Odds & Fairness
What can be in every pack, how the draw is made, how you can verify after the fact that we did not put a thumb on it, and the legal analysis of why this is a retail sale of collectibles rather than a wager.
1. Every pack publishes its odds
Each pack page lists every card that can come out of that pack and the probability of each, in full, before purchase, to anyone — signed in or not. Nothing is placed behind an account, a purchase, or a scroll-to-accept.
Those odds are not marketing copy written alongside the pack. They are read from the same table the draw itself reads, which is why a pack cannot advertise one pool and open from another. The published pool is stored under its own cryptographic fingerprint, and that stored text is never changed.
We publish odds because a customer buying a randomised outcome is entitled to know its distribution. We regard concealing it as unfair within the meaning of California's Unfair Competition Law (Cal. Bus. & Prof. Code § 17200) whether or not any statute has yet said so specifically.
2. How a pack is drawn — commit and reveal
Before the draw. The server generates a secret random seed from a cryptographic source, not a time-seeded generator, and records its SHA-256 hash — the commitment — together with the fingerprint of the exact pool it will draw from. A server seed is never reused between openings. Once the commitment is recorded, neither the seed nor the pool can be changed without its hash ceasing to match.
The draw. The cards are drawn from the server seed and a client seed. Depending on the opening, the client seed was made by your browser, typed by you, or made by our server; it is recorded with the opening and printed on its proof or receipt.
What holds for every opening. Every opening's server seed is revealed once it has opened, and the SHA-256 of that seed equals the commitment recorded before the draw. Every card opened since 20 August 2026 is drawn by the published procedure from that seed, the client seed and the other inputs printed on its proof or receipt, and the published pool whose hash is printed there too. So anyone can re-derive each of those cards and see that the recorded result is exactly what that seed and that pool produce. A seed or a pool changed after its commitment would fail that check. The only openings this does not cover are the first four, made on 19 August 2026 before pools were stored under their hash: their proofs carry no pool hash, so their cards cannot be checked against a published pool, although their server seeds do match their commitments. Section 9 writes out every step, with a worked example, and section 8 says where each opening's proof is.
When the commitment comes first. When the commitment is shown to you before your seed is made — most packs you open with tokens, and seats you join after a battle is made — it also proves we could not choose our seed knowing yours. In some cases the commitment is not shown to you before the opening — for example daily packs, some free packs that open automatically, the seat of whoever creates a battle, and openings made up to 29 September 2026. Those still re-derive exactly, but the order of the commitment and your seed does not give you that extra protection. And where our server makes your client seed, as it does for daily packs and for most battle seats, the commitment alone does not show that seed was not chosen for its result.
The keepsake is not checkable this way. In a battle or group opening with more than one player, each player's keepsake card (section 6) is not drawn from a published pool. It is picked from the low-value cards in our catalogue at the prices of the moment the battle settles — never worth more than the cheapest card drawn in that battle, and if none is that cheap, from that battle's own cheapest draws — listed by Value, highest first, then by card id; its position in that list is the 32-bit FNV-1a hash of the text serverSeed:keepsake:seatIndex, modulo the length of the list. That list is not published under a hash, so which keepsake a seat received cannot be re-derived and checked the way the cards drawn can.
Old proofs stay valid. Because the pool is stored under its own fingerprint, a later re-pricing or pool revision can never retroactively make an old proof appear invalid.
3. What “Value” means, and what it does not
Value is an estimate of secondary-market price, derived from a dated snapshot of third-party sales data and displayed with the date of that snapshot. Graded cards are priced from observed sales at that grade where such sales exist; where they do not, the figure is derived from adjacent grades and is labelled as derived.
Value is not an appraisal, not a guarantee, not a promise of future price, and not an offer to buy. Buy-back offers are separate, explicit, time-limited, and shown as their own number — see the Terms.
Prices of collectibles move, sometimes sharply and sometimes downward. Nothing on the Service should be read as investment advice or as a representation that any card will hold or gain value.
4. Membership and progression never change the odds
No membership tier, level, season, achievement or Adventure alters what is inside a pack or the probability of drawing it. A paid tier that improved the draw would render the published odds untrue for everyone who is not in it, which is precisely the deception publishing odds is meant to prevent.
Tiers buy allowances, shipping benefits and cosmetic status. They do not buy luck. This is enforced in code and checked automatically before release, not left to policy.
5. Grading, condition and authenticity
Graded cards are described by the grade assigned by the third-party grading company named on the slab, together with its certification number where available. We do not assign grades ourselves and we do not represent that any grader's opinion is correct.
Ungraded cards are described by condition on a stated scale. Condition assessment is a judgement and reasonable assessors differ. Where a card you receive is materially worse than described, the remedy is in the Terms.
We do not knowingly deal in counterfeit or altered cards. A card later shown to be counterfeit will be bought back at the price you effectively paid for it.
6. Battles, community packs, and what a win is worth
Both sides open the same packs, from the same table. A battle draws every seat from the pack odds published on that pack's own page — the same table a solo opening reads. There is no separate battle pool, no per-seat adjustment, and no code path anywhere that can ask whether a seat belongs to a customer or to a bot before it draws. A bot seat is drawn by the identical call with the identical odds.
One commitment covers the whole battle. One secret seed is generated in the step that creates the battle; its SHA-256 hash and the fingerprint of every pack in the sequence are on the battle from then on, so the seed cannot be changed afterwards to suit who joins. Each seat has its own client seed, recorded on the receipt. Each card is drawn from HMAC-SHA256(server seed, battle id : seat index : that seat's seed : pack position): the HMAC's output is hashed once more into a number, and that number picks the card by weight — section 9 below writes out every step. Section 2 says when the order of the commitment and a seat's seed adds protection, and when it does not.
Afterwards, the seed is revealed and anyone can re-derive the whole battle — every card, for every seat, in order. Hash the revealed seed and check it against the commitment recorded on the battle when it was made. A mismatch would be proof of tampering. The pack pools are stored under their own fingerprints, so a later re-pricing cannot make an old battle's proof stop verifying.
Winning is close to a coin flip, and we would rather say so. Both sides open the same packs from the same odds, so neither side is favoured and the house has no stake in who wins. Simulated over 2,000 battles against our own bots, the two sides won 772 and 755, with the rest tied. What a battle is decided by is which side's draws happened to land higher — nothing else.
Where the house edge actually is. It is in the pot, not in the outcome. A community pack returns about 83% of its price in card Value, so the pot two seats fight over is worth about 83% of what the two of them paid in. Winning a 1v1 therefore returns about 1.66x your own entry rather than 2x. Your expected return is the same whether you open a pack alone or fight somebody over it — the battle moves Value sideways between players and takes nothing extra.
Ties, and the card everybody keeps. If two sides tie exactly, nothing moves: each player is credited what their own packs were worth. And in every battle and group opening with more than one player, every player receives one identified keepsake card, worth up to 50 cents and never more than the cheapest card drawn in that battle, funded out of the pot before shares are worked out; section 2 says how it is picked, and why it cannot be checked like the cards drawn. On a one-seat opening there is nothing to divide, so the cards drawn go straight to your Vault instead and no tokens are paid.
A creator picks the cards and the shape of the odds. Nobody picks a card's chance by hand. No endpoint on this Service accepts a weight, a probability or a price from a pack's creator, so a creator cannot tilt a pack toward or away from any one card. Each card's chance falls as its Value rises — how steeply is the one thing the creator chooses, from ten fixed positions we define, and the pack is priced to return 83% at every one of them. At the middle position, and at the Values the odds were set at, every card in a pack contributes the same expected Value — a card worth ten times as much comes up ten times less often. Values move after a pack is published and its odds do not, so on any given day that proportion is approximate; the table on the pack's own page is exact, and it is the table the draw uses. One consequence worth knowing before it surprises you: because of that rule a pack's expected Value is the harmonic mean of its cards', which is dominated by its cheapest card. Adding one cheap common to a pool of grails makes it a cheap pack.
Graded cards in community packs are limited to grades we stock. Public graded price feeds report a printing's price under a related printing's identity often enough that we will not price a community pack from them unsupervised. A creator may only use a card in a grade that appears in cardhouse's own packs, where the row was chosen by a person and we have a real cost against it.
7. Why this is not gambling, and not a sweepstakes
Under California law a wager requires consideration, chance and a prize contingent on the outcome — that is, the risk of losing what was staked. The following describes why that structure is absent here.
You always receive goods. Every pack delivers real collectible cards of stated Value. There is no outcome in which the customer receives nothing. What varies is which goods, and how much they are worth on a secondary market we do not control — the same variability that attaches to a sealed booster box bought at retail.
Nothing pays out in cash. Tokens are closed-loop value, spendable only here, on our own goods and services. They cannot be withdrawn, transferred, gifted or converted to money, and neither can anything drawn from a pack: selling a card back pays tokens. There are no cash prizes and no cash equivalents.
There is one currency, not two. cardhouse has no second, "promotional" or "sweeps" currency that is redeemable when the first is not, no free-entry alternative method of participation, and no prize draw of any kind. It is a retail sale of randomised collectibles, with the distribution published rather than left to the manufacturer.
Nothing simulates a casino. There are no slot machines, reels, wheels, card tables, dice, house games or wagers, and no presentation of an opening that imitates one.
Applying the statute. California's AB 831, in force since 1 January 2026, prohibits online sweepstakes games that combine (i) a dual-currency system, (ii) simulation of casino-style gambling, and (iii) awards of cash or cash equivalents — and extends liability to platforms and payment processors that support them. cardhouse satisfies none of the three limbs, and the absence of any one of them would be sufficient. Nor is the Service a lottery under Cal. Penal Code § 319, which requires a distribution of property by chance among persons who have paid for a chance at that property, rather than a purchase of property itself.
What this section is. It is a description of how the Service is built and the reasoning that follows from it. It is not legal advice, it is not a warranty, and it does not bind any court or regulator. It is published so that the reasoning can be checked against the product rather than asserted.
8. Check us
Every opening you make is kept on our server with its commitment, its seeds, its pool fingerprint and its cards. For a pack opened with tokens or as a free pack, its proof, with a Check these cards button, is on the proof sheet you can open once the pack has opened. The sheet holds your recent pulls of this visit only — the latest and up to eleven before it — and is emptied when you reload, leave or sign out; a daily pack has no proof sheet on the site. The site does not yet list your past openings' proofs, but you can ask us at support@cardhouse.live for the proof of any opening — the seeds, the pool hash and the cards — for as long as we keep its record (see the Privacy Policy: generally seven years for financial records), and re-derive it yourself with section 9. A battle's receipt, with its seeds, every card drawn and a Check this card button on each, stays on that battle's own page.
If a verification ever fails, tell us at support@cardhouse.live. A broken proof would be the most serious defect this Service could have, and we would treat a report of one accordingly.
Section 9 below is the whole procedure, step by step, so that you can do it without us — by hand, in a few lines of code, or with the Check this card button on a battle's receipt and the Check these cards button on the proof sheet of one of your recent pulls this visit, which run the same steps in your own browser and show you every value along the way.
cardhouse is not affiliated with, endorsed by or sponsored by The Pokémon Company International, Nintendo, Creatures Inc., GAME FREAK inc., or any grading company. All trademarks are the property of their respective owners, and cards are resold under the first-sale doctrine.
9. Re-derive a card yourself, step by step
This is exactly what our server does to turn seeds into cards, written so that you can repeat it by hand, in a few lines of code, or with the check buttons on a battle's receipt and on the proof sheet of one of your recent pulls this visit. Each step is in plain words first and as a formula second. Text is always hashed as UTF-8; "hex" means lowercase hexadecimal. Seats, packs, blocks and positions in a list are counted from 0; the numbers drawn from the blocks are counted from 1 (number 1 is the first).
A. A battle card — every battle, and a community pack opened on its own. A battle pack gives each seat one card, so it uses one number.
Step A1 — check the seed. Hash the revealed server seed (the 64-character text exactly as shown) with SHA-256. It must equal the commitment recorded on the battle in the step that made it. SHA-256(serverSeed) = commitment
Step A2 — check the odds. Open the pack's pool link on the receipt, /api/pools/<pool hash>. The SHA-256 of that exact text must equal the pool hash. Its outcomes list, in the order written, is the list the draw walks.
Step A3 — write the HMAC input. Join four things with colons: the battle id, the seat number, that seat's client seed, and the pack's position in the battle. The battle id is the internal id printed on the receipt, beginning bat_ — not the CHB code. Seats and packs count from 0. input = battleId + ":" + seatIndex + ":" + clientSeed + ":" + packIndex
Step A4 — key it with the seed. Compute HMAC-SHA256 with the revealed server seed (as text) as the key and the input from A3 as the message, and write the result as 64 hex characters. We call that result m. m is what comes out of the HMAC; the input string is not m. m = hex(HMAC-SHA256(key = serverSeed, message = input))
Step A5 — hash m into a block. Take the 64 characters of m, add a colon and a zero, and hash that text with SHA-256. Call the result d. d = SHA-256(m + ":0")
Step A6 — turn d into a number. Read the first 8 hex characters of d (its first 4 bytes) as one whole number, big-endian, and divide by 232 = 4,294,967,296. That is r, at least 0 and less than 1. r = int(d[0..8], 16) / 4294967296
Step A7 — find the card. Add up the weights of every outcome in the pool: that is W. Work out r × W. Go down the outcomes in the order they are listed, keeping a running total of their weights. The card is the first outcome whose running total is greater than or equal to r × W. If rounding leaves none, it is the last outcome on the list. If the weights add up to zero or less, or their total is not a number at all (a weight that is not a number), the card is also the last outcome, without walking. The weights do not have to add up to 1; only their proportions matter.
B. A pack opening — the packs on the Packs page, whether opened with tokens, as a free pack or as a daily pack. No HMAC is used here.
Step B1 — check the seed and the odds, as in A1 and A2: the SHA-256 of the revealed server seed must equal the commitment on the proof, and the pool text must hash to the pool hash. Section 2 says when that commitment was also shown to you before your seed was made. A pack's pool text has a product.companions number (how many companion cards the pack holds), a companions list and a feature list.
Step B2 — write the message. Join the revealed server seed, the client seed and the nonce with colons. The client seed is the one recorded on the proof (for a daily pack, a daily:… seed our server picked). message = serverSeed + ":" + clientSeed + ":" + nonce Its SHA-256 is the "outcome digest" on your proof; it confirms the message, but the cards come from the blocks in B3, not from the digest itself.
Step B3 — make the numbers. Hash the message followed by a colon and a block number, starting at 0. Each block gives eight numbers, one from each 8 hex characters, read as in A6, in order. Block 0 gives numbers 1–8, block 1 (":1") numbers 9–16, and so on. blockk = SHA-256(message + ":" + k)
Step B4 — the companion cards come first. If product.companions is N, numbers 1 to N pick the companions, one number each, in order. With L cards in the companions list, the number r picks position floor(r × L), counting from 0 (never past the last). Every companion is equally likely. These are the first N cards of your pack. position = min(L − 1, floor(r × L))
Step B5 — the feature card comes last. The next number, number N + 1, is walked through the feature list exactly as in A7. It is the last card in the pack. A five-card pack (four companions and a feature) uses numbers 1 to 5, all from block 0.
Worked example — a battle card. These values were made up for this page; they are not anybody's battle, so there is no pool link to fetch. Server seed 8fa172b8298904107b777d2733563668c45063ea29c5c258c93ffb13fc619832, whose SHA-256 is the commitment cc189503962e2cd57d08cf34d590caf9e634d0be73376cf370a235bc1b97bf9b. Battle id bat_example_00000001, seat 0, client seed example, pack 0. The pool lists four outcomes in this order: sv3pt5-1 (weight 4), sv3pt5-4 (weight 2.5), sv3pt5-6 (weight 1), sv3pt5-199 (weight 0.5), so W = 8.
A3: input = bat_example_00000001:0:example:0. A4: m = e887d0cb9a28760c00afc3cb185c66147249580803deacca3b8cb8ef68d23669. A5: d = SHA-256(m + ":0") = b5268d7fa5993ccdc5036f20e6c9c4e90eb28a027ea107b8a8ebf9ce2c2b019a. A6: b5268d7f = 3,039,202,687, and 3,039,202,687 ÷ 4,294,967,296 = r = 0.7076195178087801. A7: r × W = 5.660956142470241. Running totals: 4 after sv3pt5-1 (less than 5.66), 6.5 after sv3pt5-4 (reaches it) — the card is sv3pt5-4.
Worked example — a five-card pack. Same made-up server seed, client seed example, nonce 1. The pool has four companions to draw (N = 4) from a list of six (L = 6), and a feature list weighted 6, 3 and 1 (W = 10). B2: message = 8fa172b8298904107b777d2733563668c45063ea29c5c258c93ffb13fc619832:example:1. B3: block 0 = SHA-256(message + ":0") = 531f079f83e5595e2340c5fdf95ae7affc7287eb915988ed65e35d7d5e0e384b.
Its first five numbers: 531f079f → 0.32469222671352327, 83e5595e → 0.515218339394778, 2340c5fd → 0.1377071135211736, f95ae7af → 0.9740433504339308, fc7287eb → 0.9861226032953709. B4: the companions are at positions floor(0.3247 × 6) = 1, floor(0.5152 × 6) = 3, floor(0.1377 × 6) = 0 and floor(0.9740 × 6) = 5 of the companions list. B5: r × W = 9.86122603295371; running totals 6, 9, 10 — the feature card is the third on the feature list.
Questions about anything on this page: support@cardhouse.live