unseea

A UN SEEA-EA decision planning tool built on the geo-agent runtime and the GLEN h3 data catalog.

SEEA Ecosystem Accounting is the UN statistical standard (adopted March 2021) for organising biophysical and economic information about ecosystems into five reconciling accounts. The standard ships a worked toy example, SEEALand — 250 hectares, six ecosystem types, one land-cover conversion, carried through all five accounts.

This app is SEEALand for any polygon on Earth, with real data, computed live — plus the scenario branch the standard never runs. Draw an area, get its five SEEA accounts; then propose a land-use change and watch the accounts recompute into the standard's own vocabulary: managed expansion, managed reduction, degradation, enhancement, revaluation.

Status

Phase 1 — the account library. The five accounts are implemented behind a nine-call API and reproduce SEEALand end to end against the standard's own published figures, with no network. No application code yet. Start with DESIGN.md, then the roadmap in issue #20.

import unseea

eng  = unseea.connect(fixture="research/seealand-fixture/")   # or local=True, or mcp=url
eaa  = eng.eaa()
acct = eng.asset_account(eaa, 2020).check()

acct.tables["asset"]   # the SEEA-shaped table: entries down, ecosystem types across
acct.sql               # the exact SQL that produced it, re-runnable
acct.checks            # opening + entries = closing; area + volume + price = ΔNPV
acct.provenance        # executor, parameters, ECT coverage, sources, licences

$ pip install -e ".[test]" && pytest -q
157 passed

Call	Returns
connect(fixture=… / local=True / mcp=url)	An engine bound to an executor
eaa(...)	An accounting-area handle
extent_account(eaa, date)	Area per ET: opening, additions, reductions, closing
extent_change_matrix(eaa, d0, d1)	The from→to matrix, with margins
condition_account(eaa, date)	All three stages: variables, indicators, index
landscape_metric(eaa, variable, k)	C1 metrics over a k-ring — needs an h3 executor (#7)
services_physical(eaa, date)	Physical supply and use
services_monetary(eaa, date, prices)	Monetary supply and use, and GEP
asset_account(eaa, d0, d1)	NPV per ET plus the area/volume/price decomposition

The fixture executor is the conformance tier: it runs the SEEALand CSVs through DuckDB with no network, and CI fails loudly if GEP drifts from $83,125 or forest ΔNPV from −$116,366. The local and mcp executors are declared but not yet implemented — each raises an error naming the issue that blocks it.

📖 Documentation site → — begins with a walkthrough of SEEALand, the standard's worked toy example, traced through all five accounts.

Phases 1–2 need no new data ingest — deliberately, so the account-arithmetic correctness work runs in parallel with data acquisition rather than behind it.

Contents

Path	What
src/unseea/	The account library: five accounts as SQL, three executors, parameter sets and vocabularies
tests/	The conformance tier: SEEALand reproduced cell by cell, plus the library contract
DESIGN.md	The design document: what SEEA EA is, how ARIES for SEEA compares, the hard problems, the interaction model, and a phased plan
ARCHITECTURE.md	How the system is layered: the account-compiler contract, the binding/parameter/vocabulary seam, licence and private-data policy, test tiers, and the rule deciding what belongs here versus in an application built on it
DATA.md	Best-in-class data acquisition plan: UNSD Tier framework, per-account source selection, licence analysis, and how it lines up against the data-workflows tracker
PROVENANCE.md	The planned layer list, with justification: for each layer, is it named in SEEA/UNSD guidance, used by ARIES, used in a published national account, or our own judgement — plus the roadmap issue tracking it
research/seea-ea-reference.md	Distilled controlled vocabularies and table structures from the standard — the source for system-prompt.md and the app's lookup tables
research/	Source documents: the SEEA EA standard (official 2024 ed.), UNSD Guidelines on Biophysical Modelling, UNSD Monetary Valuation technical report, and the official SEEA supplements — including the SEEALand worked example as CSV test fixtures

Why this stack

SEEA builds accounts up from basic spatial units and states that its area-weighted aggregation is invariant to the resolution the data are collated at (SEEA EA (2024 official ed.), Annex I, Table AI.5, footnote c). An H3 tessellation with area-weighted rollup is therefore not an approximation of the SEEA spatial model — it is that model, implemented. Every account reduces to a GROUP BY over hex cells joined to an accounting area, which is what makes live scenario recomputation feasible at all.

Nearest prior art in the lab

landscape-frontiers — same pattern: fork geo-agent-template, configure three files (layers-input.json, system-prompt.md, k8s/), keep the computation in SQL against the duckdb-geo MCP, and let fidelity scale with the catalog ingest.