Integration is the constraint
Water-resources management is rarely limited by the absence of a model, but by how badly independently built models, datasets, scales, and assumptions combine into something a decision can rest on.
Failure modes
- Fragmented ownership
- Models are built independently by the specialists who understand them best, then asked to interoperate through spreadsheets and file handoffs that preserve none of their structure.
- Incompatible semantics
- The same term means different things across disciplines. Demand, storage, loss, and reuse are each specific to whoever defined them, and those definitions rarely travel with the numbers.
- Scale mismatch
- Daily against monthly, reach against basin, parcel against region. Reconciling resolution is treated as a preprocessing chore rather than an analytical decision.
- Flattened uncertainty
- Ranges collapse to point values at every boundary crossing. By the time analysis reaches a decision-maker, the confidence quantified upstream has been quietly discarded.
- Lost provenance
- Six months after a recommendation, no one can reconstruct which assumption, which dataset version, or which model run produced a specific number.
- Integration latency
- When re-running a connected analysis costs weeks, teams ask fewer questions. The real loss is not the delay but the analyses never attempted and the dependencies never found.
Changed information does not propagate through connected systems, so knowledge gaps and unrecognized dependencies stay hidden until after the decision is made.
Shared foundations around independently owned science
The answer is not one model to replace the others. It is a shared foundation that specialist models can connect through while their authors retain scientific control.
- Shared semantics
- Common definitions and versioned data contracts, so a quantity means the same thing on both sides of a handoff.
- Retained ownership
- Domain specialists keep authorship and scientific control of their own models. Integration is a contract to meet, not a takeover.
- Explicit uncertainty
- Uncertainty crosses boundaries as a first-class property rather than collapsing into a single number at each step.
- Provenance and lineage
- Every output traces back to its inputs, assumptions, source documentation, and the model version that produced it.
- Versioned evidence
- Planning versions are preserved rather than overwritten, so past decisions stay reconstructible in the context that produced them.
- Auditable workflows
- The path from evidence to recommendation is inspectable by someone who was not in the room when it was built.
A working foundation, not a slide
These principles are not aspirational. Pacaya Group has designed and implemented the foundation of a modular platform built around them, connecting independently developed models, datasets, evidence, and decision workflows.
The internals stay private — the work is proprietary and still developing — but the shape of it is the point: shared contracts and semantics, versioned evidence, uncertainty that survives handoffs, and decision outputs auditable back to their sources.
It grew out of a research problem rather than a software premise. The goal was never to replace specialist science, but to remove the integration tax that stops specialist science from adding up.
A structure teams can grow into
Separating shared integration infrastructure from specialist science does more than speed up analysis: it gives a multidisciplinary team a way to grow without losing coherence at the boundaries.
Specialists can lead their own domains, publish under their own names, and mentor early-career researchers while still contributing to a common analytical foundation through shared contracts, traceable assumptions, and versioned evidence.
As traceable models, observations, and validated outcomes accumulate, that foundation becomes the substrate for the next generation of methods — hybrid physics-and-learning approaches, surrogate modeling, anomaly detection, and decision support that can show its work.