# Synthetic study n=1000 — public data package

Controlled synthetic benchmark generated with the **current CHRS engine**
(repository commit `5a576e5b887723d4480b205aede460f5a90b5164`, 21 August 2026).
It supersedes the archived 13 August 2026 public run, which predates two
generic allocation remediations (RES-001 fractional-capacity resolution and
FAC-001 equal-cost facility-tie resolution).

Everything in this package is synthetic. No Barcelona or other real-city data
is included, and no result here is an estimate of any real population.

## Run identity

| Field | Value |
| --- | --- |
| Run ID | `20260821T094703766361+0000_urban_heat_website_n1000_8cae197593` |
| Profile | `urban_heat_website_n1000` |
| Configuration hash | `8cae197593ab19b2e4a0efaf825a874720e08a00843e3f9d45d32244627f61a5` |
| Source commit | `5a576e5b887723d4480b205aede460f5a90b5164` |
| Parameter set | `climate_health_illustrative_v2` |
| Base seed | `20260101` |
| Population | 1,000 synthetic residents |
| Scenarios | CHRS-S0 … CHRS-S6 |
| Replications | 30 paired replications per scenario (210 total) |
| Warm-up / analysis | 14 days warm-up · 30-day analysis window |

The configuration hash is identical to the August 2026 run: the frozen
experiment configuration was reused unchanged; only the engine code moved
forward.

## Files

| File | Contents |
| --- | --- |
| `run_manifest.json` | Full engine manifest: provenance fields, artifact hashes, limitations. |
| `scenario_summary.csv` | Scenario-level means, medians, stochastic ranges (p2.5–p97.5) for key outcomes. |
| `scenario_replications.csv` | All 210 replication-level analysis-window results. |
| `equity_results.csv` | Subgroup person-day tallies and rates across all seven subgroup dimensions, with precision warnings. |
| `synthetic_population.csv` | The 1,000-resident synthetic population (attributes fixed at synthesis; display coordinates are schematic). |
| `facilities.csv` | The 8-facility synthetic service network. Display coordinates are schematic; capacity semantics are model-internal. |
| `config-snapshot/` | Frozen scenario YAMLs (S0–S6), parameter registry snapshot, profile snapshot, and the experiment file used to launch the run. |

## How this package was produced

1. `chrs run --config experiments/urban_heat_website_n1000.yaml --output-root <v2 work dir>`
   executed against the current repository working tree (commit above), with
   outputs redirected outside the research repository.
2. `chrs analyze <run_id>` produced the summary tables.
3. A website build script (`work/build_website_data.py`, distributed with the
   site source) re-synthesised the population deterministically from
   `base_seed=20260101`, replication 0 — the same construction
   `DailySimulation` performs — and assembled this package plus the compact
   JSON used by the demo page (`assets/data/chrs-demo-data.js`).

Derived display-only fields, documented so nothing is silently relabelled:

- `vulnerability_tier` — display bins on the synthetic vulnerability score
  (Low < 0.33 ≤ Moderate < 0.66 ≤ High).
- `transport_access` — Good if the resident has a private vehicle, Limited if
  transit-dependent without a vehicle, otherwise Standard.
- `x`,`y` coordinates (population and facilities) — schematic layout derived
  from residential zone plus a stable per-agent offset. Not model geography.
- Replay pacing values in the demo page (heat glow intensity, animation
  speed) are illustrative visual knobs scaled from scenario tmax means; they
  carry no scientific meaning.

## What these numbers can and cannot support

These values describe how the current model behaves under controlled,
declared assumptions. They are useful for inspecting mechanisms — how heat
raises need, how finite capacity converts demand into served, delayed and
unmet care, and how intervention mechanisms change the pathway. They are not:

- estimates of intervention effectiveness in any real city;
- epidemiological estimates of need, unmet care or capacity;
- validated predictions of any kind.

Monte Carlo intervals describe stochastic simulation uncertainty under frozen
synthetic assumptions, not real-world confidence intervals.