SANKOFA · THE METHOD
How this record is built

What counts, where it came from, and what to do when we get it wrong.

Sankofa keeps a record of documented harm to Black communities in America, and a map of where people can still build a life. A record is only worth what its method is worth — so the method is published here in full, including the parts that are weak. If something on this site is wrong, there is a route on this page to change it, and it is meant to be used.

01The hard boundary

The most important thing on this page, and the thing most likely to be misread if it is not said first.

What Sankofa is

  • A record of documented incidents. Each entry traces to a public record, a named dataset or published reporting, and the source is shown on the entry itself.
  • A finding aid. Missing Black children and adults, so cases that fell out of the news are still findable by anyone looking.
  • A geography. Where documented events happened, and separately, where the cost of living and local wages still leave something at the end of the month.
  • Incomplete, and openly so. The record is growing. Absence of a case means nobody has entered it yet — never that nothing happened there.

What Sankofa is not

  • Not a finding of guilt. Nothing here is a legal conclusion about any person or organisation. A documented report is a report.
  • Not a rate. Counts are raw incident counts. They are not per-capita, not adjusted for population, and cannot be read as “worse” or “safer.”
  • Not a measure of bias. Where a place appears often, that reflects reporting, record-keeping and population — not a demonstrated pattern of intent.
  • Not a database of individuals. No private addresses, no live locations, no tracking of any person, named or otherwise.
The single most common misreading

A city with many entries is not a more dangerous city. It is a better-documented one. Large metros generate more records, have more newsrooms covering them, and have more people in them. A small town with one entry and a large city with forty cannot be compared on those numbers, and this site will never rank them that way.

The same applies to institutions. A hospital that appears twice is a hospital where two documented reports named it. It is not the second-worst hospital in a league table, and there is no league table.

02What gets in

An entry has to clear all four. If it clears three, it does not go in.

  1. It traces to a citable source.

    A government or law-enforcement record, an established dataset, court filings, or reporting from an identifiable outlet. A source that cannot be linked or named is not a source. Social-media claims are not sources on their own — they are a reason to go looking for one.

  2. The event is documented, not alleged in passing.

    Something has to have been recorded by someone with a reason to record it accurately: a case number, a filing, an investigation, a published account. Rumour and secondhand summary do not qualify no matter how widely repeated.

  3. It is placeable.

    A record with no location cannot go on a map without inventing one. Where a location is approximate, that is stated rather than smoothed over.

  4. It concerns harm to, or the safety of, Black communities.

    That is the scope. It is a deliberately narrow archive, and narrowing it is what keeps it accurate.

03Where the data comes from

Every source, and — more usefully — what each one cannot tell you. A dataset used outside what it can support is how good data produces false conclusions.

NamUsFederal · official

The US Department of Justice's National Missing and Unidentified Persons System. The authoritative public record for missing-person cases in the United States.

What it cannot tell you: NamUs is opt-in. Many cases are never entered, and entry rates differ by agency and by state. A county with few NamUs cases may have few cases, or may simply not use NamUs.
The Washington Post — Fatal ForcePress dataset

A continuously maintained national database of fatal police shootings, compiled from local reporting, public records and agency statements.

What it cannot tell you: it counts shootings, not all deaths in custody, and not non-fatal encounters. Reading it as “police violence” overall overstates what was measured. It also begins in 2015 — it says nothing about before.
Court filings and published investigationsPublic record

Filed complaints, consent decrees, agency findings and settlements. Used for institution-level entries.

What it cannot tell you: a filing is an allegation until it is resolved, and a settlement is explicitly not an admission. Where an entry rests on a filing rather than a finding, the entry says so — and it stays that way even if the outcome seems obvious.
Historical recordArchival

Documented historical violence and displacement — the entries that give the map its depth in time rather than only its present.

What it cannot tell you: historical records were kept by the institutions of their time, which frequently had reason to undercount. The gaps in this layer are not neutral, and a thin century is not a quiet one.
Cost of living, rent and wage dataPublic statistics

Used only in the Where to move section, which is a separate question from the record and is kept separate on purpose.

What it cannot tell you: metro averages hide neighbourhoods. A city that averages well can contain places that do not, and no figure on that page should be treated as advice about a specific address.

04The counting rules

How a number on this site is produced, so anyone can check whether it means what they think it means.

  1. One documented incident, one count.

    Not one count per article, per victim or per agency involved. Where several sources describe the same event, it remains one entry with several sources attached.

  2. Counts are never converted to rates.

    No per-100,000 figures, no population adjustment, no ranking of places against each other. The data is not collected uniformly enough for a rate to be honest, and a dishonest rate is more damaging than no rate.

  3. An institution is named only at two or more documented reports.

    A single report is kept with the case, where it belongs. The two-report threshold exists so that one contested account cannot, by itself, attach a name to a map.

  4. Jurisdictions and record sources are counted separately from institutions.

    This distinction matters more than it looks. A city where cases occurred is not an accused party. A database the records came from is not an accused party. Both are listed apart from named institutions so the two can never be read as the same thing.

  5. Filtered views recount from scratch.

    Every figure reflects the current view, not the whole archive. Narrow the filters and the counts fall — that is the counter working, not the record shrinking.

05What is deliberately not published

Choices made against the project's own interest, because an archive of harm can very easily become a source of it.

06Corrections and removal

The part that makes the rest of it a record instead of an accusation.

If an entry is wrong, incomplete, or should not be here

Anyone can ask for a correction — the person or family involved, a named institution, a journalist, or someone who simply spotted an error. You do not need standing and you do not need a lawyer. Requests are read by a person, not a form.

  1. Send the entry. The page URL or the case name is enough. If you know the entry ID, include it.
  2. Say what is wrong. A wrong date, a wrong location, a misattributed institution, an outcome the record has not caught up with, or an entry that should never have been published.
  3. Point at something if you can. A document, a link, a case number. If you cannot, send it anyway — the entry gets re-checked against its original source either way.
  4. What happens next. Factual errors are corrected. Entries that fail the inclusion rules in section 02 are removed. Where an outcome has changed — charges dropped, a conviction overturned, a case closed — the entry is updated to reflect it rather than quietly deleted, because the update is itself part of the record.
  5. Contested but sourced. If an entry is accurate to its source and the source is disputed, the dispute is noted on the entry. Both stay visible. Nothing is silently changed to end an argument.
[email protected]

07Known limits

Stated here rather than discovered by someone else later.

If you want to use this in your own work

Cite the source, not Sankofa. Every entry names where it came from — quote that. This site is a finding aid that points at public records; the records are the evidence, and they are what will hold up.

Researchers, journalists and organisations working in this area are welcome to get in touch about the underlying data, including the parts that are not published. Questions about the method are the most useful mail this project receives.