CashoutGuard: fraud prevention for rewards, GPT, faucet, PTC and offerwall sites.

Changelog

What changed in CashoutGuard, newest first. Every change ships to all plans the same day.

November 2026

A guide, a search and a calmer dashboard

  • A quick guide on the overview: four steps to get value (connect your site, see who is flagged and why, correct mistakes, turn on blocking) next to the questions owners ask most. Close it once you know your way; the full Guide stays in the menu, in all twelve languages.
  • Search the whole dashboard with Ctrl K (⌘ K on a Mac): any page, or an account by its id, email, IP or payout wallet, with its decision and score as you type.
  • The menu folds to its icons with the round button on its edge, and the dark theme sets the menu apart from the page.
  • Accounts and Events take a single date button: today, the last 7, 30 or 90 days, or your own range.

Act on many at once

  • Fraud rings: block a ring straight from its card, without opening it first.
  • Accounts filtered to Review or Block: allowlist or block the accounts on the page at once, after a confirmation that says how many.
  • Cashout queue: select every cashout with no risk signal in one click, then approve them together.

Your own reversals tune the rules

  • When the conversions flagged for a reason are reversed far less than your other conversions, Rules & lists suggests a lower weight for it. When a reason you weighed down is reversed far more, it suggests your default weight back. Each suggestion shows your own figures, the bell names the reason once, and you decide.

More device checks on offer clicks

  • The browser script now prepares its device checks while the page is idle, so an offer click sends its scan within milliseconds instead of about a third of a second, before the user leaves for the offer.
  • The website and the dashboard work again on older iPhones (Safari before 16.4), where pages could fail to load and charts to draw.

Easier setup

  • The postback relay says when it is still waiting for its first postback and updates on its own when the offerwall's test postback arrives.
  • The Install page opens your site in one click for the live check.

Offer and publisher profiles

  • Filter Events by an offer, or search its id, for its profile: payout per conversion, which publishers send it, the share flagged against your site, conversions per day, why they were flagged and the most flagged accounts. Every offer in the overview, Traffic, Accuracy and the events table opens it.
  • Both the offer and the publisher profile show the click-to-conversion time, from under 30 seconds to over a day, with the median. Bots and farms pile up in the first group; real users spread over the time a task takes.
  • For sites that report reversals, both profiles show the share the network reversed, over the conversions old enough to have been reversed, and the chart shows it for each time group. Exports carry a new seconds_from_click column.

Accuracy that reads right

  • The reversal-rate tile measures the decisions taken at the time against what they let through. Accounts caught weeks later have their own tile and no longer dilute it.
  • The period stops at your usual reversal wait, so the latest conversions no longer count as clean before the network had time to reverse them.
  • A new panel shows how much of each wait between the offer click and the conversion was reversed.

Opt-in rule: converted long after the click

  • A new rule for conversions that come 20 minutes to a day after their click, off until you set it in Rules & lists. When your own reversals show slow conversions are reversed much more, Rules & lists suggests it with your figures and what it would add, and the bell tells you once. Games and apps convert hours later by design, so it is never on by default.

Fraud rings by date

  • The rings page has a period: 7, 30 or 90 days, or your own dates. The ring page and its block button keep the period you chose.

Integration checks

  • The bell tells you when conversions, offer clicks or cashouts stop arriving for two days while the rest of your events keep coming. The Install page shows when each kind of event last arrived.
  • Not installed yet? Sites & keys and the overview offer the free integration next to the install guide.

Emails you control

  • Settings > Company has a switch to turn off CashoutGuard's emails (reports, tips and alerts) for your company. Payment emails still go out and the bell keeps every alert. Every report and tip also carries a one-click unsubscribe link.

Fewer false alarms

  • One VPN or hosting connection on an account with weeks of clean history now flags that event, as "always block" promises, instead of holding the whole account for 30 days.

Face ID (pilot), set up by you

  • A new Face ID page in the dashboard asks users for a short video selfie (look at the camera, turn the head as asked) and finds the same face on several accounts. You choose the events (signup, login, offer click, conversion, cashout), once per account or every time, everyone or only from a risk score, and which publishers, offers and countries it covers.
  • You also choose what it changes: nothing, review or block while the check is missing, after three failed tries, or when the same face is on other accounts of your site or of the network (other sites are never named). /v1/evaluate answers face_check with the link to send the user to.
  • It runs on our own servers: face prints are encrypted, kept 12 months and never given to you, camera images are never stored, and users see a consent screen first. Off by default.

Postbacks from Offer18, Affise, TUNE, Everflow, Swaarm and Trackier

  • The postback relay now knows the macros of six affiliate tracking platforms. Choose the sender's platform when you add a relay: the page shows which tracking link parameter carries your own id, what the platform fills in each field and a postback URL to copy, and the relay reads those macros whatever your parameters are called.
  • Swaarm's #{...} variables are kept as they are (they used to cut the URL at the #), and an Affise status 3 (declined) is learned as a reversal.

One phone posing as many devices

  • New signal device_clones: a phone that opens every account in a private browser window draws a new canvas each session, so each account looked like a new device. Scans now keep a profile of everything but the canvas; 12 or more other new accounts on one profile and network in 7 days, most of them with their own canvas, is one phone cloning itself (review, block from 30 accounts). Identical phones of real users draw the same canvas and are left alone. In its first week it found two such phones, about 215 accounts across three offerwalls.
  • Through the shared memory, a phone another customer's engine already caught cloning is caught on your site from its third account, and at block level when it had 30 or more there.

What each publisher cost you

  • Traffic, the overview and the weekly report now say how much each publisher's conversions credited to accounts that are flagged now, caught then or later, which is what you can claim back. The amount opens those conversions on the Events page (new filter: Account flagged now), ready to export with their transaction ids.

Weekly report numbers that add up

  • The weekly report's total of risky conversions now counts what its publisher table counts: conversions flagged when they happened, plus those of accounts flagged since (a phone farm is often caught after its first conversions). Only dollar amounts are added up, and an account you allowlisted is left out, so no publisher's row comes to more than the total above it. On one network the total went from $34.97 to $150.18 for the same week.
  • A site that already shows the "Protected by CashoutGuard" badge, even with its own link campaign, is no longer asked in the report to add it.

Reasons your own tools can show

  • Every reason in the API answer and in the account lookup now has label, its meaning in plain English, and help, a link to its explanation in the docs. The event.blocked and account.blocked webhooks carry the labels too.

Integration problems, named and cleared

  • A user id that arrives as an unfilled macro ([USER_ID], {clickid}) is everyone a publisher sends: it is scored but kept apart, never linked to your real users, and shown on the Install page and in the weekly report with the value received.
  • "New id on every visit" names the publisher most of those ids come from, and the warnings about two ids per user and lost scans clear as soon as the integration is fixed instead of a week later.
  • A device scan that reached us a moment before its click, when the page stopped waiting for it, now joins that click, as an unverified scan.

October 2026

The rest of a farm wave

  • On networks with publisher-scoped accounts, a new account is now flagged for review when most new accounts its publisher sent to the same offer in the last 6 hours (15 or more) were caught with a cloud phone, an emulator or a spoofed device, even if its own click came without a device scan. Measured on 21 days of a network's traffic: 139 more farm accounts, none of them later approved by the network.

Shared fraud memory: chargebacks and payout wallets

  • The memory websites on CashoutGuard share now includes payout wallets and chargebacks. A wallet that another website's engine blocked is flagged like an email, unless it is shared by many honest users (an exchange deposit address). An email, wallet or device with chargebacks at two other customers is flagged for review; one customer's reports alone never flag anyone elsewhere, and chargebacks alone never block.
  • Only which identifier matched travels, never the website or its users' data. Each website can turn the shared memory off in Settings; it then neither gives nor receives.

The API says when a check was IP-only

  • Every /v1/evaluate answer now has coverage: "device" when the device check ran for the event, "ip_only" when it did not, with a hint that says what an IP-only check cannot see. Rotating residential and mobile proxies exit through real home IPs that pass any IP list; the device behind them (its clock, its real IP, the other accounts on it) is what gives them away. Asked for by a customer testing with IPs only.

Email history in Bulk check

  • Bulk check now shows each email's online history: on how many services it appears in public breach data and since which year, with the oldest services named (dating and adult sites are counted, never named), and whether it shows up in infostealer logs. The public email checker shows the count and the year. Asked for by a customer.
  • Each email now gets a profile: since which year it exists as far as any source knows (breach data and our own network), its public Gravatar profile and linked accounts, other CashoutGuard sites with the same address and whether they blocked it, how old a company domain is (a domain registered days ago is a red flag), and whether the address looks machine-generated.
  • Bulk check also asks each address's own mail server whether the mailbox exists, without sending anything: a made-up Gmail or Outlook address now reads "Mailbox does not exist". Servers that accept any address, and Yahoo, give no answer and are left alone.

Proxy farms behind one clean IP

  • New signal proxy_farm_ip: three or more accounts on one IP within a day whose browser clocks all name the same other country (for example a Lagos clock on every account behind one Australian home-broadband IP). A residential or mobile proxy looks clean to IP lists; a farm sharing one exit does not. It counts as VPN/proxy, so sites that always block VPNs block it. Asked for by a customer.
  • New "always block" switch in Rules & lists, Country fraud: blocks a browser clock or leaked real address from a lower-paying country behind a high-paying country's IP (a Nigerian clock on an Australian IP, not a US clock on a Canadian one). Off by default.
  • Phone makers' browsers (Phoenix on Tecno and Infinix, Mi Browser and similar) run an older engine than the Chrome version they name; that no longer reads as a disguised browser.

Correction notices and fewer false positives

  • Correction notices: when we fix a detection mistake, you get an email and a dashboard notification listing the accounts you blocked or whose cashout you denied that the fix clears, with what was wrong and how to undo it. Asked for by a customer.
  • Mobile networks are no longer called datacenters: a public "bad networks" list we used also contained mobile carriers (MTN and Airtel Nigeria among them). Carriers are now told apart from hosting, and past events were re-checked.
  • Phones are not robots: holding Backspace on an Android keyboard repeats at a steady pace and no longer reads as machine typing.
  • Identical phones are not one person: the same popular phone model on the same mobile network no longer links strangers' accounts.
  • Firefox privacy mode: Firefox's fingerprinting protection reports an Icelandic clock wherever the user is; it no longer counts as a clock from another country.
  • Offerwall postbacks: the relay understands BitLabs' current macros ([%USER:UID%], [%VALUE:CURRENCY%]...) and AdGate's {state}, treats their reconciliations and rejections as reversals, and passes offer names along.
  • Free tool: check up to 25 IP addresses at once, without an account.

Clearer sub-processor list

  • The privacy policy now names every provider that already runs part of the service: Resend for emails (EU), CCPayment for crypto payments, and Telegram for our own team's account alerts. No new provider was added and nothing changes in how your users' data is handled: the browser script still talks only to our servers.

Bots that fill forms like a person, and setups without server code

  • Scripted page use: the script now notices a form filled by automation even when the browser hides it: a pointer that only jumps onto the buttons, keys typed with machine-regular gaps, or events dispatched by code. Only counts and timings leave the page, never which keys or what was typed. Part of the Bots switch.
  • Risk per kind of fraud: every API answer now has categories with a 0-100 risk for bot, multiple_accounts, vpn_proxy, emulator and tampering, so you can hold bots and only review VPNs, for example.
  • Impossible graphics: a phone whose graphics run through Direct3D is a Windows PC (an emulator, or a desktop posing as a phone for mobile offers), and a Windows profile on an Apple chip or a Mac on Direct3D is an anti-detect browser copying one OS onto another.
  • No server code: on pages where users are logged in, send account_id with collect() and each user is scored a few minutes later, without any server call (holding an offer or a cashout before it is paid still needs the server call).
  • Trial ending email: when only the script is installed, it shows what the script saw and the one line that turns scoring on.

Device data from app WebViews, and offerwall networks

  • Cloud phones: Android containers rented on server boards are now caught by their GPU, which no real phone has, whatever model they claim. Part of the Emulators switch.
  • Phone farm displays: a recent flagship phone whose screen was forced to 16:9 (360x640), the way farm software streams and controls dozens of real phones from one PC, is its own reason.
  • Fake Galaxy phones: a Galaxy S21 to S25, A34, A35, A54 or A55 that reports another chip's GPU is flagged as a spoofed device.
  • Android SDK 1.1: an auto-clicker switched on while the user earns (the way playtime and tap offers are farmed) is its own reason. Only the auto-clicker's package leaves the phone; screen readers, password managers and other accessibility apps are never read out.
  • Slow in-app browsers: send account_id with collect() and a scan that arrives after the page moved on still joins that account's click from the same IP.
  • Phone checks: send phone on signup and we flag VoIP and virtual numbers, invalid numbers, a number from another country than the IP, and the same number on several accounts (whatever its format).
  • Shared IP reputation: an IP whose accounts most other websites on CashoutGuard blocked this week now counts here too. Which websites is never shown.
  • In-app browsers: when an app never sends the scan, the script now resends it on its own as a small GET, so offer clicks from publisher apps carry device data again.
  • An offer click without a scan of its own uses the same account's scan from the last 30 minutes on the same network.
  • Install page: the sources or publishers whose clicks mostly arrive without a device scan, each linked to those clicks. Events has a "Without device scan" filter, and every publisher profile shows its share.
  • The overview warns when fewer than 60% of a week's offer clicks carry a scan, and you get an alert the same day if your script stops reporting.
  • Offerwall networks get their own wording and settings across the dashboard: publishers instead of sources, conversions instead of cashouts, and "only fraudulent conversions" as the recommended enforce scope.
  • Phone farms on residential proxies: one device on 3 or more accounts, with a clock from another country behind home IPs of a high-payout one, now scores 85 on a conversion.
  • Fewer false positives: the Phantom crypto wallet is no longer read as a bot, Pixel 10a phones are no longer read as spoofed, and carrier IPs shared across one country's time zones no longer count as a proxy.
  • Accuracy waits for 10 reported reversals before showing rates, so one chargeback never reads as 0% or 100%.
  • Docs: a scan page for network click URLs and how to send reversed and valid outcomes every day. New guides: offer clicks from app WebViews without device data, and cloud phones on offerwalls.

Check anything, without code

  • Same offer completed twice from one browser with another account (same site, same publisher or another publisher of a network) now scores 85 on the conversion: duplicate payouts are blocked in enforce mode. Part of the Multi-accounts switch.
  • Enforce spike alert: if enforce mode blocks far more than usual in an hour, you get a notification at once, so real users are never blocked for hours unnoticed.
  • Always block: switches in Rules & lists to block VPNs, Tor, datacenter IPs, multi-accounts, emulators, bots or disposable emails whatever the score. On every plan.
  • "Protected by CashoutGuard" badge for your site (light or dark), with the code to copy on the Install page.
  • Overview date filter: last 24 hours, 7, 30 or 90 days, or your own from–to. Every number, chart and list on the overview follows it, and the link keeps the period.
  • Enforce mode now shows on the overview what it stopped: accounts, signups, logins, offer clicks, conversions and cashouts, with the money held.
  • Cashout status API (POST /v1/cashouts/status): when you pay, hold or deny a cashout in your own admin, the cashout queue follows. The evaluate answer now includes cashout_id.
  • IP and email lookup API (GET /v1/lookup): reputation for any IP or email without an event, included in every plan (Free 1,000 a day, Pro 500,000). Lookups never count as active users.
  • Pricing page: a calculator that compares per-lookup pricing with ours, using your own numbers.
  • Free integration on every plan: no developer needed, we set up the script, the server calls and your offerwall postbacks for you. Ask from the Install page.
  • Pricing FAQ: we charge per monthly active user, never per lookup. Every event of the same user is included.
  • Postback relay: point an offerwall at CashoutGuard instead of your site. Each conversion is scored and forwarded unchanged to your postback URL, or held back in Enforce mode. No code to change.
  • Bulk check: paste up to 500 IP addresses or emails and get a verdict for each, plus how many of your accounts use it. Download the result as CSV.
  • Searching Accounts for an IP or an email now also says what it is: VPN, Tor, hosting, disposable domain, or an alias of another address.
  • Free email checker on the website: disposable domains, domains without a mail server, Gmail dot and plus aliases.
  • If only the browser script is installed, the overview shows what it already sees (VPN and hosting visitors, IPs shared by several devices) until the server API is connected.
  • The install page shows what your integration sends for each event type, and which missing field turns off which checks.

Work as a team

  • Alerts to Discord, Slack or Telegram for every newly blocked account (up to 20 an hour), with a test button.
  • A read-only Viewer role, and the owner can require two-factor authentication for everyone.
  • Review queue: Allowlist and Block buttons right in the Review and Block lists.
  • Saved views on the Events page.
  • Rotating the secret key no longer breaks your integration: the old key keeps working for 24 hours (or revoke it at once if it leaked).
  • Traffic compares each period with the one before.
  • Your month now starts on the day your plan started, like your invoice: active users, usage alerts and extra-user billing all follow it, and the billing page shows the dates.
  • Privacy: device fingerprints, links between accounts and idle end-user accounts are deleted automatically 12 months after their last activity (see the Privacy Policy, section 5).

September 2026 — Launch

Blocking that acts, and learns

  • Enforce per event type: act on cashouts only, for example, while the rest stays in shadow mode.
  • Allowlisting a flagged account teaches CashoutGuard: the rules page suggests a lower weight for reasons that keep causing false positives on your site. "This was legitimate" is one click from any flagged event.
  • A shared fingerprint alone (identical phones, nothing else in common) no longer blocks by itself.
  • Residential proxies are caught sooner: an IP seen with many time zones anywhere on the service counts, without revealing where it was seen.
  • New signal flagged_on_network: the same email or device was blocked on another website protected by CashoutGuard. The other website is never named.
  • New webhook event.blocked, fraud-spike emails, an activity log in Settings, account search by email, IP or wallet, and CSV export of accounts.

Refer & earn, support and contact

  • Refer & earn: share your link and earn 25% of what the customers you bring pay, for 12 months. They get 30 days of Growth free. Withdraw from $50 by PayPal or USDT.
  • A Support page inside the dashboard and a Contact page on the website, both answered by a person, usually within a few hours.
  • Switch between shadow and enforce mode, and between light and dark theme, from the top bar.

Billing that handles every case

  • More for the same price: Starter covers 25,000 monthly active users with the linked-accounts graph and velocity checks, Growth 100,000 and Pro 400,000. Above that, Pro users pay $1 per 1,000 extra users and are never cut off.
  • Change plan at any time from the dashboard. Upgrades charge only the difference for the rest of the month. Downgrades turn unused time into credit.
  • Cancel with one click and keep your plan until the end of the paid period, or resume before that date.
  • Failed payments show a clear notice with "Pay now" and "Update card". A declined upgrade never changes your plan.
  • Pay by card (monthly or yearly, Stripe) or with crypto (USDT, USDC, BTC, ETH and more), prepaid for 1, 3 or 12 months. 12 months cost the price of 10.
  • Invoices with your company name, address and tax ID, downloadable as PDF. Add your VAT or tax ID at checkout.
  • Yearly card plans: 12 months for the price of 10. Switch between monthly and yearly at any time.

Stronger VPN and device detection

  • Hidden VPNs: browser timezone and language compared with the IP's country, WebRTC IP leaks, and one IP seen with many timezones (residential proxy pattern).
  • Official server lists of the largest VPN providers, hosting networks, Tor exit nodes and iCloud Private Relay, refreshed every day.
  • Spoofed phones: device model checked against its graphics chip, China-only models connecting from abroad, and default emulator profiles.

Built for offerwalls and GPT sites

  • Offer names everywhere: overview, accounts, events and fraud rings.
  • Offer checks: completed too fast for the offer, conversions without a click, click and conversion from different countries, earnings far above the account's usual day, duplicate transaction IDs.
  • Conversions inherit the device and context of their click, so server-to-server postbacks are scored like browser events.
  • Offer catalog sync so every event shows the offer's real name.

Dashboard

  • Notes and activity on every account: your team writes why it approved or blocked someone, and every block, allowlist and cashout decision is recorded with who and when.
  • Clearer setup errors: agent.js explains in the browser console when a domain is not allowed or a key is wrong, and localhost always works for testing.
  • Rules per signal: switch any signal off, give it your own weight, or make it always send users to review or block. Each signal shows how often it fired in the last 7 days.
  • Export events to Excel: a real .xlsx file with headers in your language, dates, coloured decisions and filters. CSV is still there for scripts.
  • Team: invite your fraud and support team by email. Only the owner manages billing and the team.
  • Risk by source: see which publisher, offerwall or campaign sends the risky traffic.
  • Offerwall network mode: accounts sent as "publisher:user" are only linked inside the same publisher.
  • Send a signed test webhook from Sites and see if your endpoint accepted it.
  • Traffic view by country, network type, device model and VPN share.
  • Fraud rings: accounts grouped by shared device, payout address, email or IP, with the money at risk and one-click blocking.
  • Cashout review queue, rules and lists, webhooks and a 7-day shadow-mode report by email.
  • 12 languages, including Arabic and Urdu written right to left.

Security

  • Two-factor authentication for every team member: authenticator app codes plus 8 one-time recovery codes (Settings → Security).
  • Email confirmation before the dashboard and the free trial unlock.
  • Round-the-clock health monitoring and error alerts for our team.
  • A public security page and a vulnerability disclosure contact.