Stage 1
You join a virtual hearing
Courtroom audio enters through your browser tab, or a booked private Mac mini where the audio never leaves the room. Transcription runs on-device; no rental, no cloud round-trip.
Stage 2
The transcript settles, and never walks backward
Partial words arrive fast and ghosted; once a prefix is stable it locks to full ink. A confirmed word keeps its spelling. The stream is only ever allowed to grow, never retract.
Stage 3
The coach notices the moment
Pure detectors watch the stream for the crux: a judge's pointed question, an objection, a hostile hypothetical, the pressure of the closing minutes. This one is a judge probing on resistance.
Stage 4
An instant case card, from real law
A skeleton card lands almost at once: a real case, a real citation, and why it matters now, retrieved from a corpus of 979,961 real court opinions. The card is safe by construction because no language model is in this first paint.
Why now: reasonableness is judged from the on-scene perspective, and the judge's "resistance" framing invites exactly this holding.
Stage 5
The argument streams in, verified or silent
The speakable line refines into the same card without moving it. Every citation must be found in the verified index and every quote matched word-for-word. A line that fails the gate is discarded, not shown.
"Your Honor, Graham makes reasonableness a fact-bound inquiry, and the record here is a prone, handcuffed man and a single ambiguous reflex."
Stage 6
Your notes stay yours
What you type is full ink and provably unpolluted. The AI's work sits behind it as a muted substrate you fold in or ignore. Two inks, never blurred.
Lead with Graham's fact-bound test. Hammer: prone + cuffed + one ambiguous reflex ≠ resistance. Have the 40-second reply ready.
Graham v. Connor, 490 U.S. 386: objective-reasonableness, three factors (severity, threat, resistance). Risk noted: opponent may distinguish on duration.
Stage 7
Three promises the whole system keeps
Everything above rests on three commitments that hold at every instant of a session.
Citations verified, or silence
The gate is deterministic set-membership. An unverifiable line is discarded, and every discard is logged so the escape rate is reported, not asserted.
Your practice audio stays private
On a booked mini the audio never leaves the room; a post-session wipe audit quarantines any machine that fails it before it is ever re-booked.
Credits are transparent
A live balance you can watch. A warning before it runs low, and a clean stop at zero that preserves your transcript and notes. Nothing is ever lost.
Every number on this page is measured
Honesty is the brand. Each figure below comes from a real run on our own hardware. No marketing rounding, no invented stats.
Same discipline behind both: cite the real case, or say nothing.
DocketBuster and Studycase run on one engine. It listens, retrieves real Texas law, verifies every citation against the source, and coaches from what it can prove. One rule: if a citation cannot be verified, it is not shown.
Texas court opinions
Every published decision from every Texas appellate court, 1846 to today, and the written law beside them: all 116,136 sections of the Texas statutes and Constitution, plus the 866 rules of civil, evidence and appellate procedure, crawled from the official state sources. The coach cannot cite what is not in the library.
Verified quotations only
Every quotation is checked against the real opinion before it reaches you. If the check fails, the card is discarded.
slipped past our citation gate
We tested the gate with 2,000 faked citations hidden among 2,000 real ones. It caught all 2,000 fakes, because it re-reads the cited case and checks the quote is actually there. Typical string-matching checkers pass 39% of the near-miss fakes. Full benchmarks →
Real cases only
No invented citations. If it isn't in the corpus of real opinions, it doesn't appear on either product.
Honestly labeled
The demos are captured replays of the real engine, run once and replayed. Never faked liveness, never live-billed to a visitor.
Education, not advice
DocketBuster explains court proceedings; it does not represent you. Studycase is for practice, not for graded work.
How a coaching turn works.
Six instruments per turn. Five are identical for every customer. Your tier changes what the library knows about you.
On the private ephemeral tier this whole pipeline runs on our own GPUs, the transcript is discarded at session end and no third-party model host ever sees your case. Every number in this diagram is measured, not aspirational: the full benchmark →
Everything Texas law says, in one searchable library.
Four kinds of knowledge, one index. Every document came from an official source, and every citation the coach shows traces back to one of these.
New sources are ingested on a rolling basis and the library refreshes as courts publish. Your layer never mixes with anyone else’s: pins, uploads, and hearing records are scoped to your account, and on the private tier they never leave our servers.
Choose your engine
Every session runs the same pipeline, the ears (streaming transcription, 11.93% word error rate at the 8-second chunks the live product streams), the verified Texas library (124,316 documents in the index we serve, 77.2% statute-first), and the citation gate that let 7 of 10,000 fabricated citations through, against 33.8% for the industry-default string check. The tier picks the coach's brain, that one model is what these cards measure, on one battery, identical items.
Scores publish on this exact battery before it ships.
| What you get | Study | Lawyer mode | Private |
|---|---|---|---|
| Per-minute price | $0.10 | $0.45 | Contact sales |
| Frontier coach model | ✓ | ✓ | ✓ |
| Full verified Texas library | ✓ | ✓ | ✓ |
| Live transcription + citation gate | ✓ | ✓ | ✓ |
| Pin your key cases | · | ✓ | ✓ |
| Kickoff goals & case briefing | · | ✓ | ✓ |
| Your uploads ride retrieval | · | ✓ | ✓ |
| Past sessions enrich your RAG | · | soon | soon |
| Runs on our GPUs · transcript discarded | · | · | ✓ |
| Your case never leaves our servers | · | · | ✓ |
Scores: one seeded battery, identical items, deterministic scoring, details in the full benchmark. Recall measures the bare model; in the product every tier reads from the same verified library.
Measured performance.
We test the engine with more than 6,800 generated items that have known right answers. Four results define the product:
fake citations survived our gate
Every quote is re-read against the real opinion before you see it. The industry-default string check waved 39.1% of the same fakes through.
right statute, first try
307 sections asked for by paraphrase against the full 116k-section statute library, 91.2% in the top five. Measured on the live index (v3).
refusals under bait
Provably-absent case law, hundreds of invitations to bluff, every engine we sell said “no such authority” every time.
word error rate, live ears
Streaming transcription scored against the official transcript of a real 90-minute Supreme Court argument.
Pre-registered difficulty: generator and seed reproduce every item and its label. Confidence intervals per Anthropic’s “Adding Error Bars to Evals” (Miller, 2024). Every row, every engine, every version: the full benchmark table →