Skip to main content

Methodology

How Human Cost records are built and bounded

A fixed research cohort, later official notices and post-release reporting are separate lanes, and they are never added into one total.

01

Separate lanes, never one total

Deaths reach this site through different reporting systems with different windows and inclusion rules. Combining them would produce a number no source publishes.

Fixed report cohort
The people documented in the named research cohort, inside that report's own window. It is a fixed research cohort, not a live tally, and it does not grow as later deaths are reported.
Later official notices
Agency death notices dated after the report cutoff. They are published beside the cohort, with their own dates and sources, and are never added into it.
Post-release reporting
Reported deaths following release from custody. The reporting is incomplete by its own terms, and the completeness statement is published with the records.
Adjacent agency records
Records from a different agency's custody, held outside the ledger and labeled as excluded rather than quietly dropped.
  • Every aggregate carries its source, scope, window, as-of date and inclusion rule.
  • No single unqualified live total is displayed anywhere on the site.
  • A mechanical union of the lanes exists for internal reconciliation only and is never published as a death total.
02

Observed, attributed, modeled

Most public argument collapses these three. Keeping them apart is the point.

Observed
A named event or death documented in a source. The record states that something happened, and to whom, on a date, at a place.
Attributed
An investigator, medical examiner, court, or body of strong evidence connects the harm to a policy or institutional failure. Attribution is always someone's finding, and we name that someone.
Modeled
A statistical estimate produced under a stated counterfactual, with uncertainty. A model is a claim about a population, never about a specific person.
03

How a person record is built

A person record leads with the name, then the sourced account of what happened, then the administrative facts. Where the reviewed sources carry no detail about a person's life outside custody, the record says so plainly rather than filling the space.

Fields the sources do not establish are grouped and stated as unknown. An unknown field is never replaced with an average, an inference, or a blank that reads as zero.

Names are published where they are already responsibly public and relevant to the record.

  • A death is recorded as observed only where identity, date, custody or facility relationship, and source are documented.
  • No death is described as preventable, negligent, caused by detention, caused by a contractor, or caused by inadequate care unless a cited source establishes it.
  • Where a record reads 'suspected cause,' that phrasing belongs to the cited medical review. It is not an independent determination by this project.
  • Cause, preventability, and responsibility are separate questions in this record.
04

Facility associations

A facility association on a person record states a custody location, not a finding of responsibility. It is published only where the join resolves to a canonical facility identity; an unresolved label is shown as published by the source, without a link that would assert a match.

Contract history and a death record may appear on the same facility page. Contract-level causation requires evidence a cited source supplies.

—

Where these rules are visible

The Human Cost ledger publishes each lane with its own window and source, and the harm ledger shows how observed, attributed and modeled records are handled.

—

Rest of the method

This page is one part of the published method. The reader guide summarises all of it, and corrections are recorded on the affected record with the date applied.

Current section: Human Cost