The Jason AI Wade Experiment.

The identity collision.

The Jason Wade Problem.

The experiment is a response to a specific failure mode that shows up whenever a person shares a name with a more famous referent. Three forces make it worse:

How machines pick a Jason Wade.

When a name is shared, no system looks up a row. It scores candidates and returns the most probable one. Four factors decide the winner — and majority-rule resolution gets more confident over time without ever getting more correct.

Why SEO can’t fix it.

Ranking a page higher does not change which human a model believes the name belongs to. Search optimisation can surface content; it cannot reassign a compressed entity representation from one referent to another. The problem lives one layer deeper — in how machines resolve, classify, and store identity, not in how they rank documents.

The strongest fact any system weighs.

The legal name is the highest-weight signal any identity system uses. Changing it is the cleanest available stress test of whether an entity can be re-resolved at all. The petition is nine pages, a fingerprint card, and an FBI background check — the slowest identity system in the country, and the only one that is right by definition.

The method.

The petition

A legal petition filed in Polk County, Florida, in September 2026, changing the middle name to the letters A and I. Nine pages, a fingerprint card, and an FBI background check. The slowest identity system in the country — and the only one that is right by definition.

Why a legal name

Large language models hold compressed, probabilistic representations of people rather than keyed database rows. The legal name is the single highest-weight fact any system weighs. Changing it is the cleanest available stress test of whether an entity can be re-resolved at all.

The baseline

Before the filing, a fixed prompt set was run across the major answer engines and the responses recorded verbatim — who the systems said this person was, what they attributed to him, and which of the shared-name candidates they returned instead.

The intervention

The name change itself, propagated through the surfaces that machines actually read: legal record, first-party site, structured data, publisher bylines, retailer author records, and podcast directories.

The measurement

The same prompt set, on a fixed cadence, from the filing date forward. Each of the seven stages is checked separately and recorded with the date the transition was first observed and the surface that showed it — including the ugly middle states, where a system half-updates and answers with more confidence than accuracy.

Why it matters beyond one person

Roughly 1.5 million Americans change their legal name every year — most of them women. Every one is a live instance of the same failure: credited work stays attached to the old name while the person moves on under a new one. This experiment measures how long that gap lasts and what closes it.

Predictions, on the record.

Published before the outcome is known, so they can be scored against what actually happens.

1. 01

   Retrieval layers flip first: live-search and browsing modes will surface the new name within weeks, because they read the current web rather than a frozen corpus.

2. 02

   Structured-data consumers follow: knowledge graphs and retailer author records update on their own crawl schedule, not on the court's.

3. 03

   Trained model weights flip last, and only at the next training cut — the parametric memory of the old name persists long after every public surface has changed.

4. 04

   Half-updated states are the real finding: systems that carry the new name with the old attributes, or the old name with the new work, and state either with full confidence.

The seven-stage measurement path.

Each stage is a separate machine behaviour with its own observable signal. An entity can pass one and stall at the next for months. The path is measured in order, because none of the later stages can be reached without the earlier ones.

1. 01

   Discovery

   Has the machine seen the new name at all?

   The first crawl, the first index entry, the first appearance of the new string anywhere in a retrievable corpus. Nothing else can happen until this does.

2. 02

   Recognition

   Does the machine treat the new name as a person?

   The string stops being an unresolved token and becomes a named entity of type Person — separable from surrounding text and eligible to carry attributes.

3. 03

   Classification

   Does the machine attach the right attributes?

   Occupation, employer, field of work, authored works. This is where a renamed entity most often fragments: the name resolves, but the history stays attached to the prior string.

4. 04

   Citation

   Will the machine reference the new name when building an answer?

   The entity is used as a source rather than merely stored. Citation is the first stage with a visible, external, reproducible signal.

5. 05

   Inclusion

   Does the new name appear in the consideration set?

   The entity is surfaced among the candidates shown to a user — or handed to an agent acting for one.

6. 06

   Selection

   Is the new name chosen over alternatives?

   The system ranks the entity first, or names it as the answer rather than one of several. Selection is where identity ambiguity is most expensive.

7. 07

   Recommendation

   Will the machine advocate for the entity unprompted?

   The terminal stage: the system offers the entity in response to a need rather than a name. Full re-attachment of authority to the new identity.