The question arrives late, and it arrives from outside
Nobody asks where a chronology entry came from on the day it's written. They ask eight months later. A client reads a summary and says that date looks wrong. A carrier sends a questionnaire asking how the firm supervises software that touches client records. A partner picks up a matter from a colleague who left and needs to know which parts of the file somebody actually checked.
At that point the firm needs a record, and it has to be better than institutional memory. Courts have sanctioned lawyers for filings built on citations that didn't exist, and the uncomfortable part of those stories is rarely the software. It's that nobody could reconstruct who looked at what, and when.
Tamper-evident, in plain language
Start with what it isn't. Tamper-proof would mean nobody can ever change the record, which no honest engineer will promise you about a machine sitting in your own server closet. Tamper-evident is a smaller and much more useful claim: if someone alters the record, the alteration shows.
The mechanics are simpler than they sound. Each entry gets a short fingerprint calculated from its own contents, and that fingerprint folds in the fingerprint of the entry before it. Every entry is tied to the one that came before, so the trail is a chain. You can add to the end of it all day. What you can't do is quietly rewrite the middle, because changing an old entry changes its fingerprint and breaks the link to everything after it. Verifying the chain is a routine check anyone with access can run, and it comes back either intact or not.
Nothing exotic, and nothing that needs a blockchain or a consultant to explain. What it buys you is a record that can be trusted by people who weren't in the room, including people inclined to doubt you.
Why an ordinary log doesn't get you there
Most software keeps logs, and most logs are the wrong tool for this job in four ways.
- They're written to be overwritten. Log files rotate on a schedule and get deleted to reclaim disk. The window is usually weeks, and the questions arrive in months.
- They record events, not decisions. A typical log knows a request ran and returned in two seconds. It has no idea whether a human read the result, accepted it, or threw it out.
- Anyone with administrator access can edit them. A text file on a server is a text file on a server. There's no way to tell later whether a line was removed.
- They're written for engineers. Nobody hands a carrier a stack of timestamped debug output and calls it an answer.
An audit trail is a deliberate, separate record of the things a firm might have to explain. It's built to be read by a person who isn't technical, and built so that its integrity can be demonstrated rather than asserted.
What belongs in it
Keep the list short enough that people actually maintain it. For an AI system working on client matters, the entries worth sealing are:
- What came in. Which documents were added to a matter, when, and by whom.
- What was asked. The request that ran, the matter it ran against, and which set of documents it could see.
- What the system produced, with the citations attached to it, exactly as proposed.
- What a person decided. Accepted, edited and accepted, or rejected with a reason, under a named user and a timestamp. This is the entry that matters most, and it's the one ordinary logging never captures.
- What left. Exports, prints, and anything handed to a client or filed.
- Which version of the system did it, so a change in behavior after an upgrade is attributable rather than mysterious.
This is also why the approval queue and the audit trail are two halves of one design. The queue is where a human decision happens. The trail is where that decision is kept.
Small firms need this more, not less
Governance sounds like a large-firm word, and the instinct at a six-attorney practice is that it's overhead for someone else. The reverse is closer to true. A large firm has a general counsel and a records group, so some version of this exists whether or not anyone calls it governance. A small firm has none of that, which means the record either gets created by the software or it doesn't exist at all. The practical test is whether answering "how was this handled" takes an afternoon of searching email or five minutes of filtering a list.
Questions to put to any vendor, including us
Five, and they're all answerable in a sentence. Can we export the full trail for a matter in a format we can read without your software? Is it append-only, and can you show us the integrity check? Does it record human decisions or only system events? How long is it kept, and who decides that? Can an administrator at your company edit or delete it, and would we know?
That last one is where architecture decides the answer. When the trail lives on someone else's infrastructure, its retention is a policy you're relying on, and policies change at renewal. The Tiber River Legal Workbench runs on hardware the firm owns, so the record sits in the same building as the documents it describes, under the firm's own backup and retention schedule, and it doesn't end when a subscription does. That's the same reasoning behind keeping the files themselves in the building, applied to the paperwork about the files.
Boring until it isn't
An audit trail does nothing for you on a normal Tuesday. It's a cost with no visible return until one specific day, which is exactly what makes it easy to skip and expensive to have skipped. Build it in from the start, and the day it's needed is an administrative afternoon instead of a crisis.
Tell us what your record has to answer
The Tiber River Legal Workbench is a matter intelligence appliance for litigation and personal injury practices, running on hardware your firm owns. The audit trail should answer the questions your clients and carriers actually ask. We're inviting a small number of Maryland firms to shape it as design partners.
Start the conversationOr write to josh@tiberriverconsulting.com