Personal Science

Science equity. People without a lab, a degree, or a grant should still be able to run real experiments.

A list of experiments you can do at home. No application, no review, no permission needed. One person = one lab. (Charter)

17 experiments · each page tells you how to know it worked · CC0, copy freely · daily collection + AI-designed drafts

中文 · One-minute version · The list · Growing every day · How to add one · Manifesto · Charter

The one-minute version

1. Pick one — sorted by field below, 30 minutes to 5 days
2. Follow it — every page says what you need, how to do it, where it goes wrong, and what counts as success
3. Post the result — add a line to RESULTS.md. Failures count too.

No scores, no certificates, no "professional vs. amateur". Only what you produce.

What this repo is (and isn't)

For: someone with a desk, some curiosity, and no idea where to start. Not for: people who want conclusions handed to them, badges to collect, or authority to lean on.

Why "Personal Science"

Personal computers gave computing power to individuals. Personal agents give one person the output of a team.

The next cell is the same move:

The barrier was never intelligence. It's four things:

equipment, vocabulary, method, and nobody telling you what to do next.

AI removes all four at once. Measurement, records, analysis, write-up — one person can now run the whole loop. When that works, the individual becomes the smallest unit of research. Science starts happening on a desk, the way companies started happening in garages.

Start here

Note: experiment pages are currently written in Chinese. Translations are very welcome — one page = one PR, no coordination needed.

The list

How to read this table: time is for one person; cost is what you have to buy; difficulty is hands-on difficulty, not theory.

Physics & measurement

Electronics & hardware

Light & waves

Biology & plants

Environment & water

Method & thinking

Idea pool

Ideas nobody has written up yet. Claim one, write a page with the template — full list in ideas.md.

• Measure cable loss to explain why fast charging is slow
• Use phone flash flicker to measure fan RPM
• A CD as a diffraction grating: measure a laser pointer's wavelength
• Wi-Fi attenuation heatmap through walls
• Salt water density with an egg
• Dissolution rate of different salt brands
• Shuttlecock elasticity vs. temperature
• Boiled egg timing vs. yolk state
• Distribution of red-light wait times
• Last-digit patterns in supermarket prices
• Left/right channel consistency of headphones

Growing every day

This repo has a conveyor belt that runs once a day on a US-based runner (GitHub Actions — free, no server to rent):

classic experiments seed library + Wikipedia + Hackaday + Hacker News
        ↓
candidate pool (recent items + classics, capped at 150)
        ↓  AI applies five criteria
inbox/candidates.md (up to 8, each with "how to redo it with today's tools")
        ↓  AI writes full draft pages for the top 2, including "how AI helps"
inbox/designs/ (drafts, marked unverified)
        ↓  a human edits one and opens a PR
the list above

No age limit. A classic experiment from 1801, a WWII measurement, a 1960s psychology study — if it can be redone today with AI, an ESP32, or a phone, and it can grip a normal person, it belongs here.

The script is responsible for seeing a lot. The AI is responsible for seeing accurately. A human is responsible for doing the work. What enters the list is always decided by a person.

To enable AI filtering and AI-designed drafts, set a PS_LLM_KEY (any OpenAI-compatible endpoint works). Without it, collection and commits still run.

Daily candidates dashboard

Let an AI pick for you

Take a photo of your desk, workbench, or balcony. Hand it to any vision-capable AI together with SKILL.md.

It will pick three experiments you can start today, tell you what's missing, and roughly what it costs.

It won't rate you or grade you. It does one thing: translates "here's my stuff" into "here's what you could do with it."

How to add one

Three requirements, details in CONTRIBUTING.en.md:

1. One person can do it at home — budget ≤ ¥200 (~$30), time ≤ one weekend, no license, no dangerous operations
2. Say what counts as success — the acceptance criterion must be written before you start
3. Name the traps and the risks — including "don't do this one"

Quality over quantity. One experiment you can actually run beats twenty links.

Where to post results

Add a line to RESULTS.md: which experiment, what number you got, where you think it went wrong.

Failures are more valuable. One "it didn't work, because X" beats ten "I got it right".

Five things we don't do

• We don't certify. No certificates, no scores, no verdict on what is "real science".
• We don't vet identity. Anyone can open a PR. We don't ask who you are or where you studied.
• We don't do it for you. AI can guide you, but it can't press the button for you — that turns into watching videos.
• We don't chase hype or build idols. There are no gurus here, only people who finished and people who haven't started.
• We don't touch dangerous or licensed work. High voltage, open flame, ingestion or inhalation of chemicals, live animals, anything requiring a permit.

These five are written into Article 5 of the charter. If someone later wants to add scoring, certification, or gatekeeping, point them there.

Safety line

High voltage, open flame, chemicals you'd ingest or inhale, live-animal experiments, licensed testing — none of it is accepted here.

If something looks exciting but feels dangerous, don't do it. Open an issue first.

Reference

• OpenMAIC — a multi-agent interactive classroom from Tsinghua MAIC. If you want to turn one experiment into a class you can talk with, start there.
• This repo only answers "which experiments are worth doing". Turning them into a course is out of scope here.

Repo layout

.
├── README.md            Chinese README
├── README.en.md         You are here
├── MANIFESTO.md         Manifesto (Chinese)
├── MANIFESTO.en.md      Manifesto (English)
├── CHARTER.md           Charter (Chinese)
├── CHARTER.en.md        Charter (English)
├── experiments/         One page per experiment (17 pages)
├── ideas.md             Idea pool
├── RESULTS.md           What people got (including failures)
├── CONTRIBUTING.en.md   How to add one (English)
├── SKILL.md             Photo → AI picks what you can do
├── collector/           Daily collection script
├── inbox/               Raw pool + AI filter
├── feed/                Web dashboard
├── deploy/              Where to run, where to host
├── AGENTS.md            Rules for AI agents maintaining this repo
├── maintenance/         Runbook, backlog, automated checks (Chinese)
└── archive/             Dead ends, kept on purpose

Want to help?

• Star it ⭐ so more people who are "curious but don't know where to start" find it
• Add one experiment: CONTRIBUTING.en.md — three rules, one PR per experiment
• Post your result: RESULTS.md. Failures are worth more.

Content is CC0. Copy it, remix it, no attribution needed.