Skip to main content
    Transparency

    The honest number right now is no number.

    Scam Finder started with an empty database. A report goes here once a month, counted from real records. Until there’s something to count, these lines stay blank.

    Monthly reportPeriod: none yet

    • Completed checks

      Not enough activity to report this metric

    • Published experiences

      Not enough activity to report this metric

    • Upheld corrections

      Not enough activity to report this metric

    • Time to first case update

      Not enough activity to report this metric

    • Checks with insufficient coverage

      Not enough activity to report this metric

    Methodology version
    None published
    Coverage limitations
    None to report yet
    Policy changes
    None to report yet
    The definitions

    The rules are public before the numbers are.

    Every figure will state its unit, date range, denominator, exclusions and calculation version. Each one uses what’s visible and retained on the day it’s counted.

    1. Completed checks

      Checks that produced a complete or partial report during the month. Attempts are shown separately.

      Unit
      Distinct check runs
      Period
      One calendar month, UTC
      Denominator
      None. This is a count.
      Left out
      Retries, cancelled and failed jobs, test and monitoring runs

      Split into checks people asked for and scheduled refreshes, so a refresh never passes as a person helped.

    2. Published experiences

      Reviews visible on Scam Finder at the moment the month’s snapshot is taken.

      Unit
      Reviews, one per review
      Period
      A single cutoff at month end
      Denominator
      None. This is a count.
      Left out
      Revisions, replies, and anything withdrawn or removed by the cutoff

    3. Upheld corrections

      Material correction cases resolved in the month where the challenge was upheld or partly upheld.

      Unit
      Distinct cases
      Period
      One calendar month, UTC
      Denominator
      None. This is a count.
      Left out
      Linked duplicates, which count as one case

    4. Time to first case update

      The median time from receiving a case to the first substantive update on it.

      Unit
      Median duration
      Period
      Cases closed in the month
      Denominator
      Eligible closed cases that had such an update
      Left out
      The automatic receipt, which isn’t an update

      Closed cases with no update aren’t dropped quietly. They’re reported separately, or held back if the group is too small.

    5. Checks with insufficient coverage

      The share of completed checks where required coverage was missing.

      Unit
      Percentage of reports
      Period
      One calendar month, UTC
      Denominator
      Completed-check reports in the same month
      Left out
      Nothing further. With a zero or withheld denominator, no rate is shown.

      This measures how often evidence ran out. It says nothing about how safe any website is.

    Blank lines

    A gap will always say why it’s a gap.

    Too little happenedNot enough activity to report this metric
    There’s nothing to count, or too little to mean anything. Every line on this page reads like this today.
    Held backGroup too small to publish
    A small group could point to a real person or case. Related breakdowns go too, because subtraction would reveal the hidden figure.
    No rateNo rate for this period
    The denominator is zero or withheld. Printing 0% would claim something that never happened.
    Report lateGeneration failed. Showing the last valid month.
    The last safe report stays up with its age marked. A failed month is never filled in with an estimate.
    CorrectedCorrected. Replaces the earlier figures for this month.
    A wrong number gets a dated replacement. If someone’s private data is removed, reports that depended on it can be withdrawn.

    How a month gets printed.

    1. The month ends

      Each report covers the last full calendar month in UTC.

    2. Records are counted

      Figures come from the underlying event and publication records, so a broken running tally can’t change them.

    3. Totals are tested

      The sums run against test data with known answers before anything goes out.

    4. One version goes live

      The report is released whole and versioned. No AI writes or explains the numbers.

    Flattering numbers stay off the bill.

    Traffic and warning counts look impressive and prove very little. A count of users would need a defined way of measuring them, and there isn’t one. Money saved would be a guess.

    Nothing here identifies a person or names a website’s complaints. Reports also avoid overlapping releases that could expose a single case.

    SCAM FINDER

    LINES THAT STAY OFF THIS REPORT

    • People helpedNOT MEASURED
    • Money savedNOT INFERRED
    • Live accuracyNOT CLAIMED
    • Complaints per websiteNOT PUBLISHED
    • Names, cases, sanctionsNOT PUBLISHED
    • Model prompts, security detailNOT PUBLISHED

    NO ESTIMATES. NO ROUNDING UP.

    Every line starts with one real check.

    Got a website you’re unsure about? A completed check is counted once, in next month’s report, and your details aren’t.