How an Online Tarot Draw Works: Fair, Random and Verifiable

You tap a row of face-down cards, pick three and flip them over. Then the doubts arrive. Was that really random? Could the site have arranged the result in advance? Would the cards have been different if you had let the site deal? This guide explains, in plain language, how a well-built online tarot draw system works, using ORAE's draw engine as the example. You don't need any maths background, just a few key ideas, and you'll be able to judge any draw system for yourself.

Three Questions a Good Online Tarot Draw Must Answer

1. How does the deck get mixed? Where the randomness comes from and whether the shuffle is done correctly.
2. Could the result be tampered with? Whether the site could swap cards after you make your choice.
3. Can it be checked afterwards? Whether the same reading can be recomputed and confirmed.

One clarification first: these questions are about whether the draw is fair, not whether tarot is accurate. That second question is about psychology and interpretation, and we cover it in Is Tarot Accurate?

Where Randomness Comes From: Math.random vs Secure Randomness

Computers follow instructions, so most "random" numbers are really pseudo-random. A program starts from a value called a seed and runs it through a fixed formula to produce a long stream of numbers that look scrambled. The same seed always gives the same stream.

Math.random is JavaScript's built-in random function. It's handy for animations and casual games, but the language specification doesn't require it to be cryptographically secure, and implementations are tuned for speed. In principle, someone who sees enough of its output can predict what comes next. And when a draw happens entirely in your browser with Math.random, there's usually no record anyone can check later.

A CSPRNG (cryptographically secure pseudo-random number generator) is different. The operating system gathers hard-to-predict input from hardware and system events and produces numbers that can't be predicted even after you've seen the earlier ones. Encrypted connections and online banking rely on this kind of generator.

On ORAE, when you press "Shuffle the deck", the server asks the operating system's CSPRNG for a 256-bit seed, one of roughly 1.16×10⁷⁷ possible values. That seed becomes the key for HMAC-SHA256 run over an increasing counter, which produces a long stream of random numbers. This design is called a DRBG (deterministic random bit generator). Anyone who knows the seed can recompute everything; anyone who doesn't can't predict a thing.

The Shuffle: Why Fisher–Yates Is the Standard

Random numbers are only half the job; you also need a correct shuffle. The Fisher–Yates shuffle is simple. Start with the last card, swap it with a randomly chosen card from those not yet fixed, then fix that position. Move one card forward and repeat until you reach the first. A 78-card deck takes 77 swaps, and every possible order is exactly equally likely.

Two common mistakes quietly break that guarantee.

Mistake one: swapping each card with any card in the deck

With just three cards, this method has 27 equally likely paths spread over 6 possible orders. Because 27 isn't divisible by 6, some orders come up with probability 5/27 (about 18.5%) and others 4/27 (about 14.8%), when each should be 1/6 (about 16.7%).

Mistake two: modulo bias

Suppose a site picks a card by taking a random number from 0 to 255 and using the remainder after dividing by 78. Since 256 = 3×78 + 22, the first 22 cards each get a 4/256 chance (about 1.56%) and the rest only 3/256 (about 1.17%), so those first cards are a third more likely to turn up. ORAE uses rejection sampling instead: whenever it needs to choose among n cards, any random number outside the largest exact multiple of n is thrown away and redrawn, so every outcome is exactly equally likely.

One more detail for the curious. There are about 1.13×10¹¹⁵ ways to order 78 cards, more than the number of 256-bit seeds, so not every complete deck order can occur. But you only ever see up to 12 cards, about 8.5×10²⁵ possible results including reversals, and each of those results corresponds to an enormous number of seeds. In practice the outcome is uniform.

Commit First, Reveal Later: What the Seed Hash Does

Picture a magician who writes the answer on a slip of paper, seals it in an envelope and hands it to someone in the audience before the trick begins. A SHA-256 hash is that envelope. The same seed always produces the same 64-character hash, but working backwards from the hash to the seed is practically impossible.

Here's how that plays out on ORAE:

1. You press "Shuffle the deck". The server creates and stores the seed and computes its SHA-256 hash, the commitment. Your browser receives the hash, not the seed.
2. The shuffle, pick or cut screen. Before a single card is dealt, the full hash is shown at the bottom ("Seed hash, published before the draw"). You can screenshot or copy it.
3. You pick or cut. Your choice goes back to the server, which shuffles with the seed that was already fixed and takes your cards; the reveal screen shows the first 16 characters of the hash again.
4. The result page. Open "Draw verification" to see the draw engine version, pinned meaning versions, the rule set version, the full seed hash and the seed itself.

To check it yourself, convert the seed's hexadecimal text back into its 32 bytes and run SHA-256 on them. The result should match the seed hash exactly, which shows the seed was fixed before you chose and wasn't swapped afterwards.

One honest limit: a commitment proves the seed was fixed before you chose and wasn't replaced afterwards; that the seed came from secure randomness in the first place rests on the system's published design. For a strict check, note the hash shown on screen before the cards are dealt.

Does Picking or Cutting Change the Odds?

No. The shuffled order lives only on the server, out of sight. With "Pick by hand" you choose positions; with "Interactive cut" you choose where to split the deck. Either way, those positions point into a deck that has already been uniformly shuffled and that you cannot see.

Think of a friend who thoroughly shuffles 78 cards and spreads them face down. Whether you favour the left, the right or always take the first card, every card has a 1-in-78 chance of landing in the first position. ORAE's tests specifically check a user who always picks the same spot, and the cards still come out evenly. Mathematically, all three methods are equivalent to auto-dealing.

So why offer a choice at all? Because ritual helps people focus. Like the veil behind The High Priestess, the cards stay unknown until they're turned, and that moment of waiting is part of the experience.

Your question text and whether you have an account are not inputs to the draw either. Neither has any effect on which cards appear.

Upright or Reversed: How Reversals Are Decided

Each card's orientation is decided separately. The system draws a number from 0 to 999,999; below 500,000 the card is reversed, so by default every card has an even 50% chance. Every drawing method uses the same amount of randomness for this step, so the positions you pick have no effect on orientation. With "Upright only", every card comes out upright.

At those odds, three upright cards in a row happens 1 time in 8 (12.5%), so at least one reversal turns up 87.5% of the time. In a ten-card spread, an exact five-five split happens only about 24.6% of the time. Reversals are normal; see upright and reversed cards for how to read them. With a physical deck, reversals come from turning part of the deck while shuffling, so the ratio isn't necessarily half; our guide to shuffling and cutting tarot cards explains how.

What "Verifiable" Actually Means

• It can be recomputed. The same seed, the same canonical deck order and the same picks or cut always give the same result from the same draw function. ORAE stores the result, your picks or cut, a snapshot of the deck order, and the engine, meaning and rule versions, so every draw can be traced.
• It's drawn once. Double clicks, network retries and page refreshes all return the same result. Asking again means starting a new reading, not rewriting the old one.
• It isn't a promise of accuracy. Verifiable means the draw wasn't tampered with. It doesn't mean the cards predict your future. For why readings sometimes feel uncanny, see why tarot feels spot on.

Weak vs Good Draw-System Practices

	Weak practice	Good practice
Randomness	Math.random, or the clock as a seed	The operating system's CSPRNG
Shuffle	Swap with any card, sort randomly, use a plain remainder	Fisher–Yates with rejection sampling
Deck order	Visible in the page source before you pick	Shuffled in advance and kept hidden
Your question	Used as the seed, so the same question always gives the same cards	Not an input to the draw
Refresh	Reloading deals new cards	One draw per reading
Reversals	Unknown ratio	Decided per card, ratio disclosed
Audit trail	Nothing recorded	Hash first, seed revealed, versions stored
Explanation	None	Method published

You can use this table to check any site. For a wider checklist, see how to choose an online tarot site.

A Fair Draw Is Where a Good Reading Starts

The scales of Justice ask that things be weighed as they really are, and the Wheel of Fortune reminds us that chance is always turning. A good draw system makes the turning fair and transparent. What the cards mean for you still comes down to your own reflection and choices.

To see the verification details for yourself, do a reading on the online tarot page and open "Draw verification" on the result page. For something quicker, draw a single card.