A small controlled OnlyDetect search test returned the intended record for all ten exact handles, all ten handles prefixed with @, and all ten handles with one final-character substitution. When we applied an incompatible offer filter, none of the ten intended records appeared, yet nine searches still returned other results.
The useful finding is that a nonempty result page does not prove the intended profile was found. This is a 40-case software experiment, not a platform-wide accuracy score or a study of real users' searches.
The test protocol
We selected the first ten eligible directory records ordered by lowercased handle. Eligibility meant active, indexable, unsuppressed, and validated within 180 days. We queried the public search function read-only with anonymous permissions and a limit of 48 results.
For each target, we tested four conditions: the exact handle; an @ prefix; the final character replaced with z, or y when it was already z; and the exact handle combined with an incompatible free or paid offer filter. The experiment record and SQL protocol preserve the definitions.
Results from the first 48 returned records
Scroll across the table on smaller screens.
| Input condition | Cases | Intended record found | At least one result |
|---|---|---|---|
| Exact handle | 10 | 10 | 10 |
Handle with @ prefix | 10 | 10 | 10 |
| Final-character substitution | 10 | 10 | 10 |
| Incompatible offer filter | 10 | 0 | 9 |
The last row is especially important. Other matching records can remain after a filter excludes the intended target. A fan should check the destination identity instead of assuming any plausible card is the requested account.
What this suggests when a search disappoints
If you know the public handle, start with it and remove uncertain filters. Then inspect the intended record's identity and current source. Add a condition back only when it is relevant and supported.
In this experiment, one particular typo pattern still retrieved every target. That does not establish that arbitrary misspellings, reversed names, or old aliases will work. The username guide explains a more reliable approach based on published identifiers.
Why this is not an accuracy benchmark
The ten targets were selected from records we already knew existed in the directory. The experiment did not include a representative set of unknown names, missing accounts, languages, or real search behavior. It also did not measure the rank of the intended result beyond presence in the first 48.
We did not inspect private search histories or publish individual query strings. The controlled transformations were generated from public directory records solely for this test.
How to extend the experiment
A broader evaluation would define its tasks in advance, include absent targets, distinguish identity precision from recall, and measure several query patterns across a justified sample. Until then, the practical conclusion remains narrow: filters can remove a known target, and other results can still look superficially successful.