SYKQD is made by people using machines, and both parts are true everywhere. Below is what we declare, section by section — written by hand in one file, not measured by a tool and not audited by anyone else.
44%
Declared AI involvement
Across the whole site, averaged by section.
74%
Human involvement
Editing, rejecting, rewriting, drawing over.
4/5
AI depth — Deep
How many layers of the making a machine touched at all.
These are project-declared figures, not an independent audit. The single number is a weighted formula, kept inspectable: 15% ideation + 15% user journey + 20% creative influence + 20% production + 20% final output + 10% × (100 − human involvement).
Ideation
18%
Whose idea the thing was in the first place.
Weight 15%
User journey
36%
Who decided the order, the doors, the pacing.
Weight 15%
Creative influence
41%
Tone, art direction, jokes, restraint.
Weight 20%
Production
64%
Who actually generated the pixels and the prose.
Weight 20%
Final output
59%
Who assembled and shipped what you can see.
Weight 20%
Human involvement
74%
How much a person edited, rejected, rewrote or drew over it.
Weight 10% · counted inverted
SYKQD
14%
Surface · 1/5
The properties, the name, the point of the whole thing. Argued about by people over a long time.
Home
51%
Deep · 4/5
Structure decided by us; the motion, the copy polish and the code written with AI assistance.
Laium Atlas
54%
Deep · 4/5
Ten belts and their meaning are ours. Belt and city illustrations are AI-generated to a tight brief, then culled.
Someplace2Write
42%
Deep · 4/5
Written from real working life, drafted and tightened with AI, edited line by line by a person.
Ric · Rac · Iys
47%
Throughout · 5/5
Character canon is fixed and human-authored. Jokes are co-written; the rendering is drawn in code.
Daily Dots
48%
Deep · 4/5
The six scopes and the reading logic are hand-written rules — deterministic, not a model.
Situations
40%
Deep · 4/5
Questions were designed and rewritten by us. The reading is a rules engine, not a generated answer.
Enterprise
53%
Throughout · 5/5
Find a cause builds its blueprint deterministically. The Questionnaire Lab offers a template draft and, on request, an AI-assisted draft. Public pages are only read when you press the button.
Google Drive — content source
Only when a signed-in editor opens the library panel and picks a folder.
Reads folder and file names, types and links, and stores those references. Read-only: nothing in Drive is copied, moved, renamed or deleted. Drive content is never sent to a model.
Public page reading — Find a cause
Only when you press ‘Analyse public presence’, and never automatically.
Opens up to three public pages you listed, quotes what it saw, and keeps that evidence separate from any reading of it. No logins, no crawling, no claim about compliance and no guess at anything sensitive.
AI-assisted questionnaire drafting
Only when you press ‘AI-assisted draft’ in the Questionnaire Lab.
Sends the fields you typed on that page to a model on our server and checks the reply against a schema before showing it. Failures fall back to the template draft. The result is a draft, not validated research.
A separate question
Around the site you'll find a small prompt asking whether something felt machine-made, and a slider from all-human to all-machine. Those answers are audience perception. They are stored apart from the declared numbers on this page and never averaged into them — the gap between the two is the interesting part, and collapsing it would waste it.
Perception responses currently stay on your own device. When they move to our side, they will still be reported separately.