Controlled synthetic benchmark

What happens when heat raises need, but care stays finite?

The synthetic benchmark exists precisely because it is not a city. One thousand modelled residents run through seven scenario contrasts, thirty paired replications each, so every mechanism — heat, seeking, access, capacity — can be inspected in isolation from real-world evidence.

Synthetic benchmark Not Barcelona data Model-conditional
Reference conditions
schematic view · residents distributed for visual clarity
Schematic replay · 1,000 study residents
Day 1 / 7
Care need0
Served0
Delayed0
Unmet0

About the replay: the animation makes the model easier to read. Residents move slightly to reveal the city layout, while care-need cases follow highlighted service pathways.

Current release: an inspectable model study.

The population, facilities, climate and scenarios are synthetic. They make the current implementation concrete while empirical calibration and historical validation remain future work. These values describe model behaviour under controlled assumptions — not intervention effectiveness in any real city.

Run provenance

Frozen source commit 5a576e5 · parameter set climate_health_illustrative_v2 · 1,000 synthetic residents · 30 paired replications per scenario · regenerated 21 August 2026 with the current engine, including RES-001 fractional-capacity resolution and FAC-001 equal-cost tie resolution. No empirical calibration or external validation.

See what changes Open the underlying run dataCompare with Barcelona
01 / Exploration

Start with a city small enough to understand.

Before any real population enters the model, the mechanism needs a population small enough to inspect. These 1,000 synthetic residents vary in vulnerability, transport and location, which exposes whether the pathways behave coherently.

1,000study residents
—aged 65+
—high vulnerability
—limited transport
8service facilities
01Heat exposure
02Vulnerability
03Care need
04Care seeking
05Spatial access
06Finite capacity
07Served / delayed / unmet
02 / Description

What changes when one part of the system changes?

Here the same population runs under different conditions. Each scenario uses 30 paired replications over a 30-day analysis window after a 14-day warm-up. The values are means across runs; the ranges show run-to-run variation within the current model.

ServedDelayedUnmetBars compare only these three displayed ledger categories.

Do the patterns differ across socioeconomic groups?

Mean unmet person-days per 1,000 need person-days across 30 replications.

Subgroup pattern
A null result worth reading. Targeted transport support alone (CHRS-S4) leaves the unmet-need rate statistically indistinguishable from extreme heat alone in this configuration. The mechanism activates, but capacity — not transport — binds here. Only when combined with extended hours and mobile units (CHRS-S5) does access support translate into materially less unmet care. That is what a mechanism-isolating benchmark is for.
03 / Explanation

Averages hide the path that produced them.

R

Study resident

Loading profile…

One resident is shown because aggregate rates can hide the mechanism. Following the same model resident across two scenarios makes it easier to see how access, disruption and capacity combine. The profile comes from the generated study population; the two paths are illustrative narratives consistent with the aggregate results, not recorded individual trajectories.

CHRS-S6 · Compound heat + electricity disruption

High heat-related need is generated by the engine.
The resident would normally use their nearest service.
Electricity outage closes five of the eight fixed facilities (no backup power).
Limited transport constrains alternative access.
Outcome: —

CHRS-S5 · Combined access package

The paired resident experiences care need in the same population.
Extended hours raise effective capacity.
Mobile units and targeted transport support are active.
A service pathway remains available.
Outcome: —
04 / Inspect the run

Check the run before drawing conclusions.

The files below expose the population, frozen scenario assumptions, all 210 replication results, equity tables across seven subgroup dimensions and full run provenance for CHRS run 20260821T094703766361+0000. They are here so the run can be checked, reproduced and questioned.

Study populationCSV · 1,000 residents FacilitiesCSV · 8 services Experiment definitionYAML · frozen n=1000 design Replication resultsCSV · 210 paired replications Scenario summaryCSV · means and stochastic ranges Equity resultsCSV · seven subgroup dimensions Run manifestJSON · provenance and limitations Data READMEMD · provenance and interpretation Source repositoryGitHub · engine and docs ↗
What this run can support These files show how CHRS behaves under its current assumptions. Connecting those assumptions to observed evidence — need anchors, utilisation targets, operational capacity — is the next research stage, documented on the Barcelona pilot page.