{
    "the_headline": "One of these layers is a government record and the other is a survey. Treating them as one dataset is the mistake this endpoint exists to prevent.",
    "official_layer": {
        "source": "City of San Antonio - SAPD Calls for Service (driving events only)",
        "included": "Accident Major,Accident Minor,Wrong Way Driver",
        "excluded": "Traffic Stop, Traffic Violation and DWI are excluded because they count enforcement activity rather than events, and Traffic Related is excluded as too vague to mean anything about driving.",
        "why": "Calls for service is police workload. Only call types describing an event that happened are counted; types describing an enforcement action are not, because a ZIP with more traffic stops has more policing rather than worse driving.",
        "no_denominator": "A busier ZIP collects more of everything and there is no traffic count here to divide by, so this is where police were called and not where driving is worst.",
        "hour_of_day": "Hour of day is the hour the call reached police. Unlike a crime report date, that is essentially the hour the event happened, because somebody dials while it is in front of them.",
        "by_month_is_not_normalised": "by_month is a raw count per calendar month and February is three days shorter than January. Divide by days in month before comparing months or February will read as roughly eight per cent safer every year, which is the length of the month rather than a change in driving. Nothing on the page displays this series for that reason.",
        "weekday_counts_are_uneven": "The archive is not a whole number of weeks, so one or two weekdays occur once more than the others across it. The difference is under one per cent and is left uncorrected in matrix; do not read a sub-one-per-cent gap between two weekdays as a finding.",
        "hours_are_evenly_represented": "Every day in the window contributes all 24 hours, so by_hour needs no exposure correction. This is not true of by_month or of the weekday axis.",
        "hour_of_day_contrast": "This is the one thing this dataset does better than the crime data on the same site. A NIBRS offense carries the date it was reported, so an hour-of-day reading there measures when people file paperwork. A dispatch time is when somebody dialled, which for a crash or a wrong-way driver is while it is happening. by_hour and matrix are published for that reason and would not be defensible on the offense data."
    },
    "resident_layer": {
        "verified": false,
        "why_unverified": "Nothing is checked against any official source because none is published: the City has no crash dataset, the State crash database is behind a login, and no feed anywhere flags a road rage incident.",
        "self_selection": "People who report are not a random sample of drivers. A road can rise up the list because more of its drivers know this page exists.",
        "unlock_threshold": 250,
        "unlocked": false,
        "display_floor": 5,
        "withheld": null
    },
    "not_published": {
        "identifiers": "No license plate, vehicle description, name or photograph is collected anywhere by this tool. The form has no free-text field, so none can exist to publish.",
        "day_resolution": "Reports are stored with a day for de-duplication only. Nothing served here is finer than a calendar month, because a day plus one arterial plus a severe category could identify one real incident.",
        "local_streets": "Only classified arterials can be reported. Residential streets are excluded so a report cannot point at a household.",
        "raw_reports": "Individual submissions are never served, and raw rows are deleted after 90 days."
    },
    "caveats": {
        "two_layers_never_summed": "The \/calls and \/reports layers measure different populations and must never be added together. A call is somebody who phoned the police; a report is somebody who did not, which is nearly everyone, because you cannot call 911 about being tailgated.",
        "reports_are_unverified": "This is not crime data and not a crash record. It is what people chose to report, and nothing in it has been verified against any official source, because no such source is published.",
        "self_selection": "Self-selection is the dominant effect. People who report are not a random sample of drivers, so a road with more reports may simply carry more people who know this page exists.",
        "no_denominator": "There is no denominator. Without traffic volume a busy road will always accumulate more reports than a quiet one, which says nothing about which is worse. Roads here are ranked by most REPORTED, never by most dangerous.",
        "arterials_only": "Only classified arterials can be reported. Local and residential streets are deliberately excluded so that a report cannot point at a household.",
        "month_resolution": "Nothing published here is finer than a calendar month, and no report is ever shown individually.",
        "floored": "Roads below the reporting floor are withheld and counted in the withheld total, so nothing disappears silently.",
        "calls_are_events_only": "The calls layer counts only Accident Major, Accident Minor and Wrong Way Driver. Traffic stops, traffic violations and DWI are excluded because they count enforcement activity rather than events, and Traffic Related is excluded as too vague."
    }
}