In daily use · Los Angeles
The System / reference

This page documents the system end to end: the agents, the pipelines they compose, the verification gate, and the trial tier.

It is the reference for the technically curious; the short version lives on How it works. Built and run independently by Jaden Van, a paralegal, today inside the Law Offices of Allison B. Margolin, PLC in Beverly Hills, where it was built, is in daily use, and supported a federal trial start to finish. The system is fitted to how your firm already works. Everything goes back to your firm’s attorney to review and sign; the final call is always your firm’s.

The work this takes off your desk
  • Discovery audits
  • Motion drafts
  • Hearing prep
  • Discovery responses
  • Court billing
  • New-matter intake
  • Trial preparation
  • And the one your firm needs that is not on this list. That is the one we build for you.
or read the whole system, top to bottom.
01 Why this exists

Start with the week you actually have.

It is Friday. A discovery dump just landed, three hundred and some pages. You have a prelim in twelve days, a motion you have not started, and three deadlines you are carrying in your head, because the calendar does not catch everything. You could hand the dump to someone. You would still have to check all of it yourself.

So the workload is only half of it. The other half is quieter, and it is the half nobody puts on a website. It is the set of things that can go wrong with your name on them.

This system is built around those fears, not around features. Here is what it is, and then here is each fear, answered, in order.

01It invents a case that does not exist, and you file it.
02It is your bar license on the line, not the software's.
03Your client's file quietly feeds someone's training data.
04You stand in front of a judge less prepared than you should be.
The four fears · each one answered below, in order
02 What this actually is

Not a single chatbot prompted per task. Three plain ideas do the work.

A single general-purpose assistant is hard to check, because there is no fixed spec for what it was supposed to do. This system is built the other way: the work is broken into small jobs, run in a fixed order, and checked by something that did not do the work. Here is the whole shape on one line, then the three ideas underneath it.

From
Your systems
email, files, the calendar, the court record
Small workers
Agents
one job each, run in parallel
In fixed order
A pipeline
the same steps, every time
Draft
Work product
built, not yet trusted
Checked by
The gate
reviewers who never saw it written
To
Your desk
for your judgment, your signature
Idea one

An agent

One small worker with one job. It reads discovery, or it checks citations, or it profiles a judge, and nothing else. A do-everything assistant cannot be checked, because you can never say exactly what it was supposed to do. A worker with one job can. Small jobs are checkable jobs.

Idea two

A pipeline

A fixed order of those workers that turns an input into a finished document. The same steps, in the same order, every single time. So the work is repeatable instead of improvised, and when it gets better it stays better, instead of depending on who was at the keyboard that day.

Idea three

The gate

Between the draft and your desk sits a separate set of reviewers that never saw the draft written, reading the source documents the drafters never opened. It is the system checking its own work before you ever see it. Section 04 shows the day it caught a finished filing that was wrong.

The discipline underneath

Every worker hands back the same typed slip. No worker is allowed to pretend.

You do not need to read code to trust the system. You need to know it cannot quietly fail. So every hand-off uses one fixed form, and the form forces an honest answer: it worked, it did not have enough to go on, or it errored. There is no fourth option where a worker guesses and reports success.

This is the real envelope the agents pass to each other. Plain, strict, and the same for all seventeen.

{ "schema": "dispatch-envelope/v1", "primitive": "case-research", "status": "ok", // ok · insufficient_data · error "summary": "Ellison: 1 case, prelim set, 2 open gaps", "data": { … }, // the result; shape known to the caller "artifact_path": ".../Ellison_snapshot.json", "issues": [], "error": null }
03 The agents

Seventeen workers. Grouped by what they do.

These are the agents the firm needs today. The heart of the engagement is the one we build for your workflow, a single new worker, authored by hand, dropped into the same fixed form as the rest.
Research & record
case-research
Pulls one matter's full state from every firm system into a single brief.
fact-ledger
The source of truth for a case. Every fact validated before it is written.
chronology
Builds the dated timeline of events from the record.
sample-mining
Finds the firm's best prior version of the document you are about to write.
People & adversary
person-intelligence
Profiles a judge, a DA, an officer, down to financial disclosures for recusal.
says INSUFFICIENT DATA, never guesses
adversary-modeling
Predicts a named DA's and the judge's attack, then pre-writes the rebuttal.
every forecast tagged as prediction
Drafting & format
document-pour
Pours finished text onto California pleading paper, the firm's own template.
outbound-emails
Drafts the external email or letter in the firm's voice. Never sends on its own.
client-redaction
Burns in victim and witness PII to make a client-safe copy of discovery.
format-scrub
Catches stale firm identity and stray formatting before anything ships.
pdf-form-fill
Fills the court's PDF forms, fields mapped, no retyping.
Court operations
master-sheet-rw
Reads and writes the firm's master calendar, the one source of dates.
filing-channel-lookup
Knows how each court and case type actually wants the filing made.
examination-prep
Builds the witness examination outline, cross or direct.
Verification
citation-verifier
Checks every citation against the public record. Extraction runs on the machine.
pre-delivery-verifier
The gate. Dispatches the six reviewers before anything reaches your desk.
Click any agent to open what it does, how it works, and where it stops.
+ one, for setup cold-start-interview configures the whole system to a new firm on day one. It does setup, not case work, so it sits apart from the sixteen above. That is the seventeenth.
04 The pipelines

Eight finished jobs, each one a fixed run of those agents.

Eight is what the practice runs on today. The ninth is yours, the repeatable job your firm does by hand every week, composed from the same agents and held to the same gate.
motion-pipeline
Draft a motion, end to end, on pleading paper.
halts for your strategy
open the run ▸
discovery-audit-pipeline
Turn a discovery dump into a defense-ready audit memo.
open the run ▸
discovery-response-pipeline
Respond to served discovery: interrogatories, admissions, production.
halts for your narrative
open the run ▸
hearing-snapshot-pipeline
One page of prep for an upcoming hearing, everything at the podium.
open the run ▸
remote-appearance-pipeline
Get a client approved to appear remotely, forms and cover letter included.
open the run ▸
matter-intel-daily-brief-pipeline
The morning brief: today's calendar, what is urgent, what ran overnight.
open the run ▸
ebpace-billing-pipeline
Walk the county billing portal, field by field, from a minute order.
open the run ▸
matter-open-pipeline
Open a new matter and stand up its file the firm's way.
open the run ▸
What it reads from

It works inside the systems your practice already lives in.

Gmail The master sheet Drive Dropbox discovery CourtListener Court portals
Any system you already use, we wire to by its API or connector, and build the connector where one does not exist yet. Some case-management systems have no off-the-shelf connector today. That is a build, not a wall.
05 Watch it work

See it work on a real case, step by step.

This is the discovery-audit pipeline running on a synthetic case. Press play, and watch the compute clock: it races through the bulk reading, then drops to real time at the one moment that matters, when the body camera contradicts the report.

run 8F3A-ELLISON People v. M. Ellison synthetic

One count: possession of fentanyl for sale under Health and Safety Code section 11351. The fentanyl came out of the center console, and the whole case turns on whether that search was good. So the audit goes straight at the search, and at the one sentence in the report that justifies it.

discovery-audit-pipelinerun 8F3A-ELLISON
0:00computeready
01Gatherproduction in
02Cross-checkreport vs camera
03Verifysix reviewers
04Authoritygood-law check
05Memoready
One count: possession for sale. The whole case turns on whether that search was good. Press play, and watch the clock.
0files
0pages
0facts
0findings
0cites
arrest_report.pdf BWC_VANCE_0314.mp4 DASH_U12_0314.mp4 cad_log.pdf lab_report.pdf
Police report · arrest narrativep.5 ¶4
The 48.6 grams of fentanyl came out of the center console. The report states K-9 Argus gave a positive alert, and only then was the compartment searched, and on that basis the vehicle was searched without a warrant.
BWC · Dep. Vanceconf 0.96
02:09:22Ellison ordered out of vehicle
02:11:07console search begins
02:12:44"we'll sort out the PC later"
02:14:22K-9 unit arrives, per CAD #011
02:15:36"alert" claimed at open door
Cross-source contradiction · 02:11:07 → 02:14:22
The search preceded the dog by roughly three minutes. The alert was written after the fact, and with it the sole basis for the warrantless search.
Finding 1 · favorable and material · this is Brady
one honest flag
Body-camera audio at 02:12:44 is partly ambiguous on speaker ID. Surfaced, not buried. You decide whether it matters.
to your desk Ellison · discovery audit reaches you with the contradiction at the top, every claim pinned to its source.
Inspect the catch
Finding 1 · alert post-dated the search, per the record
close ✕
Police report · arrest narrative
arrest_report.pdf · p.5 · ¶4 · DA-000116
The report recites: "K-9 Argus gave a positive alert to the vehicle. Based on the alert and my training and experience, I then conducted a search of the passenger compartment" and located a compartment containing suspected narcotics. On that basis the vehicle was searched without a warrant.
Body-worn camera · Dep. Vance, with CAD log
BWC_VANCE_0314.mp4 · 02:09:22 to 02:15:36 · conf 0.96
02:09:22Ellison ordered out of vehicle
02:11:07console search begins (BWC 09:20)
02:12:44"we'll sort out the PC later" (BWC 10:57)
02:14:22K-9 unit arrives, per CAD #011
02:15:36"alert" claimed at open door
Roughly three minutes separate the console search from the dog's arrival. Probable cause cannot rest on an alert that had not yet happened. The body camera and the CAD log foreclose the report's sequence, and the on-tape "sort out the PC later" impeaches it. Favorable and material: this is Brady.
Authority pinned: Katz v. United States (1967) · Rodriguez v. United States (2015), every cite verified live.
Synthetic case. The cross-source check and the verification are real pipeline behavior.
Synthetic case, real pipeline, real output. The patterns are exactly what the system does. No client data is ever shown.
06 Why it cannot quietly go wrong

The reviewers never saw it written. And they read the source the drafters never did.

This is the answer to the first fear, the one about a confident, finished, completely wrong document. The defense is structural, not a promise to be careful. A draft is checked by a separate gate of reviewers, each with no memory of how it was written, each applying one rubric only.

citation
Every cite, against the public record.
reads · the draft
factual-recital
Charges, counts, enhancement scope, dates.
reads · the charging papers
completeness
Every required section present.
reads · the draft
procedural
Filing channel, signer, service math.
reads · the draft
format
Pleading format, typography, firm identity.
reads · the draft
voice
House writing rules, no machine tells.
reads · the draft
Five of the six can pass a document that is uniformly wrong, because they only ever read the draft, and the draft agrees with itself. The sixth reads the Information the drafters never opened, and reconciles every charge, count, enhancement, and date back to the source paper.
▲ the day it caught a finished filing

A felony filing, marked finished and ready, carried four defects no internal check would surface, because an internal check only compares the draft to itself.

01A blank declaration where sworn facts were supposed to be.
02A sentencing enhancement scoped to every count, when the charging document attached it to two.
03An incident date that did not match the charged conduct.
04A service email addressed to the wrong recipient.
Five reviewers passed it. None of them read the Information. This one read the Information.
A real matter in the firm's active caseload. The client name and case number are not shown here, for the same reason this whole page exists. The defects, and the catch, are exactly as they happened. No invented statistics.
you stay the lawyer

It stops and asks you first.

Before it writes a word of argument, the motion pipeline halts and puts the strategy in front of the attorney. Her decision becomes the drafting directive. It does not draft until she says go. You are not reviewing the machine's theory of the case. It is building yours.

motion-pipeline · Phase 0Gstrategy gate · HALT
trust, then verify

Re-run the receipts yourself.

Every citation is checked against the public court record, and the check is re-runnable: the job ID stays on the work, so a skeptic can run it again and get the same answer. When a source is thin, it says so, instead of dressing it up.

11 citations · verified livejob id retained
07 Confidentiality and data handling

How client files are handled.

The work is handled by one person, not an anonymous team, an offshore queue, or a platform without visibility into how it operates. What follows is exactly how files are treated, in two categories: what can be checked directly, and what the signed agreement makes binding.

What you can verify yourselfin the architecture
Your draft is never uploaded to check its citations. The text is read on the machine, and only the list of citations leaves it for verification.via citation-verifier · local extraction (eyecite)
Your files are never used to train any model. The tools are run under terms that forbid training on your data. Your matters are not anyone's training set.no-training terms · your data is never a training input
It wires into the systems you already use. The work happens in your email, your drive, your case files, by connector, and where a system has no connector yet, that connector gets built, rather than copying everything into a separate product.connector-native: Gmail, Drive, Dropbox, Sheets; others built on request
There is a built-in redaction step. Before a document goes to a client, victim and witness information is burned in, not left to a human remembering.via client-redaction
What the agreement makes bindingin writing, on signing
Your matters are held in strict confidence. Nothing about your case is disclosed to anyone except as law compels, in a pitch, a sample, or anywhere else, and privilege runs through your firm. Every client example on this page is synthetic for exactly that reason; the one real filing shown is public court record, redacted to initials.engagement agreement · confidentiality
Your files stay yours, and are deleted on request. Everything held for your matters is returned or destroyed when you ask, confirmed in writing.engagement agreement · return and deletion
It is written into the signed agreement, not asserted informally. These commitments are part of the Confidentiality and Data-Handling Commitment executed at the start of the engagement, so they are binding rather than a reassurance.signed at the start of the engagement

The technical items above can be checked directly. The rest is written into the agreement you sign. One person, with a name, is accountable for the work: Jaden Van.

Jaden Van
dvan@usc.edu · (909) 287-4983 · Los Angeles
built and run independently · today inside the Law Offices of Allison B. Margolin, PLC
08 Why it only gets better

A paralegal's learning walks out the door. This does not.

When you train a person and they leave, the training leaves with them. Here, every correction is captured, mechanically, in three places, and re-checked forever. This is the part that makes the system an investment that compounds rather than a tool you rent.

It happens
A correction
a miss, a wrong call, a thing the attorney fixes by hand
Same day
Logged as a patch
written down with a target for what "fixed" must mean
Then
Fixed at the source
the rule, the pipeline, or the template, not a one-off
Forever
Frozen as a test
the exact failure, locked into a set that re-runs
And it returns to the system, so the next draft, and every draft after it, is held to the lesson. It cannot quit, and it cannot forget.

When the attorney corrects something, that correction is captured three ways at once. Any one of them would help. Together they make a mistake structurally hard to repeat. The loop is also a fact about the operator: none of these tools was commissioned. They were built unprompted, out of doing the firm's own work, and they keep being extended the same way, which is why the system compounds instead of staying fixed.

capture · one

An enforced rule

The correction becomes a rule that an independent reviewer checks on every future draft. Not a note someone has to remember. A gate that runs.

capture · two

A frozen regression case

The exact failure is locked into a test set that re-runs, so it can never come back quietly. If it ever reappears, the test fails loudly.

capture · three

A reusable sample

The corrected work becomes the template the next draft is mined from, so the improvement spreads to new matters instead of staying on one.

19of 19, last full run
Nineteen cases sit in that test set, and every one is seeded from a real thing that went wrong here, not a synthetic benchmark. The filing defect from the section above is the first case in it. It cannot return without the system catching it.
Honest mechanics: 12 of the 19 run by script in about a minute; the other 7 are dispatched by hand. It is a real, re-runnable regression set, not a one-button CI suite. We do not pretend it is more automated than it is.
09 The proof you can hold

Everything else here is synthetic, on purpose. This is not.

A skeptic should ask whether the polished demo is a flattering mock-up. So here is the real thing to measure it against: an actual motion this firm filed, redacted to initials, pulled from the public court record. The synthetic suppression motion in the run above and this real one are the same work.

Real deliverable · filed, public record · redacted to initials for display Los Angeles Superior Court
Law Offices of Allison B. Margolin, PLC
Allison B. Margolin · Attorney for Defendant
THE PEOPLE OF THE STATE OF CALIFORNIA,

  Plaintiff,
  v.

D. M.,

  Defendant.
Case No. [REDACTED]

NOTICE OF MOTION TO SUPPRESS TANGIBLE AND INTANGIBLE EVIDENCE (Pen. Code, § 1538.5); MOTION TO SUPPRESS AND DISMISS

Dept: 4
Introductionthe theory, in two sentences

Officer P. initiated a traffic stop of a vehicle owned by Mr. M., citing an alleged missing front license plate (Veh. Code, § 5200(a)) and inoperable brake lights (Veh. Code, § 24250(a)). Nothing indicated the vehicle contained contraband or evidence of a crime, and nothing indicated Mr. M. was under the influence (Ex. 3, police narrative).

Statement of factsthe body camera is the spine; every fact pinned to an exhibit

The body camera contradicts the stated basis for the stop. The front plate was present, and dispatch confirmed registration and insurance current and valid in Mr. M.'s name (Ex. 4, p.2 to 3). Officer P. claimed an odor of marijuana, yet no smoke left the vehicle and no marijuana or paraphernalia was recovered. A sealed alcoholic beverage sat in the backseat, but Mr. M.'s blood alcohol content was 0.0 percent (Ex. 3, p.5).

exhibitEx. 4 · p.2 to 3 · plate present

Without articulating probable cause, Officer P. removed Mr. M. from the vehicle and patted him down. By the officer's own report, he struck Mr. M. twice in the face during the arrest (Ex. 2 at [04:28]). Only then were the vehicle and its contents searched.

body camEx. 2 · 04:28
Argumentthe burden placed on the People, every step cited to authority

The warrantless search and arrest were unreasonable. The Fourth Amendment bars warrantless searches subject only to narrow exceptions (Katz v. United States (1967) 389 U.S. 347, 357), and the People bear the burden of justifying one (Schneckloth v. Bustamonte (1973) 412 U.S. 218). Probable cause requires more than a hunch (Ornelas v. United States (1996) 517 U.S. 690, 696).

The stop lacked a valid basis, because the cited violations were not observed and the registration was valid (People v. Reyes (2011) 196 Cal.App.4th 856). The custodial arrest violated Miranda and lacked probable cause (People v. Espino (2016) 247 Cal.App.4th 746, 758 to 760).

And it preserves the constitutional claimthe Racial Justice Act, raised early, not waived

The allegations indicate a violation of the Racial Justice Act, Penal Code section 745(a), shown through the totality of the stop and arrest (see Young v. Superior Court (2022) 79 Cal.App.5th 138). Suppression is governed by Penal Code section 1538.5 (People v. Williams (1999) 20 Cal.4th 119, 126 to 127).


filed copy
redacted
redaction defendant to initials; case number and officer names removed source public court record, Los Angeles Superior Court authority every citation verified live

Read the synthetic Ellison motion and this real one side by side. They are the same work. That is the entire reason this one is here, redacted, in the open: so the demo has something real to be measured against.

10 The depth tier

When it goes to trial, the system goes with you.

Synthetic reconstruction
This tier is bespoke, built per engagement, and it supported a full federal trial. You cannot see that one: it is under a protective order, and nothing from it leaves the firm. So what follows is a synthetic reconstruction of the same capability patterns, on an invented case, so it can be shown at all. Synthetic case, real patterns.
United States v. D. Okafor · No. 2:25-cr-00481-AC (C.D. Cal.) · Hon. A. Calloway · 21 U.S.C. §§ 846, 841(a)(1), (b)(1)(C)
The decision memo Should the client take the stand? The call that can decide the trial. synthetic reconstruction · prepared for the trial team
Memorandum · attorney work productprivileged in the real matter · synthetic here
Re: United States v. D. Okafor, whether the defendant should testify
Bottom line: do not call the defendant on the current record. The causation defense is won on the government's own expert, who already conceded the death "may be multifactorial," and it does not need him. Putting him on the stand opens the proffer and the text thread, and converts a clean reasonable-doubt posture into a credibility contest. Condition and fallback below.
What only his testimony can give
  • An innocent account of the relationship with J.R., that it was friendship, not a sale.
  • That he did not know the pill contained fentanyl.
  • Context for the texts the jury will otherwise read the government's way.
  • A person for the jury to acquit, instead of a name in a chat log.
What it puts at risk
  • The proffer. The moment he testifies, his prior statement to the agents is fair game, including the parts kept out.
  • The full text thread, no longer limited to the friendly reading.
  • A credibility contest with Abara, the cooperator, that he is positioned to lose.
  • The causation concession. One agreement that "he took what I gave him" feeds the but-for element the defense is contesting.
The cross, as the system rehearsed itprosecution · defendant on the stand
QYou and J.R. were close.
AYes.
QClose enough that he came to you when he wanted something.
ASometimes.
QThe night before he died, he texted you, "you got me?"
AYes.
QAnd you answered, "I got you."
AYes, but it was not like that,
QYou got him. In March, you sat down with the agents, with your lawyer, and you told them where the pills came from.
AI did.
QSo you knew there were pills. You knew the source. You just want this jury to believe you did not know what was in them.
AI did not,
QYou did not ask.
ANo.
QYou never ask. You move what you are handed, and you do not ask what is in it. Nothing further.
Every answer is true. Every answer is a brick in the government's wall. He cannot give the friendly account of the relationship without putting the proffer and the full thread in play, and the closing sequence turns his own honesty into the knowledge the government has to prove. The causation defense never needed him to say a word.
Recommendation

Do not call him on the current record. We win this case on the government's own toxicologist, not on the defendant's credibility, and we should not trade the stronger ground for the weaker.

Condition. Reassess only if the Marchetti cross fails and the causation case survives intact. If the jury then needs an affirmative account, the fallback is a narrowly scoped direct, limited to the relationship and rehearsed against the cross above, never an open account.

Ex. A · the "I got you" text thread Ex. B · defendant proffer, 3/14 (sealed) Ex. C · Abara ROI, 9/14 vs 2/03 (changed story) Ex. D · Marchetti report at 14, "may be multifactorial"
Synthetic reconstruction. The pattern, weighing the testify decision, rehearsing the cross that decides it, and writing the call into a work-product memo, is exactly the judgment the trial tier produces. The case and people are invented. Because this mirrors recent real work, this synthetic version was reviewed and cleared by Allison Margolin before publication.
Exhibit 1 The system red-teamed the firm's own opening and wrote the prosecution's rebuttal to it. prosecution red-team · vs. defense opening draft v3
2 vulnerabilities flagged · run 4m 12s
Vulnerability 1 · the toxicology argument contradicts itself severity: high
"The 11 ng/mL femoral figure is inflated by postmortem redistribution and cannot be taken at face value."
our draft v3, at 14
"Femoral blood is the gold-standard draw site precisely because redistribution affects it least. The government's own lab chose it."
our draft v3, at 22
the government's closing, as the system wrote it against us
"The defense told you the femoral number was inflated. Eight pages later they told you femoral is the most reliable draw site there is. Both cannot be true, and they said both."
repairCut the at-14 redistribution sentence. The tolerance and multifactorial arguments run through our own expert. Redistribution belongs in the cross of the government's witness, not in our affirmative theory. Caught at draft three, not at the podium.
Exhibit 3 Every promise in the opening, on the hook. opening draft v4 · 11 promises indexed
Opening promises ledger nothing is said at the podium without a source pin and a witness who will deliver it 11 promises · 5 verified · 2 supported · 2 unverified
promise, as it will be spokensourcedelivers at trialstatus
1"The medical examiner could not say fentanyl alone killed J.R."Autopsy Op. ¶6 · BS#0412cross of MEverified ✓
2"Dr. Marchetti has testified nine times. Nine times for the government."Salgado Tr. 1133:5 · CVMarchetti cross, arc 1verified ✓
3"The pill came from one source that night."Voss proffer 3/12 · BS#1644Voss crosssupported
4"Abara's story changed after the plea offer."ROI 9/14/24 vs 2/03/25Abara crossverified ✓
5"No one ever tested the residue on the foil."lab intake log · BS#1873stipulation pendingunverified
Exhibit 2 · expert conflict matrix

A conclusion that never moves.

The government's toxicologist swore "fentanyl was the sole but-for cause" in this case and, word for word, in a 2023 case with different facts. Identical language. A template, sworn to twice.

212 pp. transcript · 19 concessions pinned · under Burrage
Exhibit 4 · binder assembly

The cross-binder builds itself.

A script pulls every pinned concession from the transcript and assembles the tab-keyed cross binder, so the examination outline and the paper behind it stay in lockstep.

build script · 19 tabs · 212 pp. → one binder
Exhibit 5 · venire research

Panel analysis, inside the ethics rule.

The venire is researched under ABA Opinion 466, the rule that governs reviewing jurors online, so the work stays inside the line that protects it.

panel of 60 · batch 3 of 5 · ABA Op. 466
Trial-grade support, scoped per engagement. Under the hood it is two ported primitives, chronology and examination-prep (both adapted from Anthropic’s open claude-for-legal skills, rebuilt around the firm), plus a war room hand-built for the case. It is not a packaged pipeline, and it is not the pilot. When you have the trial, we scope it then. The patterns above are exactly what the system did on a real one. The case is invented so it can be shown at all.
11 Start here

Running it on one real matter.

Pick one real matter, a discovery audit or a motion. It runs on your case, and the work product reads the way the real motion above reads.

The first matter is free: one discovery audit or one motion, about two weeks start to finish, yours to keep.

There is no scripted follow-up and no fee to negotiate at this stage. It is one piece of real work, on a matter you choose, and you decide what it is worth from there.

the intake
Three things, and a file.
Or forward one file directly. dvan@usc.edu
Work runs directly with Jaden Van, a paralegal who built the system and runs it independently. It operates today inside Allison B. Margolin’s active criminal-defense practice, on real deadlines; that practice does not supervise work done for your firm. Your own attorney directs, reviews, and signs every deliverable; the judgment stays with your firm. One person is reachable and accountable for the work, not a platform or a queue.

Synthetic cases, real patterns. Every capability on this page traces to a working skill in daily use. No client data is ever shown.