Data

The site is backed by an immutable versioned analytical snapshot. The complete 108,644,312-sample Parquet archive is published on Hugging Face; verified native Prometheus and Grafana copies remain private preservation artifacts.

Derived snapshot v3

Download files

Exact-data release

The public dataset package contains:

raw/          exact sample timestamps, values, and original labels
derived/      reconciled five-minute through daily analysis products
validation/   third-party comparison tables and provenance
*.json/*.csv  schemas, export provenance, and SHA-256 manifests

The native Prometheus TSDB and Grafana database remain private: those operational formats can contain credentials, users, or unrelated service metadata. Public raw Parquet preserves the eight application metric families, exact millisecond timestamps, values, every original label, and a source_series field that distinguishes direct from historically imported series.

Licenses:

• Code: MIT
• Published data: CC BY 4.0

Both the complete Git history and the publication candidate tree passed a Gitleaks scan. The ignored live Prometheus configuration is neither tracked nor included in the release.

Example analysis

import pandas as pd

daily = pd.read_csv("daily.csv", parse_dates=["local_date"])
complete = daily[
    daily["complete_day"]
    & (daily["lighting_period"] == "pre_lights")
]

weekday = (
    complete.groupby(["weekday", "direction"])["mean_flow"]
    .mean()
    .unstack()
)
print(weekday)

Use lighting_period == "commissioning" or "illuminated" for the other optical regimes. Omitting the filter deliberately combines incompatible nighttime detector conditions.

External validation snapshot v1

This additive snapshot contains the high-level comparison with MTC and Caltrans public statistics. It does not contain or imply calibrated vehicle counts.