Rules
Standings rules and drop rates.
Two things are published before anything is given out: how a standing is calculated, and what chance a booster has of being what. Both are the figures the server actually uses.
First written on 22 September 2026; the wording may change. The drop rates are the figures the server applies today. The standings rules describe how each standing is calculated and what it gives in the game.
On this page
Standings rules
Every reward is in the game: INK, boosters and season cosmetics. None of them can be bought, and none ever converts into money or anything of monetary value.
1. Free entry, and no stake of any kind
Every mode is free. No payment is ever needed to play, to enter or to be rewarded. An optional PRO subscription is coming soon. It adds market tools, cosmetics, a bigger free-booster reserve (you keep up to 10 free boosters instead of 3, so if you come back once a day you can find up to 6 instead of 3) and the PRO track of the season pass, which has extra boosters and INK. It never sells a booster, never speeds one up, never changes the drop rates, never raises the daily limit of 11 boosters, and never improves your place in a standing. Nothing else is for sale: no game currency, no booster on its own, no card, no cosmetic.
A player never puts up anything: no money, INK, card or anything else of value. There is no entry fee, no ticket, no position to take and no counterparty. A player can only receive. Wikilore is a free game of skill. It is never a prediction market, and picks are never traded between players.
2. Who can enter
- 18 and over. Age is declared at sign-up. An account that turns out to belong to a minor is closed.
- One account per person. Holding or using several accounts is prohibited, and so is having someone else play an account.
- You sign in with Google. Every week counts in the standings, with or without X. A linked X account is needed only to receive a reward booster, and it must be linked before the rewarded period starts: otherwise you get the INK and the title, not the booster. X is also required to trade, directly or publicly, and to message a player who isn't your friend.
- The service is not open to people or territories covered by restrictive measures that would prohibit it.
3. What is forecast, and what never is
A Wikilore forecast is always and only about how much a Wikipedia article is read in a given window. It is never about a match, a goal, a transfer, an election, a trial, a share price or the weather. You don't guess who wins. You guess who gets read.
A pick is also never about a period whose views have already been published. Every pick window closes before the period it measures begins: the leagues and Up or Down close Sunday 23:59 UTC for a week that starts the next day, and the Trend Call closes at 12:00 UTC for views counted from 12:00 to 18:00 UTC that same day. Those views are read from the public hourly files of Wikimedia, which appear about two hours after each hour. When you choose, the answer doesn't exist anywhere yet. This applies to themed leagues exactly as it does to the open one.
4. How each standing is calculated
- The Open League (weekly)
- Three picks from a published list of 16 subjects (since the lists of 19 October 2026; the week of 12 October keeps 15 subjects, and the week of 5 October 30 subjects and five picks). The list is published Monday 12:00 UTC and picks close the following Sunday 23:59 UTC. The measured week begins the next morning, Monday 00:00 UTC, and runs to Sunday 23:59 UTC. It is resolved on the Monday 12:00 UTC that follows. Each subject's growth is the week's views divided by seven times its daily average over the previous 28 days. Each subject also has an expected growth, computed by the server from the days published before the close and frozen at the close (how it is computed). The 16 subjects are ranked by real growth divided by expected growth. A pick scores its position: 16 points for first, 1 for last, ties share the average. Three picks give a score between 6 and 45. Three random picks are worth 25.5 ± 7.4, and that line is published on the standings.
- Themed leagues
- The same rules, on a list of 16 subjects limited to one theme, with a floor of 1,000 views a day and no subject reused within eight weeks. Each themed league has its own standing and its own rewards; leagues are never added together. The seven themes are Football, Screen, Music, Games, Minds, Politics and Tech & business, and the list is closed. Games has been suspended since 24 September 2026 and publishes no list, because too few of its subjects stay above the floor all year. It comes back when a quarterly test says so. The politics list is suspended for the week of a major national election, on a calendar published in advance.
- Trend Call (daily)
- Five pairs a day, the same for everyone. The two subjects of a pair are matched to within 5% of each other's readers over the seven days known when the pairs come out. The pairs of a day come out the day before at 12:00 UTC and close at 12:00 UTC on the day itself, so everyone has 24 hours. They are about that day's views from 12:00 to 18:00 UTC. At the close, the server publishes an expected figure for each subject: its views over 12:00–18:00 UTC as the server expects them, from the hours already published that morning, its usual day, its size and its recent trend (how it is computed); a provisional figure is shown as the morning's hours come in, and the figure at 12:00 UTC is the one that counts. The winner of a pair is the subject whose real views land further above its expected figure: the most-read subject does not necessarily win. A tie, a death or missing data makes the pair neutral. The result is published the next day at 12:00 UTC. Ten INK per correct pick. The week's bar is at least 25 correct picks out of the week's 35 (part 5). The bar is out of the week's 35 pairs, not out of the pairs you played, so missing three days puts it out of reach. That is on purpose: the Trend Call is a daily appointment.
- Up or Down (weekly)
- Thirty subjects, the same for everyone, drawn Monday 12:00 UTC among subjects whose recent week sits between 0.85 and 1.2 times their usual week. Answers close Sunday 23:59 UTC. The question for each subject is better or worse than expected?: at the close the server publishes an expected ratio, next week's views divided by this week's, from the days already published (how it is computed). At the result, each subject gets its real ratio divided by its expected ratio; the fifteen highest of the list are "better than expected", the fifteen lowest "worse". There is no neutral zone; a deceased subject leaves the list and counts for nobody. A subject left without an answer counts as wrong. Five INK per right answer, twenty-five more for a perfect list.
- Face Down (weekly points)
- One point per counted win. A duel counts only against a server-matched opponent who is not a friend, and only for the first five counted duels of the game day, won or lost. This way neither screen time nor a farm of accounts decides the standing. The week runs from Monday 12:00 UTC to the next Monday 12:00 UTC.
5. Rolling standings and their rewards
- Leagues and Up or Down. The standing is the sum of the last four weekly scores (an unplayed week counts as zero), with at least three weeks played in the window to appear in it.
- Each league (each one apart; they are never added together) gives 20 INK for every week you play, when its result comes out. The same Monday, on the four-week standing: the top tenth get a Chapter and 100 INK; the top hundredth get a Codex instead of the Chapter, with the same 100 INK.
- Up or Down gives 5 INK per right answer and 25 more for a perfect list. Each Monday a result comes out, the top tenth of the four-week standing get a Chapter.
- Trend Call. This one has a different, simpler rule, with no rolling window. Ten INK per correct pick, and streak boosters: 3 correct in a row gives a Page, 7 a Chapter, 12 a Codex. A perfect day (the day's five pairs all counted and all right) gives 25 more INK. A week at 25 correct or more out of its 35 gives a Codex and, the first time in the season, the title "Trend reader · All Hallows 2026". The week is settled once its seven days are settled, on the following Monday after 12:00 UTC. The high bar is what keeps a farm of fake accounts out: chance clears 25 out of 35 about one week in 120.
- Face Down. On the week's points, the top tenth get a Chapter; the top hundredth get a Codex instead and, the first time in the season, the title "Duel elite · All Hallows 2026".
- The top tenth is the first tenth of the players in the standing, rounded up, so a standing of 5 rewards one player. The top hundredth is the first hundredth, rounded down, so it takes 100 players in a standing to reward one.
- Equal points in a league. The player whose picks beat their expected growth by more goes first: the average, over their counted picks in the four weeks, of each pick's real growth ÷ expected growth (a subject who died is left out). It is shown on the standing. There is no draw and no bonus for sealing early. A week of another size (30 subjects on 5 October, 15 on 12 October) counts in the standing at the scale of 16: its points keep the same place between the lowest and the highest possible score.
- A tie at the cut. Players still tied (in a league, on points and on the tie-break) share the same rank, and that shared rank is the one that counts: if it falls inside the top tenth, each of them gets the reward. Nobody above the tie gets less, and no draw is ever used.
- Reward boosters go onto the shelf (or wait for a space there), outside the daily limit of 11, with their tier fixed and the usual INK at opening. They need an X account linked before the rewarded period starts (the first pairs of the Trend Call week, the first list of a four-week standing, the Monday of a Face Down week); INK and titles don't. Each reward is given once per account and per period. Season titles are exclusive and never sold.
6. How "expected" is computed
Every subject in a league, in Up or Down and in the Trend Call is judged against what the server expects of it, not against the other subjects' size. The expected figures are computed by formulas that are frozen in dated files, published by the server at /api/events/formulas, and changed only by a dated decision. They read nothing but public view counts.
- Leagues: expected growth = exp of a weighted sum of five ingredients: momentum (the log of the last seven published days over the subject's 27-day base), last three days (the same over the last three), rise (the log of the last seven days over the 21 before them), spike (the log of the highest day of the base over its median) and size (the log of the base). The list is ranked by real growth ÷ expected growth.
- Up or Down: expected ratio = exp of a weighted sum of five ingredients: over base (the log of this week's published days over the 28 days before), over median (the same over the median of the previous whole weeks), slope (the log of the week's last published days over its first three), over last week (the same over the week before) and spike (the log of the highest day over the median). "Better" is the top half of the list by real ratio ÷ expected ratio.
- Trend Call (version 2): expected views over 12:00–18:00 UTC = a fixed factor × the subject's usual 12–18 UTC views (the median of the previous days) × exp of a weighted sum of fifteen ingredients: today's published hours against the usual, the last three hours, the hours shown the day before, size, the 7-day average, the 7-day trend, the same weekday a week and two weeks ago, the month's spike, the last three days' spike, a recent rise, and five yes-or-no marks (a weekly pattern, a spiky month, a spike under way, a rise under way, a sport, screen or music subject). The hourly counts come from the public hourly files of Wikimedia, agent "user".
-
Coefficients (frozen 8 October 2026)
- Leagues: exp(0.399 + 1.187 × momentum + 0.537 × last three days − 0.754 × rise − 0.123 × spike − 0.042 × size).
- Up or Down: exp(0.067 − 0.141 × over base − 0.085 × over median + 0.189 × slope − 0.032 × over last week − 0.127 × spike).
- Trend Call: factor 0.899; today's published hours against the usual 0.456, the last three hours 0.202, the hours shown the day before 0.048, size −0.19, the 7-day average 0.169, the 7-day trend 0.056, the same weekday a week and two weeks ago 0.035, the month's spike −0.019, the last three days' spike −0.063, a recent rise −0.05; marks: a weekly pattern 0.022, a spiky month 0.005, a spike under way 0.046, a rise under way −0.053, a sport, screen or music subject −0.016.
The same figures, as the server reads them:
/api/events/formulas. - Why: measured on real views, the most-read subject of a pair could be guessed 72% of the time by reading the public hourly files, following the displayed rise in a league finished near the top almost every month, and answering "down" everywhere in Up or Down was right 59% of the time. Against the expected figures, no program reading public views does better than chance within the published margin. Each formula is held to a public test on at least four weeks of data it has never seen, with no program above 52% per pair.
7. Deaths are neutralised
If a subject dies between the publication of a list and the end of the measured week, the pick is worth the median position, so it neither gains nor loses. In Trend Call a death voids the pair: it gives no point and has no effect on the streak. Detection happens at resolution, from the article's categories and the Wikidata date of death. A death recorded later is corrected within seven days.
8. Protection of the measurement
Artificially causing views of a Wikipedia article, by any means, to influence a standing or a card's rank is prohibited. It also doesn't work: every listed subject has a view floor, and each subject's week is compared with its own earlier split between desktop and mobile readers. That comparison is measured and shown to the team; no exclusion is automatic, a human looks first (part 9).
Also prohibited: multi-accounting, automation of any kind (script, extension, artificial intelligence) to play or to help answer, collusion between players, knowingly exploiting a bug, and trading cards, INK or an account for money or anything of monetary value, inside or outside the game. This does not cover card-for-card trades inside the game, which follow their own rules.
9. Detection, sanctions and disputes
Statistical checks run weekly before any standing reward is given: response times, success on close pairs, agreement with public data, and clusters of accounts. Thresholds are not published; the principle is.
No sanction is automatic: a statistical flag triggers a human check, never a direct exclusion. Sanctions are graduated: a warning, victory boosters withheld, removal from a standing, a sale cancelled, a reward withheld, exclusion from standings and their rewards on a second flag, and account closure for multi-accounting, automation or attacks on the service. A sanctioned player is told, and can dispute it at the contact address.
10. Everything is published
We publish the list with its week number, everyone's picks at close, the final figures and the calls that produced them, the daily Trend Call results, and the formulas above. If we find a calculation error, we correct it and publish the correction within seven days.
Drop rates
These are the figures the server applies. This project will never publish a probability that differs from the one in the code. A booster's tier is drawn on the server the moment the booster appears, from a secret and the booster's number. It is shown before opening and can never be re-rolled.
The four tiers
| Tier | Chance | Cards | Timer (victory booster) | Guarantee |
|---|---|---|---|---|
| Note | 70% | 5 | 2 min | None (drawn evenly across the pool) |
| Page | 22% | 5 | 6 min | At least 1 Rare or better |
| Chapter | 7% | 6 | 12 min | At least 1 Epic or better, plus 1 Rare or better |
| Codex | 1% | 7 | 18 min | 1 Legendary, plus 1 Epic or better |
A booster is worth 5.1 cards on average. Cards outside a guarantee are drawn across the whole pool; a guaranteed card is drawn among the cards of that rank or higher. In every slot, a Legendary card is less likely to come out than any other card: 3.5 times less in a Note or a Page (it weighs 2 where every other card weighs 7), 2.5 times less in a Chapter or a Codex (2 against 5). The Codex's guaranteed Legendary is drawn among Legendaries only, so it is always one.
How many Legendaries a booster really holds
The guarantees above say what a booster holds at least. They don't say how often it holds more, and one line is easy to misread. A slot guaranteed “Epic or better” is drawn among the Epic and Legendary cards, each Legendary at its lower weight, so that slot comes out Legendary about one time in eleven. A Codex carries such a slot on top of its guaranteed Legendary.
That is why we publish the figures below as well. They come from simulating 100,000 openings of each tier with the same draw the server runs.
| Tier | Legendaries on average | None | Exactly 1 | 2 or more |
|---|---|---|---|---|
| Note | 0.05 | 95.3% | 4.6% | 0.1% |
| Page | 0.06 | 94.1% | 5.8% | 0.1% |
| Chapter | 0.18 | 83.6% | 15.4% | 1.0% |
| Codex | 1.16 | — | 84.7% | 15.3% |
Three Legendaries in one Codex happens about 1% of the time. A Note holds one about once in twenty-one, a Page about once in seventeen, a Chapter about once in six. Nothing guarantees it, and nothing prevents it.
What is in the pool being drawn from
Cards are drawn among people and fiction, except for one “world” slot in every second booster, reserved for the Concept, Place and Curiosity types. That slot first draws its type, one chance in three for each, then a card of that type. A Curiosity card never comes out of any other slot, and never from a guarantee. Across about 13,800 cards (about 2,200 of them Curiosity), each rank's share inside its own type follows the published cuts: about 55% Common, 30% Rare, 12% Epic and 3% Legendary, or 50 / 30 / 15 / 5 for the two types with fewer than 500 cards. Because a Legendary weighs less, it makes up about 1% of the cards in a Note, not 3%.
Where boosters come from
- One per Face Down duel won against a server-matched opponent. It lands in a slot and ripens on the timer of its tier above, never more than twenty minutes. At most ten a day, reset at 12:00 UTC, however much you play. That cap is what limits the flow of cards; the timers don't. Past the tenth, a win still pays its 25 INK.
- A free booster every 4 hours for everyone, in its own reserve of three (ten for PRO subscribers, once PRO opens), apart from the shelf. Its tier is drawn with the same chances as a victory booster (the table above), on the server, when it arrives, and it is shown before you open it. It counts toward the guarantee counters. Victory boosters and free boosters share a common limit of 11 a day (12:00 UTC).
- IQ tier rewards: boosters for each tier of the Face Down ladder you reach. You get a Codex for every tier from 2 to 10 (nine Codex for the ten tiers), and more as tiers are added on top. Once per tier and per account, for an account linked to X, outside the daily cap, onto the shelf.
- The reward road: the Face Down road gives boosters on some of its steps, 14 Chapters, 4 Pages and 4 Codex, for everyone, X linked or not. Once per step and per account, outside the daily cap, onto the shelf.
- Trend Call streaks: 3 correct in a row gives a Page, 7 a Chapter, 12 a Codex. They go onto the shelf, never into the slots. A wrong pick resets the streak; a neutralised pair neither breaks nor extends it.
- Sign-up: 5 boosters with a fixed, published composition (3 Note, 1 Page, 1 Chapter, about 26 cards, at least one Epic guaranteed), plus 250 INK.
- Your first duel: a guided Face Down duel against The Archivist, for anyone who arrives without an account. It ends with one Chapter, win or lose, drawn by the server, and this Chapter always holds a Legendary: its last card, in place of the “Epic or better”, drawn evenly among the Legendaries that are neither Icon nor Curiosity. Its cards go to the account you create next. That Legendary stays on the account for a while: it can't be sold, traded or discarded until 30 days have passed and X is linked to the account. It plays in duels and counts for its level from the start. This keeps accounts from being made just to collect Legendaries. Once per account, and once per device: playing it again brings back the same booster.
- The shelf: everything you are given waits there: the sign-up boosters, the Trend Call streaks, the IQ tier rewards and the Chapters of the reward road. They have no timer, nothing is lost, and you open them whenever you want. It holds 10. When it is full, a reward waits for a space. Boosters won in duels land in the slots instead, on their short timers, and the booster every 4 hours has its own reserve.
- Duels against The Archivist give INK only, 150 a day at most: no victory booster and no IQ. The guided first duel is the one exception, above.
A booster is never sold on its own, whether for money, INK, an advertisement or anything else. A PRO subscription, once it opens, keeps up to 10 free boosters waiting instead of 3, at the same pace (one every 4 hours), with the same drop rates and the same daily limit of 11. The PRO track of the season pass adds its own boosters, a fixed number per season, outside that limit. Opening one sooner only ever costs INK, never money, and INK can't be bought.
INK and boosters
- Every booster you open gives a little INK: Note 2, Page 5, Chapter 15, Codex 40. The Chapter of the first duel gives none.
- A victory booster still on its timer can be opened now for INK. The price is per minute left, with a minimum: Note 10 a minute (at least 10), Page 10 (at least 25), Chapter 15 (at least 50), Codex 20 (at least 100). It goes down as the timer runs, and a ready booster is always free.
- Its tier is drawn and shown before you pay, and paying changes nothing about what's inside. It adds no booster to your day: the cap of ten still holds, and the timers never get longer than twenty minutes. Shelf boosters have no timer, so there's nothing to speed up.
The Lucky Bindery
After you open a booster, the Lucky Bindery sometimes opens: on average once every 30 openings (the number is drawn between 20 and 40 each time), if you have at least five spare cards of one rank. It stays on your shelf for 24 hours, ten binds a day.
| You give | You get |
|---|---|
| 5 spare Commons | a Rare, 95% · an Epic, 5% |
| 5 spare Rares | an Epic, 98% · a Legendary, 2% |
The card is drawn evenly among all the cards of that rank, and it can be one you already have. Only spare copies go in: you always keep one of each card, and a card on sale or in a trade stays out. Nothing is ever lost: you always get a card of a higher rank. Epics can't be bound. It is free, and nothing about it is for sale.
Guarantee counters
Two counters run per player, on victory boosters and free boosters together:
- A Chapter at the latest every 20 boosters.
- A Codex at the latest every 100 boosters.
The counter resets when the tier it guarantees is drawn, whether it was drawn by chance or by the guarantee. Trend Call boosters do not move the counters: their tier is set by the streak, not by a draw.
Icon cards
Icon is an appearance, not a rank: thirteen cards, on a hand-written list, wear the gold and the crown and keep the values of their real rank. They have no separate draw. An Icon card comes out like any other card of its rank. Thirteen cards among the roughly 11,600 people and fiction cards make 0.11% per card drawn from them and about 0.57% per booster, about one booster in 176.