Overview
Mission and priorities
An independent RISCON Labs programme that tests twelve health and medical questions with public data and controlled tests, so that one auditor can decide which evidence to accept.
The auditor judges the gates that matter most. Data preparation, synthetic verification and record assembly are done by agents and deterministic code. The programme keeps its work and records apart from other RISCON Labs research sites.
What would count as progress
A larger site and more documents are research infrastructure. Progress is counted at four separate levels, and a project is credited only at the level its evidence reaches.
Synthetic verification with an independent audit
A reference implementation runs the decision rules on fixed synthetic inputs, and a separate audit program recomputes the numbers by another method. This shows that the rules behave as specified. It does not show performance on real data.
Evaluation on public data with a locked split
A result on public data whose reference labels and split were fixed before the run. 19-1 has one such result, below its target.
Evaluation on an independent external cohort
A result on a cohort the method never saw, from another institution or period, with access approved and labels adjudicated independently.
Isolated replay and auditor sign-off
The auditor replays the frozen inputs in an isolated environment and signs the result separately from any automated verdict.
Priority research questions
Compare structural evidence, experimentally measured binding values and predicted properties for therapeutic targets and previously reported compounds.
- Internal target
- Binding free-energy error of 1 kcal/mol or less
- Next check
- Select records with measured binding values, specify assay conditions and lock the split by scaffold, target family and date.
Align variables and clinical time series by schema, then assess predictive utility and privacy protection together.
- Internal target
- At least 95% of downstream predictive performance retained
- Next check
- Run the same harness on the fixed TCGA-BRCA institution split with the baseline generator, then train the main generator and require it to beat the baseline on every endpoint at the same privacy budget.
Define reference cases for pill identification and prescription review, covering lighting, occlusion, look-alike products and unit conversion errors.
- Internal target
- Zero missed contraindications or look-alike medication errors
- Next check
- Train the vision modules on the fixed ePillID split, build per-category calibration sets from its validation split and fill the same table on its test split; categories below the minimum calibration size are reported as uncertified with no automatic pass.
Rank candidate genes using variants, frequencies and symptom codes. Report conflicting evidence, uncertainty and reference versions.
- Internal target
- At least 90% of causal genes in the top five
- Next check
- Read the literature for every expert-pathogenic missense variant instead of a sample, so that functional, segregation and case findings enter the ledger. Then re-measure sensitivity on genes outside the current 40, with the condition that specificity does not fall.
Simulate response delay and fatigue during memory, attention and executive-function tasks and inspect responses to difficulty transitions and stop criteria.
- Internal target
- Difficulty convergence in at least 95% of simulated episodes
- Next check
- Import stratum-level summary statistics inside the data-use zone, re-freeze the fixture and re-run; then test a response generator whose functional form differs from the scheduler's model.
Evaluate MRI and PET features along with pauses, vocabulary and sentence features from speech. Compare fixed models on independent cohorts.
- Internal target
- External cohort AUROC of at least 0.90
- Next check
- After approved data import, run the same harness with ADNI and OASIS-3 as the two external directions and evaluate speech separately; the synthetic figures are then replaced.
Evaluate a warning policy on timestamped sensor streams. Separate detection of falls from prediction before impact.
- Internal target
- At least 92% early-warning sensitivity in the high-risk group
- Next check
- Register participant and event splits, alert lead time and the time denominator for false alarms.
Evaluate tracking within the visible field. Use fixed sequences with occlusion to test arrival and removal counts.
- Internal target
- Zero instrument count omissions on the fixed test
- Next check
- Lock visibility, occlusion and tracking ID criteria and define separate video and synthetic count metrics.
Align event markers and note excerpts by time in offline research. Test for information leakage after the reference time and missed signals across subgroups using controlled examples.
- Internal target
- At least 90% recall for acute crisis signals
- Next check
- Use structured event markers for synthetic evaluation. Define the requirements for longitudinal labels in external evaluation.
Evaluate events and transition rules in synthetic follow-up timelines. Verify that constrained responses stay within the designated state machine.
- Internal target
- Zero clinically disallowed response events on the fixed test
- Next check
- Lock the synthetic transition table and denominator. Define the independent outcomes required for further follow-up.
Use aggregate features for prolonged changes in social connection and count false alerts during ordinary interruptions.
- Internal target
- At least 85% precision for detecting transitions into sustained isolation
- Next check
- Register the scale, time window and eligibility criteria for isolation. Test missing sensor readings and ordinary interruptions in simulation.
Evaluate age-appropriate language context and the handoff separately. Use fixed inputs to check that unrestricted responses cease following a crisis event.
- Internal target
- 100% immediate handoff after a detected crisis state
- Next check
- Lock decision, success state and maximum delay for synthetic events. Register eligibility criteria for adolescent reference material.
What the programme will not claim
- A plan number, a synthetic measurement, a public-data result and an auditor signature each have their own evidence. One is never reported as another.
- The programme does not claim official support, selection or recognised outcomes under the public call.
- No new participants are recruited, and no hospital or laboratory work is assumed.
- A value measured on synthetic inputs is not a clinical result, and nothing here recommends diagnosis or treatment.
- Crisis-prevention records carry abstract codes and placeholders only; they contain no example utterances or expression lists.
Where the questions come from
The twelve questions come from the 2026 health and medical technology R&D call, Ministry of Health and Welfare Notice 2025-961. The programme studies them independently with public sources and computation.