The Centurion Project Exposed

A new kind of political machinery has been quietly built

Behind a referendum push framed as Albertans simply talking to their families about independence sits a sophisticated piece of political infrastructure. The Centurion Project, launched by longtime Alberta organizer David Parker, is the local deployment of a Michigan-built platform called 10X Votes, owned by a private company called Voteatron. Stripped of branding, what it amounts to is a relational organizing platform, a system that converts a movement's supporters into the unpaid labor force that compiles a high-fidelity political-affinity graph of an entire voting population.

The pitch is disarming. You sign up, you find ten people you already know in a database of three million voting-age Albertans, you "claim" them, and you take responsibility for whether they vote in the October 2026 referendum. Parker positions this as a return to classical citizenship, the Greek and Roman idea of the citizen as someone who shows up, takes responsibility, holds others accountable. The Centurion Project's own materials describe their mission as recruiting, equipping, and mobilizing community leaders across Alberta to win sovereignty.

That description is accurate, as far as it goes. What it omits is that the architecture underneath the friendly civic language has properties that do not dissolve when the referendum ends, that are owned by a private vendor in another country, and that, in their durable form, are something quite different from grassroots organizing as it has historically been understood.

Four layers, stacked, with private ownership at the base

The system can be reduced to four functional layers. Each one looks innocuous in isolation. The interesting properties emerge from their composition.

At the bottom sits the input layer: scraped public data assembled into a single profile database. Voter rolls where available, Canada 411-style directories, postal-code-to-riding mappings, registry data. Parker says three million records for Alberta; the parent product advertises 1.7 million, two million, and 1.2 million "low propensity voters" in Michigan, Pennsylvania, and Wisconsin respectively. Names, addresses, electoral divisions. No phone numbers. He calls it a digital phone book without phone numbers.

On top sits the compilation layer. Every supporter is given a search interface into the database and invited to claim people they personally know. A claim is exclusive (once your aunt is claimed, no one else can claim her), which forces the volunteer base to specialize along its real-world social-network boundaries rather than overlapping. On top of each claim sits a three-question survey: do they support independence, did they sign the petition, will they vote. Supporter, undecided, opponent. Parker estimates that what would cost twelve dollars per identification by phone now costs roughly nothing per identification by claim. The labor is donated; the trust channel is pre-existing.

Above that sits the inference layer. Parker's stated target (18,900 users contributing roughly ten surveys each) produces about 189,000 ground-truth labels. He says explicitly that with that volume of labeled data, "you can model out the entire Alberta population." That is not a metaphor. With that many labels cross-joined to the demographic and geographic features already in the database, a classifier can score the remaining 2.8 million unsurveyed Albertans with usable accuracy.

The top layer is activation. Near the referendum, a fourth state appears in the system (has-voted), and the same network of claimers becomes a get-out-the-vote enforcement mesh. Reminders flow through pre-existing relationships rather than through cold calls.

Asymmetric warfare wearing the clothes of civic virtue

Several mechanisms operate inside this architecture at once, and each is worth naming.

Cost-asymmetric warfare. Door-knocking yields about two identifications per hour. Phone-banking costs roughly twelve dollars per identification and is degrading, because almost no one answers an unknown number anymore. Claiming yields ten identifications per fifteen minutes at zero marginal cost. A movement that adopts this model competes against opponents using legacy methods at roughly a forty-to-one cost ratio. Parker is unusually frank about this; he calls the other approaches obsolete.

Trust-channel substitution. The conventional canvasser is a stranger asking a personal question, and immunity to that channel is now near-total. Substituting "your sister-in-law texts you" for "an unknown number calls you" exploits the one remaining un-saturated communication path in modern political life. The medium is not new. The systematization of it is.

Compartmentalization as both privacy theater and security architecture. The claim that "even admins cannot see who you claimed" is a meaningful design choice. It raises the cost of subpoena, breach, or insider leakage of the relationship graph. But it is a design choice, not a cryptographic guarantee. The database is centrally hosted, the developers can see it, and the politically sensitive part (the survey results) is visible to admins. The privacy property is partial. More importantly, it functions to make participation feel low-risk to the volunteer, which is the actual operational requirement.

Decentralization as anti-fragility. Parker's stated goal of "a hundred leaders with a thousand each" rather than one leader with a hundred thousand is taken almost directly from how movement infrastructure has operated on the political left since the 1960s. It hardens the movement against decapitation strikes (investigations, deplatforming, defamation suits, leadership burnout) by ensuring that no single removable node carries the load.

Civic identity reframing. The Greco-Roman citizenship rhetoric is doing real work. It converts what would otherwise feel like uncompensated political-data labor into a moral duty owed by free people to their polity. Reframing the activity as virtue is what moves it from "annoying chore" to "thing my whole social circle is doing."

Six things artificial intelligence does with the claim graph

The claim graph is the most interesting asset because it is the part the system uniquely creates. The list below moves from the unobjectionable to the structurally consequential.

Door-knocking is a flow. The trained model is a stock

This is the part that matters most, and the part the activist conversation around the project does not address.

When a door-knocker spends a Tuesday talking to two hundred households, the output is a clipboard with checkmarks, a few names in their head, and a vague impression of which streets felt warm. That output decays on three timescales at once. It decays physically: the clipboard gets shredded, the volunteer moves on, the memories fade. It decays contextually: the question it answered is bound to a specific moment that does not repeat. It decays organizationally: institutional knowledge about how to use it evaporates. A door-knocking operation is a flow. It produces value while it is happening, and what it leaves behind is mostly residue.

A trained classifier is a different kind of object. Once you fit a model to (demographic features, geographic features, registry features) → (supporter, undecided, opponent), you have frozen the relationship between inputs and political identity into a small artifact: a few megabytes of weights and parameters. That artifact is portable: copy it, license it, sell it. It is re-runnable: feed it any new record with the right input fields, indefinitely, at near-zero marginal cost. And, as the legal point Parker himself surfaces when describing the destroyed petition signatures, it is durable. Authorities can compel destruction of the petition. They cannot easily compel destruction of derivatives. A model trained on the data is a derivative.

The features the model relies on are stable in a way the labels are not. Postal code, age band, residence type, registry signals. These exist for every Albertan now and for every Albertan who turns eighteen, moves into the province, or buys a house in the next decade. The 189,000 labeled records are the one-time investment. The scoring capability they unlock is a continuing yield. The system does not just classify the three million people whose records seeded it; it classifies anyone who shows up with the same features.

It is also re-fittable. The same labeled data, the same features, the same volunteer infrastructure can be turned to adjacent questions with surprisingly little additional work. Once a population is scored on Alberta sovereignty support, the model is a usable proxy for low trust in federal institutions, willingness to attend a rally, donation propensity for related causes. It is not sovereignty-specific in any deep sense. It is a classifier of a particular political identity that happens to express itself as sovereignty support in this referendum and will express itself as something else in the next one. Each future political question that touches similar identity contours can fine-tune from this base rather than starting cold.

Privatization in the precise sense, not the obvious one

The shape of what is happening here is best described as the platformization of political organizing infrastructure, with privatization as the structural consequence. Three things are happening simultaneously.

The relational organizing layer (the act of talking to your friends and family about politics) is being moved onto a digital platform that captures, structures, and stores that activity. The platform is privately owned and not accountable to the activists who feed it. The data assets accumulated have transferable, durable value beyond any single campaign, value that accrues to the platform owner, not to the volunteers.

This is privatization in the same sense that the gig economy privatized the dispatch layer of taxis, the booking layer of room rentals, the discovery layer of the open web. Voteatron-style platforms do not invent political organizing. They privatize the coordination infrastructure that organizing now runs on.

The thing this displaces is older than the consultancy model and is worth naming. Political organizing infrastructure used to live inside membership institutions: political parties with riding associations and AGMs, trade unions with elected leadership, churches with congregational governance, civic associations with members who could show up and vote on direction. Those institutions had real flaws and were never as democratic as their charters claimed. But a person who contributed labor and money to them got at least nominal voting rights at the next leadership review. A person who contributes labor and survey data to the Centurion Project gets nothing that looks like that. The platform operator has no governance relationship with them. There is no AGM. There is no member-controlled list. The decentralization Parker emphasizes is decentralization at the human-leader layer, where it does real work hardening the movement against decapitation. At the infrastructure layer underneath, the system is unusually centralized: one platform, one database, one model, one corporate owner.

The mismatch between consent given and infrastructure built

None of this is a claim of bad intent. Parker has been candid about the architecture. The Centurion Project is publicly explained. 10X Votes openly markets itself. There is no secret. The interesting question is not whether something hidden is being done; it is whether the people contributing to it understand what kind of thing they are contributing to.

The volunteers' implicit consent is bounded in time and purpose in a way the asset is not. A person who claims their family in May 2026 is consenting, in their own mind, to help a sovereignty referendum in October 2026. They are not consenting to be a permanent training-data contributor for any future political-identity classifier built on top of that infrastructure, because they do not know that is what they are contributing to. The temporal mismatch between the consent ("help me win this vote") and the artifact ("a durable, transferable, re-fittable model of Albertan political identity") is the part that should make a careful observer pay close attention.

The referendum is the visible event. The model is the thing that outlives it. It outlives Parker's stated intent to step away. It outlives the Centurion Project as an organization if the Centurion Project dissolves. It is owned, at the substrate level, by a private company in another country with which the volunteer base has no relationship at all.

That is the asymmetry. None of the individual pieces are scandalous. The composition is what deserves attention. And the lesson generalizes well beyond Alberta, because this is the shape political organizing now takes, and it is going to keep taking it, and the question the next decade will turn on is whether the form of accountability that membership institutions used to provide can be reconstructed inside platforms that have no equivalent of members.