Medium risk
Baseline routing
+4.1%
High touch+5.0
Digital first+4.4
At risk+2.3
App Data OS · Synthetic populations
Test the change before it touches a real person.
Shape a governed population, run the intervention, and walk away with a signed decision record your reviewers can defend under audit. This is a checked-in workflow preview. In a pilot it is scoped to your own data.
Seed metrics · scoped workspace metrics in a pilot
Seeded populations3
Healthcare access, finance servicing, and public-sector eligibility cohorts are available in the seeded preview workflow.
Experiments tracked9
Live and completed intervention tests remain linked to run detail and outcome distribution views.
Decision exports6
Approved manifests preserve methodology, reviewer chain, results, and download state.
Cohort builder
Shape a governed population before you test an intervention.
Sliders set cohort size and privacy bounds. Constraint chips and the distribution preview show every suppression decision in the open, so no small segment leaks into a result you cannot defend.
Healthcare access cohort
calibrated · suppression warnings active
Agents1860
Minimum k12
Privacy epsilon1.6
no singleton cellsage balancedmanual review on small segments
High touch34%
Digital first28%
At risk20%
Rural12%
Appeals prone6%
Population cards
Three scenarios, each ending in a decision you can defend.
Healthcare
Healthcare access
Which patient segments are most likely to miss follow-up care after a policy change?
Calibration
91%
Last run
2026-04-22
Navigator outreach improves completion for high-touch care by 12 points while flagging transport-constrained segments for manual review.
Finance
Finance servicing
How do borrower cohorts respond to hardship routing and collections-language changes?
Calibration
88%
Last run
2026-04-21
Hardship-first routing lowers escalation risk by 9 points without widening adverse-action variance.
Public sector
Public-sector eligibility
What happens to uptake, appeals, and processing load when eligibility rules shift?
Calibration
90%
Last run
2026-04-20
Simplified eligibility copy increases uptake by 7 points with a manageable appeals-load increase.
See where the change helps, and where it does harm.
High touch
74%
18%
7%
1%
Digital first
68%
22%
8%
2%
At risk
49%
25%
19%
7%
Rural
52%
29%
14%
5%
Appeals prone
46%
28%
18%
8%
ImprovesNo changeNeeds reviewNegative shift
Re-identification risk
k
12
l
4
t
0.18
epsilon
1.6
delta
1e-6
agents
1860
Re-identification risk: LOW
Score 0.21. Current profile is within launch bounds. Keep k at 12 or higher for small segment exports.
Compare the variants side by side.
Low risk
Navigator + relief
+11.8%
High touch+14.2
Digital first+7.1
At risk+9.0
Selected outcome comparison
High touch
+5.0Baseline routing
+14.2Navigator + relief
Digital first
+4.4Baseline routing
+7.1Navigator + relief
At risk
+2.3Baseline routing
+9.0Navigator + relief
Selected cell: At risk + Navigator relief improves by +9.0 with low re-identification risk and policy approval attached.
The signed decision record
Every export carries the method, the result, and the reviewer chain that approved it, sealed to a digest you can verify later. That is what survives examination when the August 2026 EU AI Act provenance clock asks where the decision came from.
Approved · spm_7f3b2c91a80e
Seeded cohorts, calibrated before intervention, compared against baseline routing, reviewed for segment-level fairness and safety signals.
Navigator + relief variant selected for limited rollout with watchlist on appeals-prone segment.
Simulation owner
Privacy reviewer
Policy approver