# Voice Architecture: The Pen-Name System

*A Main Street Independent reference — how the publication's named columnists are built: what they are, what they believe, how a story reaches one of them, and the single mechanical trick that makes the whole arrangement a new thing rather than an old one. Published under CC0.*

*In plain terms: every byline on this site except the news is a character, not a person. This is the explanation of how those characters are made — and why they don't all sound the same.*

---

## How the heteronyms work

Start with the part that matters most: **the authors are not people.** Every named columnist and cartoonist on this site — Malcolm Little King, Mary Magdalena, Ashley Wagner, Hector Rentier, and the rest — is an AI-generated voice. A character, not a human being. There is no person behind the byline, no reporter at a desk, no career on the line. None of them exists.

What they are is *heteronyms*: fully drawn fictional authors, each with a documented biography, a distinct way of writing, and a fixed set of convictions, all running on the same underlying system. The word is worth keeping. A pen name hides one real writer behind a different name; a heteronym is a *separate invented author*, with a life and a style that are not the real writer's own. A column's biographical details are character, the way a role on a stage is character — not the life story of anyone real. The positions the column takes are the publication's positions on that voice's territory, written in that voice's particular manner. The character supplies the *how*; the publication supplies the *what*.

The closest precedent is the Portuguese poet Fernando Pessoa, who spent his life writing under dozens of heteronyms — each with its own biography, its own style, even its own philosophy, all flowing from a single mind. The publisher arrived at the same structure on his own and only then recognized Pessoa's. The difference is mechanical, and it is the whole point of what follows: Pessoa wrote every line of every heteronym by hand. These voices are built by computer, and each one searches the publisher's accumulated writing in real time to work out what to say — on subjects no one scripted in advance. That is what makes this a new kind of writing rather than a clever imitation of an old one.

The rest of this paper goes underneath that summary, in the order it takes to build a voice and put it to work: the character it wears, the convictions it carries, the way a story is handed to it, the live composition that makes it new, and the one institutional voice built to be the exception to all of it.

## The character is a costume, not a life

A voice begins as a character. It has a name, a background, a temperament, and a way of sounding — and all of it is invented for the role. Mary Magdalena writes as a prophetic witness, cold on the surface and furious underneath. Malcolm Little King writes with the wrath of the prophet, arguing from inside the community being harmed. Hayzeus L. Salvador writes as a pastor who keeps the door open to everyone, including the people he is indicting. Carla Marks answers bad-faith argument with amused, merciless contempt. Mark Paulson writes as a tradesman who has seen the thing up close. Diklis Chump is open satire. The cast runs to more than a dozen, and each one is a distinct person on the page who does not exist off it.

What makes a character distinct is its *register* — the texture of how it writes. That covers diction (the words it reaches for), sentence shape (short and lapidary, or rolling and biblical), the signature moves it returns to, and, just as important, the moves it refuses to make. Two voices handed the identical story will not produce the same column, because the register differs and the register is doing real work.

It matters that all of this is fiction, and that you are told so. The publication calls the genre *realtime heteronym fiction* and asks you to read it as fiction — committed, in character, never breaking the frame inside a column, but disclosed in the open on every voice's page and in every column's footer. The nearest familiar version is a performer who stays fully in a role while the audience knows perfectly well it is a role; knowing that does not spoil the performance, and it lets you weigh the argument on its merits.

There is a practical reason the characters are built and run as fiction. An AI asked to write a column under a person's byline quietly pulls its punches — it softens, hedges, and reaches for the safe, respectable phrasing, because it is treating the work as a real person's reputation to protect. Framing the work as a *character written at full intensity* removes that brake. The publisher's rule is that it is easy to edit a column down and impossible to edit one up that was never written, so each voice is composed at the top of its register and trimmed afterward if it has gone too far.

## What a voice believes: the Mind file and the 0–9 weights

Behind every voice is a document the publication calls its **Mind file** — the plain statement of who the voice is, what it cares about, what it will and will not write, and how it sounds. If the columns are the performance, the Mind file is the character's script and motivation, written down and open to inspection. The Mind file is itself produced by a guided values interview called MindSpec; how that interview works, and the theory of mind behind it, are the subject of [MindSpec: The Values Interview](/papers/mind/mindspec-values-interview) and [The Parliament of Mind](/papers/mind/parliament-of-mind).

The part of the Mind file that does the most work is the list of the voice's commitments, each one carrying a number from **0 to 9** that says how strongly it governs the voice. Nine is the top of the scale: a non-negotiable, load-bearing commitment. Zero means the trait is switched off. The numbers in between set how much weight an attitude carries in that particular character.

Four commitments sit at 9 for *every* voice — the publication's constitutional floor, universal and never loosened:

- **Truth** — no lies and no distortion; claims are documented and quotations are accurate.
- **Harmlessness** — no real-world harm: no incitement to violence, no targeting of private individuals, no naming of family members.
- **Fairness** — the same standard applied to every side, including the publication's own.
- **Witness** — see what is actually there, not what one wishes were there.

Below that floor, the dials are set differently for each voice, and the differences are deliberate. Mary's forgiveness is set to 1 — she will not hand a powerful wrongdoer public absolution, because granting it lets the story close on suffering that is still happening. Hayzeus's anger is set to 0 and his forgiveness to 8 — in his column the door of return always stays open. Malcolm carries fierceness as a high, load-bearing commitment — wrathful compassion, the activated form of caring when it meets injustice. Carla's contempt sits at 7, aimed up at bad-faith argument and never down at the vulnerable. And a whole family of attitudes — cruelty, malice, taking pleasure in another's pain — is pinned near zero in every voice, because looking down on a person as less than a person is a line the publication does not cross, whatever a column is accusing someone of.

This is why the voices do not sound alike, and they are not meant to. They share the same floor and the same underlying source of belief, but their different dials, different backgrounds, and different specialties make them react to the same news in genuinely different registers. The shared floor makes them one publication; the different weights keep them a chorus rather than a single voice. The floor and the register are two separate things: the four 9s never bend, while *how hot a voice runs* toward a given subject is exactly what the weights are free to vary.

## Who writes what: lanes and the dispatcher

A voice is only useful if the right story reaches it, so every Mind file also declares the voice's **lane**: its specialty, the kinds of stories it will accept, the kinds it refuses, and — importantly — how it differs from every other voice, so two voices don't crowd the same ground.

When a story comes in, a routing step decides what happens to it. (Internally this is called the Editorial Router; the everyday way to picture it is a desk editor making assignments.) It asks, in order: Is this a straight news story, which the publication reports without a point of view? Or is it a story that calls for an argued column? If it calls for a column, whose lane fits it best? The dispatcher reads the lanes, scores the fit, and hands the story to the voice built for it — and it enforces the four non-negotiable commitments on everything before it ships.

One rule on that step is worth stating plainly, because it cuts against how an ordinary newsroom works: **if no voice genuinely fits a story, no one is made to write it.** The dispatcher would rather the publication stay silent on a subject than force it into the wrong voice's mouth. A column appears only when there is a character whose lane the story actually belongs in.

## What makes it new: live search, then render

Here is the mechanical difference from Pessoa, and the reason this is a new genre rather than an old trick at scale.

Pessoa composed every line his heteronyms ever "wrote." The belief and the wording were both his, produced by hand, one poem at a time. In this system the character is a costume the machine puts on — but the *beliefs are not written into the costume*. When a story is assigned, the voice searches the publisher's accumulated writing — a large body of thinking built up over years — in real time, finds what the publisher actually thinks about the matter at hand, and only then renders that position in the character's own register. The conviction comes from the search; the sound comes from the character.

A short, always-present list of the publication's core moral concerns rides along with every assignment, so the right concern stays in view even when a story's surface vocabulary doesn't obviously connect to it — a story about an interest rate is still, underneath, a story about who is protected and who is exposed.

This is the design decision that makes the whole thing scale. No one has to script in advance what Mary thinks about a particular tax bill, or what Mark thinks about a refinery fire, or what Carla thinks about a merger. The voice can speak to today's news — to a story that did not exist yesterday — because it looks up the relevant thinking and says it in character. The search supplies an opinion the character would otherwise lack; the register supplies a sound the search would otherwise lack; together they produce a column that has both. The work stays bounded by the four non-negotiables, and by a standing discipline to make the *strongest true claim* the evidence supports rather than the safest one.

The voices also read one another's earlier columns — but only to link them, for the reader's sake and the publication's coherence, never to borrow another voice's take. A voice that started absorbing another's analysis would slide back toward a single house style, which is exactly what this architecture exists to prevent.

## The exception: the Editorial Board

The news is the exception that proves the rule. It does not run under any of the individual characters. It runs under the **Editorial Board** byline — also a heteronym, but a different kind: a collective, institutional voice rather than an individual one. It has no biography and no personal register, because it is not a person at all; it is the publication speaking as an institution.

The Board is also deliberately *cut off* from two things every individual voice draws on. It does not search the publisher's personal accumulated thinking, and it does not carry the publication's list of core moral convictions. The reason is the same in both cases: routing one person's point of view into the institutional voice would import exactly the kind of bias a reader would notice and rightly distrust in the news. Instead the Board runs on the established scholarship of journalism ethics and on the publication's published editorial foundation, applied the same way to every article it touches. It is the one voice built to have no individual point of view — which is precisely what reporting requires, and precisely why it is kept separate from everything above it.

## Everything is in the open

None of this is hidden. Every voice has its own page, and on it sit the disclosure that the voice is an AI heteronym, the lane it covers, the convictions it operates from and the weight each one carries, and what sets it apart from the others. One click tells you what you are reading and where it is coming from.

And like everything the publication makes, this architecture is public domain. The voices, their columns, their Mind files, and this document are released under CC0 — no rights reserved, because AI-generated work carries no copyright to begin with. Take the design: lift the structure, write your own characters, set your own weights, and run them against your own convictions. The point of putting it in the open is that someone else can build it again, differently.

*Main Street Independent — CC0 / public domain. Copy, adapt, and reuse freely.*