RISCON Labs / Lab overview · Status as of 25 October 2026
RISCON Labs:
Work, Assets and Road to 2030
Download PDF A4 · for email
Each cluster is one family site; the ringed point at its center is that site’s central question. Drag the field to turn it, and choose a ring or a name to read it here.무리 하나가 사이트 하나이고, 가운데 고리 점이 그 사이트의 중심 질문입니다. 끌어서 돌려 보고, 고리나 이름을 누르면 여기에 뜹니다.
01 / 08
Why we exist
Answers pass to ASI. The question stays human.
RISCON Labs assumes that artificial superintelligence shows its first signs around 2030. We prepare for that age in two ways. We build question libraries, sites where hard questions in one field are collected, structured and checked against their sources. And we train human ASI auditors: people who can question a machine’s judgment and correct it.
We are a small human team that works through AI orchestration. A human director decides what to ask and where to go. Many model agents write, translate, check and publish, under rules that keep every claim traceable to its source and every author checked by a different model family.
02 / 08
How we build
Every record follows the same cycle: gather questions, expose their premises and flaws, connect and compress them, demand an answer, audit the answer, and turn what we find into the next question.
- Source first. Writing starts from primary sources: court decisions, statutes, clinical trial records, market data. Each claim is checked against that source before it is published.
- Different families check each other. The model that writes a text is never the one that checks it for errors. A model from another family audits it.
- Publication gates. A batch goes public only when its confirmed error rate in a fresh random sample is 5% or lower. Errors that would break trust on sight are held to a stricter rule. These are a reversed holding, a wrong statute or case number, or a translation that says the opposite of its source. For these, the target is zero across all records, checked in full rather than by sampling.
- English first. New writing is composed in English, with a reviewed Korean companion. Our earlier Korean writing is being translated into English, and each translation is checked against the original.
- Corrections stay visible. When we fix a published record, the fix and its reason go on record.
03 / 08
The family of sites
Fourteen public question libraries. Eleven are built around one central question; HEBE, Cosmic Press and SIMUL open with a guiding line instead.
| Site | Field | Central question |
|---|---|---|
| Meridian | Decision research | What has to be decided now, and what change would require changing that decision? |
| PRO SE | Court without a lawyer | When I go to court without a lawyer, what must I check first and leave on the record so the judge can follow my arguments? |
| ASI 2030 | ASI scenarios | In 2030, when ASI makes the judgments, what must I ask to be able to correct them? |
| AEGIS | Defense decisions | What must be protected, and what preparation is needed within the given resources and constraints? |
| Skeptron | Signs and timing | What new perspectives and choices does the sign I have just sensed open up for my question? |
| Dionyas | Creative IP | Where, to whom and with what meaning is this work received, and what turns that reception into lasting value? |
| WSTiger | Market judgment | What is moving the market, and under what evidence and invalidation conditions does a judgment about it hold? |
| ARCHON | Who decides | In this organization, who decides, on what grounds, and under whose responsibility? |
| NADIR | Organizational decline | How far has your organization fallen, and what will decide which path it takes from here? |
| GAEKJU | People’s flaws | Do the pointless busywork and idle talk that never reach anyone above really hurt performance, and what comes out of those flaws? |
| IATHEON | Cancer research | What evidence brings targeted therapy for gastric and breast cancer closer to a cure, and what must be checked next? |
| HEBE | Obesity · hair loss | Obesity and hair loss, read one question at a time |
| Cosmic Press | Research space | Many lives. One question. |
| SIMUL | Organizational simulation | The same decision, in another company. |
Alongside them, the RISCON OA Program trains auditors on a ladder of levels from L0 to L6.
04 / 08
What we hold today
-
PRO SE, Korean Supreme Court law in English and Korean.
About 1,378 structured decision-ledger entries, 800 case and practice cards, 1,551 preparation questions, 137 legal briefs and 28 case studies. The library covers civil, criminal, violent crime, family, juvenile and military law. Each record is read against the full text of the decision. As far as we have found, no other public source offers Korean case law written in English first, as structured commentary across every field, with each record checked against its source.
-
Meridian, a library of 1,135 decision cards.
Each card is the seed of a full research thesis.
-
WSTiger, 867 market notes in English.
Each note gives its evidence and the condition that would prove it wrong.
-
IATHEON and HEBE, medical evidence libraries.
Paper summaries, clinical trial summaries and judgment commentary, read question by question.
-
Seven sites in expansion.
AEGIS, Skeptron, GAEKJU, ARCHON, NADIR and ASI 2030, together with Ordo, a new site in preparation. The first wave of 7,405 new English records is under source checking.
Counts updated 2 November 2026.
05 / 08
In progress
Translation of our Korean writing into English across the family. Source and fact checks are running. Sites go live one by one as each passes its gate.
PRO SE expansion, in monthly waves. Decision-ledger entries come first, because every other PRO SE format draws its decisions from the ledger.
Meridian theses. Before the full schedule starts, a four-thesis trial is testing how seed quality and writing rules affect the factual error rate.
A fatal-error method. We are building our own way to find the errors that would break trust on sight. Code checks against the primary source run first; a model from another family then checks only what the code flags. We measure how many planted errors the method catches, so that “zero fatal errors” is a measured claim.
Status update pending
06 / 08
Targets and schedule
| Target | Scale | When |
|---|---|---|
| English editions of existing Korean writing | All family sites | Late October to November 2026 |
| PRO SE decision ledger | 3,000 decisions | By mid-2027 |
| PRO SE cards | 10,000 (case law 70 : practice 30) | By mid-2027 |
| PRO SE briefs, case studies, law-reform studies, questions | 1,200 / 500 / 1,000 / 1,500 | By mid-2027 |
| Meridian theses | One thesis per card, 1,135 in total | In waves over about a year, then ongoing |
| IATHEON evidence base | 20,000 de-identified patient records from public cancer research datasets (gastric and breast cancer) | In progress |
| Seven-site expansion | About 31,700 new English records | In waves, from late 2026 |
| SIMUL companies | 100 fully orchestrated companies with virtual offices | Toward 2030 |
When: Status update pending
07 / 08
The SIMUL vision
SIMUL began as a simulation office: synthetic organizations whose histories, meetings and decisions are generated so we can study how organizations decide, grow and fail.
The next step moves into the real world. When a SIMUL company has a working product, it launches publicly under its own domain as a company run end to end by AI orchestration, and it says so on the site: a fully orchestrated company built by simul.risconlabs.com. A coordinating agent acts as its CEO. Its synthetic office stays behind the scenes as the back office, and what the public sees is the product, the site and the channels that carry it. We watch both at once: the office inside and the company outside.
Our aim is 100 such companies and offices. We hope at least one becomes a unicorn, measured by the value recognized in its own time rather than by the investment it has raised.
08 / 08
Principles we keep
We do not pretend. Synthetic material is labeled as synthetic. An AI-run company says it is AI-run.
We judge work by whether it is right, not by whether it persuades.
One visible error can discredit everything else, so we check hardest where an error would be most damaging.
Questions, sources, versions and audits are kept for every record. The record is the asset.