Metrics table
The table is reloaded from the platform when the page opens (values are cached there for up to 10 minutes). The same data in machine-readable form: /api/public/tech-metrics (JSON).
| Metric | Value | Period | How it is measured |
|---|---|---|---|
| API response time — all requests | P50 27 ms · P95 108 ms (4-day data) | last 7 days | From the server receiving a request to sending the response. File uploads and AI chat are excluded. |
| API response time — all requests | P50 27 ms · P95 108 ms (4-day data) | last 30 days | Same, 30-day window. |
| Server error (5xx) rate | 0% (4-day data) | last 30 days | Share of responses that ended with a 5xx status. |
| Public partner API (/api/v1) response time | P50 42 ms · P95 111 ms (4-day data) | last 7 days | Only requests made with an API key or OAuth token. |
| Postback and webhook delivery | being measured | last 30 days | Messages sent to partner endpoints: share that received a 2xx response. |
| Status change → postback delivered | being measured | last 30 days | From the order status change to a 2xx response from the partner server. |
| Orders counted twice | 0% | last 90 days | Delivered orders whose pair was also delivered (same phone + same product within 24 hours, or the same external order ID). |
| Repeat orders created | 0.2% | last 90 days | A second order from the same phone for the same product within 24 hours (e.g. a double click). |
| Meta lead → order uniqueness | 100% | all time | Meta lead-form leads that became orders: one lead, at most one order. |
| Age of status data | median 3 min · P95 5 min (4-day data) | last 7 days | Sampled at random moments: time since the last successful status check. |
| Payout speed | median 14 min · 67% within 1 hour · 96% within 24 hours | last 90 days | From a withdrawal request to the approved payout. |
| Site speed (Lighthouse, mobile) | Performance 84/100 · LCP 2.5 s · CLS 0.02 · SEO 100/100 · Best practices 100/100 · Accessibility 96/100 | measured on 2026-11-18 (3 runs, median) | Home page, Google Lighthouse mobile profile. |
| Orders with a recorded click | being measured | last 30 days | Share of orders placed on a link page (/l/…) where the click log has a click by the same visitor within 24 hours. Loss = 100% − this share. Shown from 30 orders. |
| Bot click share | 0.3% | last 30 days | Link page opens identified as bots, link previews or prefetch. They are excluded from click statistics. |
How we measure
- API response time. The duration of every request is measured inside the server — from receiving the request to sending the response. Values go into an in-memory histogram and are written to the database as a daily aggregate every 5 minutes. P50 means half of the requests are faster, P95 means 95% are faster. Histogram buckets are about 20% wide, so the ms values are accurate to roughly ±10%. Network time between the user and the server is not included. Excluded: file uploads, AI assistant chat and voice messages, the MCP stream and file downloads — they depend on size and external services.
- Server errors. Share of those requests that ended with a 5xx status.
- Postbacks and webhooks. Every delivery attempt is recorded: the partner server's response code, the number of attempts and the delivery time. If the response is not 2xx, the message is retried (after 1, 5, 15, 60 minutes and so on, 8 attempts over about 2 days). Until the sample is large enough, the row says "being measured".
- Duplicate orders. Computed from the order database for the last 90 days (see the section below).
- Payout speed. Approved withdrawal requests: from the request to the approved payout. Amounts and the number of payouts are not published.
- Site speed. Google Lighthouse, mobile profile (simulated slow network and CPU), home page, 3 runs, median of each metric. The date is shown in the row.
How duplicate protection works
- One phone — one product — 24 hours. A second order from the same phone for the same product within 24 hours is rejected; one phone can order at most 3 different products within 24 hours. The rule is the same on every channel: link page, shop, API and AI assistant.
- Idempotency-Key in the API. A request repeated with the same key does not create a new order — the existing one is returned. Even if the connection drops and the request is sent again, there is no duplicate.
- Meta lead forms. Each lead ID is unique in the database: one lead becomes at most one order.
- On the call-center and delivery side. If an order is rejected there as a duplicate, it is marked as not created and earns no commission.
- Commission only for delivered orders. The commission is credited when the order becomes "delivered"; if the status changes afterwards, it is reversed.
- Webhook messages. Every message has its own ID (
evt_…), so the receiver can drop repeats. Incoming status messages are also checked by ID, and re-applying a status does not change the result.
"Repeat orders created" in the table are identical requests that arrived a few milliseconds apart (a double click). "Counted twice" are cases where both orders of such a pair ended up delivered.
Click and order tracking
- Clicks are counted server-side. When a partner link (
/l/…) opens, the counter increases on the server — it does not depend on a browser pixel or an ad blocker. - Click log. Every click goes into a lightweight log: bots, link previews and prefetch are flagged separately, and a repeat open by the same visitor within 30 minutes is not counted as unique. Raw IPs are not stored — the visitor key is hashed with a salt that changes every day. Period clicks, unique clicks and CR in the partner dashboard come from this log.
- click_id and sub1–sub5. Tracker parameters on the link (
?click_id=…&sub1=…) are stored with the order and returned in the postback. - A postback on every status change. When an order goes new → confirmed → delivered or cancelled, your tracker (Keitaro, Binom and others) receives a message.
- From your own site — via the API. You send the order from your server and attach your own ID (external_id).
Tracking accuracy. For every order placed on a link page we check whether the log has a click by the same visitor within 24 hours. The "Orders with a recorded click" row is that share; tracking loss = 100% minus it. Visitors who opened the page and left without ordering are not a loss — that is ordinary conversion. A person who switches networks (say, Wi-Fi to mobile data) may look like two visitors, so the metric tends to overstate the loss rather than hide it. Until 30 orders are collected the row says "being measured". Setup guide: API, postback and MCP.
Limitations
- These metrics are measured by the platform itself and have not been externally audited. The only independent source is the monitoring link on this page (when shown).
- We do not publish an uptime percentage ourselves: that should come from external monitoring.
- API response time is server-side time. What you wait on your phone also depends on network quality and distance.
- Measurement started in November 2026 (the exact date is shown under the table): until the 7- and 30-day windows are full, each value shows how many days of data it covers.
- The number of orders, requests and payouts, and all amounts, are intentionally not published — only percentages and times.
How data is protected: Security and data protection. What the platform can and cannot do: Technical capabilities. Criteria for evaluating a CPA network technically: Technical checklist; changes: Changelog.
FAQ
Where do these numbers come from?
From the platform's own measurements: API response time is measured inside the server on every request; the rest (postbacks, duplicates, payout speed) is computed from order and delivery records. The table refreshes when the page opens; the same data is available as JSON at /api/public/tech-metrics.
Why do some rows say "being measured"?
There is not enough data for that metric yet, or it cannot be measured honestly (for example, tracking accuracy until 30 orders are collected). We do not publish estimates.
If a customer orders twice, is the commission paid twice?
No. A second order from the same phone for the same product within 24 hours is rejected, the API supports Idempotency-Key, and Meta lead IDs are unique. Commission is credited only for delivered orders. The share of orders counted twice is shown in the table.
How fast are postbacks delivered?
A status change is queued immediately and the sender runs every 15–20 seconds. If the partner server does not respond, the message is retried 8 times over about 2 days. The measured median and P95 are in the table (once the sample is large enough).
How do I know if the site is down?
If an independent monitoring link is shown on this page — there. We do not publish an uptime percentage ourselves.