Method
How this archive is made
Every claim here carries a source grade, and the grades are the point. A cell a maker confirmed and a cell the compiler guessed at should never look alike.
Source grades
- M · Maker-confirmed
The maker stated it, in interview or in writing.
- P · Published source
A maker’s listing, a venue page, a review or scholarship says it.
- F · First-hand
The compiler experienced the work.
- I · Inferred
The compiler’s judgment from indirect evidence. No direct statement exists.
- ? · Not yet established
Nobody has answered this. It is an open question, not a no.
The rule that governs the coding
Silence in a description is not evidence of absence. Marketing copy mentions props in about three per cent of these listings. Coding everything unmentioned as no would have manufactured several hundred confident negatives out of nothing. Where the evidence does not positively establish presence or absence, the cell is unknown.
A table that is a fifth unknown and honest is worth more than one that is complete and quietly invented. The unknown cells are also the agenda: they are the questions to put to the people who made and answered these calls.
Corrections
Every practitioner interviewed was told at the outset that the finished work would come back to them to be checked for error. That undertaking stands, and this page is where corrections are recorded rather than quietly absorbed.
AJ Danna, October 2026. A summary of the conversation stated that he had wanted an option to detonate the bomb and destroy the town. That was wrong, and wrong in a way that matters: the desire was his audience’s, expressed after playing, and he had deliberately excluded mass death from the design — partly because visiting the National Atomic Testing Museum humanized the era for him, partly because his uncle had worked at Mercury. The finding survived and reversed direction: participants asked for the atrocity, the maker declined. The error stays on the record here rather than being removed.
Found something wrong? Experienced a work listed here? Corrections and first-hand accounts are the mechanism by which this archive improves.
Findings, for people who make this work
Eight things that emerged from thirteen interviews and twenty-six calls. They are craft, not theory — several were arrived at independently by practitioners who have never spoken to each other.
1. The six-second constant governs unmarked silence only
Trapman-O’Brien and Markham independently named roughly six seconds as the longest silence a phone listener will tolerate. Nicky Perfect, who has held people alive on the line, breaks that ceiling deliberately: she announces the silence, tells the person she is still there, and counts to ten. Nothing ruptures. The limit is not on silence but on silence the listener cannot account for. Declaring the pause does the work the face would have done.
2. Connection before direction, arrived at three times independently
Perfect calls it connection before direction. McInnes, running an improvised comedy, says the lost object is a MacGuffin and the real work of the first call is finding what the person wants from the week. Sela designs only for ease and kindness and will not play resistance. Crisis negotiation, improvised comedy and phone theater are not in conversation with each other. They converge on the same ordering rule.
3. Consent is an event for the designer and a state for the participant
Markham must offer three to five refusals before anyone takes one, because compliance is the default. Allen Otto reads the same fact from the other side: repeated permission can function as cover, accumulating a record of agreement that licenses the thing it was meant to guard. Both are true, and which one obtains depends entirely on whether the participant retains the asking. The phone preserves only the event. Candle House shows what the event looks like when it is built properly — a waiver, a rehearsed safe phrase, a spoken acknowledgment at the head of the call, and named exits for medical and psychological emergency, all established before the fiction starts. Finding 08 sets that out in full. Neiden also supplies the principle that keeps this from collapsing into caution: Candle House makes “work that keeps you safe but doesn’t keep you comfortable.” Safety and comfort are separable, and conflating them produces either a piece with nothing at stake or a piece that hurts people.
4. Refusal has a cost, and the cost can be designed down
The Telelibrary asks whether you consent to be recorded and scripts the refusal as a line that is pleasurable to say. Trapman-O’Brien also built the system persona precisely because listeners felt too guilty to stop him. Making the interlocutor a machine lowers the social price of no. This is a design solution to Markham’s finding, and it implies something uncomfortable: a warm human practitioner extracts more compliance than a system does.
5. The forward pass is not new. The non-consenting subject is
Telelibrary callers describe what is outside their window and later callers hear it; its Compliment Mailbox lets users leave messages for people who have not called yet. The Program now resolves not to a live operator but to recordings of its own earlier participants. Ant Hampton’s Showing Without Going invites users to contribute connections that later users encounter. In all three, the author consents and the recipient is merely surprised. In an earlier version of my own fall project, the person described was a different, identifiable person who never agreed. That was the novelty and the ethical problem at once — and Whitney E. George’s Eavesdropping, a continuous telephone installation whose opening instruction is to pick up and stay quiet, shows what the honest version looks like. It puts the listener inside a conversation they were not invited into and makes that the subject rather than the byproduct. Sumner’s formulation — that overhearing begins without the consent of the listener — is the whole premise of the work. A mechanic I treated as a liability is somebody else’s thesis. The Program supplies a quieter version of the same drift: at launch its automated tree resolved to a live human operator, and it now resolves to recordings of participants from that original run saying what they were most looking forward to. The connection the piece was built to deliver has been replaced by an archive of the people who once received it. Its architecture is also instructive for a different reason: the archive a listener hears was produced by two participants working through a fifteen-day structure with deliberately asymmetrical information, each holding objects and instructions the other did not. The recordings of their real calls become the piece. Performance and documentation are not separate layers there, which is the condition this research arrived at by accident.
6. Consent can be made durable by moving it before the call
Two companies solve Finding 03 by moving consent out of the call entirely, and they do it by opposite means. Perspicuum Immersive requires an application, a waiver and an information form before Perfection, asking about your worst insecurities and proudest moments, and uses what you write to personalize the call — then publishes the sentence that makes it an advance rather than a disclaimer: you can avoid some of its triggers by leaving certain things off the form. The participant defines what the piece is permitted to know, and therefore what it can do to them. Consent as input. Candle House runs the same logic in reverse. Before CLAWS the participant receives an onboarding email and a recorded audio orientation, is cast in advance as a volunteer helpline representative, is told to be alone in a dimly lit room with headphones and a charged phone, is given their own first line to say when the chime sounds, and is given the safe phrase in writing, in capitals, with its exact consequence stated: say it at any time and the call will conclude. Consent as briefing. What both share is the property that matters here. Markham must offer three to five refusals during a call, and every one of those depends on the participant retaining the offer long enough to use it. A form you filled in and a safe phrase you read in an email do not. They persist in a filing cabinet and an inbox, outside the participant’s memory, which is exactly where they need to be for someone in the condition I was in. I still have the CLAWS email. I could look up my own safe word today. That is what a durable consent architecture looks like, and neither company describes it in those terms.
7. Anticipation, experience and retention come apart
Candle House occasionally offers an unannounced piece to its followers called a slow dance: you send a song, the operator listens to it, calls, and improvises a poem out of the lyrics and what they know of you. I had wanted this for a long time. I have since verified the date from my own records: 9 July 2026 at 7:51 PM, the day of my surgery, and by some distance the least likely hour in the whole period for me to have been able to receive or answer anything. Interviewing Evan Neiden on 27 July, I mentioned hoping to experience it one day, and learned that I already had, eighteen days earlier. The wanting was intact and had never stopped. The having left nothing. Anticipation, experience and retention are usually treated as one continuous thing; here they separated cleanly, and the desire outlasted both the event and any memory of it. For a form that trades on what a caller carries away, this is the sharpest question the research produced: what is a one-on-one work for, if the longing it was meant to satisfy can survive the satisfying of it entirely intact?
8. When encoding fails, the archive becomes the memory
For eighteen days I have file metadata, recordings, correspondence and a program bearing my own contribution, and no retained experience of any of it. The timeline in Figure 1 was not reconstructed from recall. It was reconstructed from timestamps. This is not only my situation; it is the situation of every participant in an intimate one-on-one work once the call ends, differing in degree rather than in kind.
These come out of the compiler’s own research; the full account, and the circumstances that produced it, are at Cameo Wood.