By Bill Hartzer · Published
Six working papers, reproduced in the book as Appendices A through F and published here in editable form so that a template can become your template rather than staying somebody else's. Each is short, each is meant to be marked up, and none of them assumes anything about which systems you use or whether you use any at all.
The six working papers
What each contains
Appendix A — AI-use log template
A single ruled page with six columns — date, matter, system and version, purpose, what was submitted, and what was done to verify the output — plus a header block naming the engagement, the retaining firm, and the protocol in force. It is designed to be filled in a line at a time over months, not reconstructed in an afternoon.
The four rules printed above the grid are the ones that decide whether the log is worth anything. Enter the row when the work happens, not later. Describe the purpose in the language of the task rather than the tool. Record verification as an act, naming the source consulted, not as a checkbox. And keep one log per matter, because a single running log across engagements becomes producible in all of them at once.
The form deliberately has no field for the prompt text itself. Prompts live where the system keeps them, exportable and complete; a paraphrase transcribed into a log is a second, worse version of a record that already exists, and the difference between the two is exactly what an examiner will spend an afternoon on.
Appendix B — Engagement letter provisions
Nine model provisions, each drafted to sit inside an ordinary engagement letter without renumbering the rest of it. They cover the protocol in force, sequence, independent authority, the record and its production, permitted accounts and service tiers, persons working under the expert's direction, materials supplied by counsel, disclosure, and the schedule on which the record is exported.
Two of the nine protect the retaining firm rather than the expert, and they are the two most often left out. The provision on materials supplied by counsel establishes that the expert does not submit privileged or protected material to any system without written confirmation that doing so is permitted — which puts the decision where the privilege actually sits. The provision on persons working under the expert's direction extends the protocol to associates, contract reviewers, and anyone else whose work reaches the report, because an expert's independence is not preserved by a rule that stops at the expert.
Each provision is annotated with the clause of the Standard it implements and a short note on what changes if it is negotiated away. Firms adopting the Standard across a practice will find the annotations more useful than the text; the text is a starting draft, and the annotations are what let a general counsel decide which departures are survivable.
Appendix C — The verification checklist
The checklist divides every assertion in a draft report into six categories of checkable thing — citations, figures, dates, quotations, specifications, and characterizations of a record — and states, for each, what counts as having checked it. A citation is checked when the source has been opened, not when it has been confirmed to exist. A figure is checked when it has been recomputed from the underlying data, not when it has been compared against the same output that produced it.
Each assertion resolves to one of four dispositions: verified against a named source, corrected, hedged with the uncertainty stated in the report itself, or removed. There is no fifth disposition, and in particular there is no probably fine. The discipline of the checklist is that it forces every assertion to a terminal state before the report is served.
The escalation procedure covers the case that causes the most trouble in practice: a source that cannot be located. The checklist treats an unlocatable source as a failure of the assertion, not of the search, and the assertion comes out. That is a harder rule than most practitioners apply, and it is the rule that makes the verification pass mean something when it is described under oath.
Appendix D — Deposition question sets
Two question sets that run in opposite directions. The first is for retaining counsel preparing an expert before the report is served — the questions that surface a problem while it can still be fixed. The second is for examining counsel taking the deposition of an opposing expert, sequenced so that each answer constrains the next.
The seven questions are the core of the examining set. They establish, in order, what systems were used, when in the sequence they were used, what entered them, what came out, what was done to check it, what record exists, and where that record is now. They are written to be asked of an expert who used nothing at all, because the answers are as informative in that case as in any other.
The four trap questions are the ones that sound devastating in a transcript and establish nothing. Each is set out with the reason it fails, the objection available, and the reformulation that asks the same thing in a form that actually holds. An examiner who asks the trap version gets a clean answer and loses the topic; an examiner who asks the reformulation gets an answer worth having.
Appendix E — Reproducible analysis specifications
Specifications for seven analyses that recur across disciplines — deduplication, date range extraction, keyword and phrase incidence, entity co-occurrence, sequence reconstruction, numeric reconciliation, and sampling for review. Each specification states what the analysis must take in, what it must emit, and what must be recorded alongside the output for the result to be reproducible.
The specifications are deliberately written against the task rather than against any language or library. An analysis that satisfies the specification in Python satisfies it in R, in a spreadsheet formula, or in a query, and that portability is the point: the obligation under the Standard is that the method returns identical results when repeated and is preserved, not that it was written in any particular way.
The five checks are for the situation most experts are actually in — holding a script they did not write and cannot fully read. They are: confirm the input is what you think it is, run it twice on the same input and compare, run it on a small set whose answer you already know, examine the boundary rows rather than the middle, and reconcile the totals against a count you obtained another way. None of the five requires reading the code, and together they catch the failures that matter. The runnable implementations live here on the site.
Appendix F — Preservation and production protocol
The day-one folder is the whole protocol in one instruction: at the moment of retention, before any work begins, create the folder that will hold the record for this matter and put the engagement letter in it. Everything else in the appendix follows from having somewhere for the record to go from the start.
The export cadence covers what most practitioners get wrong by omission. Conversation histories held by a vendor are not a record you control; accounts get closed, retention windows expire, and terms change. The appendix specifies a monthly export in the provider's own complete format, stored in the matter folder, with a note of what was exported and when — so that the record survives the account.
The production section addresses what to do when a request arrives, including the sequence for asserting an objection without destroying the material the objection covers. It closes with the cross-matter problem: a log, a folder, or an account that spans engagements can become discoverable in all of them, and the separation that prevents that has to be established at the beginning, because it cannot be established afterward.
Requesting the working papers
Write to record@theexpertrecord.com with the word papers in the subject line and they will be sent to you. There is no charge and no obligation. If you would rather not send an email, every one of them is reproduced in full in the book.
Terms
Adapt the working papers freely for use in your own practice. The one restriction is the same as the Standard's: do not distribute an altered version of the Standard itself under its name. Everything else here is a template, and templates are meant to be edited.