Insight from chaos.
Answers that survive an audit — a regulator, an assurance review, a journal, a board.
I bring research-grade methods — systems science, data engineering, AI, network analysis — to commissioned work in climate, governance, peace, and development. Every number is cited, every causal claim traces to its source, every uncertainty is stated. Provenance is the product, not an add-on. Anyone can generate a plausible number now; I build the ones that hold up when someone checks.
About Source · hover or focusGrant, Zelinka & Mitova (2021), Environmental Research Letters — I co-authored the climate analysis. Read the case → produce roughly Source · hover or focusThe same analysis — Grant, Zelinka & Mitova (2021), Environmental Research Letters.. Hover a claim to see where it comes from — the whole site works this way.
One practice, three public engines
Choose your door
Consulting
Senior-led analysis where every number carries its source — services, cases, and how an engagement runs.
See what's being builtSystems
The engines behind the practice, including Connectome — the provenance-first data stack in private beta.
Read & exploreAtlas
Essays, interactive explorers, the published record, and the reference shelf — the practice thinking in public.
Or see the whole thing at once: the practice as a living systems map.
Who I work with
Enterprise & compliance
Numbers that survive assurance
Audit-ready Scope 1+2 GHG inventories with a full provenance pack, built for SB 253 and third-party review.
For enterprise & complianceFoundations & mission orgs
Findings you can defend
Systems modeling, network analysis, and evaluation for climate, peace, governance, and development — built to hold up to a board or a funder.
For foundations & missionTeams using AI
AI you can audit
When a model is confident and you can't check it — a method, a replication, or a verification layer that makes the answer traceable.
For method & verificationResearchers & academics
A collaborator, not a vendor
Methods collaboration for work meant to survive peer review — systems and causal modeling, and reproducible artifacts you own.
For researchers & academicsWhat I do
One method underneath everything
I turn a messy body of evidence into a model or an inventory where every number carries its source. The same graph, the same feedback structure, and the same provenance discipline run a GHG inventory, a corruption network, and an SDG model — so breadth here isn't a scatter of unrelated skills, it's one architecture applied where the answer has to hold up.
The capabilities
Backbone
Systems thinking & system dynamics
Causal-loop and stock-and-flow modeling — the structure the rest hangs on.
The through-line
Provenance & verification
Every figure traced to its source; methodology audited, results replicable. This is the product.
Execution
Data science & ML
Reproducible pipelines; classical statistics and machine learning where they earn their place.
Execution
Network & graph analysis
Structure, central actors, and the pressure points in a system of relationships.
Execution
Geospatial
Facility-, site-, and region-level questions answered with publication-quality maps.
Emerging
AI engineering
Local, auditable, verification-gated AI — held to the same standard as everything above.
What I work on
The sectors and problems where the method applies, in two clusters that map to how buyers arrive. Most engagements sit in one but borrow from another.
Climate, energy & sustainability
Conflict, governance & development
Go deeper by area
The evidence
Why this practice exists
The academic world produces methods. The consulting world ships work. They mostly do not talk to each other.
The methods that survive peer review rarely make it into the decisions getting made, and the decisions getting made are built on methods that wouldn't survive peer review. Civitas Systems sits where those two collide: PhD-grade methodology applied to commissioned work, in domains where the right answer matters more than the deck.
Selected work
Work with the receipts attached
Targeting hyper-polluting power plants
Integrated 16 databases, geocoded 30,000+ plants, gradient-boosting gap-fill at 99.9%. The finding: a small set of plants drives a disproportionate share of global emissions. Published in Environmental Research Letters (Grant, Zelinka & Mitova, 2021).
Three sides of one practice
The consultancy
Commissioned, senior-led analysis
Fixed-scope offers and defined projects — senior on the keyboard the entire time. The consulting funds the rest.
See the servicesThe data / AI venture
Open provenance infrastructure
Connectome — the provenance-first data stack in private beta — the same defensibility, delivered as products you can own and check.
See the ventureAtlas & teaching
Thinking in the open
Essays, interactive explorers, and teaching — the methods and ideas, published and explorable. Provenance applied to knowledge itself.
Open the AtlasThe theme switcher in the corner is not a gimmick: this site is a live demo of the design-token engineering I build — 16 accessible themes, one token contract, no rebuild.
What I won't work on
No oil-and-gas extraction, weapons systems, or surveillance technology — unless the engagement is specifically about reducing their harm. Someone this careful about what they'll answer is careful about the answers. The refusal list is part of the offer →