality of Supreme Court practice. By analyzing the actual docket, identifying exactly how the Solicitor General and the petitioners are framing their arguments, and updating the standing analysis to reflect the current Court's hostility toward attenuated climate injuries, it provides highly actionable intelligence. The inclusion of practical tail risks, like the recusal issue, makes it a much more sophisticated and useful tool for anticipating the case's outcome.
That seems about right.
ChatGPT + CoCounsel
I also wondered how the latest ChatGPT model would perform if it were given the same CoCounsel report with which I provided Opus. As it turned out, at least on this problem, ChatGPT did better. The memo it produced is both sharper and more pedagogical. It makes better use of the case research provided by CoCounsel. Here's the document for those who want to look at this in greater detail.
And here's a screen capture of the first page. Notice how it provides a clear conceptual framing and a useful answer. The key cases are all on that first page.
And here, later in the memo, ChatGPT engages in close analysis of precedents. Below you can see it analyzing precedents on jurisdiction over arguably interlocutory decisions and noting precedents supporting the idea that something known as the Cox 4th exception cannot "swallow the finality rule." It's a well written and clear analysis that integrates discussion of doctrine and policy.
Overall, ChatGPT + CoCounsel gives hope that a combination of high-end legal research models (CoCounsel) coupled with very strong but now-no-longer-frontier reasoning models (ChatGPT 5.5 Thinking High) can produce very strong legal analysis that should be ready for partner review. Whether ChatGPT combined with slightly less fancy legal research models like CourtListener or possibly Midpage would do as well is an experiment for another day. So too is a serious experiment seeing how to pair frontier model with other legal research services such as Bloomberg Law. So many experiments, so much research, so little time.
Appendix: What About Fable + CoCounsel?
After writing this blog entry, it occurred to me that I had not tested what might be an even more powerful tool for legal analysis: Fable + CoCounsel. What would be ideal, of course, is if Claude could directly access CoCounsel without a lot of so-very-2022 cutting and pasting. And, in theory, Claude can do this. (See my recent blog entry Unleash Claude for Legal). There is a CoCounsel connector for Claude that should let you be able to access CoCounsel resources from within Claude or even perhaps from within Microsoft Word using Claude for Word with a CoCounsel connector. See, look!
Alas. At present, legal academics do not have access to this direct approach. So, rather than quitting my job and joining a big lawfirm just to gain access to this capability, I used an older school approach. With only a trace of shame, I downloaded the Deep Research done by CoCounsel and then uploaded it, along with the other documents to Claude Fable. I then gave it a prompt similar to what I have already shown several times in this blog entry.
The resulting memo is satisfactory, but the writing is clunky. It appears to have influenced by the ugly compendium style of the CoCounsel Deep Research production. Sometimes it's a little hard to put one's finger on precisely what is wrong with the memo so I gave Fable the earlier production and its Fable+CoCounsel work and asked it a leading question. I also asked it to see if it could figure out how to morph its report to get the best of both worlds: the knowledge of CoCounsel and its own original more analytic style.
I'm now going to show you an alternative memo on the topic. In my opinion it is much more useful for attorneys because of the way it is written. Explain clearly why the alternative memo is much more useful even if the analysis is no better. (If you disagree, you can tell me). Then write a prompt that would morph a memo written in the style of the thing you just produced to the style of the alternative.
Fable then explained what was wrong with its own work. The "alternative" it mentions was the original memo it produced without input from CoCounsel.
Ghost Editor Note: v2 vs Original Memo
I agree with you, and the reason isn't analytic quality — on pure doctrine the two memos overlap heavily (both end at Atlantic Richfield vs. Cox category four, plus petitioner-side Article III injury). The alternative is more useful because of what it treats as the unit of analysis: mine is a memo about the law; yours is a memo about the fight.
The concrete differences:
- It's anchored in the docket, not the doctrine. Part V reports what each actor actually filed — petitioners' brief (May 14, Shanmugam), the SG's amicus (May 21), Boulder's BIO, the July 27 due date — and, critically, where the filings diverge from the abstract analysis ("the statutory fight has shifted from Cox to Atlantic Richfield"; the SG "conspicuously declines to endorse" the Cox argument). An attorney can act on that: it tells you where the battleground has moved, which arguments allied parties won't make, and what's still unfiled. My memo cited the same authorities but as free-floating doctrine.
- Every argument has an owner. "Petitioners' argument writes itself…"; "Boulder will answer…". Litigators think in moves and counter-moves, and prose organized that way is directly reusable for brief-writing, moot prep, or client advice. My memo mostly stated rules and then evaluated them in an authorial voice.
- Headers state conclusions. "The Massachusetts Analogy Is Weaker than It First Appears"; "ASARCO Relocates the Standing Inquiry to the Petitioner"; "Whose Standing, and When." A partner skimming the table of contents gets the entire analysis. My headers ("Background," "Discussion," "I. Statutory Jurisdiction") are filing labels that carry no information.
- Doctrine is given a trajectory, not a snapshot. The special-solicitude section traces Massachusetts → AEP's 4–4 → UARG → Texas (Gorsuch concurrence) → Murthy → the Fifth Circuit, and concludes "the direction of travel … is unmistakable." Static rule statements tell you what courts held; trajectory tells you how an argument will land with this Court — which is what a litigation decision actually requires.
- It maps how the issues interact. Part IV ("nested, not parallel") is a decision tree with payoffs: if ASARCO applies, county standing is irrelevant; a jurisdictional dismissal is "a de facto win for Boulder"; the incentives around dictum on municipal standing are asymmetric. My memo treated the two questions as parallel silos and never told the reader who wins under each branch.
One honest caveat: some of v2's advantage is scope, not style — it covers plaintiff-side standing and special solicitude, which I omitted, and it had the petition-stage and merits filings to mine. But even holding scope constant, the stylistic claims stand.
I've attached Fable's morphing prompt here.
I then used the morphing prompt to see if I could coax Fable back into the lovely style of its original Fable-alone memo.
Use that prompt to redraft the memo you just produced. Don't "cheat" and use the alternative memo as a target. Just use the memo you just created and the morphing prompt. Put the result in a new .docx file
The result was dreadful. Here's what I told it.
Goodness that is awful. It reeks of AI generated prose and is cryptic to boot. "Finality doctrine has a direction of travel." That kind of aphorism is not what a lawyer wants to see in a memo. Plus the audience seems to migrate. First is seems objective. Then it seems like it some kind of bench memo written by a SCOTUS clerk: "Article III does not bind state courts, but it binds this Court whenever its appellate jurisdiction is invoked:" It is basically unreadable. Reread that original memo. The one I really liked. The only difference should be that this version has the benefit of additional research from the "Image a case" core dump that came out of Westlaw Cocounsel. Can you please try to get back to the consistent, readable, analytic style of that earlier work. And Belcher-proof it at the end so the worst AI tropes are vanquished.
The result was much better. I'm attaching it below. As Fable confessed, "however, Fair criticism of the last attempt — the morphing prompt produced exactly the tics it was supposed to cure. This version takes a different approach: your v2 memo is the document, reproduced verbatim, with one addition." In other words, it cheated by using the earlier work and thus making the methodology essentially non-replicable. It may be that with some additional work, we can do better using Fable + CoCounsel than Fable + CourtListener and it may be that some additional human stewardship such as that provided here is required. For now, however, the biggest bang for the buck appears to have come from just using Fable with basic connectors such as CourtListener, Midpage and/or Consensus.