The self-narration seal — Constello never asks people about themselves
Sealed August 11, 2026, at the end of the Overtone stress-test conversation. Ethan: “okay, I feel like constello is strictly better enough to me to devote to now — I was also considering the possibility of integrating something like overtone’s interview process in constello as well, but that’s been all but closed by this conversation for me.”
The seal, worded precisely: Constello never solicits self-narration as signal. No input channel, ever, where a person is asked about themselves and the answer feeds a reading. The person is asked about their things; the reading infers the person.1 Conversation itself is everywhere in Constello — the coax, collection chats, amendments — what is sealed is only the self-report channel. The operational form is the object-questions law, binding hardest on the mint’s interviewer: never “what kind of person are you,” “what matters to you,” “tell me your story” — only “what do you keep,” “show me,” “which one,” “since when,” “what happened when you almost lost it.”
This is a refusal of the permanent kind, not the deferred kind — rare in Constello, where refused moats are normally deferred, not renounced. The reasoning is recorded here because someone will eventually propose reopening it, probably in the form “just add a quick interview.”
Why the interview channel is structurally worse
The unreliable narrator. Decades of self-report research: people are not lying about their preferences, the reasons just aren’t introspectively available. Ask someone what they value and you get the press release. An interview is a higher-resolution instrument for reading the press release.
Ethan’s steelman, and what survives it. He pushed back that a good interviewer never asks about preferences at all: “you just talk to them, and here the stories that matter to them, the way they conduct themselves, you can get everything you need just from interaction.” True — people leak constantly, and skilled conversation harvests the leak, not the testimony. What survives the steelman is the frame: every word of a matchmaking interview is produced inside a context the person knows is evaluative — the highest-stakes performance frame a human can occupy. Conduct dresses up too. The residue of the distinction, once the steelman burns off the rest: not “talking vs. things,” but material produced inside the evaluative frame vs. material that pre-exists it. The voicemail kept through three phones wasn’t kept at anyone. Constello’s own coax conversation stays on the right side of this line only because it points at the pre-existing stuff — the conversation is a flashlight, not the deposit.
The articulacy tax. For identity questions, ease and depth sit on opposite ends of one axis. Palatable prompts (“what’s your perfect Sunday?”) harvest the culture’s stock answers; deep prompts price out everyone but the introspective. Ethan: “the questions would have to be extremely open-ended to accommodate the breadth of depth in response, and even then it only rewards people with the verbal readiness and self-conception to realize themselves.” And open-endedness doesn’t distribute the burden, it transfers it — the blank page is the maximum rate of the tax, since supplying structure to a blank space is exactly the unequally-distributed skill. So an interview instrument’s signal quality is a function of the user’s articulacy: every match it produces is a match on narration ability confounded with everything it tried to measure. Ethan, on Overtone as a user rather than a founder: “no I think people will have quite a hard time writing themselves properly, and so Im actually quite excited. When the only medium of expression is written, the moat I have to others is intense. Ill likely meet people I like very much on there for this very reason.” An interview product serves the top decile of self-narrators — the population that was already served. Constello exists for the other ninety percent: people whose worlds are real, dense, carried for years, and unnarratable by them. They don’t have to write themselves; they point at what they kept, and the reading does the articulation on their behalf.
Object questions don’t live on the ease/depth axis at all. “What have you never deleted?” is both easy and deep, simultaneously, with no tension: easy because it’s pure recall — no composition, a person who can’t write two good sentences answers it perfectly; deep because the difficulty was already paid, years ago, wordlessly, every time the thing was kept while everything around it got purged. The depth is stored in the object, not produced by the answer. An interview’s questions have to generate significance in the moment from someone who may not be able to; object questions only have to locate significance already sitting there.
Why sealing costs nothing
Take the interview’s genuinely best material — the stories that matter to someone. A story that matters is a story the person keeps re-telling, which makes it a carried thing, reachable by object question: “what story do you keep telling?” This generalizes: anything an identity-interview could surface that is actually real — the story, the grudge, the joke they always make — is real precisely because it recurs, and recurrence makes it a kept thing, already Constello’s jurisdiction. Everything true that walks through the interview door also walks through the object door. What has no object-question equivalent is exactly the material the seal exists to exclude: the composed self-summary, the aspiration-as-press-release, the answer that exists only because the question was asked.
The priced loss
The trace lags the person. Someone at the very start of becoming something new — nothing kept yet, only intention — is audible to an interview on day one and invisible to Constello until the keeping begins. Accepted without flinching, for two reasons: aspiration-talk is the single most performed register humans have, so the channel that hears it earliest is also the channel most flooded with press releases; and the score ladder’s top band already handles becoming correctly — “what it is trying to become” read from the direction of the trace, not from testimony. Constello sees becoming slightly late and almost never wrong; the interview sees it instantly and constantly wrong.
The boundary in one line
Overtone asks people to narrate themselves; Constello asks people to point at things. The mechanism stays superior only while this holds — the moment any Constello surface asks a person to narrate themselves, it has rebuilt the interview with the same flaw. The line between “what do you keep” and “what matters to you” is the entire boundary, which is why it is law in the mint harness, not a stylistic preference.