Prep Reports · Bartolus
What it is
Bartolus reads every question a judge has asked across every case in the corpus. It finds the patterns — how this judge opens, what triggers skepticism, when a long back-and-forth chain is coming — and applies them to your briefs.
The result is a customized prep report: predicted questions with brief-grounded answers, a record vulnerability audit, rejection triggers from the corpus, and a backtest score from the last time the model ran on a case like yours.
Every section is generated from the corpus and from the briefs. No generic boilerplate. If a section can't be grounded in data, it says so — and flags the limitation explicitly.
Personality, opens-with pattern, danger zone — built from every tagged question this judge has asked in the corpus, not from reputation or general knowledge.
Predicted questions with attorney-voice answers grounded in the brief. Every answer cites a specific page. Every response leads with a one-sentence short answer — the kind you can actually deliver under time pressure.
For each predicted question, the specific passage in your brief that answers it — with key phrases you can use at the podium. Added after our first attorney feedback: the attorney doesn't need us to explain the law. They need the exact language from their brief, handed to them.
Verbatim quotes from the corpus where this judge cut off an argument — with the pattern that triggered the response and what it means for your case specifically. Drawn from real arguments, not inference.
Facts a judge may probe regardless of what either side argues. A judge once asked about a person neither brief mentioned by name. That question is why this section exists — covering representation at critical hearings, correspondence that cuts against the client's characterization, and factual silences.
Opponent's likely arguments, point by point, with counters grounded in the briefs. Each counter is calibrated to what judges in this subject matter actually surface during rebuttal, not to what opposing counsel is likely to say.
On spiky topics: A topic that fires in one case out of four is not a 25% prediction — it is bimodal: 75% chance of silence, 25% chance of a deep dive. Bartolus distinguishes primary predictions from contingencies. Spiky topics are flagged as "if raised, expect an extended exchange" — not as primary predictions that crowd out what the judge is actually likely to ask.
On thin data: If Bartolus does not have enough corpus data on a specific judge in a specific subject matter, the report says so and flags it with a ⚠ caution. The report is directional, not predictive — and the distinction is labeled explicitly. We do not paper over data gaps with confident-sounding generalities.
Real examples drawn from an actual Bartolus prep report. Parties, judges, and case names anonymized.
Tends to press litigants who appear to be running inconsistent positions across proceedings — particularly when a party's prior conduct in earlier litigation undercuts the argument now being made. Has a documented pattern of shutting down arguments that depend on separating closely related legal theories when the underlying facts are essentially the same. This is a consistently active bench: expect a steady volume of questions across the full argument, not a burst at the end.
Likely to open by pressing the appellant directly on the sequencing of the two lawsuits — specifically, why the constitutional claims were not raised in the prior proceeding given that the client had already experienced all of the relevant events before the first lawsuit was filed. The City's timeline in its brief is precisely the factual predicate this judge tends to surface first.
The highest-risk moment is the statutory-constraint argument — the claim that the prior proceeding was so narrowly confined by statute that the constitutional claims could not have been brought there. This judge has a documented pattern of pushing back hard when a party argues it could not have raised claims earlier, particularly when the record suggests those claims were available and the party simply chose not to bring them. The timeline in the opposing brief — showing the relevant events preceded the first lawsuit by months — is precisely the kind of factual predicate this judge tends to surface and press on directly. Lead with the statutory text. Anchor every other argument to that dispositive point before the Court asks you to.
"I just don't even really understand why we're in such deep hay here over all of this detail, when it's pretty clear to me she doesn't have a cause of action period."
This judge cuts through elaborate doctrinal scaffolding and flatly rejects a claim when she views the threshold legal question as dispositive — expressing frustration with detailed argument that avoids confronting what she sees as a clear barrier. Do not allow argument on the statutory-constraint question to become a dense recitation of sub-factors and case distinctions without first confronting whether the opposing brief's authority forecloses the argument at the threshold. Lead with the statutory bar, and anchor every other point to that dispositive issue.
"Your client had already experienced all of the relevant events before the first lawsuit was even filed. Walk me through why those claims weren't included in that proceeding."
Your Honor, the client was statutorily confined to challenging specific orders under the applicable review provision — a provision that did not permit adjudication of the constitutionality of an entirely separate ordinance, or claims for monetary relief against a different department. The prior proceeding could only address whether the specific notices before the commission were valid. It was not a vehicle for constitutional claims against a separate program administered by a separate office. (Br. pp. 24–26)
Reject the framing that "could have been raised earlier" is the operative question. Under res judicata, the issue is whether those claims could have been raised in the specific proceeding that was actually filed — not whether a different lawsuit was theoretically available in a different forum.
Who represented whom at a critical hearing. The record contains a signature block from a prior proceeding listing a different attorney on behalf of the client. Neither brief addresses this by name. A judge asking about continuity of representation — or about what was said at that hearing — surfaces a factual inconsistency no one has prepared an answer for.
Key correspondence that cuts against the client's characterization. The opposing brief quotes an internal communication showing the client was informed of the relevant regulatory exposure months before the first proceeding was filed. The client's brief characterizes this as routine notice. A judge reading the communication directly may not reach the same characterization.
An unsubstantiated factual claim. The client's core narrative depends on what a prior tribunal verbally communicated — but there is no transcript. Asserting this at argument without record support invites a sharp demand for documentation. One judge's only substantive intervention in the actual argument was exactly this demand. Prepare a precise answer before it comes.
After argument, we run the transcript against the prep report and score coverage. The primary grade is driven by what judges actually asked that the report failed to prepare for — not by how many predictions fired. Over-preparation costs nothing. Under-preparation is a product failure. Every miss becomes a specific architectural improvement.
Res judicata · Fifth Circuit · Panel of three · Anonymized
The report correctly identified the core battleground issues and the dominant judge's questioning style — including the specific interrogation technique he uses when two sets of pleadings share the same factual nucleus. However, one panelist asked exactly one question and was otherwise silent, having been crowded out by a co-judge's extended-engagement pattern. That prediction failure led directly to panel dynamics modeling, now standard in every report.| Judge | Predicted | Hits | Near hits | Misses | Hit rate | Notes |
|---|---|---|---|---|---|---|
| Judge A | 4 | 4 | 0 | 0 | 100% | Dominated the argument. Record-focused, extended-engagement style was precisely predicted. Back-and-forth probing on complaint overlap, timeline, and structural barriers matched exactly what the report anticipated. All four predicted questions hit. |
| Judge B | 6 | 3 | 2 | 1 | 50% | Less active than predicted. Questions matched her expected pattern — efficiency concerns, limiting principles, socratic framing — but she deferred substantially to Judge A throughout. The report correctly identified her style; it overestimated her volume. |
| Judge C | 6 | 0 | 1 | 5 | 0% | Asked exactly one question — a sharp procedural challenge demanding documentation for an unsubstantiated factual claim made from the lectern — and was otherwise silent. Historical patterns predicted active textualist engagement. Judge A's dominance suppressed participation entirely. |
The miss: Judge C had been categorized as an active questioner in prior cases — a textualist with a pattern of pressing for record support and referencing opposing briefs directly. We predicted six questions. He asked one, and it was entirely procedural.
Why it happened: Judge A dominated the argument so thoroughly — extended back-and-forth chains, record deep dives, multi-question threads — that Judge C and Judge B had little room to intervene. The report treated the panel as three independent questioners. It was actually one dominant questioner and two observers.
The fix: Prep reports now model panel dominance as a distinct variable. If a judge with an extended-engagement pattern is on the panel, predicted question counts for other panelists are adjusted downward, and the adjustment is labeled explicitly. When Judge A is on your panel, prepare for one judge's questions in depth — not three. The report will tell you which judge that is and why.
We picked a decided Fifth Circuit appeal and generated the full report blind: both briefs plus the corpus, with this case excluded from every judge profile the model could see. Then we scored the report against the argument transcript. Your report arrives as a formatted PDF — these are pages from this one, followed by what the panel actually asked.
Black bars are redactions: parties, counsel, docket, dates, amounts, page numbers, and source-case names. Everything else appears exactly as generated.
How each panel judge questions — built from the judge's record, not reputation.
Actual report page · Real case — anonymizedIn re · No. · Panel: , , · You are Appellant
expects lawyers to defend apparent contradictions in their legal theories, particularly when they claim to defer to one jurisdictional forum while simultaneously challenging another's authority over similar matters. Rather than accepting formal procedural compliance, demands detailed explanations of how legal theories work in practice and their real-world implications. This judge listens selectively but pounces with precision when spotting inconsistencies or weaknesses in an attorney's position.
will likely begin by challenging the apparent contradiction in 's position — arguing the bankruptcy court lacked jurisdiction while simultaneously claiming it should have abstained from exercising jurisdiction it supposedly didn't have.
The abstention arguments on pages present the greatest vulnerability. will likely hammer the logical inconsistency of seeking mandatory abstention (which presupposes jurisdiction exists) while arguing the court had no jurisdiction at all.
Moments this judge shut an argument down — quoted verbatim from prior arguments — and what to do instead.
Actual report page · Real case — anonymizedIn re · No. · Panel: , , · You are Appellant
"…sounds like a policy argument rather than a textual argument. I think there's no reason to treat the two situations differently."— , No. (5th Cir. )
Don't argue that the bankruptcy court should have abstained based solely on policy grounds like judicial efficiency or forum convenience. Judge will expect concrete textual analysis of jurisdictional statutes and abstention requirements. Frame abstention arguments around specific statutory language (28 U.S.C. § 1334) and Fifth Circuit precedent on mandatory/permissive abstention factors, not generalized policy preferences for state courts handling state law claims.
The question, why it's coming, and an answer grounded in your brief — with page citations.
Actual report page · Real case — anonymizedIn re · No. · Panel: , , · You are Appellant
will test arguments by focusing on practical realities that undermine a party's position. This question exposes the disconnect between 's conduct at the hearing and his later claims.
"You were present at the hearing, you heard the judge allocate specific amounts to you and your attorneys separately, you agreed to it, and you waived your right to appeal. What changed between that moment and when you filed this lawsuit six months later?"
What changed was my attorneys' breach of their promise. At the hearing, I agreed based on their assurance that the attorney fee allocation would be shared per our agreement. … The breach and my discovery of it occurred after the Order — that's what gives rise to these claims. I couldn't have known at the hearing they would renege on their promises. (App. Br. , )
[ In re — Fifth Circuit ]
Three predictions from the report, next to the transcript.
Actual argument transcript · Real case — anonymized"You were present at the hearing … you agreed to it, and you waived your right to appeal. What changed between that moment and when you filed this lawsuit six months later?"
"[I]n your theory, what should have done? Let's say he's upset at that moment he sees the order. What should he have done?"
Demands you explain the practical mechanics and real-world implications of procedural or technical legal arguments, not just their formal compliance.
"He should tell his lawyers what to say in court, and then the lawyers are supposed to dutifully report what the client said, even though the lawyers themselves disagree and obviously want more money."
The backtest scored this exchange as pivotal to the argument.
"…the record shows you were present at the hearing, heard the court allocate funds separately to you and to your attorneys, and consented to that allocation. How is this not simply a collateral attack on the bankruptcy court's judgment?"
"…[W]hat he's really challenging is the integrity, the fiduciary responsibility … of his lawyers. And that's not really a bankruptcy issue. … [H]e's challenging … the actions of his lawyer, not the actions of the bankruptcy judge."
Scored against the transcript · 26 predicted questions · Real case — anonymized
The report predicted the substantive concerns that dominated argument — the collateral-attack framing, the conduct-at-the-hearing problem, the fiduciary-duty theory — and the questioning style of both active judges. It also over-predicted technical jurisdictional questions that never came. Over-preparation costs an attorney nothing. The grade turns on the two topics the panel raised that the report missed.The miss: Both active judges spent much of the argument on a person the report never flagged — a separately appearing attorney whose presence at the critical hearing bore directly on whether the client had independent representation. The second miss was pre-hearing correspondence about what that hearing would decide. Both were sitting in the record.
The fix: That failure became a standing section of every report — the Record Vulnerability Audit: who represented whom at critical hearings, correspondence that cuts against the client's characterization, factual silences a judge can drop into. The section described in § 01 exists because of this case.
Get started
Available for arguments in any federal circuit and the Supreme Court. Turnaround is typically 48–72 hours once briefs are received.
We send the report before you pay. If it doesn't satisfy you, delete it. You pay nothing.