Overview
This is written for the attorney on either side of an internet defamation case, and it starts from one claim: a case of this kind divides by what each element turns on, and that division decides whether retaining a technical expert is worth anything at all.
I am not an attorney, and the legal judgments in a matter are not mine to make. What I do is sort the record, and the sorting is the useful part, because the six things that have to be established in a case like this behave very differently once you ask what evidence each of them rests on.
Sort them by platform and nothing separates. Sort them by what settles them and the picture is stark. Four are supported by records and closed by something else. Two are not technical questions in any sense. None is settled by primary records standing alone, and an expert who tells you otherwise about any of the six is describing a capability that does not exist.
Three states, and only two of them are mine
Every element page on this site carries one of three verdicts.
- The record establishes it. The primary records dispose of the point on their own, and a second examiner handed the same material arrives where the first one did.
- Record plus testimony. The records take the point most of the way and then stop. Discovery, a custodian, or a witness with direct knowledge covers whatever is left.
- Not a technical question. The answer belongs to the law or to the jury. Nobody settles it by analysis, and an expert who volunteers to has drafted the other side's cross.
Read across the six element pages and the first state does not appear once. It exists because the methods half earns it repeatedly. The elements half does not, and the reason is structural: records are made by machines for operational purposes, and every element ends in a human judgment no machine was recording.
The four the record supports and cannot close
Four elements sit in the middle state, each for a different reason.
Publication and reach. One person other than the subject receiving the communication satisfies this element, and a technical record almost never identifies a person at all. Access logs hold requests. Platforms hold delivery events, plus counters whose definitions belong to them. Everything an expert produces about audience therefore goes to the scope of the injury, not to whether publication occurred.
Republication and amplification. A version diff is a fact: this text was at this address on this date, that text was there on that one. Whether a change amounts to republication is a conclusion drawn on top of the fact, and the two get run together constantly. The record dates copies. It does not enumerate them, and no method does.
Attribution to an author. The chain from content to a named human runs through five links and only four are records. Account records, address assignments, timing correlation and device artifacts narrow the field to an account and a connection. The last step, that a particular person was at that keyboard, has never been supplied by a record.
Falsity and the contemporaneous record. Captures, archives, revision histories and time anchors that are neither my clock nor the publisher's can fix what a page said and when, sometimes to the minute. Whether the statement was true is a question about the world, answered by registries, filings and witnesses rather than by anything in my discipline.
The two that are not technical questions
Identification of the plaintiff and reputational damages both look, from a distance, like measurement problems. Neither is one.
Identification asks whether the content refers to the plaintiff. Much about that is observable: whether a name sits in the page markup and not only in the rendered text, what anchor text third parties used pointing at the page, what commenters said underneath it, what a search result looked like on a date. All of it is evidence a fact-finder can weigh. None of it establishes what a reasonable reader understood the content to mean, and that understanding is the element. The gap between the two is the gap between an exhibit and an overreach.
Damages are the harder case, because the market supplies a number on request. Reputation is not a measurable quantity. Traffic, conversion, rating movement and branded query volume are proxies, and every method converting a proxy into a figure runs through an instrument, a counterfactual, a control set and an extrapolation, of which only the instrument is observed. The most useful thing an expert can produce in that situation, and the thing least often produced, is a reasoned refusal to put a number on it.
What will not work
Every page here names something that fails, because failures are more useful than capabilities. Five recur.
- An audience figure offered as proof of publication. It answers a question the element never asked, and starts a fight over an estimate nobody needed.
- An address offered as an identification. An IP address identifies a connection. Shared office networks, carrier-grade address translation, VPN exit nodes and household routers sit between the address and a person.
- A screenshot offered as the page. It records what one browser displayed to one examiner, not what a server served, and the difference is where a competent opponent works.
- A stylometric attribution standing on its own. There is no published error rate for authorship attribution on short platform texts, and none at all for text drafted with a model's help.
- A vendor traffic estimate standing in for exposure on a low-traffic page. The one large measured comparison found anywhere in this research put a widely used estimator roughly 94% above the sessions the sites themselves tracked, and least accurate on the smallest properties, which describes most pages in this subject.
The order the questions come in, and the clock underneath them
Counsel weighing an element needs four answers, and they arrive in a fixed order.
- What does the element turn on? A record, a witness, or a judgment. This decides everything after it.
- Does a record of that kind exist, and who holds it? Almost never the plaintiff. Usually a platform, a host, an access provider, or the opposing party.
- What does legal process to that holder return? A separate question from whether the record exists, and the one most often assumed rather than asked.
- Is an expert worth retaining for this element? On two of the six, the honest answer is frequently no.
Underneath all four sits a calendar indifferent to the answers. Server logs rotate in days or weeks and no rule preserves them. Platform preservation, where an operator publishes a window at all, runs 90 days. The gap between what could have been proved and what can be proved is created in the first sixty days, by a preservation demand that either named specific record types or did not, and nothing in the second year of a case recovers it.
These pages describe what each element rests on and where it stops. If one of them leads you to conclude that an expert has nothing to add to your matter, it has done its job.
Published