Rewrite event paper around Cool City heat-safe routes
This commit is contained in:
1 parent
b99f5ee2d4
commit
8e598dea25
5 files changed
+160
-157
No files matched your search
@@ -1,214 +1,170 @@
|
||||
# PulseCast: Turning a Noisy Footfall Index into a Weekly Staffing Plan
|
||||
# Cool City: A Public Heat-Safe Route and Cooling-Gap Layer for Heilbronn
|
||||
|
||||
**Author:** Noel Sabia (ORCID 0009-0008-3064-7573)
|
||||
**Affiliation:** —
|
||||
**Date:** 14 November 2026
|
||||
**Artifact:** https://pulse.noelsabia.dev · **API:** https://api-pulse.noelsabia.dev · **Code:** to be released after the hackathon
|
||||
**Prototype:** https://pulse.noelsabia.dev · **Cool City route:** `/cool` (bundle TBD) · **DOI:** TBD
|
||||
|
||||
## Abstract
|
||||
|
||||
Inner-city store managers must decide next week's shifts without knowing next week's footfall. We present PulseCast, a decision-first staffing copilot built on eight passive Bluetooth footfall counters in a ~580 m × 280 m cluster in Heilbronn, Germany. The system fuses five months of 15-minute device counts with Open-Meteo weather and Baden-Württemberg public holidays, trains a global gradient-boosted model on a time-based split, and converts the forecast into a weekly shift plan through a transparent staffing heuristic. On a 28-day holdout, the weather-aware 7-day model reaches MAE 3.32, WAPE 27.5%, and R² 0.777, versus 5.40, 44.7%, and 0.462 for a seasonal-naive baseline; the day-ahead model reaches 3.36, 27.8%, and 0.763. The plan is generated in 19 ms cold on v1.1 (contract < 2 s), and the deployed system handles 120/120 concurrent SSE clients with zero errors and a p95 event latency of 331 ms. A code-level audit of the public 2026 firmware commit e8be8bf shows a ~10-second scan, a 16-bit dedup key, and a 0.7 multiplier; the deployed 2025 binary may differ, and the data confirm the 0.7 scaling and per-wake export, not the scan duration or truncation. We therefore frame the result as a relative activity index that supports staffing decisions, not a city-wide people counter, and state what calibration and privacy work absolute claims would require.
|
||||
Heat is a measurable public-health threat, and Heilbronn already knows where its cool places are, but not how a person reaches them. Cool City is a public heat-safe route built on the existing eight-sensor pulse. A weather-aware global model forecasts the relative footfall index, reaching MAE 3.32, WAPE 27.5%, and R² 0.777 on a 28-day holdout versus 5.40, 44.7%, and 0.462 for a seasonal-naive baseline. A route and shade layer combines OpenStreetMap walkable ways, the city tree cadastre, and solar position to offer shortest, shaded, and accessible routes with modeled shade coverage, direct-sun minutes, step-free continuity, water, and benches. A cooling-gap layer ranks the eight sensor catchments by relative footfall, shade deficit, access deficit, and—where sourced—vulnerability. The existing retail staffing plan remains one commercial application. Shade is modeled, the cooling gap is relative rather than a measured temperature, and the underlying pulse is estimated devices, not people; route and city values are TBD until the route/shade bundle is frozen.
|
||||
|
||||
## 1. Introduction
|
||||
## 1. Problem
|
||||
|
||||
A store manager on Kaiserstraße knows the weather forecast and the staff roster. She does not know how many people will walk past next Saturday. Raw counts are not a decision; the manager needs a forecast, an explicit staffing rule, and a way to test a scenario. A dashboard that shows history does not close that gap.
|
||||
Heat is not a comfort issue. The Robert Koch Institute estimated about **16,000 heat-related deaths in Germany in 2026 through calendar week 37** (95% prediction interval 14,400–17,600; 19.3 per 100,000) [1]. That is a model estimate, not a count, and it is not a Heilbronn-specific figure. Heilbronn’s commissioned FITNAH-3D report gives a citywide physiological equivalent temperature (PET) range of **16.1–44.5 °C**, with unshaded sealed street space exceeding **41 °C PET** [2]. The report does not say that every inner-city street reaches 41 °C; the claim is about unshaded sealed surfaces.
|
||||
|
||||
The Schwarz Digits *Pulse of the City* challenge asks teams to turn privacy-preserving pedestrian counts from Heilbronn into predictive tooling for advertising, store locations, and staffing [1]. We chose staffing: eight fixed locations, five months of multi-sensor history, and a clear daily/weekly rhythm. Store-location ranking would need city-wide coverage that eight sensors in a 580 m cluster cannot provide; a city-wide heatmap would be a visualization, not a measurement.
|
||||
The city already publishes **almost 60 cool places** on a map, with an address list [3]. That is a useful inventory, but it is not a routing or time-aware service: it tells a person where cool places are, not how to reach one at 16:00 with shade, water, benches, and step-free access. A local councillor argues that the map and care-home information do not reach alone-living older people, who carry the greatest risk; we quote that as her assertion, not as an evaluated finding [4]. The 2022–23 summer zones were discontinued, according to local reporting [4].
|
||||
|
||||
PulseCast makes three contributions. It treats the decision as the product: the forecast feeds a weekly shift plan. It uses a global, weather-aware gradient-boosted model with a time-based holdout and a seasonal-naive baseline, rather than a model the data cannot support. It is deployed as a three-service system on SKE with a cached fallback and a live scenario layer. It also reports a code-level audit showing that the published value is a short-window count of randomized BLE addresses scaled by 0.7, not a 15-minute count of people.
|
||||
This is not a city that has done nothing. Heilbronn adopted a climate-adaptation concept in April 2023 with 31 measures across five fields [8] and published a 2025 spatial climate-adaptation concept [9]. The gap is narrower and more actionable: **the person-level route**. The city knows where the cool places are; it does not yet give a person a safe way there, hour by hour, with the access needs that determine whether the route is usable.
|
||||
|
||||
The report also serves the Schwarz Digits criteria: **problem and benefit** (a manager gets a plan, not a dashboard), **originality** (decision-first tooling plus a measurement audit), **feasibility and future** (live on SKE, 120-client load test, calibration roadmap), **reliability** (120/120 connections, zero errors, cached fallback), **understanding** (Sections 2.2 and 6 state what the index does and does not measure), **teamwork** (data/ML, Go API, Nuxt UI, and infra workstreams against a written contract, with human integration), **documentation** (code, contract, metrics, and this report), and **technology** (Vue/Nuxt + Go + Onyx + FastAPI on namespace-only SKE). For the final pitch, the technical evidence supports the 50% idea/implementation weight, the honest limits support the storytelling, and the calibration/privacy roadmap is the 20% vision.
|
||||
The existing eight-sensor pulse is the evidence spine, but it is a **relative device index**, not a count of people. Cool City does not hide that. It uses the index to forecast relative demand and to rank cooling gaps, and it states the measurement limits wherever a decision depends on them. The contribution is the public decision layer over a noisy city index: three routes for a persona, a relative cooling-gap ranking for the city, and the retail staffing plan as one commercial tenant.
|
||||
|
||||
We do not claim a production people counter, a city-wide heatmap, annual seasonality, live sensor data, or a proven return on investment. We claim a working decision loop, a credible relative forecast, and an honest account of the measurement limits.
|
||||
## 2. Data and measurement audit
|
||||
|
||||
## 2. Data
|
||||
The delivered PAX CSV contains 192,464 rows: 96,232 density readings and 96,232 battery readings, from 12 device IDs. After merging the two `density-01` IDs and excluding three test sensors, eight usable locations remain in a ~580 m × 280 m inner-city cluster. The **12 June model window** (12 June–7 November 2026) contains **83,156 density values**; the **18 June coverage panel** contains **82,170 readings** and is used for coverage statistics.
|
||||
|
||||
### 2.1 The delivered PAX data
|
||||
The public 2026 Code for Heilbronn `lorax` firmware (commit `e8be8bf`) implements a ~10-second passive BLE scan, a 16-bit dedup key, a 0.7 multiplier, and a 900-second sleep [11]. The deployed 2025 binary may differ. The data confirm the **0.7 scaling and per-wake export**, not the scan duration or the 16-bit truncation: every one of the 83,156 values in the 12 June model window is an exact multiple of 0.7; the reconstructed count has median 12, p95 71, max 418, and **2.70% of those values are zero**.
|
||||
|
||||
The dataset contains 192,464 CSV rows: 96,232 density readings and 96,232 battery readings. There are 12 device IDs; after merging the two `density-01` IDs and excluding three test sensors, eight usable locations remain in a ~580 m × 280 m cluster: `density-01`, `hn-goetzentuem-01`, `hn-hagenbuchersee-01`, `hn-hagenbuchersee-02`, `hn-marahaus-01`, `hn-neckarufer-01`, `hn-soleo-01`, and `hn-stadtgalerie-01`.
|
||||
Coverage is uneven. In the 18 June panel, only **28.2%** of 15-minute slots have all eight sensors; the modal count is seven (4,117 slots). Aggregating to 30, 60, or 120 minutes raises complete coverage to 63.4%, 69.7%, and 71.0%. Across all 96,232 density rows, 5.18% are zero, 10.36% are <1, and 39.57% are <5; in the 12 June window the shares are 2.70%, 8.32%, and 35.86%. Gaps reach four to six days per sensor and are not random. The 12 June panel has median 8.4, p95 49.7, and p99 79.8; across all 96,232 rows the values are 7.7, 46.9, and 78.4.
|
||||
|
||||
The 12 June model window (12 June–7 November 2026) has a median cadence of 15.3 minutes; values range 0–292.6, with median 8.4, p95 49.7, p99 79.8. Across all 96,232 density rows, the median is 7.7, p95 46.9, p99 78.4 and the multi-sensor mean is ~14.4. In the 12 June model window, the inner city peaks 18:00–21:00 local, top 19:00; Saturday is busiest on average (mean 19.1) [2]. The sensors are LoRaWAN-connected PAX counters that passively sniff Bluetooth Low Energy (BLE) advertisements and transmit aggregated device counts [1, 2]. The counts are estimates, not a census: device penetration, Bluetooth settings, and the scanning design all matter.
|
||||
The forecast is a global `HistGradientBoostingRegressor` trained on hourly means of the published index, with calendar, lag, Open-Meteo weather [12], and sensor-identity features. On the 28-day holdout, the production 7-day model reaches **MAE 3.32, WAPE 27.5%, and R² 0.777**; the day-ahead weather model reaches **3.36, 27.8%, and 0.763**; the seasonal-naive baseline is **5.40, 44.7%, and 0.462**. Weather is the main driver: removing it raises day-ahead MAE from 3.36 to 3.96. The holiday flag changes 3.96 to 3.84, a small effect. These are single-site, 28-day results.
|
||||
|
||||
### 2.2 What one value actually is
|
||||
| Audit layer | Finding | Population / caveat |
|
||||
|---|---|---|
|
||||
| Sensor network | 8 usable locations, ~15 min cadence, ~580 m × 280 m | Inner-city cluster, not city-wide |
|
||||
| Values | `0.7 × U`; 100% multiples of 0.7 | 12 June model window (83,156 values) |
|
||||
| Coverage | 28.2% all-eight slots; 4,117 modal-seven slots | 18 June coverage panel (82,170 readings) |
|
||||
| Zeros / low counts | 5.18% zero, 10.36% <1, 39.57% <5 | All 96,232 density rows |
|
||||
| Model | MAE 3.32, WAPE 27.5%, R² 0.777 | 28-day holdout; production 7-day model |
|
||||
| Unit | Estimated devices, not people; 0.7 factor unvalidated | The core caveat for every downstream layer |
|
||||
|
||||
A code-level audit of the public 2026 Code for Heilbronn `lorax` firmware (commit e8be8bf) changes the interpretation of the numbers [3, 4]. The node scans BLE passively for about 10 seconds, counts distinct locally administered addresses, truncates the last two address bytes into a 16-bit ID, multiplies by 0.7, and sleeps for 900 seconds. The deployed 2025 binary may differ; the data confirm the 0.7 scaling and per-wake export, not the scan duration or truncation.
|
||||
*Table 1. The measurement audit. Cool City inherits every limit in this table; it does not turn the index into people or into temperature.*
|
||||
|
||||
In the public 2026 firmware, the published value is `0.7 × U`, where `U` is the number of distinct randomized BLE addresses seen in a ~10-second scan. The 15-minute cadence is the sleep cycle, not the integration window; the effective duty cycle is ~1.1% in that code. Every one of the 83,156 values in the **12 June model window** is a multiple of 0.7; the reconstructed count has median 12, p95 71, max 418, and 2.70% of those values are zero. A hardware replacement creates a scale break in `density-01`; the legacy and current series are not pooled [5].
|
||||

|
||||
|
||||
This is a relative activity index, not a count of people. A 30-second passer-by has a ~4.4% chance of overlapping a scan; a one-hour lingerer can be counted up to four times. The index conflates flow and dwell, cannot distinguish phones from watches/beacons/cars, and the people-per-device ratio is unknown. The 0.7 factor is a provisioning default, not a validated penetration correction.
|
||||
*Figure 1. Forecast versus actual for one sensor over the 28-day holdout. The model captures the daily/weekly shape but misses spikes; the band is a planning aid, not a calibrated 95% interval unless coverage is reported.*
|
||||
|
||||
> **Practical consequence.** The model predicts the index well enough to compare relative peaks and drive a staffing rule. It does not predict people, and it does not support absolute footfall or revenue claims without calibration.
|
||||
## 3. Cool City
|
||||
|
||||
### 2.3 Coverage, gaps, and context
|
||||
### 3.1 A person, three routes, and a clock
|
||||
|
||||
Coverage is uneven. In the **18 June coverage panel (82,170 readings)**, only 28.2% of 15-minute slots have all eight sensors; the modal count is seven (4,117 slots). Aggregating to 30, 60, or 120 minutes raises complete coverage to 63.4%, 69.7%, and 71.0% [5]. Across all **96,232 density rows**, 5.18% are zero, 10.36% are <1, and 39.57% are <5; in the **12 June model window** the shares are 2.70%, 8.32%, and 35.86%. The worst month falls to 23.3% coverage at one location in the 18 June panel. Gaps reach four to six days; the cause is undocumented, but missing periods differ systematically from full-coverage periods [2, 5]. Weather comes from Open-Meteo: historical archive for training, current/7-day forecast for the dashboard. Temperature and precipitation are the useful features; cloud and wind did not improve the short test. Baden-Württemberg holidays are a feature and badge; they added little and are context, not a large driver [2].
|
||||
The demo persona is **Ruth, 78, who lives alone**. She is a clearly labelled composite, not a real person; the prototype can later use a real local story with consent. Her trip is **Stadtgalerie → Marktplatz**: short, ordinary, and exactly the kind of trip that heat turns into a risk. The route geometry and place names must be verified on OSM before the demo.
|
||||
|
||||
| Sensor | Readings | Mean | Median | p95 | Max | Max daily gap |
|
||||
|---|---:|---:|---:|---:|---:|---:|
|
||||
| hn-stadtgalerie-01 | 8,982 | 21.1 | 8.4 | 67.2 | 137.2 | 6 days |
|
||||
| hn-neckarufer-01 | 9,527 | 19.2 | 11.9 | 61.6 | 255.5 | 5 days |
|
||||
| hn-hagenbuchersee-02 | 10,099 | 19.3 | 11.2 | 64.4 | 200.9 | 4 days |
|
||||
| hn-marahaus-01 | 9,254 | 13.7 | 9.1 | 38.5 | 194.6 | 6 days |
|
||||
| hn-hagenbuchersee-01 | 12,345 | 13.0 | 6.3 | 49.7 | 292.6 | 4 days |
|
||||
| hn-goetzentuem-01 | 9,952 | 11.1 | 8.4 | 29.4 | 210.7 | 5 days |
|
||||
| hn-soleo-01 | 10,327 | 10.6 | 8.4 | 28.0 | 111.3 | 4 days |
|
||||
| density-01 | 12,670 | 9.9 | 7.7 | 28.0 | 63.0 | 35 days |
|
||||
Cool City offers three routes:
|
||||
|
||||
*Table 1. Per-sensor coverage and value statistics. Except for the `density-01` max-gap entry, statistics use the 12 June–7 November 2026 model window; the `density-01` 35-day gap spans the longer history and hardware replacement.*
|
||||
1. **Shortest** — the fastest path, with no shade or access preference.
|
||||
2. **Shaded** — the cool route via the Deutschhof passage / Kiliansplatz (verify the exact geometry).
|
||||
3. **Accessible** — a route that prefers step-free continuity, water, benches, and toilets (verify tags and continuity).
|
||||
|
||||

|
||||
A time slider at **10:00, 12:00, 14:00, and 16:00** changes the recommendation as the sun moves. Each route card shows modeled shade coverage, direct-sun minutes, distance, duration, step-free status, water stops, and bench stops. The shade values are **modeled**, not measured. The audience vote is a comprehension device, not a data source: the room picks a route for Ruth, and the stage shows the split next to the model’s recommendation. Most people take the shortest; the model takes the cool, accessible one.
|
||||
|
||||
*Figure 1. The eight sensor locations in the ~580 m × 280 m inner-city cluster. Measured points only; no city-wide surface.*
|
||||
### 3.2 Route and shade model
|
||||
|
||||

|
||||
|
||||
*Figure 2. Mean index by hour and weekday across the eight sensors. The peak is 18:00–21:00 local; Saturday is busiest on average.*
|
||||
|
||||
## 3. Method
|
||||
|
||||
### 3.1 Preprocessing and features
|
||||
|
||||
We parse timestamps as UTC, convert to `Europe/Berlin`, handle daylight saving, treat the two `density-01` IDs as one logical location but do not pool the legacy integer series with the 0.7-scaled replacement (the model uses the current device only), exclude the three test sensors, and aggregate to hourly means. The target is the hourly mean of the published index, not people.
|
||||
|
||||
Features are calendar (hour, weekday, month, weekend, holiday), lags (24, 48, 168 hours, plus 24-hour and 7-day rolling means), temperature and precipitation from Open-Meteo, and sensor identity. All lags and rolling windows end at the forecast issue time; the time-based split prevents leakage.
|
||||
|
||||
### 3.2 Model and validation
|
||||
|
||||
We train a global `HistGradientBoostingRegressor` across all eight sensors. Train 12 June–10 October 2026; test the last 28 days, 10 October–7 November. Baseline: seasonal naive, same hour one week earlier. We report MAE, WAPE, and R². The global model pools strength across sensors while keeping sensor identity; with eight series and five months, a deep sequence model is not justified.
|
||||
|
||||
The model serves 15 min, 1 h, 48 h, and 7 d. The 15 min/1 h horizons are nowcast/replay because the CSV is static; 48 h/7 d are planning horizons, with 7 d the hero.
|
||||
|
||||
### 3.3 From forecast to shift plan
|
||||
|
||||
The ML model predicts the published index; a deterministic heuristic turns it into staff:
|
||||
The route layer uses the OpenStreetMap walkable network (footways, pedestrian streets, paths, steps, crossings, buildings, trees, benches, drinking water, toilets, and wheelchair tags; ODbL) [10], the city tree cadastre from January 2026 (crown diameter; dl-de/by-2.0) [5], and solar position for the selected hour. Buildings and tree crowns cast modeled shade; the route cost combines travel time with penalties for direct sun, steps, and missing water/bench stops:
|
||||
|
||||
```text
|
||||
forecast index per hour
|
||||
→ calibration factor k (unvalidated)
|
||||
→ visitors per staff per hour (service standard)
|
||||
→ required staff = ceil(index × k / standard × safety factor)
|
||||
→ minimum staff, shift length, opening hours
|
||||
→ shift blocks + coverage score + estimated cost
|
||||
cost(route) = travel_time
|
||||
+ λ_sun × direct_sun_minutes
|
||||
+ λ_step × step_free_penalty
|
||||
+ λ_water × water_gap
|
||||
```
|
||||
|
||||
The assumptions are visible and adjustable: calibration factor `k`, visitors per staff hour, labor cost per hour, minimum staff, shift length, and safety factor. `k` converts index units to visitors and is **unvalidated**; setting `k = 1` would treat one index unit as one visitor, which the audit says is not justified. The manager can edit assumptions and regenerate the plan with a scenario. This is decision support, not an optimizer with proven ROI; we have no POS, sales, queue, or labor-cost data.
|
||||
The weights `λ` and the route bundle are **TBD** until the route/shade data is frozen. The route is not a temperature model: it reports shade coverage and direct-sun minutes, not “degrees cooler.” A temperature claim would require a validated radiant-temperature model and field data. The exact route metrics are also TBD; Table 2 defines the slots they will fill.
|
||||
|
||||
## 4. System
|
||||
| Route | Character | Shade coverage | Direct-sun min | Detour | Step-free | Water / bench stops |
|
||||
|---|---|---|---|---|---|---|
|
||||
| Shortest | Fastest path | TBD | TBD | 0 by definition | TBD | TBD |
|
||||
| Shaded | Via Deutschhof / Kiliansplatz (verify) | TBD | TBD | TBD | TBD | TBD |
|
||||
| Accessible | Step-free, water, benches (verify) | TBD | TBD | TBD | TBD | TBD |
|
||||
|
||||
PulseCast runs as three namespace-scoped Deployments on STACKIT SKE plus a Cloudflare Tunnel connector: **Nuxt 4 SSR + Onyx** for the server-rendered shell, map, charts, forecast, plan, and scenario controls; a **Go API** for sensor/series/forecast/staffing endpoints, SSE fan-out, scenario state, and a cached fallback; and a **FastAPI ML service** for `/predict` and `/simulate`, loading a model baked into the image at build time.
|
||||
*Table 2. Route cards. Values are TBD until the route/shade bundle is frozen; the paper does not invent route metrics.*
|
||||
|
||||
The namespace does not allow Ingress, Jobs, CronJobs, cert-manager, or external-dns. We expose the app through a Cloudflare Tunnel, terminate TLS at Cloudflare, and run the batch at build time. The model and fallback JSON are baked into the images; there is no database or PVC. If FastAPI is unavailable, Go serves the cached forecast and labels it. SSE carries scenario changes with a 15-second heartbeat; polling is the fallback. The plan is generated in 19 ms cold on v1.1; the contract is under two seconds.
|
||||

|
||||
|
||||
The public app is at https://pulse.noelsabia.dev and the API at https://api-pulse.noelsabia.dev. The v1.1 images are 16.9 MB (api), 249 MB (web), and 952 MB (ml). In the v1.1 120-client Cloudflare k6 run, 120/120 SSE connections opened and closed with zero connection errors; p95 event latency 331 ms, p95 scenario latency 646 ms, presenter POST 49.60 ms, over 4m57.9s. The system runs non-root with resource limits and a read-only root filesystem. The public dashboard is read-only; `/control` uses basic auth. We store no PII or MAC addresses in the application. This is a hackathon deployment: one API replica, no SSO, no HA, no live ingestion. We document those gaps.
|
||||
*Figure 2. OpenStreetMap base network for the focus area (14 November 2026 snapshot; ODbL). Orange = footway, red = pedestrian street, purple = steps, grey = buildings, green = trees, blue = drinking water, brown = benches, pink crosses = labelled anchor places. This is a base network, not the final route or shade layer.*
|
||||
|
||||

|
||||
### 3.3 Cooling-gap index
|
||||
|
||||
*Figure 3. Three services plus `cloudflared`: Nuxt SSR, Go API/SSE and cache, FastAPI ML. Build-time batch; cached fallback; no Ingress, cert-manager, or Jobs.*
|
||||
The city layer uses the eight sensor catchments (300–500 m). It computes a **relative cooling-gap index**:
|
||||
|
||||

|
||||
```text
|
||||
CG = normalize(footfall) × normalize(shade deficit)
|
||||
× normalize(access deficit) × normalize(vulnerability)
|
||||
```
|
||||
|
||||
*Figure 4. The v1.1 plan screen for `scenario=sunny-saturday-market`, `standard=standard` (live API query, 14 Nov 2026; reconstructed for print): Saturday peak 5 staff, 172 scheduled staff-hours, €2,752 labour cost, 100% coverage, versus a €5,600 flat schedule. The live screen also includes a “why this plan” context card; the app is at https://pulse.noelsabia.dev.*
|
||||
`footfall` is the published relative device index; `shade deficit` comes from the modeled shade layer; `access deficit` combines missing water, benches, step-free continuity, and toilets; `vulnerability` is included **only where sourced** and is otherwise set to 1. Population by age is available at planning-district level [6], and disability statistics are citywide (11,730 people in 2023; the city says “almost 12,000”) [7]. Those are aggregate inputs, not block-level diagnoses. The index is normalized 0–1 and shown as a rank with a one-line reason. The weights and the final ranking are **TBD** until the city bundle is frozen. It is a relative model, not a measured temperature map, and it must not be presented as a city-wide heatmap.
|
||||
|
||||
## 5. Evaluation
|
||||
### 3.4 Retail staffing as one application
|
||||
|
||||
### 5.1 Forecast accuracy
|
||||
The existing staffing plan remains a separate module. The same forecast schedules store staff; a retailer pays for that module. In this paper it is a commercial tenant, not the hero. The live plan is currently a snapshot with unstable scenario/window behavior, so we do not report its plan numbers here; the module is described qualitatively until the plan window and scenarios are stabilized. The Cool City paper’s contribution is the public route and cooling-gap layer, not a retailer dashboard.
|
||||
|
||||
Table 2 reports the 28-day holdout from the generated `metrics.json`. The weather-aware 7-day model reduces MAE by about 19% relative to the no-weather 7-day model (4.08 → 3.32) and by about 39% relative to seasonal naive (5.40 → 3.32). WAPE falls from 44.7% to 27.5%, and R² rises from 0.462 to 0.777. The day-ahead weather model reaches 3.36, 27.8%, and 0.763; the longer horizon is slightly stronger (3.32 vs 3.36 MAE) and remains well above the baseline.
|
||||
## 4. Evaluation
|
||||
|
||||
| Model | MAE | WAPE | R² |
|
||||
|---|---:|---:|---:|
|
||||
| Seasonal naive (same hour last week) | 5.40 | 44.7% | 0.462 |
|
||||
| Day-ahead, no weather | 3.96 | 32.8% | 0.706 |
|
||||
| Day-ahead, + temperature/rain | 3.36 | 27.8% | 0.763 |
|
||||
| 7-day-ahead, no weather | 4.08 | 33.8% | 0.697 |
|
||||
| 7-day-ahead, + temperature/rain (production) | **3.32** | **27.5%** | **0.777** |
|
||||
**Route-level metrics.** The route evaluation will report shade coverage (%), direct-sun minutes, detour versus the shortest route, step-free continuity, water stops, and bench stops for each route at each time. All values are **TBD** until the route/shade bundle is frozen. The target direction is explicit: more shade, fewer direct-sun minutes, acceptable detour, continuous step-free access, and at least one water/bench option on the accessible route.
|
||||
|
||||
*Table 2. Model comparison on the last 28 days, from the generated `metrics.json` (`n = 5,350`). The mission brief’s 3.36/27.8/0.779 was the quick baseline; the canonical generated metrics are the production 7-day 3.32/27.5/0.777 and day-ahead 3.36/27.8/0.763. The target is the published index, not people.*
|
||||
**City-level ranking.** The cooling-gap index will be reported as a rank across the eight catchments with a one-line reason, plus the input layers. The rank is **TBD**. The evaluation will show the sensitivity of the rank to the shade/access weights and to the vulnerability input; a rank that flips under plausible weights is not a decision-grade result.
|
||||
|
||||

|
||||
**System metrics.** The inherited SSE path was measured at **120/120 connections opened and closed, zero errors, p95 event 331 ms, p95 scenario 646 ms, presenter POST 49.60 ms, over 4m57.9s** on v1.1. The Cool City route page and vote endpoint are new: their latency, vote tally accuracy, and stage behavior are **TBD**. The retail plan module’s cold generation is 19 ms; that number belongs to the retail module, not the route layer.
|
||||
|
||||
*Figure 5. Forecast versus actual for one sensor over the 28-day holdout. The model captures the daily/weekly shape but misses spikes; the band is a planning aid, not a calibrated 95% interval unless coverage is reported.*
|
||||
**Comprehension test plan.** A small task-based test with older adults and caregivers will ask participants to (1) choose a route for Ruth at 16:00, (2) explain why it is cooler or more accessible, (3) compare the 10:00 and 16:00 recommendations, and (4) find water, benches, and step-free access. The sample, recruitment, and results are **TBD**; this is a plan, not a completed study.
|
||||
|
||||
### 5.2 Ablation and per-sensor behavior
|
||||
| Evaluation layer | Metrics | Status |
|
||||
|---|---|---|
|
||||
| Route | Shade coverage, direct-sun minutes, detour, step-free continuity, water/bench stops | TBD |
|
||||
| City | Cooling-gap rank, input layers, weight sensitivity | TBD |
|
||||
| System | SSE 120/120, p95 event 331 ms, p95 scenario 646 ms, presenter 49.60 ms | Measured on inherited v1.1 path |
|
||||
| Vote / route page | Open success, tally accuracy, route render latency | TBD |
|
||||
| Comprehension | Task success, explanation quality, route preference | Plan only; TBD |
|
||||
|
||||
Weather is the main driver. In the day-ahead ablation, removing it raises MAE 3.36 → 3.96. The holiday flag changes 3.96 → 3.84 (0.12); it is context, not a large lift. The production 7-day model is 3.32 with weather and 4.08 without. These are single-site, 28-day results.
|
||||
*Table 3. Evaluation status. Shade is modeled; the cooling gap is relative; no temperature or causal claim is made.*
|
||||
|
||||
Per-sensor errors vary (Table 3): Stadtgalerie has the highest MAE (5.83); Hagenbuchersee-01 the highest WAPE (38.6%); density-01 the lowest MAE (2.26). The global model is pooled; local errors are not.
|
||||
## 5. Privacy and limits
|
||||
|
||||
| Sensor | MAE | WAPE | R² |
|
||||
|---|---:|---:|---:|
|
||||
| density-01 | 2.26 | 30.8% | 0.722 |
|
||||
| hn-goetzentuem-01 | 2.40 | 26.9% | 0.718 |
|
||||
| hn-hagenbuchersee-01 | 2.47 | 38.6% | 0.602 |
|
||||
| hn-marahaus-01 | 2.67 | 29.8% | 0.671 |
|
||||
| hn-soleo-01 | 3.19 | 20.2% | 0.548 |
|
||||
| hn-neckarufer-01 | 3.63 | 28.2% | 0.695 |
|
||||
| hn-hagenbuchersee-02 | 4.12 | 31.1% | 0.731 |
|
||||
| hn-stadtgalerie-01 | 5.83 | 25.3% | 0.765 |
|
||||
Cool City uses aggregate device counts only. It does not store MAC addresses or PII, does not require accounts, and does not upload a visitor’s location. The audience vote is aggregate; a random session ID in local storage is enough. A production deployment needs a DPIA and a documented legal basis before collecting MAC-derived counts, with the city as controller and a legal review before launch. The existing index is estimated devices, not people; device penetration is unknown; the 0.7 factor is unvalidated; and only 28.2% of 15-minute slots in the 18 June panel have all eight sensors.
|
||||
|
||||
*Table 3. Per-sensor holdout error for the production 7-day model (`n ≈ 669` per sensor).*
|
||||
The route layer has its own limits. Shade is modeled from buildings and tree crowns, not measured with radiometers; tree cadastre and OSM coverage are incomplete; OSM is a 14 November 2026 snapshot, not an official city dataset; and the route geometry and place names still need verification. The cooling-gap index is a relative ranking, not a measured temperature map, and it does not establish that an intervention will cause a health improvement. We make no “4 °C cooler” claim and no causal claim for shade, water, or bench interventions. The current prototype is a decision-support layer over a noisy index, not a validated public-health instrument.
|
||||
|
||||
### 5.3 System metrics and decision evidence
|
||||
## 6. Future
|
||||
|
||||
| Metric | Value |
|
||||
|---|---:|
|
||||
| Plan generation latency | 19 ms cold on v1.1 (contract < 2 s) |
|
||||
| Concurrent SSE clients | 120/120 opened and closed; 0 connection errors |
|
||||
| SSE p95 event latency | 331 ms |
|
||||
| SSE p95 scenario latency | 646 ms |
|
||||
| Presenter POST latency | 49.60 ms |
|
||||
| Duration | 4m57.9s |
|
||||
| Image sizes (api / web / ml) | 16.9 MB / 249 MB / 952 MB |
|
||||
| Cold deployment time | not measured |
|
||||
The next step is a city-owned public trust: the city controls the route and cooling-gap layer, the underlying data is open where licensing allows (tree cadastre, OSM, city cool places, aggregate population), and the retail module is a commercial tenant rather than the product’s owner. The route/shade bundle and the cooling-gap weights should be published with the paper, tested with older adults and caregivers, and transferred to a second city with a different street network. Field validation of modeled shade and a small route-choice study would turn the current design into evidence. The long-term vision is a public heat-safe wayfinding layer that any city can adopt, with the retail staffing module as one paying application.
|
||||
|
||||
*Table 4. System metrics from the v1.1 deployment and the v1.1 120-client Cloudflare k6 run. Cold deployment time was not measured.*
|
||||
## 7. TBD register
|
||||
|
||||
The v1.1 120-client Cloudflare run is the measured result: 120/120 connections opened and closed, zero errors, p95 event 331 ms. The plan screen compares the recommended schedule to a flat schedule on coverage and cost under stated assumptions; this is a heuristic comparison, not an ROI claim.
|
||||
| Item | Why it is TBD | Next step |
|
||||
|---|---|---|
|
||||
| `/cool` route deployment | The route/shade bundle is not frozen | Verify the live `/cool` page and fallback |
|
||||
| Route geometry (Deutschhof / Kiliansplatz, step-free continuity) | OSM tags and passage geometry need field verification | Walk the route; check steps, crossings, water, benches |
|
||||
| Route metrics (shade %, direct-sun min, detour, duration, stops) | Route/shade bundle not frozen | Freeze the bundle; report values on the four time slices |
|
||||
| Cooling-gap weights and rank | City bundle and weight sensitivity not frozen | Publish weights; run sensitivity; report rank and reason |
|
||||
| Vote / route-page latency | New endpoint not measured | Run the same k6/SSE protocol; report open success and p95 |
|
||||
| Comprehension test | Plan only; no participants yet | Recruit older adults/caregivers with the city; publish tasks and results |
|
||||
| DOI | Event draft has no DOI yet | Mint on Zenodo after the event |
|
||||
| OSM/tree coverage | 14 Nov 2026 OSM snapshot; cadastre may be incomplete | Document query date; compare with city tree data |
|
||||
|
||||
### 5.4 What the evaluation does and does not show
|
||||
|
||||
The evaluation shows that a global, weather-aware model beats a seasonal-naive baseline on a time-based holdout in one cluster. It does not show that the model predicts people, generalizes to another city, that weather is causal, or that the plan improves profit. The 28-day test is short, weather is observed, and the target is a ~10-second BLE index. It is a relative decision-support signal, not a breakthrough.
|
||||
|
||||
## 6. Discussion and Limitations
|
||||
|
||||
The main limitation is the measurement. The sensor samples ~10 seconds out of every 15 minutes, counts only randomized BLE addresses, truncates to 16 bits, multiplies by 0.7, and cannot distinguish people from devices. The value is a relative index, not a census. Penetration is unknown; a single factor cannot absorb device ownership, multiple devices per person, Bluetooth settings, and the short scan window.
|
||||
|
||||
Coverage is also a limitation. Eight sensors in 580 m × 280 m cannot support a city-wide heatmap or store-location recommender. Only 28.2% of 15-minute slots have all eight sensors; gaps are not random. The 28-day holdout is short, weather is observed, and there is no event log, POS, or labor-cost data. The deployment has one API replica, no SSO/HA, and no live ingestion; 15 min/1 h are replay/nowcast.
|
||||
|
||||
**Privacy.** The app uses aggregate counts only, stores no MAC addresses or PII, and does not redistribute the raw challenge CSV. The sensing layer is upstream of this application. A production deployment would need a DPIA and a documented legal basis before collecting MAC-derived counts.
|
||||
|
||||
**What to keep:** decision-first framing, global model, time-based holdout, seasonal-naive baseline, transparent heuristic, cached fallback, build-time batch, honest limitations. **What to replace:** sensing (thermal/radar/camera/probe/manual), calibration (site-specific, with uncertainty), privacy (no raw MAC retention, central DP, k-threshold), spatial model (network-aware hierarchical field), forecast (quantile GBM, forecast weather, conformal calibration), platform (auth, observability, SLOs) — in that order. Fix the measurement before scaling the model.
|
||||
|
||||
## 7. Related Work
|
||||
|
||||
Passive Bluetooth/Wi-Fi sensing consistently finds a small, biased sample: counts track relative trends, absolute levels need local calibration [6, 7]. MAC randomization reduces linkability but does not make passive counting a census, and fingerprinting can defeat it [7]. Our audit is consistent and makes the sampling design explicit. Gradient-boosted trees remain a strong baseline for tabular, covariate-rich, small-data problems [8]. Weather affects pedestrian activity, but its magnitude is context-dependent [9]; we treat it as predictive, not causal.
|
||||
|
||||
## 8. Conclusion
|
||||
|
||||
PulseCast shows that a small, weather-aware global model can turn a noisy inner-city footfall index into a weekly staffing plan that beats a seasonal-naive baseline and runs in 19 ms cold on a real deployment. It also shows that the index is not a people counter: each value is a short BLE scan scaled by 0.7, sampled every 15 minutes, with unknown penetration and incomplete coverage. The honest conclusion is narrower and more useful than a city-scale claim: the decision loop works, the relative forecast is credible, and the next step is calibration and a privacy-first sensing layer. We would rather ship a plan a manager can use and state what it does not measure than promise a city-wide heatmap the data cannot support.
|
||||
|
||||
## Artifacts, data, and AI disclosure
|
||||
|
||||
The live app is https://pulse.noelsabia.dev (API: https://api-pulse.noelsabia.dev); the code repository will be released after the hackathon. The raw challenge CSV is not redistributed. Derived aggregates, the model card, and the metrics JSON can be shared subject to the organizers’ permission. The app stores no MACs or PII; the sensing caveats are in Section 2.2 and the privacy box is in Section 6.
|
||||
|
||||
AI tools were used for drafting and editing. The authors checked each number and citation; some sources are project documentation or estimates. AI tools are not authors and take no responsibility for the content.
|
||||
*Table 4. Open items. These are deliberately not filled with plausible-looking numbers.*
|
||||
|
||||
## References
|
||||
|
||||
[1] Schwarz Digits / Code for Heilbronn. *The Pulse of the City*, challenge brief, 2026. Local: `OKF_Vault/Schwarz Digits/FCH Challenge The _Pulse of the City_.pdf`.
|
||||
[1] Robert Koch-Institut. *Hitzemortalität in Deutschland 2026*. https://www.rki.de/DE/Themen/Gesundheit-und-Gesellschaft/Gesundheitliche-Einflussfaktoren-A-Z/H/Hitze/Bericht_Hitzemortalitaet.html
|
||||
|
||||
[2] PulseCast project. `OKF_Vault/mission/data.md`, `mission/mvp.md`, `mission/architecture.md`, 2026.
|
||||
[2] Stadt Heilbronn. *FITNAH-3D-Abschlussbericht zur Raumstruktur- und Klimaanpassung*, 2025. https://www.agenda21-hn.de/wp-content/uploads/2025/07/034_2025_DS_034_2025_Anlage_5_Abschlussber_Raumstrukt_Klimaanpassung.pdf
|
||||
|
||||
[3] Code for Heilbronn. “Passantenzahlen messen mit LORA,” 18 September 2025. https://codeforheilbronn.de/blog/2025/09/passantenzahlen-messen-mit-lora/
|
||||
[3] Stadt Heilbronn. *Gesundheit und Hitze: kühle Orte*. https://www.heilbronn.de/leben/gesundheit/staedtisches-gesundheitsamt/gesundheit-und-hitze.html
|
||||
|
||||
[4] Code for Heilbronn. `lorax` / `sensor-pax` firmware and backend, commit `e8be8bf5c11e1b76c88c3f176382c6df99eea989` (develop, 18 June 2026). https://codeberg.org/cfhn/lorax/commit/e8be8bf5c11e1b76c88c3f176382c6df99eea989; upstream `ESP32-Paxcounter` and `libpax`, https://github.com/cyberman54/ESP32-Paxcounter and https://github.com/dbinfrago/libpax.
|
||||
[4] Rhein-Neckar-Zeitung. “Hitze in Heilbronn: und nicht genug Hilfe,” 2026. https://www.rnz.de/region/sinsheim-kraichgau_artikel,-Kampf-gegen-hohe-Temperaturen-Hitze-in-Heilbronn-und-nicht-genug-Hilfe-_arid,2401501.html
|
||||
|
||||
[5] PulseCast data-quality analysis, `paper/research/07-data-quality.md`, 2026.
|
||||
[5] Stadt Heilbronn. *Baumkataster Januar 2026*, dl-de/by-2.0. https://opendata.heilbronn.de/dataset/baumkataster
|
||||
|
||||
[6] A. Kurkcu and K. Ozbay. “Estimating pedestrian densities, wait times, and flows with Wi-Fi and Bluetooth sensors.” *Transportation Research Record* 2644(1):72–82, 2017. DOI: 10.3141/2644-09.
|
||||
[6] Stadt Heilbronn. *Bevölkerung nach Altersgruppen*, Planungsbezirke. https://opendata.heilbronn.de/dataset/bev%C3%B6lkerung-nach-altersgruppen-11
|
||||
|
||||
[7] J. Martin, T. Mayberry, C. Donahue, L. Foppe, L. Brown, C. Riggins, E. C. Rye, and D. Brown. “A study of MAC address randomization in mobile devices and when it fails.” *Proceedings on Privacy Enhancing Technologies* 2017(4):365–383. DOI: 10.1515/popets-2017-0054.
|
||||
[7] Stadt Heilbronn. *Menschen mit Behinderung* (11,730 in 2023); city page “knapp 12.000.” https://opendata.heilbronn.de/dataset/menschen-mit-behinderung · https://www.heilbronn.de/leben/barrierefreiheit-inklusion-leben-mit-behinderung.html
|
||||
|
||||
[8] J. H. Friedman. “Greedy function approximation: a gradient boosting machine.” *Annals of Statistics* 29(5):1189–1232, 2001. DOI: 10.1214/aos/1013203451.
|
||||
[8] Stadt Heilbronn. *Klimaschutzteilkonzept zur Anpassung an den Klimawandel*, adopted 27 April 2023, 31 measures. https://www.heilbronn.de/umwelt-mobilitaet/klima-klimaschutz/anpassung-an-den-klimawandel.html
|
||||
|
||||
[9] L. F. Miranda-Moreno and A. C. Lahti. “Temporal trends and the effect of weather on pedestrian volumes: a case study of Montreal, Canada.” *Transportation Research Part D* 22:54–59, 2013. DOI: 10.1016/j.trd.2013.02.008.
|
||||
[9] Stadt Heilbronn. “Hitze im Blick: konkretisiertes Klimaanpassungskonzept,” 2025. https://www.heilbronn.de/kommunikation/pressemitteilungen/details/artikel/hitze-im-blick-konkretisiertes-klimaanpassungskonzept-zeigt-wege-fuer-besseres-stadtklima.html
|
||||
|
||||
[10] OpenStreetMap contributors. *OpenStreetMap*, ODbL. https://www.openstreetmap.org/copyright
|
||||
|
||||
[11] Code for Heilbronn. `lorax` / `sensor-pax` firmware, commit `e8be8bf5c11e1b76c88c3f176382c6df99eea989` (develop, 18 June 2026). https://codeberg.org/cfhn/lorax/commit/e8be8bf5c11e1b76c88c3f176382c6df99eea989
|
||||
|
||||
[12] Open-Meteo. *Historical and forecast weather APIs*, CC BY 4.0. https://open-meteo.com/
|
||||
Reference in new issue
Block a user