How CityOps measures requests.
Source coverage, refresh health, metric contracts, and limitations are part of every interpretation.
NYC Open Data 311
- Dataset
- erm2-nwe9 ↗
- Supported source window
- Jan 1–31, 2025
- Problems
- Noise - Residential · HEAT/HOT WATER · Illegal Parking
- Loaded records
- 5,000
- Oldest / newest loaded
- 2025-01-01 / 2025-02-01
- Last successful refresh
- 2025-02-01T08:00:00Z
Latest ingestion status
- Run status
- succeeded
- Rows seen
- 181,205
- Inserted / updated
- 181,205 / 0
- Quarantined
- 0
Definitions before charts
| Metric | Definition | Interpretation limit |
|---|---|---|
| Requests created | Distinct request IDs with source NYC-local created time in the half-open selected range. | Reports received, not confirmed incidents. |
| Currently open | Selected requests whose latest ingested status is normalized as open. | A current status cannot reconstruct historical backlog. |
| Time to close | Recorded closed time minus created time when both timestamps are valid and nonnegative. | Open requests have no completed duration. |
| Seven-day closure | Mature creation-cohort requests with valid closure recorded within seven days, divided by all eligible requests. | Seven days is an analytical threshold, not an NYC SLA. |
| Period change | Current count against the preceding equal-length period. | No percentage is shown when the prior count is zero. |
| Problem share | Problem count divided by all requests under the current filters. | Changes when filters change. |
Why the risk queue is hidden
logreg-2025-jan-v1 · held-out 2025-01-27 to 2025-02-01 (exclusive)
The test period contained 3 seven-day unresolved outcomes among 17,289 requests. Logistic regression PR-AUC was 0.000513 versus 0.000554 for the grouped historical-rate baseline. Both had 0% precision among the top 100 ranked requests. The publication gate failed. The full training, validation, calibration, and subgroup evaluation is recorded in analytics/modeling/report.json in the repository.
Source time and data limitations
NYC publishes floating civil timestamps in this dataset. CityOps preserves them as NYC-local timestamps without inventing a UTC offset. Durations near daylight saving transitions inherit that ambiguity. Latest status does not reveal every prior reopening or reassignment. Coordinates may be absent or invalid. Counts reflect reporting behavior, population, and awareness as well as conditions.
This deployment is a bounded historical case study for three problems in January 2025. Its source-check timestamp records when the slice was last reconciled; it does not mean new 311 requests are being imported daily. Model features use only creation-time values.
See the analyses ↗