HUMAN OS

Four directions — sleep, food, closeness and protection — started as four separate ideas: a managed shutdown (Sleep OFF), “last meal” (Last Meal), a consent interface (Consent UI) and safety without weapons (Safe Mode). I merged them into one system: HUMAN OS — “an operating system for biological life”. The project line: “Ancient body. Modern world.”

This is not a scrolling landing page. One viewport, scenes replace one another, the central object is a human and their material world: a face, hands, food, fabric. The interface gradually gives way to the body: DATA → BODY. The drama follows the arc MEASURE → MODEL → OPTIMIZE → FAIL → REVEAL: the system first seems authoritative, then shows interdependencies, tries to optimise the human — and admits that care is not optimisation.

The shots are taken from the running build at 1440×900. In reality these are interactive scenes: switches, sliders, mutual confirmation, sound on request.

The whole experience is a single dramatic arc. Every chapter ends with a return to the main map, and the map changes: at first it is strict and confident, by the finale — open and incomplete.

1. Boot. Three loading scenes: species — Homo sapiens, origin — about 300,000 years ago, environment — industrial and digital. “Evolutionary legacy meets modern environment”.
2. Main System. Four processes around a central object. “System model / structured”, dependencies unmapped. After each chapter it changes: dependencies detected → variables interdependent → boundary permeable → model incomplete.
3. Four chapters. Sleep, Fuel, Connect, Protect — 8 screens each. Each has its own visual carrier: the body, material, a pair of people, the space around the body.
4. Optimization failed. 5 screens: calculating the optimal state → conflicting variables → objective function unresolved → optimal state undefined → optimization failed. This is not a crash: the system admits its limit.
5. Reveal. 4 screens: “Model limit detected”, “A human is not a problem to solve”, “Optimization is about control. Care is about response”, “You are not a machine. You are a living system. Take care of it.”

The chapters differ in form but obey the same principles. Each has its own hard rule that forbids turning it into a familiar health interface.

Between chapters, relations are shown (no more than 1–3 at a time): related, potential, contextual, interdependent, in tension, unresolved. Each is marked with a reliability level — OBSERVED, INFERRED or SIMULATED. The words “priority”, “recommended”, “best next”, “causes”, “fixes” are forbidden in the interface.

The project is recognised not by colour, glitch or grid, but by how the system thinks, doubts and admits the limits of its own model. For that I formulated 10 rules — the Human Interface Language.

• 01 / Body before interface
• 02 / Data is a model
• 03 / The system can be wrong
• 04 / Motion must have a cause
• 05 / Data → Body
• 06 / Biology overrides UI
• 07 / Error has meaning
• 08 / 100% is not a human state
• 09 / No moral judgment
• 10 / Care ≠ Optimization

13 familiar elements are forbidden in the interface: health, wellness, readiness and risk scores, compatibility %, calorie target, macro %, daily goal, streak, progress %, best next step, recommended and priority rank. Green does not mean “good”, red does not mean “bad”.

Two readiness tests. The blind test: remove the name, the modules and the accent colours and ask “what does this interface want from the person?” — the right answer is “nothing specific, it observes and describes conditions”. The anti-sci-fi test: if the interface could be renamed “Orbital Defense System” without changing its language, it must be redone.

The project was built in stages. Each stage ended with a document and a check, not just a picture: about 64,000 words in 36 files in total — from the concept to the acceptance criteria.

1. The original task: four themes — sleep, food, closeness, defence — as one noticeable authorial project.
2. One system instead of four cases; the formula “operating system for biological life”.
3. A full 25-screen script, then vertical slice v0.1: Boot → Main System → Sleep → Main System → Final.
4. Sleep as the reference chapter: 9 key-frame candidates, a photo selection matrix, 3 sleep-restriction variants, a clickable prototype.
5. Fuel, Connect and Protect on the same pattern; a transition map between modules and rules for showing relations.
6. A handoff package for development: screen inventory, state machine, components, tokens, motion, responsive rules, copy, acceptance criteria. Built from it with the help of Codex.

The site is a static Next.js export: the host needs neither Node.js nor a database. The whole script is described by types and a finite-state machine, so each of the 49 states is reachable and testable.

The repository also has a server-side layer that classifies the sleep state through TypeSafe Jev (typed states and a confidence threshold); it is not part of the static version — it is groundwork for extension.

The project is speculative design, not health-tech. During the work 13 formulations that looked scientific were replaced with more honest ones. For example:

Three reliability levels for every value: OBSERVED — a user choice or a known value, INFERRED — a plausible direction of effect, SIMULATED — an artistic model with the mandatory label “simulated system model”.