Population generation
Production of 100 million synthetic records across 1,000 controlled batches.
MADRE-H360 / Public research brief / July 2026
From population-scale engineering to life-course architecture.
H360 v1 established a governed experimental population of 100 million synthetic records and demonstrated that VEHICLE Systems Lab can generate, batch, preserve, inventory and query synthetic data at population scale. A formal scientific review now moves the project into a more demanding phase: H360 v2 will restart from the first synthetic human and build coherent lives before rebuilding scale.
H360 v1: preserved / read-only
100M records: retained for bounded research use
H360 v2: architecture under construction
MADRE: governed reader, not life generator
100Msynthetic H360 records preserved as the experimental v1 population.
1,000controlled batches of 100,000 records each.
0missing batches in the completed population inventory audit.
v2life-course architecture begins from a clean generative order.
Scientific transition / Truth before sunk cost
We will not protect a version because of the time invested. We will protect the truth of the project. H360 v1 is preserved as a learning laboratory. H360 v2 begins as a life-course architecture.
The objectives, constraints and omissions discovered during H360 v1 are not discarded. They become the scientific requirements of H360 v2. VEHICLE Systems Lab will continue the work with discipline, transparency and sustained commitment until identity, biography, family, territory, education, health, work, economy, memory and visual continuity can exist together through time.
H360 v1 / Verified operational foundation
What H360 v1 achieved
H360 v1 is not being deleted or disowned. It remains a material research result. Its value is strongest where evidence exists: large-scale synthetic-data engineering, governance, traceability and pre-deployment testing.
Preserved originals
Original population records remained immutable while derived research layers were created separately.
Traceable lineage
Identifiers, manifests, ledgers, hashes and batch-level records support reproducibility and audit.
Cloud-scale storage
Private storage, inventory structure and governed metadata-access routes were established.
Derived-layer experience
Habitat, memory and economy work revealed both the power and limits of independent injections.
Scientific self-correction
The laboratory detected an architectural limit before presenting v1 as a complete model of human life.
The closure concerns the independent-layer generative order. It does not invalidate the population engine, batching system, cloud infrastructure, lineage controls or the lessons produced by the research.
Scientific boundary / Claims must match evidence
What H360 v1 can support - and what it does not claim
Supported scope
- Population-scale data-pipeline and storage experiments.
- Batching, indexing, sampling, inventory and retrieval tests.
- Lineage, manifest, ledger and hash-verification workflows.
- Governed metadata-access and interface experiments.
- Synthetic scenario preparation before expensive deployment.
- Research into contradictions produced by layered data systems.
Not claimed
- A fully validated model of human life.
- Complete longitudinal biographical coherence.
- Clinically valid psychological or cognitive diagnosis.
- Joint causal validity across education, occupation and economy.
- Consciousness, sentience or equivalence to real persons.
- External validation against representative real-world populations.
VEHICLE Systems Lab ecosystem
Data-conscious synthetic humans.
Test before investing.
H360 v1 remains a governed, read-only synthetic research population for testing data architecture, storage, indexing, routing, interface behavior and deployment assumptions before organizations commit expensive compute, infrastructure or product capital.
“Data-conscious” means designed with explicit attention to provenance, lineage, uncertainty, governance and data limitations. It does not mean conscious or sentient.
Bounded use / 100 million retained records
The 100 million remain useful when the question is correctly framed
Infrastructure sizing
Estimate storage, partitioning, compression, indexing and retrieval behavior before procurement or scale-up.
Data-pipeline stress tests
Exercise ingestion, transformation, validation, batch orchestration and recovery procedures at realistic volume.
Governance rehearsal
Test manifests, provenance, audit logs, access policies and rollback without exposing private real-person data.
Interface evaluation
Evaluate dashboards, search, sampling and aggregate access across a large fictional population.
Model-routing experiments
Benchmark embeddings, clustering, retrieval, routing and controlled aggregation over a large synthetic corpus.
Pre-investment contradiction tests
Detect data-model weaknesses before they become expensive infrastructure, integration or product failures.
H360 v1 should be used as an experimental synthetic-data population with declared limits, not as proof that 100 million complete human lives have already been modeled.
H360 v2 / Life-Course Architecture
H360 v2 will begin again from the first synthetic human.
The new population will not be assembled by assigning independent thematic labels and reconciling them afterward. Each person will emerge from one temporal life-course kernel connected to a finite world, households, institutions, events, constraints and opportunities.
World before population
Territories, households, resources, institutions and capacity must exist before people are generated at scale.
Life before labels
Current attributes must be consequences of ordered events, not isolated assignments.
Domains evolve together
Education, work, economy, health, family and psychology interact through time.
No false precision
Evidence, assumptions, inference and stochastic variation remain distinguishable.
Biography grounds appearance
Image, video and 3D systems receive a structured scene contract derived from life state.
Scale only after gates
One human, one family, 1K, 10K, 100K, 1M, 10M and 100M - only after validation.
Clean reconstruction / No patch chain
The new generative order
Foundation
Constitution of the H360 world, historical period, laws, territories, institutions, capacities and scarcity.
Households
Reciprocal family relations, homes, caregiving, resources, languages and territorial location.
Birth and baseline
Identity, cohort, body, health baseline, functional capacities, inheritance and early context.
Life events
Care, education, illness, migration, relationships, work, crises, opportunities and decisions through time.
Derived states
Education, occupation, economy, memory and present conditions are calculated from the event history.
Visual contract
Scene generation receives age, context, clothing, environment, activity, relationships and uncertainty.
Scale gates
No increase in population size is approved while micro, family, institutional, temporal or statistical coherence fails.
NVIDIA relevance / Proposed experimental direction
Why accelerated computing matters more after the scientific reset
The revised challenge is not merely to store 100 million rows. It is to generate, validate, update and query large temporal graphs while preserving deterministic lineage, uncertainty and cross-domain constraints. This is a substantially harder and more defensible accelerated-computing problem.
Temporal population simulation
Parallel event processing across individuals, households, institutions and territories.
Constraint validation
Large-scale detection of impossible ages, relationships, capacities, transitions and institutional overload.
Graph and memory workloads
Neighborhood retrieval, temporal aggregation, embeddings and governed multi-perspective queries.
Visual consistency evaluation
Batch validation of scene contracts, identity continuity and multimodal outputs across life stages.
This page describes a research direction prepared for NVIDIA Innovation Lab consideration. It does not state or imply NVIDIA endorsement, validation, funding or partnership.
MADRE / Governed access boundary
MADRE reads validated life layers. MADRE does not assign the lives.
VEHICLE/H360 designs, generates and validates the synthetic world and its life-course population. MADRE receives governed read access only after a version is frozen, registered and audited. This separation protects causal provenance and prevents the query layer from silently rewriting the population it interprets.
Documentation / Public English edition
Scientific Closure and Architectural Restart Report
The report documents the work completed, the evidence boundary, the reason for the reset, the assets preserved, the H360 v2 constitution and the validation gates required before scale is rebuilt.
VEHICLE H360 - Scientific Closure and Architectural Restart v1.0
English public research edition. PDF. July 2026.
Public safety and research integrity
What this page does not expose
No raw population
No individual H360 population records are published here.
No factories or seeds
Private generation code, prompts, seeds and restricted operational assets remain private.
No false human claim
The project does not claim sentience, clinical validity or equivalence to real persons.