English · 中文

How Accurate Is CountPilot? Our Test Results, Including the Bugs

A counter is only as good as the cards it misses without telling you. These are the numbers from our own recordings of real tables — including the mistakes we found along the way.

CountPilot guides · Updated 8 October 2026

What we measured

TestResult
Hands where the opening cards were caughtabout 97% (558 hands, 9 clean shoes)
Hands checked card by card against the physical cards88 hands, about 271 cards
Hand-total digits read correctly (held-out recording)122 of 122
Readings per secondabout 6
Recorded real-table frames used for regression tests39,872

These measurements are from Playtech regular seated tables on a 2940×1912 screen, where the digits are about 21 pixels tall. On smaller screens digits get smaller, and the app tells you when they are too small to read reliably.

The errors we found — and why statistics didn't catch them

Every serious mistake we found looked completely normal on screen. The catch rate, the card distribution and the digit accuracy were all fine while these were happening. They were found by checking hands card by card against the physical cards:

What it still can't see

How the error range is calculated

Each missed card is random, so it doesn't push the count one way, but it adds uncertainty. With a Hi-Lo standard deviation of about 0.877 per card, missing n cards adds roughly 0.877 × √n to the running count, divided by the decks remaining for the true count. That is the ± you see next to every true count.

Check that counting aids are allowed where you play — many casinos' terms prohibit them.

Try it on a real table. CountPilot reads the live table for you and shows the true count with its error range, bet sizing and strategy prompts in a small window over the game. Free 30-minute trial, macOS and Windows.
Try it free in your browser

← All guides