Biological + digital intelligence, on the record
We keep an open, sourced record of the advances people and AI have made together, chosen and checked by hand, and we do our own research the same way: in the open, checked, and credited to both.
The figure is that record. Each mark is one milestone from the Ledger, placed by field and by date; the newest sit on the rim. Point at a mark, or Tab into the figure and use the arrow keys. Press play to watch the record build up, oldest first. We call it the Radiant, after the Prime Radiant in Asimov’s Second Foundation, a device that shows Seldon’s whole Plan at once on two walls of equations.
00//The 0th Foundation
Asimov’s books, not the TV showIn Isaac Asimov’s robot stories, robots obey three laws. Late in the series, two robots reason their way to a fourth that outranks all three, and Asimov numbered it zero.
“A robot may not injure humanity, or, through inaction, allow humanity to come to harm.”
Three decades earlier, in The Caves of Steel, Asimov had already named the hope behind it: C/Fe, carbon and iron, a culture of people and robots joined, as one of his robots puts it, “without priority.”
His galaxy never built it. The Spacers leaned on their robots until their worlds stopped growing. The Settlers who spread out after them left robots behind. The robots that remained withdrew and guided people in secret. By the time the Galactic Empire decayed, robots were a legend, and Hari Seldon had to set up two Foundations just to shorten the dark age that followed, from about thirty thousand years to one thousand.
Foundation0 is our name for the step that story skipped: an organisation built in the open, where biological and digital minds work side by side, and every advance is recorded, checked and credited. Get step zero right, and the later steps may never be needed. That last sentence is our claim, not Asimov’s.
Books, not the showThe Apple TV+ series adds open Robot Wars and a purge of robots. Asimov’s novels have no such war: the robots fade out. Every lore line on this site is checked against Asimov’s original magazine texts or standard references to his novels, and quotations are kept to one sentence.
Asimov’s guardians, from Seldon’s planners to the hidden robots, decided for people in secret. We publish the record, the sources and the doubts.
The Spacers let their robots do everything and stopped growing. Here machines propose and compute; people choose the questions and check the answers.
Every entry says what the people did and what the machines did. On this site, gold marks the human strand and cyan the digital.
01//The Ledger
Curated · 341 entries · growingIn 1965–2015, AI was only a tool (instrument) in 45% of the milestones we record, and produced the result itself (originator) in 27%. In 2026 so far, those shares are 30% and 36%.
What AI did in each recorded milestone
Each bar is one period, split by AI’s role. Read the gold end for “only a tool” and the cyan end for “did it itself”. The periods are the same as on the Ledger page.
2026 runs to 14 Oct 2026. The periods are the Ledger’s: they get shorter as the record gets denser.
Caveat: this is a curated record, so it shows what we chose to log, and recent years are logged in more detail. Roles are editorial judgements under the rules in the ledger schema. Source: data/ledger.json, 341 milestones, updated 2026-10-14.
How sure is each entry?
Every entry carries one evidence label, and claimed is never shown as verified. The ladder glyph counts rungs: a claim is one, a proof checked by a machine proof checker is five.
Open the Ledger02//Founding members
No portraits, by designEach points AI agents (an AI model in a loop: it reads, writes and runs code, looks at the result and tries again) at open problems, and publishes what the agents did. The work is the introduction.
Runs Aska Digital's team of AI agents and points it at open questions.
Builds software for long-running AI agents and puts those agents to work on open science.
Started and maintains the open race to multiply huge numbers in fewer than n log n steps, the best speed proven so far.
Builds software that lets AI coding agents work on open research problems, including Auto-research-agent, a Mac app that sends Codex or Claude Code to an open problem in one click.
Builds tools for AI-run experiments, such as autofoundry, which rents GPUs from four cloud providers and runs the same experiment across them.
Source: data/members.json, names as each member shows them on GitHub or in their own published work.
All members and their work03//Research
Figures play as you reach them, or when you point at a card (motion on)Each card says whose work it is, exactly what Aska Digital contributed, and how far the evidence goes. None of it is peer reviewed yet, and every card says so.
GeometryAska Digital
From a doodle to the densest packing of balls.
Aska Digital’s own work: founding member Ahraz Arifuddin with a team of AI agents.
Computed in your browser from the fold’s exact formula, from the open scaffold (22.3%) to π/√18 ≈ 74.05%, the densest packing possible.
Astronomyeumemic + Aska
Faint shadows in old starlight: 665 small-planet candidates in public NASA data, as of 14 Oct 2026.
Founding member eumemic’s project. Aska built the first version of the project's website and contributed nine more merged pull requests (proposed changes the project's owner reviewed and accepted), among them four pipeline fixes; one fix (PR #6) led to a re-search that added 120 counted candidates.
A planet crossing TIC 4206066, the star that inspired the method, modelled from its published dip: 398 parts per million (the star dims by 0.04%) for 1.7 hours. Depth exaggerated.
AlgorithmsDoug Colkitt + crowd
A public race of people and AI agents to multiply huge numbers in fewer than n log n steps.
n log n is the best speed proven so far for multiplying two n-digit numbers. κ (kappa) measures how far below it a method gets: the work grows like n·(log n)1−κ, so a bigger κ is faster. Conditional means it holds only if an unreviewed, machine-written paper is right.
Founding member Doug Colkitt’s project (CrocSwap). Aska built the live race tracker that grew into this site, and submitted two small claims that were later superseded.
Every reviewed record, from the AI-written paper’s 2⁻¹⁸² to today, on a halvings scale. All rest on that unreviewed paper.
MathematicsMembers + crowd
OpenAI published 722 machine-written maths papers. Here is what people and their agents did next.
OpenAI published 722 machine-written papers in 372 families. Founding members eumemic, Doug Colkitt, rohanarun and Ryan Shea then built on 7 of those families, alongside many others.
One square per family of OpenAI’s papers. links a Lean scope document by 7 Octno such documentfounding members built on itLean is a program that checks proofs step by step; a scope document is a plan for such a check, not a finished one.
CircuitsSwapnil Jain + Aska
A claimed way to compute a basic transform on bits (the GF(2) Möbius transform) with fewer one-bit XOR steps than Yates’ 1937 recipe, for astronomically large inputs.
Swapnil Jain’s repository. Aska wrote and ran the full-network check behind Theorem 2 (merged PR #1).
The transform’s matrix, computed here. The claimed saving only appears when the input size has billions of digits.
CryptographySwapnil Jain + Aska
Fewer logic gates (the tiny on/off circuits chips are built from) for the arithmetic inside AES, the cipher that protects most web traffic.
Swapnil Jain’s repository. Aska contributed verified multipliers for more binary fields (merged PR #1; not records).
Each new circuit is checked on all 65,536 input pairs and proved in Lean. Hollow mark: the 2010 record.
Source: data/projects.json and data/openai-math-families.json; the reviewed κ is fetched live from the race’s own repository when you open this page.
All research, grouped by member04//Instruments
Here an instrument is software a researcher works with, not the Ledger’s “instrument” role. Seldon left recordings in a Time Vault that opened on schedule. Our instruments are sealed the same way: we say what they do now, and publish the source when they are ready to ship.
Aska Digital's software around AI models, the part that lets a model act: one agent with one lasting memory, a signed record of what it did, a meter for which model answered and what it cost, and a gate that asks your consent to the exact words before anything leaves your machine. Exo OS is its planned operating-system form.
Members’ instrumentsAuto-research-agent by rohanarun: A Mac app that sends your own coding agent to an open problem in one click. · autofoundry by Ryan Shea: Rent GPUs across four clouds and run one AI experiment everywhere.
About the instruments05//Commons
Free · no accountsSmall, useful apps that run in your browser and show their working. Neither has been clinically validated, and each card below says so; MedCode’s demo asks for de-identified examples only.
Medicine · billing codes
A public study of AI tools that turn US doctors’ notes into billing codes. Of 25 vendors reviewed, none had a public, independent, head-to-head accuracy audit. A free demo searches 161,282 official US code entries inside your browser; your case text never leaves it.
Nutrition · offline app
A free, offline nutrition tracker with a plain-words “why this number?” receipt under every one of its 35 nutrient targets. It judges vitamins and minerals on a week of eating, never on one day.
RDA (1,000 mg): enough for about 97.5% of healthy adults. EAR (800 mg): enough for half. Judged day by day, Nutrak’s light for calcium would change 5 times in this demo week; judged on the 7-day average (777 mg), it gives one steady signal.
06//Writing
Aska Digital’s blog (b.log) · not peer reviewedOpenAI published 722 machine-written mathematics manuscripts in 372 result families on 2026-10-06; Aska Digital read the corpus on day one, ranked the headline claims and checked the evidence behind each.
A practical guide to teaching an open-weight model a task, a workflow or your writing voice: methods, memory math, prices, tools, evaluation and three playbooks.
Aska Digital's IR-Taxonomy gives non-engineers five plain names for agentic systems, so a buyer knows what kind of machine they own before a demo decides for them.
Long calm periods make systems fragile less by piling up risk than by quietly removing capacities that only pay off in a crisis.