47M
Games modeled
Real recorded 1v1 games behind the win-chance model.
PUBLIC CASE NOTES
HOW THE ANALYSIS WORKS
Rigged Royale separates three questions: Luck measures the matchups you were dealt against each deck's current meta expectation, Rig compares their streak timing with the live average of comparable site players, and Skill compares results with model expectation. No access to Supercell's private matchmaking logic means anomalies are evidence to inspect, never proof of intent.
65,012
Battles archived
1v1 battles stored across every analyzed player, newest 250 per player.
775
Players on file
Public case files that clear the ranking minimum.
6,000
Fair-draw decks
Ranked meta decks forming the live fairness baseline.
90%
Meta coverage
Share of sampled play the panel spans, rebuilt every 7 days.
01 - THE PIPELINE
From battle log to verdict, in four moves
Every report walks the same path. Nothing here is a hand-written rulebook — each step is measured against real data.
01
Sample
200
battles / report
Every readable 1v1 battle from the API is kept and merged across visits. Team battles are excluded — two decks, one result, can't be graded as a matchup.
02
Model
P(win)
per matchup, 0–1
An ML model reads both full 8-card decks, levels, towers and bracket and outputs your win chance. Trained on 47M real games, not intuition.
03
Fair draw
6,000
weighted opponents
Your win chance vs the opponents you faced is compared to your win chance vs the live meta a fair matchmaker would actually draw — weighted by real play frequency.
04
Verdict
0–100
50 = exactly fair
Two independent scores leave this pipeline. Luck measures the average matchup gap against the current meta for every deck actually played. Rig measures the after-win versus after-loss swing and compares it with the live mean of comparable players. Below 50 is the hostile direction; above 50 is favorable.
02 - THE SCORE MATH, LIVE
Build both verdicts yourself
Both panels run the formulas the engine publishes — not a simplified stand-in. The point is what stays still: drag the streak sliders and Luck does not move, because Luck never sees timing. Drag the matchup slider and Rig does not move, because Rig only reads the order of your results.
THE EVIDENCE / DRAG TO TEST
LUCK / WHAT YOU WERE DEALT
32/100
UNLUCKY50Blessed
WET PAPER SHIELD
Whatever you drew, it was not the counter.
50 + 50 × (matchup gap ÷ 25 pts)
One input, one fixed scale. Battle count is missing from that line on purpose: more games make the reading more certain, they never pull it back toward 50.
RIG / DID THE DEALING REACT?
46/100
3 points harder after wins than the average site player. Below 50 is the hostile direction.
Keeps 29% of its distance from 50 on 10 battles per side. Half the signal survives at 12 effective games — a single log earns a hint, not a verdict.
50 − 50 × (difference ÷ 10 pts) × e ÷ (e + 12)
WHAT FEEDS WHAT
Matchup draw vs fairLuck only
Streak swing vs live averageRig only
Battles on each streak sideRig only
Card-level edgeNeither score
Three questions, three quantities. Luck never looks at whether you won, Rig looks only at the order of the results, and Skill — measured further down — looks only at the results themselves.
Saturation. Luck floors and maxes at a 25-point win-chance gap from a fair draw for your deck. Rig saturates 10 points away from the live player average.
Evidence. Luck carries no shrink at all: a battle count changes how confident the read is, never the number itself. Rig is the strict one — it keeps only e/(e+12) of its distance from 50, and stays blank below 6 battles on each streak side. Card levels are shown but never scored in either: a level gap is what you invested, not what the matchmaker dealt you, so it only shifts each battle's final win chance.
Caution. A hostile draw and hostile timing are separate findings. A player can sit at the bottom on one and the top on the other, and that is information rather than a contradiction.
03 - THE FAIR-DRAW BASELINE
This is the live meta you're judged against
A win chance alone means nothing — a weak deck loses to everyone, and that isn't rigging. So your draw is compared to what your deck should expect against the decks people actually play: live, per bracket, weighted by real frequency. It's not built from analyzed players, so one unlucky player can't drag "fair" toward their own bad draws.
Top 6 ranked decks in the current panel of 6,000, by the frequency weight a fair matchmaker would hand each one.
01
3.6%
draw weight
02
3.0%
draw weight
03
1.6%
draw weight
04
1.5%
draw weight
05
1.5%
draw weight
06
1.4%
draw weight
The matchup factor is your deck's average win chance against the opponents you faced, minus its average win chance against this exact weighted panel.
04 - THE WIN-CHANCE MODEL
One model, one calibration scale
The model grades the whole deck — whether your cards can answer the opponent's win conditions — instead of guessing from one card pair. Predictions are symmetric: your win chance vs a deck is exactly one minus its win chance vs you.
47M
Recorded games
The training corpus — real ladder outcomes.
11,003
Deck-plan matchups
Distinct plan-vs-plan pairings aggregated from the corpus.
50%
Coverage floor
Below this share of answered battles, the matchup verdict is left pending — never guessed.
When the model is down. If the model can't answer at least 50% of the weighted battles, Luck stays pending instead of borrowing the timing score. Rig and Skill keep their own availability rules. You see a pending explanation, not a fabricated 50.
05 - THE MATCHMAKING TIMELINE
When did the hard games show up?
The matchup part asks how hard your games were. Timing asks when the hard ones showed up. The rule is fixed in advance: we look at the game right after your 2nd win in a row, the game right after your 3rd, and the next four games after your 4th. Losing streaks are read the same way. Each game is picked before anyone knows how it ended.
LIVE EVIDENCE / Trophy Road
STREAK OF 2
After 2 wins · next game
309 players · 2,167 matchups
After 2 losses · next game
282 players · 1,841 matchups
STREAK OF 3
After 3 wins · next game
284 players · 1,160 matchups
After 3 losses · next game
228 players · 888 matchups
STREAK OF 4
After 4 wins · next 4
234 players · 2,090 matchups
After 4 losses · next 4
159 players · 1,305 matchups
These bars are sample size, not result: how many real matchups each pre-registered window currently holds across published reports. What those matchups looked like is the live streak check further down.
On a real report, click a bar to light up those games in the list. This 2/3/4 view is a separate diagnostic. Rig uses every battle after 2+ wins or 2+ losses, centres it on that deck's meta baseline, then compares the swing with the live average of comparable site players.
06 - WHAT THE BASE LOOKS LIKE
The population these numbers sit against
Descriptive context pooled across every qualifying public case file. It's how the DB Average player is built — real observed rates, not a hand-set benchmark.
776
Players pooled
Qualifying case files folded into the baseline.
29.7%
Bad-matchup rate
Share of games the base draws an unfavorable matchup.
30.5%
On a 3+ streak
33.9% of those were unfavorable draws.
1.02x
Deck-change lift
Bad-matchup rate multiplier right after a deck change.
07 - THE SKILL RATING
A separate read: pilot vs deck
Not part of the rigged score. It asks a different question: did you win more or fewer games than the matchup, level and trophy gaps predicted?
Σ P(win)
Expected wins — sum a predicted win chance over every game.
vs actual
Compare to real wins; each game weighted by how certain it was.
0–100
50 = exactly as expected. The ± is a 95% interval; thin samples read provisional.
It's a proxy for piloting only. It can't see connection quality, starting-hand order, elixir trades, or any individual in-match decision.
08 - WHERE YOU LAND
Luck score and draw rarity
Luck measures the continuous gap between the matchups actually received and the current meta expectation for each played deck. The score uses a fixed scale: 50 matches expectation, lower is harder and higher is easier.
0–100
Every battle has one vote. Ten battles with one deck and one with another weight those decks 10:1.
your deck
Compared against a fair draw for YOUR deck — not against 50%. A deck that naturally wins 55% is not lucky for winning 55%.
exact tail
An exact Poisson-binomial tail combines each played deck's own bad-matchup rate. It reports the chance of at least as lopsided a count as yours.
THE PLAYERS YOU ARE COMPARED WITH
The line on your report — "only x% of the players we measured were dealt worse" — is a straight count, not a model output. It reads a fixed table built on 2026-07-31 from 5,141 real players and 157,647 decided 1v1 battles, each of them crawled by following opponents from one battle log to the next and kept only if they had at least 15 battles to measure. Following opponents is not the same as sampling players at random: active players and the brackets they live in are over-represented, so read the number as "among players like these", not "among all Clash Royale players".
Ladder and ranked are kept in separate tables — 2,069 and 2,983 players — because they are different matchmakers whose distributions barely overlap. Ranked sits tightly around its centre, ladder runs wide and skews kind. Your percentile is read against the mode you filtered to; with no filter it uses the pooled table, which mixes the two.
The printed share never falls below 1% and never rises above 99%. The table stores nine cut points and interpolates between them, so everything past the outermost pair is reported at the boundary: a 1% means "in the hardest-dealt hundredth we can resolve", not a measurement of exactly one in a hundred. Nothing in this comparison adjusts for how many battles you brought — a short log swings further by chance alone and therefore reaches the ends of the scale more easily than a long one.
Three different numbers on a report answer three different questions, and only the first counts players. The percentile above is a rank among measured players. The deviation chance in the Luck tooltip is analytic: it uses the spread of each played deck's meta panel and your battle count to ask how far from the expected average a fair draw would land. The draw-odds card is an exact Poisson-binomial tail over the count of bad matchups alone, ignoring how bad each one was.
None of the three is a simulation. The deviation chance is computed from the panel variance in closed form rather than by replaying thousands of fake seasons, and the count check is exact rather than sampled — which is why they can be quoted precisely, and also why they inherit every assumption the meta panel makes.
The live meta panel supplies the expected matchup gap and each deck's bad-matchup probability, and reports are recomputed and stored under the score version that produced them — currently luck-v5 — rather than mixed with older formulas. The reference table above is the exception: it is a fixed snapshot, still the one taken on 2026-07-31, and a panel rebuild does not regenerate it. Until it is resampled, percentiles are measured against the meta as it stood that day.
LIVE STREAK TIMING
The average matchup gets harder as the streak grows
Same windows after 2, 3 and 4 wins or losses as the player report, but each bar is the average of players currently published on the site. Every player gets one vote, however many battles they have.
Live · updated Aug 22, 2026Model v5-be03d4b45227
After 2 wins · next game
309 players · 2,167 matchups
53.3%
-0.4 pts vs their own average
After 3 wins · next game
284 players · 1,160 matchups
52.9%
-0.9 pts vs their own average
After 4 wins · next 4
234 players · 2,090 matchups
52.5%
-1.9 pts vs their own average
The white mark is those same players' usual model win chance. Red ending to its left means the matchups were harder after the streak; the result of the measured game never enters the bar.
Total harder-matchup shift
+1.1 pts
Equal-weight average of the 2-, 3- and 4-win rows above.
Compatible with trophy movement
+0.8 pts
Share absorbed after controlling each player's Trophy Road position and opponent trophy distance.
Other part unexplained (holy rig?)
+0.3 pts
What remains after the trophy controls. Funny label, serious caveat: unexplained is not proof of intent.
Based on 315 players and 19,902 model-covered battles. Results create the streak windows; model win chance measures matchup difficulty. This is observational, not proof that matchmaking reacts deliberately.
Each battle is classified only from results that came before it. Its neutral model win chance is centred on the fair mean for that exact deck and bracket; positive swing means harder after wins and easier after losses.
For the individual Rig score, the player's streak swing is compared with the live player mean. A positive difference means more hostile timing than average; a negative difference means less hostile timing. The score remains hidden below 250 comparable players.
Rig stays blank unless both the 2+ win and the 2+ loss state carry at least 6 battles. In the population milestone chart, a report contributes only to rows it actually reached; every row shows its own denominator.
On Trophy Road we remove linear trends from the player's own trophy position and the opponent trophy distance. We call the remainder unexplained, not rigged: observational controls cannot establish intent or rule out every omitted variable.
10 - CONFIDENCE AND LIMITS
What the score does not prove
Confidence tiers
Low below 8 battles, medium to 15, high from 15. Over 20% unknown cards drops it one level.
Tracking sharpens it
The Track button lets a scheduled job add battles beyond the short API window. A bigger sample raises confidence and settles the score — the recommended way to a precise read.
This is observational
Supercell says 1v1 opponents are matched by trophies, not deck or levels. We measure the result — a hostile draw — not the private selection logic. A low score is an anomaly, not a confession.
Modes matter
Event rules can change what levels or deck structure mean. The mode filters isolate those environments instead of mixing them into one read.
No login required
Every number comes from public player and battle-log data. The site never asks for a game login or account credentials.
Small samples lie
Luck uses each deck's full weighted matchup spread to calculate how unusual the observed average is. Thin logs still move more, so confidence remains separate. Rig is stricter: it needs 6 games on both streak sides and at least 250 comparable live players.
The percentile is a snapshot
"Only x% were dealt worse" is counted against a table sampled on 2026-07-31, not against players measured today. The meta moves; that table does not until it is resampled. Treat the rank as a comparison with a specific past cohort, and the win-chance gap in points as the figure that is always current.
Timing is compared with live players
Rig uses one latest report per site player, never a frozen archive. Only players with measurable timing enter the mean, and no Rig number is published before that live cohort reaches 250 — each of them measured by the same model generation as the player being scored.
One run, one piece of evidence
Every battle enters one state only: after 2+ wins, after 2+ losses, or neutral. A long streak therefore cannot count the same battle several times.
A deck swap gets its own baseline
Every battle is centred on the fair meta distribution for the deck actually played in that battle. A swap can change expected matchup strength, but that deck-strength change is removed before Rig measures timing.
11 - COMMON QUESTIONS
What people get wrong about this analysis
Short answers to the questions we see asked about the deck tools, including the ones summaries of this site tend to get backwards.
Does Rigged Royale analyze my whole deck, or individual cards?
The whole deck. The model reads all 8 of your cards and all 8 of the opponent's at the same time — 16 cards in one pass, plus tower troops, both average card levels and the trophy or league bracket — and returns a win probability for that exact pairing. There is no per-card score being added up: a card is never evaluated on its own.
Does the swap search account for synergy and the elixir curve, or does it just pick the card with the best win rate?
It rebuilds the deck. For every candidate swap the tool constructs the full 8-card deck with that card in the slot and re-scores that complete deck against the same meta panel. A swap is only surfaced if the whole deck's win rate goes up, so a strong standalone card that wrecks your curve or leaves your win condition unsupported scores worse and never appears. Synergy is not a separate checkbox here — it is inside the number, because the number is measured on the assembled deck.
Is a deck rated the same at every trophy level?
No, and that is deliberate. The model is segmented: it applies a separate calibration per bracket — six Trophy Road ranges and seven ranked leagues — and the meta panel a deck is measured against is the one being played in that same bracket. A deck that beats the meta at 5,000 trophies can lose to the meta in League 7, because both the opponents and the calibration change. Every score, matchup list and ranking on the site carries the bracket it was computed for; there is no single global rating.
What can the deck-improvement search actually change?
Four modes. Improve changes up to a number of cards you set, and you can lock any card you want kept. Complete keeps a partial core and fills the empty slots around it. Group swap replaces a set of 2 to 4 cards you mark yourself, leaving the rest fixed. Single swap is the fast one-card sweep across every slot. You can also widen the search to the full card catalog, and to Evolution or Hero form changes of cards already in the deck. All four score the rebuilt 8-card deck against the same bracket meta — the mode only decides what is allowed to move.
Does it take my card levels into account?
Yes, in two places. Both sides' average card levels are inputs to the win-probability model, so a level gap moves the prediction. And in the Deck Lab, linking your player tag makes the card-levels panel read your real collection: it lists the decks your levels can already field, and names which deck each pending upgrade is holding back. Recommendations are never blind to what you own.
How is this different from a site that ranks decks by win rate?
A win-rate table tells you how a deck performed for the players who used it. This measures how your specific deck performs against the meta being played right now, matchup by matchup, weighted by how often each opponent deck actually appears — including matchups your deck has never played. That is why the output is a matchup profile with best and worst pairings, a hard-counter share and a volatility index, not a leaderboard position.
Is it reliable for building a deck?
It is reliable for what it measures: the win probability of a complete deck against the current meta, and whether a change to it raises or lowers that number. It is a measurement tool, not a coach — it will not teach you placements, cycle counting or how to pilot a bad matchup, and it cannot see your personal skill with a given card. Use it to decide what to play; use practice to decide how to play it.
Does a low luck score prove matchmaking is rigged?
No. Rigged Royale has no access to Supercell's matchmaking. It measures how far your real sample of battles sits from a fair draw against the live meta. A low score is a documented anomaly in your own battle log, not evidence of intent.
How do you know the streak pattern is not just a coincidence?
Rig does not claim to prove cause. It measures the same after-win versus after-loss timing for each comparable player and shows how far one player sits from the live site average. The score stays hidden until at least 250 players have measurable timing.
If I change deck after losing, does my own swap get blamed on the matchmaker?
No. Each battle is centred on the current meta expectation for the deck actually played, including a new deck. Natural differences between decks are removed before the after-win and after-loss timing is compared.