A STUDY IN BEHAVIOUR / EVIDENCE
01 / THINKING IN PRISMSOne question.
One question.
Four ways of seeing.
An idea looks different depending on the lens through which we examine it. Choose a territory, then rotate between evidence, philosophy, craft, and scepticism. The perspectives here are authored and curated, not AI responses.
CHOOSE THE SUBJECT8 TERRITORIES
ROTATE THE LENS4 PERSPECTIVES
A lens changes the interpretation—not the underlying facts. Factual claims still need evidence, even when the design becomes poetic.
EXPERIMENT / Human interpretation is contextual. The source is always one click away.
02 / CREATIVE CHEMISTRYCollide two worlds.
Collide two worlds.
Find a third question.
Pair domains that normally live in different rooms. The machine offers a proposed bridge, a serious objection, and sources to explore. It is a transparent, authored combinatorial engine—not a language model or an empirical result.
THE TWO REAGENTSRECOMBINABLE IDEAS
FIELD EXPERIMENT 002AX-03 / DX-05
IDENTITY
×
SCENT
ANALOGY IS NOT EVIDENCEUSEFUL FRICTION ↗
03 / THE PUBLIC RECORDReal doors.
Real doors.
Not invented rooms.
A curated, searchable set of confirmed public links from Sahil's site universe, Medium writing, professional work and research profiles. Unlike the reactive visual above, these destinations are actual published resources.
* The Substack publication URL is listed on the .online home, but live access could not be confirmed during this prototype's research. This is a curated snapshot, not live cross-platform synchronisation.
04 / A VISITOR BECOMES A PARTICIPANTLeave with a question.
Leave with a question.
Not a notification.
The reader is not the product. Turn the experience into a private note you can keep. This site has no account, server endpoint, telemetry script, or submission button.
A MESSAGE FROM THE INSTRUMENT
What would you study if no one were grading your curiosity?
Some ideas matter because they help us understand the world. Others matter because they help us notice what the world feels like. Not everything needs to become a career.
PRIVATE BY DEFAULT / YOU KEEP THE KEY
MY UNFINISHED QUESTIONLOCAL MEMORY
Nothing has left your browser.
TAKE HOME AN ARTIFACTBROWSER-GENERATED
Your device is rendering an evolving visual field. Capture a postcard from it as a PNG. The export contains the generated artwork and the title of your current idea—not your camera image or private note.
LocalStorage can be cleared by your browser, device, or private-mode settings. Export a text file if you want a durable copy. This is not a cross-device backup.
05 / THE BROWSER IS THE MEDIUMNot a video.
Not a video.
A live instrument.
Look underneath the performance. A mobile browser can render dynamic graphics, create images, speak text, use the camera with explicit consent, and store notes locally. Some features vary by browser and permissions.
WHAT THIS DEVICE CAN DODETECTING…
01 / Real-time graphics
A fragment shader runs on the GPU through WebGL2 where supported, with a quieter Canvas 2D fallback. The imagery is drawn anew—not streamed from a server.
TESTING
02 / Motion control
Lower the animation intensity for a calmer experience. Reduced-motion preferences are respected automatically.
48%
03 / Voice as an interface
The browser can narrate the current authored idea via available speech voices. No microphone is required. Voice quality and language vary by platform.
TESTING
04 / A mirror made of pixels
With your permission, your camera can appear in the living visual through a layered browser effect. Frames are displayed locally; this application does not upload or record them. Some browsers may ask for permission.
OFF
05 / Private, portable memory
Notes and saved reading paths can be kept in this browser's local storage where available, and exported as a file. No sign-in and no invisible cross-site archive access.
TESTING
This is a static site. It does not contain live AI inference, real-time Medium/Substack scraping, a server-side knowledge base, search embeddings, or an autonomous agent. Those require additional software and often an API, credentials, consent, ongoing costs, and content verification. The camera is strictly optional; speech recognition is intentionally not activated.