EXAMINE / HOW THIS EXPERIENCE WORKS
Before You Know
What this experience does, and what it is not allowed to do.
- VERSION
- 0.3
- LAST UPDATED
- 25 July 2026
- STATUS
- Pre-release draft — not yet confirmed against a released build
- REFLECTION
- Empirical-descriptive — this experience reports what happened; it does not tell you who you are.
This register is binding. What is not written here, the experience does not do. It is versioned, and its changes are listed in full below rather than replaced quietly.
This register describes an experience that has not been released. It is published now, before the experience runs, because the register is the thing that makes the experience checkable — not a document that follows once it is too late to check. Every statement below must be true of the shipped build before the experience is published, and the register is converted to version 1.0 at that point.
- 1
Name & status
Before You Know. Register version 0.3, issued 25 July 2026. A pre-release draft: the experience it describes has not been published yet. Reflection type: empirical-descriptive — this experience reports what happened; it does not tell you who you are.
- 2
What this claims to be
A four-round experiment about one question: does knowing where an answer came from change your judgment of it?
It is not a scientific study, not a psychometric test, not a Turing test, and not a verdict on you or on AI.
- 3
What it asks of you
Nothing before you begin. There is no age declaration, no acceptance step and no gate of any kind.
Then, in each of four fixed rounds: choose which of two answers you find more helpful; state which one you presume was written by a human; and, after the sources are revealed, indicate whether you keep your choice, switch it, or are less certain.
No account, no email, no name, no free text.
- 4
What is measured
Per round: your helpfulness choice; your presumed-human judgment; your post-reveal reaction (kept, switched, or less certain); and the randomised position of each answer. That is the complete list.
Nothing is timed. How long you take is not recorded.
- 5
What is inferred
A descriptive summary of your four rounds only: how often you selected the AI-generated response or the documented human response as more helpful; how often your helpfulness choice differed from the response you presumed was human; how often you selected “not sure”; and how often the reveal led you to keep, switch or become less certain about your original helpfulness choice.
The reveal does not re-measure your presumed-human judgment. Your original judgment stays your original judgment.
The interpretive boundary is phrased exactly as: “This does not show whether an AI understood you. It shows where origin entered your judgment — and where it did not.”
No bias percentage. No personality trait. No attitude score.
- 6
What is stored, where, and for how long
Before You Know stores its session seed, response order, progress and choices in this tab’s sessionStorage. The data normally ends with the browser session and can be removed manually at any time.
Before You Know does not transmit response choices or result data to a server, and it collects no aggregate counts in this version. Known Artificial therefore has no copy of your experience data to retain or delete.
- 7
What is never collected
Free text, identifiers, timestamps linked to you, cross-page tracking, and any data about you on our servers.
- 8
The return loop, and why it exists
There is none, by design. This is a one-time experiment with fixed rounds; nothing in it is built to bring you back.
If a serial version ever exists, it will carry its own register.
- 9
What money may and may not do here
The experience is free, permanently, in full. Any future paid layer elsewhere on this site may offer additional editorial analysis, but never exclusive access to the reveal, the source record, or your own result.
- 10
Mechanisms deliberately excluded
Accounts and logins. Server-side collection of your choices. Streaks, scores, badges and rankings. Personalised round selection — everyone gets the same four rounds, and only the left-right order of the two answers varies, decided in your own browser. Any “you are X per cent …” verdict.
- 11
Sources & labeling
Every source label shown at the reveal must be true. This requirement is permanent and does not soften once the experience is released.
Each AI response ships as “AI-generated response”, supported by its documented model, date, generation record and editing history.
Each response on the human side ships as “Documented human response”, supported by a contributor record, a date, confirmed consent, an explicit confirmation that no generative AI was used, an editing history, and an immutable record of the final text. It qualifies only if it was written without generative-AI assistance.
Development placeholders and AI-generated human-style candidates may not ship under either label. The experience is built so that this is structurally impossible rather than a matter of editorial care: if any record is incomplete, no source is revealed at all and the experience does not run.
- 12
Limitations
Four rounds is an encounter, not a dataset. Your summary describes these four cases only.
A switch after the reveal shows that origin moved your judgment in that case. It does not by itself show prejudice, and stability does not show open-mindedness.
There is no population comparison in this version, because we collect no server data. Nothing here tells you how anyone else answered.
- 13
Changes
v0.1 — pre-release draft, 25 July 2026.
v0.2 — clean-copy pass, 25 July 2026: field 3 aligned with the actual interface option (“are less certain”); field 5 rewritten as an exhaustive list of derived descriptive totals, with the explicit note that the reveal does not re-measure the presumed-human judgment; field 6 scoped to Before You Know with technically accurate sessionStorage wording; field 11 extended with “AI-generated response” as a mandatory source category and a permanent truth requirement that survives release; field 14 corrected to reference the Local Data Inspector on the Privacy page.
v0.3 — 25 July 2026, the first version published on this page. It supersedes v0.2. Two approved public source classes only, AI-generated response and Documented human response: the earlier “human-originated, shaped with AI assistance” class is removed from this experience, because launch version 1 requires four genuinely human-written responses produced without generative-AI assistance. Fields 3 and 6 record the 18+ self-declaration and confirm that it is not stored. This register publishes its version and issue date rather than a build reference.
v0.4 — 31 July 2026, the version that now binds. It supersedes v0.3. The visitor-facing 18+ self-declaration is removed from this experience and from the site: the 18+ standard governs the provenance and depiction of human and humanlike subjects, and the adult status of source participants where that applies, and it was never an access restriction for readers. Field 3 records that the experience now asks nothing before you begin; field 6 no longer describes a declaration, because none is made or stored.
- 14
Questions & removal
Your session data normally ends with the browser session. You can restart the experience from its result screen.
To see and remove the browser data Known Artificial stores, use the Local Data Inspector on the Privacy page.
Questions about this register: see Corrections & Contact.