benchmarks
Same input, every tool, scored. We ran usernames through Opsis, OSINT Industries, Sherlock, Maigret (6 configurations including paid tiers) on 2026-10-30. Every hit on a verified handle was checked by hand and classified as a real account, a redundant mirror, or a false positive. Usernames that exist nowhere were run as controls: anything a tool reports for them is wrong.
scoreboard
Verified handle (redacted): 27 real accounts across all tools. Recall is the share of those a tool found; precision is the share of its hits that were real.
| tool | real found | recall | precision | false hits | false hits / control | time | cost / search |
|---|---|---|---|---|---|---|---|
| Opsis premium | 23 / 27 | 85% | 100% | 0 | 0 (1 run) | 112s | 2 credits |
| Opsis | 19 / 27 | 70% | 100% | 0 | 0 (1 run) | 111s | 1 credit |
| OSINT Industries premium | 15 / 27 | 56% | 100% | 0 | 1 (1 run) | 54s | 3 credits |
| OSINT Industries | 14 / 27 | 52% | 100% | 0 | 1 (1 run) | 53s | 2 credits |
| Maigret all sites (-a) | 14 / 27 | 52% | 11% | 108 +5 dup | 110 (4 runs) | 6m | free |
| Sherlock | 12 / 27 | 44% | 63% | 7 | 5 (4 runs) | 105s | free |
head-to-head
false positives on usernames that don't exist
Random strings with no accounts anywhere. Every hit here is a false positive. The paid APIs were run on one control; the free tools on 4.
| tool | xj3hih12w2lc | xrlaowpjb1y6 | x7cwpg3nkvbg | thisdoesnotexist2398239 | sites reported |
|---|---|---|---|---|---|
| Opsis | – | – | – | 0 | none |
| Opsis premium | – | – | – | 0 | none |
| OSINT Industries | – | – | – | 1 | imgsrc |
| OSINT Industries premium | – | – | – | 1 | imgsrc |
| Sherlock | 7 | 3 | 4 | 5 | 7cups, aniworld, awsskillsprofile, bandcamp, boardgamegeek, contently, f3, forumguns, houzz, ourdjtalk, tiktok |
| Maigret all sites (-a) | 108 | 111 | 107 | 113 | 3rabnaar, abonentsntvplus, alienexpanse, alsh3er, amperka, aqa, artpersona, astrotalks, ausflugstippsforum, autoexperience, ayenee, baniclub … +106 more |
premium vs standard
- Opsis premium vs Opsis on the verified handle: +5 (cracked, fishbrain, fortnitetracker, leafly, malibucrew), −1 (genius). Modules failed: 84 standard, 102 premium.
- OSINT Industries premium vs OSINT Industries on the verified handle: +2 (fortnitetracker, genius), −1 (polymarket).
other usernames (unscored)
Not verified, so these show volume only. A common word like example is genuinely registered almost everywhere.
| tool | jasondraper01 | example |
|---|---|---|
| Opsis | 8 · 100s | 281 · 54s |
| Opsis premium | 6 · 111s | 316 · 113s |
| OSINT Industries | 7 · 53s | 264 · 54s |
| OSINT Industries premium | 7 · 53s | 305 · 54s |
| Sherlock | 9 · 103s | 148 · 88s |
| Maigret all sites (-a) | 110 · 6m | 599 · 4.9h |
method
- One username, sent to every tool from the same harness. Open-source tools ran locally through a residential proxy; Opsis and OSINT Industries ran through their APIs.
- Sherlock 0.16.2 (
--timeout 60). Maigret 0.6.6 with-a(all ~3,000 sites,--timeout 30 --retries 1). - Opsis API v1, standard and
premium: true. OSINT Industries API v2,timeout: 55, standard andpremium: true. - Site names are normalised across tools so the same platform counts once. Lookups that just mirror another account (e.g. a GitHub Gist page for a GitHub account) are counted as redundant, not as extra finds.
- Recall is measured against every real account any tool found. Accounts no tool found aren't counted, so true recall is lower for everyone.
caveats
- One run per tool per username. Results vary between runs: a module that times out in one run can succeed in the next.
- Runs overlapped, so times are rough. Maigret
-atook 4.9h onexample. - OSINT Industries' first run on the verified handle returned 1 and 0 hits at the timeout; it was rerun and the rerun is shown. Both runs were charged.
- The paid APIs were run on a single control username, the free tools on 4.
- Upwork hits from Opsis premium are excluded: that module returned false positives during this run and was fixed afterwards.
Raw data (verified handle redacted): username-2026-10.json