// detection intelligence, kept fresh

your stack still works.
we tell you the day it stops.

find something that worksfind what's breakingsee what changed this weekshift with a dynamic marketdiscover the next stepstop guessing which stack passesknow it before your users doknow which stack still workswatch a detector change its mindread the number, not the pitchsee the trials behind the claimcatch the drift before you shipcompare two tools on one screenspend less time re-testingget told when a config breaksfind what passes for lessget a webhook when it changesstop re-running the same testmeasure instead of guessingsee which layer caught youpick a proxy on evidencetrack a vendor over timeknow what changed last weektell a real failure from a flakeskip the tools that never passsee the whole board at oncestart from what already workscheck our arithmetic yourselfbuy the answer, not the runtimewatch the field moveknow where you stand todayplan the next test with datafind something that works

you run the proxies and the browsers. we run the fleet that measures them against named detectors, publish what came back with its trial count beside it, and keep that answer fresh so you do not have to.

we sell the knowledge. you run your own stack. we never touch your traffic.

ask for access see how we measure

we read every application and answer by email. the measured results below are free to read here.

measured 2026-09-05 · modelled 2026-09-05

toolpassesincolumbscancreepjsd&bifpjssannysoft

81%n=2260humanhumanhumanhumanhumanbot

62%n=2633botbotbotbotbothuman

52%n=2070·botbotbotbothuman

cloakbrowser50%n=1723·bothumanbothumanhuman

selenium-driverless49%n=1741·bothumanbothumanhuman

undetected-chromedriver39%n=1699·botbotbothumanhuman

playwright-extra39%n=1661botbotbotbothumanhuman

human = read as human
bot = caught
thin = too few trials to call

live, as of 2026-09-05·15 of 29 tools in this grid measured·the 6 most discriminating of 16 detectors·27,551 trials in this grid

66 tracked · 19 measuredtools tracked across every family, and how many carry a measured record

16 scoredreal detectors in the live grid

78,858definitive trials behind those records, across every family

2 of 7detection layers with a measured reading

// what changed

detection moves, and we watch it move.

vendor releases, coverage reports and service incidents, collected as they are published. the links go to the source rather than to us.

// what you get

start from the answer, at the depth you need it.

the free account costs nothing, and we read every application and answer by email.

you want to check a tool before you commit to it

the whole grid, free with an account

every tracked tool against every scoreable detector, with the pooled record and the trial count beside it, refreshed whenever the fleet runs.

ask for access →

you need to know how reliably, and which way it is moving

the per detector breakdown, on a plan

the rate, the denominator and the movement per tool and per detector. the verdict tells you whether a stack got through; this tells you whether to bet on it.

see the plans →

you need this in a pipeline, not in a browser

the api the console is built on

every figure on this page comes from a documented endpoint. keys, webhooks and metering are published, and the quote below is computed by the function that bills.

read the api docs →

// methodology

we publish a benchmark and we sell what it finds.

that is a conflict of interest, so the method is public and so is the endpoint behind every figure we print. what a verdict means, which of our own rows we flag as not comparable, and the layers we do not score at all.

read the methodology →

// how it works

from a hostname to the winning config.

name a target

ask for a hostname through the api or the console. we detect which vendors defend it and which layers they enforce.

we benchmark the field

the fleet runs real browser, proxy, fingerprint and solver configurations against the live target and scores what actually got through.

you get the current best

the winning per-site recipe comes back with its trial count and its confidence, and a webhook fires the moment it changes.

// what it costs

stop reverse-engineering detection by hand.

start from the answer. the grid, the trial counts and the current best config, kept fresh by a fleet that runs whether you are watching or not.

ask for access

// detection intelligence, kept fresh

your stack still works.
we tell you the day it stops.

find something that worksfind what's breakingsee what changed this weekshift with a dynamic marketdiscover the next stepstop guessing which stack passesknow it before your users doknow which stack still workswatch a detector change its mindread the number, not the pitchsee the trials behind the claimcatch the drift before you shipcompare two tools on one screenspend less time re-testingget told when a config breaksfind what passes for lessget a webhook when it changesstop re-running the same testmeasure instead of guessingsee which layer caught youpick a proxy on evidencetrack a vendor over timeknow what changed last weektell a real failure from a flakeskip the tools that never passsee the whole board at oncestart from what already workscheck our arithmetic yourselfbuy the answer, not the runtimewatch the field moveknow where you stand todayplan the next test with datafind something that works

we sell the knowledge. you run your own stack. we never touch your traffic.

ask for access see how we measure

we read every application and answer by email. the measured results below are free to read here.

measured 2026-09-05 · modelled 2026-09-05

toolpassesincolumbscancreepjsd&bifpjssannysoft

81%n=2260humanhumanhumanhumanhumanbot

62%n=2633botbotbotbotbothuman

52%n=2070·botbotbotbothuman

cloakbrowser50%n=1723·bothumanbothumanhuman

selenium-driverless49%n=1741·bothumanbothumanhuman

undetected-chromedriver39%n=1699·botbotbothumanhuman

playwright-extra39%n=1661botbotbotbothumanhuman

human = read as human
bot = caught
thin = too few trials to call

live, as of 2026-09-05·15 of 29 tools in this grid measured·the 6 most discriminating of 16 detectors·27,551 trials in this grid

66 tracked · 19 measuredtools tracked across every family, and how many carry a measured record

16 scoredreal detectors in the live grid

78,858definitive trials behind those records, across every family

2 of 7detection layers with a measured reading

// what changed

detection moves, and we watch it move.

vendor releases, coverage reports and service incidents, collected as they are published. the links go to the source rather than to us.

// what you get

start from the answer, at the depth you need it.

the free account costs nothing, and we read every application and answer by email.

you want to check a tool before you commit to it

the whole grid, free with an account

every tracked tool against every scoreable detector, with the pooled record and the trial count beside it, refreshed whenever the fleet runs.

ask for access →

you need to know how reliably, and which way it is moving

the per detector breakdown, on a plan

the rate, the denominator and the movement per tool and per detector. the verdict tells you whether a stack got through; this tells you whether to bet on it.

see the plans →

you need this in a pipeline, not in a browser

the api the console is built on

every figure on this page comes from a documented endpoint. keys, webhooks and metering are published, and the quote below is computed by the function that bills.

read the api docs →

// methodology

we publish a benchmark and we sell what it finds.

read the methodology →

// how it works

from a hostname to the winning config.

name a target

ask for a hostname through the api or the console. we detect which vendors defend it and which layers they enforce.

we benchmark the field

the fleet runs real browser, proxy, fingerprint and solver configurations against the live target and scores what actually got through.

you get the current best

the winning per-site recipe comes back with its trial count and its confidence, and a webhook fires the moment it changes.

// what it costs

stop reverse-engineering detection by hand.

start from the answer. the grid, the trial counts and the current best config, kept fresh by a fleet that runs whether you are watching or not.

ask for access

picking up where you left off

your stack still works.we tell you the day it stops.

detection moves, and we watch it move.

start from the answer, at the depth you need it.

the whole grid, free with an account

the per detector breakdown, on a plan

the api the console is built on

we publish a benchmark and we sell what it finds.

from a hostname to the winning config.

name a target

we benchmark the field

you get the current best

stop reverse-engineering detection by hand.

your stack still works.we tell you the day it stops.

detection moves, and we watch it move.

start from the answer, at the depth you need it.

the whole grid, free with an account

the per detector breakdown, on a plan

the api the console is built on

we publish a benchmark and we sell what it finds.

from a hostname to the winning config.

name a target

we benchmark the field

you get the current best

stop reverse-engineering detection by hand.

your stack still works.
we tell you the day it stops.

your stack still works.
we tell you the day it stops.