EDRM from start to finish — collection through case law, without vendor bias or fluff.
You already know the doctrine. This course teaches you how the work actually gets done — from the first custodial interview to the last privilege log entry, and everything the model draws in between. You'll leave able to scope, evaluate, and defend an eDiscovery matter — and pick the right tool and vendor for each stage — in plain English.
Delphoria doesn't run your collections or your processing. We help you evaluate the tools and vendors that do — so the workflow is defensible, the invoice is fair, and the story you tell a court holds up. Eight modules move through Collection, Processing, Loading, Review, Analytics, Production, Privilege Log, and the case law shaping how counsel is expected to use these tools — and, increasingly, AI. Each module pairs a lecture with a hands-on exercise on a live-fire matter scenario. No vendor bias. No hype. Just what works, what's defensible, and what actually improves outcomes.
If a custodian used it to communicate, someone can collect it — your job is to know who, with what tool, and how to prove it was done right.
Collections is where defensibility is either built or broken — and it's rarely built by the lawyer running the matter. This module walks through every meaningful data source you're likely to encounter — from a company-issued laptop to a personal cell to a WhatsApp thread on a device you can barely unlock — and the tools your vendors will reach for on each. You'll leave with a Word document of the metadata fields output by each source, a working sense of which vendors specialize where, and the ability to speak to opposing counsel about any of it without hedging.
"You've been asked to oversee a collection from a departing GC — laptop, corporate email, personal iPhone, and a Signal thread the compliance team just learned about. Two vendors have quoted the work."
Assessment for this module: Lab 01 protocol (pass/fail on defensibility checklist) + a 6-question quiz on metadata-source mapping.
Natives on the left. Structured data on the right. Everything worth defending — and every invoice worth scrutinizing — happens in the middle.
"Processing" is the least glamorous, most consequential step in the model — and it's the one you're most likely to hand to a vendor and forget about. Don't. This module makes it visual: the .msg, .xlsx, .pdf, and .zip files a custodian recognizes become rows in a database the review platform can filter, sort, and search. You'll learn how global de-duplication actually works across custodians, why embedded objects require care, and how to read a processing report (we use RelOne as the example) — including the exceptions you can push a vendor to remediate versus the ones you can't.
What the custodian handed you.
What the review tool sees.
"Vendor sent over a RelOne processing report on a 480 GB collection. You have one hour with the client on the phone. What do you tell them — and what do you tell the vendor?"
Assessment for this module: Lab 02 triage worksheet (rubric-graded) + a 5-question quiz on de-duplication logic.
The environment picks the template. The template shouldn't pick the vendor.
Loading is where processed data meets the review platform. The mechanics look different in every tool — Relativity, Reveal, Everlaw, Nuix Discover — but the underlying decisions are the same: which fields land in the tool, how they're typed, what the family relationships look like, and how the Unique-ID stays intact through the handoff. Delphoria maintains a vendor-agnostic template reference you can adapt, hand to any hosting provider, and expect a clean load in return.
"Same processed corpus. Different vendors quoted wildly different load fees. Why?"
Used well, these tools create leverage. Used poorly, margins erode.
This is the module that changes how you talk to opposing counsel. You'll leave able to defend a TAR protocol, run the statistical sampling that supports it, explain the trade-offs of Generative AI summarization, and — critically — identify the documents that should never enter a machine-learning workflow because their text is bad, their length is wrong, or their content will poison the model.
| Axis | TAR 1 · Predictive Coding | TAR 2 · Continuous Active Learning | Generative AI Summary |
|---|---|---|---|
| How it learns | Trained on a stable seed set, then applied to the whole corpus once. | Model retrains continuously as reviewers code — surfaces likely-relevant next. | Doesn't "learn" per matter — LLM applies a prompt across documents. |
| Best on | Static, well-defined corpora. Long timelines. Rich subject-matter expert time. | Most modern reviews. Reviewer time is the scarce input. | Summarization, first-pass triage, translated content, deposition prep. |
| Statistics | Control-set F1, recall/precision at cutoff — defensible with proper sampling. | Elusion/richness sampling at end — defensible, less front-loaded work. | Hallucination rate + human spot-check — no accepted statistical standard yet. |
| Cost shape | Heavy upfront (SME + control set), light back-end. | Even burn — pay per reviewer hour and per prediction cycle. | Per-document token cost + spot-check labor. Cheap at small scale. |
| Do NOT use on | Small corpora (< ~10k docs), heterogeneous file types, low-text content. | Corpora with badly extracted text or missing custodian metadata. | Documents with encrypted/redacted regions, spreadsheets, images without OCR, poisoned prompts. |
| Risk profile | Front-loaded — a bad seed set is expensive to unwind. | Steadier — coding drift is the failure mode. | Novel — hallucination, prompt injection, model-version drift. |
Delphoria posture: use the workflow the corpus and the case can support — and be able to show your work. Transparency isn't a slogan here — it's policy.
"Opposing counsel wants your TAR protocol before you've picked one. What do you send?"
The right analytic on the wrong corpus is a very expensive graph.
Analytics is where the review environment stops being a database and starts being an argument. This module dissects three families — conceptual (clustering, near-dupes, email threading), contextual (concept search, communication analytics, sentiment), and generative (LLM-based summarization, entity extraction, Q&A). You'll leave able to size each one to a matter, quote what it costs, and know when to walk away.
"120,000-doc corpus. Fixed $75k budget for analytics. Build a stack."
The set that leaves your door is the set opposing counsel gets to argue about. Get it right.
Production searches are boolean logic in a suit and tie: pull the responsive population, exclude anything on a privilege withhold, layer redactions where warranted, and hand it over in a format the receiving party can actually open. This module also covers OCR strategy for redacted documents (yes — after redaction, not before), metadata redaction, and the downstream effects on email threading that show up on the privilege log the following week.
"Opposing counsel wants a rolling production of ~4,000 docs by Friday. Half need redactions. Two are attachments to privileged parents."
The log is a document. Draft it like one.
The privilege log is where an opponent decides whether to file a motion. This module walks through the fields most jurisdictions expect, the populations you actually need to log versus the ones you can categorically claim, and the substantive impact of email-threading choices — because how you threaded the review population dictates what a "single privileged communication" even means on the log.
"Same 300 privileged emails. Two threading approaches. Two privilege logs. Compare."
The cases that decided what "reasonable" means when the discovery is digital — and, increasingly, generated.
This module surveys the case law that has actually shaped how attorneys are expected to use discovery tools and, more recently, AI. The cases are grouped by the doctrine they moved. Read them in the order presented; each group builds on the previous. You'll leave with a working sense of what a court will accept, what will draw sanctions, and where the AI questions are still open.
"You're drafting a discovery protocol post-Mata. What do you write, and what do you refuse to write?"
Certification is earned, not attended. You'll be graded on the lab deliverables — the tangible artifacts you'd hand to a client — and short quizzes that check your working vocabulary. There is no final exam. There is a capstone protocol.
| Component | What it looks like | Weight |
|---|---|---|
| Module Labs (01–08) | Rubric-graded practical deliverables — one per module. | 60% |
| Module Quizzes | Short vocabulary + concept checks after each module. | 15% |
| Capstone Protocol | Draft a full EDRM protocol for a matter of your choosing. | 20% |
| Cohort Participation | Moot arguments & peer review in Module 08. | 5% |
This program is designed for up to 20 hours of CLE credit, including at least 1 hour of ethics (Module 08). Actual credit is subject to your jurisdiction's approval. Delphoria Learning will provide the accreditation paperwork; you file with your bar. Fair is fair.
Participants who complete all eight modules and earn ≥ 80% overall receive a numbered Delphoria Certificate of Completion, issued digitally and verifiable via the Delphoria registry. Recipients may use the "DLC-EDRM 2026" credential on résumés, LinkedIn, and firm bios.
Live cohort sessions are the core of the program. Two absences are permitted; a third requires cohort transfer. Recordings are provided for makeup but do not substitute for the lab-day exercises.
Lab deliverables must be your own. You may consult firm playbooks, prior work product, and AI tools — but you must disclose AI assistance on the cover page of any deliverable that used it. See Module 08 for why this matters.
Delphoria templates, checklists, and reference documents are licensed to you for internal firm use. Redistribution beyond your firm requires written permission. White-labeled versions are available for firm-wide adoption.
Reach out before day one. We accommodate what we can — and we're transparent about what we can't. Transparency isn't a slogan here — it's policy.
I have read the syllabus for The Practitioner's EDRM: A Delphoria Learning Certificate. I understand the assessment structure, the CLE-credit process, and the policies above.