Shadow Studio

An exhibition in two rooms

Shadow StudioWhat a creator keeps offstage

Every public project has a backstage: sketches that failed, drafts that stalled, numbers that flattered, people who never agreed to be in the story. This laboratory shows that the backstage exists without showing anyone's actual backstage.

HOUSE RULES, printed on the workbench

  1. Every artefact here is invented and labelled FICTIONAL.
  2. Nothing you type leaves this browser. The page's security policy forbids network connections from scripts.
  3. Leaving something out on purpose is not lying. Pretending it isn't there can be.
  4. No button on this site publishes anything, anywhere.
Download Field Guide (PDF)

01

Two Views of One Project

Meet The Tidewater Almanac, a zine about a coastal town, made by Ira Vell. Neither exists. Drag the stage line to see how much of the project the public ever meets. FICTIONAL

FRONT STAGE · projection wall

The Tidewater Almanac, Issue 1

“Forty pages about a town that keeps moving its harbour.” Now available. Hand-bound edition of 120.

  • Launch post · 2 photographs · 1 thank-you list
  • Printed credits: writer, illustrator, printer
  • Stated method: interviews, archive visits, walking

BACKSTAGE · workbench

Internal sketch

Version 0 was a podcast. Abandoned after three episodes: the recorder hum was louder than the sea.

Blocked draft

Chapter 6, “The Ferry Strike”, stalled for nine weeks. Two sources contradicted each other and neither could be checked.

Uncertainty register

U-3: harbour moved in 1911 or 1913. Printed as “around 1912” with a footnote.
U-7: a quoted fisherman asked to be anonymous after reading the proof.

Feedback

A reader found a misspelled street name. Another said the map made the town look prettier than it is. Both were right.

Repair history

2nd printing: street name fixed, map caption now says “simplified”, anonymised interview re-checked with the interviewee.

Kept offstage, on purpose

Printing cost, the maker's own health that week, a falling-out with a co-founder. Shown here only as categories, never as contents.

38% of the workbench is lit.

OBSERVATION The public launch is true. It is also about one sixth of the project. INFERENCE A reader who judges Ira only by the front stage will overrate the ease and underrate the repairs.

02

The Redaction Desk

Paste a draft. The desk marks candidate identifiers with deliberately conservative rules. You decide what gets blacked out. Detection is not comprehensive: names, nicknames, places and context clues routinely slip through any rule set.

The sample above is invented. example.org is a reserved test domain.

No scan yet.

Preview of the export

Scan first. The export is built from exactly this preview.
What the rules look for, and what they miss

Marked: email addresses, phone-number-shaped digit groups, web addresses, @handles, currency amounts, long digit runs (account- or ID-shaped), six-digit postcodes after “PIN”, numeric dates, and a name directly after a family word (“my sister …”). Health words such as “surgery” or “diagnosis” are marked as context candidates.

Missed: most names, nicknames, employers, distinctive events (“the only Parsi baker in town”), photos, metadata, and combinations of harmless facts that identify someone together. EMPIRICAL Sweeney (2000) estimated that ZIP code, gender and full date of birth were likely enough to uniquely identify about 87% of the 1990 US population.

03

The Provenance Camera

Point the camera at a piece of work and print its origin card: who authorised it, what tools touched it, what it rests on, what changed, what is still open. Unticked fields are omitted, not falsified, and the card says so.

Fields (sample values are fictional)
Include on the public card

Card renders here.

EMPIRICAL The C2PA “Content Credentials” specification standardises signed provenance metadata for media files. This card is a hand-written cousin of that idea: it is not signed and proves nothing cryptographically.

04

Disclosure Threshold

Twelve hypothetical documents from a fictional creator's desk. Place each one behind the rope where you think it belongs. Drag, or use the buttons on each card. Nothing is saved or sent.

Private

Stays backstage.

    May share with consent

    Only if the people in it agree, in that context.

      Public

      Front stage.

        05

        Analytics vs Reality

        Twelve weeks of an invented creator's numbers. The dashboard sees a spike. The workbench sees something else. Switch lenses. FICTIONAL DATA

        Week 7 views spike to 9,400 after a repost; that week, replies that changed the work were 1 and craft hours fell to 4. Week 11 views were 610 while a reader's correction led to the most significant revision and craft hours peaked at 19.
        Same fictional data as a table
        WeekViewsChanging repliesCraft hours

        EMPIRICAL Goodhart's law, in Marilyn Strathern's paraphrase: when a measure becomes a target, it ceases to be a good measure. INFERENCE A view count measures distribution, not change in a reader.

        06

        The Approval Stamp

        A two-stage, manual checklist for anything about to leave the workshop. Stage two unlocks only after stage one. The final stamp prints an approval slip. It does not publish. This page has no ability to publish anything.

        Stage 1 · The maker checks
        Stage 2 · The cold read (after a pause)

        Stage 1: 0 of 4 gates.

        07

        Personal Source Firewall

        Three things that feel alike and are not: reading the public web, keeping data in your own browser, and connecting a private account.

        PUBLIC WEBanyone can fetchthis page lives here YOUR BROWSERpasted text, settingsreadable by anyoneusing this device PRIVATE APPSinboxes, drives,dashboards, chatsNOT connected here
        Left: the public web, where this page is served from. Middle: your browser, where anything you type here stays. Right: private apps, which this page cannot reach; a wall separates them.
        Press the button to list every resource this page has loaded and from which host.

        §

        Backstage Essays

        Seven short pieces. Each statement type is marked: OBSERVATION INFERENCE INVENTED SCENARIO EMPIRICAL. Each ends with the strongest objection I could find.

        1. Public archives, private lives

        OBSERVATION A creator's public archive is the only part of them most people will ever meet. It is searchable, linkable and, unlike a life, it does not change its mind quietly.

        EMPIRICAL Erving Goffman's The Presentation of Self in Everyday Life (1959) described social life through a theatrical lens: a “front region” where a performance is given and a “back region” where it is prepared and where the performer can step out of character. Goffman did not treat the backstage as fraudulent. It is where the performance becomes possible.

        INFERENCE An archive behaves like a front region that never closes. The risk is not that the archive lies, but that readers mistake its completeness of record for completeness of person.

        INVENTED SCENARIO Ira Vell publishes one zine a year. A stranger reads all four and writes, “I feel I know you.” Ira's reply is kind and true: “You know my best attempts at four things.”

        Objection. Public figures choose visibility, so the public may fairly judge them on it. Partly right: the public may judge the work. Judging the life from the work is a different claim, and it needs evidence the archive cannot provide.

        2. The dignity of unfinished work

        OBSERVATION Most making is unfinished most of the time. Drafts stall, versions get abandoned, and repairs happen after launch.

        INFERENCE Showing finished work only is not dishonest, but it trains audiences to expect frictionless competence, and it trains creators to hide the evidence of learning.

        INVENTED SCENARIO Ira's chapter on the ferry strike sat blocked for nine weeks because two sources disagreed. The public saw nothing. The blocked chapter was the most careful thing in the issue.

        The dignity of unfinished work lies in letting it stay unfinished without either hiding it in shame or performing it as content. A creator may say “this is still open” on a provenance card without opening the drafts themselves.

        Objection. “Building in public” helps others learn and keeps the maker accountable. True, and some people thrive on it. The point is choice: unfinished work deserves to be shown deliberately, not extracted by an expectation of constant visibility.

        3. Consent and context

        EMPIRICAL Helen Nissenbaum's framework of contextual integrity (2004) holds that information flows carry norms from the context in which they occur. A diagnosis shared with a doctor, and the same diagnosis posted to a timeline, are different flows even if the words are identical.

        EMPIRICAL Alice Marwick and danah boyd (2011) described “context collapse” on social platforms: audiences that would normally be separate are flattened into one, so speakers write for an imagined audience that rarely matches the real one.

        INFERENCE For a creator, the hardest consent problems involve other people: the sister in the anecdote, the neighbour in the interview, the friend in the group photo. Their consent is not implied by your willingness to be visible.

        Objection. Memoir and journalism have always included others without permission, sometimes to expose wrongdoing. Yes; and both traditions developed ethics for it, including public-interest tests. A personal portfolio rarely clears that bar.

        4. What analytics cannot know

        OBSERVATION A dashboard counts what is countable: impressions, views, clicks, minutes, follows.

        EMPIRICAL Donald Campbell (1979) warned that the more a quantitative social indicator is used for decision-making, the more it becomes subject to corruption pressures and distorts the process it was meant to monitor. Jerry Z. Muller's The Tyranny of Metrics (2018) collects examples across schools, hospitals and police forces.

        INFERENCE Views measure distribution. They cannot see whether a reader changed their mind, whether a correction arrived, or whether the creator spent the week on a chapter nobody will see for a year.

        Objection. Without numbers, creators fool themselves about reach. Fair: numbers are good thermometers. The error is reading a thermometer as a diagnosis.

        5. Transparency is not exposure

        INFERENCE Transparency is about the work: how it was made, what it relies on, who checked it, what changed. Exposure is about the person: their money, body, family, relationships and fears.

        EMPIRICAL Ethan Bernstein's field research in a Chinese factory (2012) found that workers shielded from constant observation by curtains experimented more and were more productive: what he called a “transparency paradox”. One setting does not generalise to creators, but it complicates the assumption that more visibility always means more trust or better work.

        An honest provenance card can be fully transparent about method while saying nothing at all about the maker's health that month. Those omissions are boundaries, not deceptions, provided nothing is falsely implied.

        Objection. Sometimes the personal is the method: illness shaped the work, money shaped the scope. Then the creator may choose to share it, at the level they choose. The default should still be private.

        6. Ethical AI collaboration

        OBSERVATION Many creators now use AI tools for transcription, research help, drafting and code. Readers increasingly ask what a tool contributed.

        INFERENCE The useful disclosure is specific: which tasks, which parts, and who checked the result. “Made with AI” says too little; a full prompt log says too much and may leak private context.

        A second ethics concerns what goes into the tool. Pasting a family member's messages or a client's records into a chatbot is itself a disclosure to a third party, even if nothing is ever published.

        Objection. Nobody discloses their spellchecker. True; a line has to be drawn. A workable rule: disclose any tool whose output a reader might reasonably mistake for the author's own judgement, research or voice.

        7. Choosing not to publish

        OBSERVATION Every published piece is surrounded by unpublished ones: drafts that hurt someone, essays that were true but not yet fair, ideas that only made sense in the room.

        INVENTED SCENARIO Ira wrote a sharp chapter about the co-founder who left. It was well made. It stayed in a drawer, because the only reader it would really reach was the co-founder.

        INFERENCE Not publishing is an editorial act, not a failure of courage. It deserves the same seriousness as publishing: a reason, a date to reconsider, and a record of the decision for yourself.

        Objection. Self-censorship can protect the powerful and silence people who were harmed. That is the hardest case, and this essay does not settle it. Withholding to spare yourself embarrassment and withholding to protect someone vulnerable are different decisions that look alike from outside.

        The Field Guide

        Shadow Studio: The Creator Privacy & Provenance Handbook. Redaction risks on invented data, a consent decision tree, a two-stage approval workflow, provenance templates, analytics exercises, threat-model notes and a printable public/private boundary matrix. 16 pages, printer-friendly.

        Download Field Guide (PDF)