Hub architecture: hubs are not measurement targets — the district is
Attribution
Prior state
Earlier framings (in places) treated 12 community hubs as themselves
the measurement target — fab-lab logs functioning as a proxy for the
district's waste / energy / material throughput.
Change
Four hubs, one per pilot (Barcelona · Boston · Santiago · Bali).
Hubs are NOT measurement targets. They are:
(1) Harvesters of municipal/district open data
(Open Data BCN, Analyze Boston, Santiago Metropolitan Region
portals, Bali Satu Data).
(2) Sensor-campaign operators filling gaps where municipal data
is silent (≥30-day colocations + longer-term deployments).
(3) Local-intelligence instantiators running quantised SLMs with
federated weights and Matryoshka embeddings.
H₀-T signal source = municipal/district open data, harvested by
the four hubs and ingested into bioregion servers. Treatment unit
is the district anchored by the hub. Synthetic control is built
per pilot from similar non-hub districts.
In-grant test = community/district scale. Bioregional H₀-T is a
designed post-grant deliverable on month-60+ data.
Reasoning
Fab labs do not produce community-scale throughput data; the
district does. Conflating the two created an unfixable
measurement-design problem and would have been visible to any
serious reviewer of the methodology.
The corrected mechanism — hubs as the human + technical
infrastructure that turns sovereign municipal data into a
federated learning substrate, with sensor campaigns filling gaps —
preserves PLANETAI's distinctive contribution (federated AI causal
inference at bioregional scale) without pretending fab-lab logs
are a waste-proxy. It also aligns the experimental unit with the
unit at which the cities themselves report data, which is what
partner cities can actually contribute.
Artifacts changed
- planetai_publish/core-ideas-paper.md
- planetai_publish/observatory/index.html
Related decisions
Edit this decision on GitHub →
New here? See how to contribute →