They hand you thirteen numbers. You guess.
Every comparison page in this category is a table somebody typed. This one is read from the competition's own responses — our retrieval fleet has been calling them for months, so their raw output is already in our findings table. The numbers below are counted from it, and the method is stated so you can disagree with it.
What the category actually returns.
median 8 · max 50
A bare array of strings. No line type, no date, no source, no rank. Dialling order is the reader's problem.
0.7% of records
Not “a poorly calibrated confidence”. There is no field for one in the schema the whole category shares.
dates: 18 / 464
So three brokers reselling one upstream feed read as three sources agreeing. That is the single most common way a wrong number looks corroborated.
Read this before quoting it. Measured on the free and scraped tier reached by our own retrieval path — the tier a prospect compares against. A paid report from one of these vendors is not in this sample and may carry more. Counted across reverse-phone (70), people-search (394).
Same data. A different question answered.
We buy and scrape from the same places they do. The difference is not the haystack — it is that we rank it, say how sure we are, and publish what happens when we are wrong.
| Question a skip-tracer actually asks | The category | Here |
|---|---|---|
| Which of these numbers do I dial first? | All 12.7 of them, unranked | Ranked, top candidate named, the rest collapsed with their score range |
| How sure are you? | No field for it — 3 of 464 | A calibrated posterior with a published reliability curve |
| Who told you that? | 23 of 464 name a source | Every value carries its sources, counted as INDEPENDENT families — three brokers on one feed is one family |
| Is it still true? | 18 of 464 carry any date | As-of on every value; line currency modelled as time-to-event (blocked on a carrier lookup, and it says so) |
| What if it is a common name? | Same weight for Smith as for Kirkendoll | Agreement on Smith is worth 0.66 bits; on Kirkendoll, 13.97 — measured against a counted population, not a flat weight |
| What if you do not know? | Returns a list anyway | Abstains, and says which evidence would settle it |
| How often are you wrong? | Not published by anyone in the category | Right-party rate published on /accuracy, including the numbers we dislike. A measured false-match rate too — for the face matcher, which is not yet wired into a trace |
Where they are currently ahead
A comparison that only flatters the author is the thing this page exists to replace, so: their coverage is far larger than ours today — they have been buying this data for a decade. Our accuracy claim is calibrated, on 97 positive labels of the 30 the harness requires, and 0 of those were confirmed by a human. Address history and relatives need a paid source we do not fire on our own account.
What we will not do is publish a number we have not measured in order to close that gap on paper.