LearnClash by the Numbers [2026]: ELO, SRS & Design

ELO-matched LearnClash duels trend toward a balanced win rate. Random matching does not. The difference is the system working.

LearnClash by the numbers is a tour of how a competitive learning app is built: ELO spread across 22 ranks, the composite matchmaker, 3-stage SRS transitions, the difficulty bands, the duel format, and prime-number topic counts. Every number below is a design choice you can verify in the app, grounded where relevant in the learning-science research cited throughout.

If you’ve wondered how ELO really places players, whether spaced repetition sticks, or why we use 37 questions per topic instead of 50, this is the source. Start a 3-minute duel on any topic to see it in action.

	LearnClash design
Ranks	22 across 8 tiers (Iron to Phoenix)
Starting ELO	1300 (Gold II, ladder average)
Matchmaking	Weighted composite: 50% ELO proximity + 50% category overlap, no hard rating gate
SRS model	3-stage SRS: wrong > 7d > known > 90d > mastered
Questions per duel	18 (6 rounds × 3 questions, 45 seconds each)
Questions per Practice session	9
Async turn window	72 hours
Topic question counts	Prime-only: 37, 43, 47, 53, 89
Difficulty bands	Easy, medium, hard (every distractor plausible by design)

How the LearnClash ELO Ladder Is Built

LearnClash’s ELO system spreads players across 22 ranks in 8 tiers, from Iron at the bottom to Phoenix at the top. Every new player starts at 1300 (Gold II), the ladder average, so the starting point sits in the middle by design. The top bracket is deliberately narrow, matching the top-bracket shape you see in chess federations.

Figure 1: The LearnClash tier structure. New players start at the Gold II ladder average; the Phoenix bracket at the top is deliberately narrow, in line with top-tier chess spreads.

The eight tiers and their ELO ranges:

Tier	ELO range
Iron I-III	100-599
Bronze I-III	600-899
Silver I-III	900-1199
Gold I-III	1200-1499
Platinum I-III	1500-1799
Emerald I-III	1800-2099
Diamond I-III	2100-2399
Phoenix	2400+

Two mechanics anchor the ladder. The ELO floor at 100 stops infinite downside, so a player on a losing streak can’t fall below starter range. And Glicko-2 rating-deviation growth marks dormant accounts as uncertain, so a returning player is re-calibrated quickly rather than matched on a stale rating. Together they keep the ladder honest: skill places you, the floor catches you, and inactivity widens your uncertainty until you play again. For the math and K-factor logic behind the placement, see our full ELO rating system explainer.

What ELO-Matched Win Rates Actually Tell Us

When LearnClash’s composite matchmaker surfaces a close ELO match with heavy category overlap, duels trend toward a balanced win rate. Random matching does not, and the fun drops fast. The matchmaker scores open duels on 50 percent ELO proximity and 50 percent category cosine similarity, with no strict rating-range gate. ELO proximity scores 1.0 at a zero-point difference and decays to 0 at a ±400-point gap; category similarity carries the rest.

Figure 2: LearnClash composite matchmaking score blends ELO proximity with category cosine similarity to keep win rates balanced, without applying a hard ELO gate.

Matchmaking also respects three rules that keep the feed playable:

• Async turn-based windows of 72 hours. A duel never expires mid-round. Either player can take their turn when they have time.
• No ads between rounds. LearnClash monetizes through premium subscriptions only. Round-to-round pacing stays uninterrupted.
• ELO updates fire after every ranked duel, not every week. You see the rating shift within seconds of the final answer.

A balanced win-rate band is what the matchmaker chases because it aligns with the “desirable difficulty” zone in learning science. Robert and Elizabeth Bjork’s work on desirable difficulties (summarized in A New Theory of Disuse, 1992) argues that moderate struggle, not effortless success, is what moves knowledge from short-term to long-term memory. A 50-50 duel feels fair. A 90-10 duel teaches nothing.

The composite is designed so that the closer two players sit on rating and the more their recent topics overlap, the higher the score, and the more balanced the expected duel. New players in calibration (K=40) and deep-topic matches where category overlap outweighs rating proximity are the cases that widen the band, which is by design rather than a failure. For the full walkthrough of K-factor calibration, topic-overlap modulation, and how ELO-matched wins compound 3-stage SRS retention, see ELO matchmaking and the balanced win-rate band.

The 3-stage SRS in Practice

LearnClash uses a named internal system we call 3-stage SRS. Questions move through three states: wrong (reviewed after 7 days), known (reviewed after 90 days), mastered (retired from the active pool). It’s different from the 1-3-7-21 interval schedule most memory blogs cite.

Figure 3: LearnClash 3-stage SRS transitions. Wrong cards return at 7 days, known cards at 90, and mastered cards exit the pool permanently.

Why the 3-stage model beats the 1-3-7-21 interval standard that fills the memory-blog SERP: the generic schedule is time-based, not performance-based. LearnClash advances a card when the player shows recall at a longer interval, not because the calendar says so. Cepeda et al. (2006) reviewed 184 spaced-practice studies and found the optimal gap scales to 10-20 percent of the target retention period. The 7-day and 90-day intervals sit inside that band for short-term and long-term retention. For the per-stage design of the curve, see the LearnClash SRS retention curve.

Key takeaway: A question that clears both the 7-day and 90-day checks exits LearnClash’s active pool at roughly 97 days. After that point, it only returns if the player resets the topic.

But the 3-stage model is also opinionated about what counts as a miss. A wrong answer demotes the card by exactly one SRS stage, not a full reset. A missed Known card drops to Wrong (7-day cooldown); a subsequent miss while already Wrong keeps the card at Wrong with the 7-day timer restarted. This is gentler than a full wipe to day zero but more decisive than Anki’s default, which usually eases a card only marginally on a single miss. It also means mastery is meaningful. A card that exits the pool has cleared both checks cleanly. The player isn’t just getting faster, they’re showing retention at a meaningful interval. Research consensus in 2026 shows spaced repetition produces roughly 200 percent better retention than massed practice, with recent meta-analyses putting long-term recall at 80 percent for spaced learners versus 30 percent for crammers.

How the Climb from Bronze to Gold Is Shaped

The climb is non-linear on purpose. Because every new player starts at 1300 (Gold II) with K-factor 40, the first 10 duels swing the rating fast as calibration finds the player’s real skill. Then the K-factor drops from 40 to 20, each duel moves the rating less, and the climb to higher tiers takes more games per rank. Early progress feels quick; later progress is earned.

Figure 4: The climb steepens past the early tiers because K-factor shifts from 40 to 20 after the first 10 duels, so each rating point is harder to earn.

The shape of a committed player’s path:

1. The first 10 duels: K-factor 40, fast calibration, wide rating swings
2. After calibration: K-factor 20, each duel moves the rating at half speed
3. Higher tiers: a wider topic portfolio is needed, because depth across many topics is what a strong matchmaker rewards
4. Top tiers: progress is slow and plateaus appear, the same shape a chess ladder shows

LearnClash doesn’t apply a hard ELO decay. Instead, after 7+ days of inactivity, Glicko-2 rating deviation (RD) grows each day. RD is the system’s confidence in your rating; larger RD means the next few duels move your ELO more aggressively until calibration tightens again. Accounts with RD above 75 are temporarily excluded from global leaderboards as a motivational lever to return.

The asymmetry matters. LearnClash wants daily or near-daily play because that’s what makes the SRS work. RD growth leaves your rating intact but signals uncertainty to the matchmaker and removes dormant accounts from public boards until the player comes back and re-calibrates.

How Easy and Hard Questions Differ by Design

Every LearnClash topic carries three difficulty bands: easy, medium, and hard. The bands are a design choice, not a label slapped on after the fact. Hard questions are hard because they point at a specific wrong-answer trap, the kind that makes you doubt your first instinct.

Figure 5: Difficulty by design. A question is hard when its distractors are close enough to the correct answer to compete with it.

What a LearnClash “hard” question looks like in practice:

Hard, History: In what year did the Berlin Wall fall? Answer: 1989. Most wrong answers pick 1990. That’s the year Germany reunified, not the year the wall fell.

The trap is narrow. Reunification happened in 1990, so 1990 is the tempting wrong answer. LearnClash’s hard questions live at the border between recall and retrieval error, which is also where the testing effect does its strongest work. Roediger and Karpicke (2006) showed that retrieval practice produces 50 percent more retention than repeated studying, and the effect is strongest when retrieval just barely succeeds.

Did you know? LearnClash’s question generator is instructed to produce every distractor as a plausible wrong answer, not filler text. A hard question is designed so that guessing is genuinely hard, even with zero prior knowledge.

For production examples of those traps, see the 12 questions people get wrong data pull. It breaks down real LearnClash misses like Cleopatra’s native Greek, banana as a botanical berry, and early samurai using bows.

The difficulty design also points to how you climb. Players stuck at a tier often assume they need to push harder on hard questions inside one topic. The matchmaker rewards the opposite: breadth across many topics. Because the composite weights category overlap at 50 percent, a player who maintains accuracy across a wide topic portfolio is matched into more learnable duels than one who drills a single subject. Depth across topics beats depth inside one topic.

Why the Round Is Short by Design

A LearnClash duel is short by design: 18 questions, 6 rounds, 45 seconds each. A Practice session is shorter still at 9 questions. The format is built to fit between dinner and a show, not to fill an evening.

Figure 6: The short-round format. A duel is 18 questions across 6 rounds; a Practice session is 9, both sized to fit a daily habit.

The short round is not an accident. When David started building LearnClash, he pulled from 12 years of daily QuizDuel with his mum. The short round was what made it fit into evenings. Long sessions fail the habit test, and a few minutes a day is what keeps spaced repetition working: the SRS only schedules effective reviews if you come back regularly. A round you can finish on a break is a round you’ll come back to tomorrow, which is the whole point of a retention engine.

The format earns the habit it’s designed for. The longest daily streak LearnClash has observed is 121 days, a player returning to the app every day for four straight months. Source: LearnClash production data, exported May 2026. That is the short round doing its job: a session small enough to finish on a break is a session you come back to tomorrow, and the day after that.

How Questions and Topics Work on LearnClash

LearnClash generates questions per topic on demand, and the scale of that catalog is the clearest proof the system works. The LearnClash production export from May 2026 held 27,131 questions across 556 topics, an average of about 49 questions per topic, every one generated on demand rather than pre-written by a content team. Source: LearnClash production data, exported May 2026. Not all topics are pre-populated; the app spins up a new topic pool when a player searches a concept that isn’t already indexed. Each topic lands with a specific, odd-numbered prime count (37, 43, 47, 53, or 89) based on breadth and difficulty target, and all 27,131 questions ship in four languages (English, German, French, Spanish).

Figure 7: How LearnClash content is built. The May 2026 export held 27,131 questions across 556 topics, generated on demand and sized with a prime-number question count to shape completion behavior.

How the catalog is structured, from the May 2026 production export:

• 27,131 questions across 556 topics, an average of about 49 questions per topic
• Topic depths range from 37 (narrow) to 89 questions (broad pillar topics)
• Topics are generated on demand, so the catalog grows as players explore new subjects
• Content ships in four languages: English, German, French, Spanish

The design matters because it shapes what duels are possible. When a LearnClash duel selects 6 topics across 18 questions, the matchmaker picks from the player’s accuracy-weighted history plus category preferences. Players with a wide topic portfolio get more varied duels; new players get narrower pools that align to their onboarding topic.

One thing worth calling out: LearnClash doesn’t pre-write questions. It generates them on demand when a player explores a new topic, then validates and stores them for re-use. That means the catalog is naturally weighted toward what players actually care about, not what a content team guessed would be popular. The long tail of topics looks nothing like a traditional quiz-app catalog. Alongside history and pop culture, you’ll find niche domains like reef aquarium husbandry, Icelandic sagas, or Formula 1 qualifying strategy, each sized with its own prime-number question count.

Which Topics Players Get Wrong: The Trivia Difficulty Index

On June 9, 2026 we ran a read-only export of every answered question in the LearnClash production database and built the Trivia Difficulty Index: a league table of real per-topic wrong rates. The headline finding: across 80,553 real answers, players get 43.5 percent of questions wrong, and the hardest subjects are not the ones you would guess. Source: LearnClash production export, June 9, 2026.

#	Topic	Wrong rate	Answers
1	Culture (anthropology)	61.5%	283
2	Law	61.1%	203
3	Art	59.9%	162
4	Pepe the Frog (meme)	58.0%	219
5	Colonization of Mars	57.1%	105
6	Famous Quotes (literature)	57.0%	172
7	Jellyfish	56.1%	212
8	AC/DC (band)	56.0%	182
9	Steve Jobs	55.0%	129
10	World War II	54.7%	276

What stands out in the full index:

• The humanities top the difficulty chart, not science. Culture, Law, and Art hold the top three spots, while most science topics sit mid-table.
• The easiest qualified topic is German Proverbs at 6.3 percent wrong, a 55-point spread to the hardest.
• On 29 of the 156 ranked topics, players miss more answers than they get right.
• The single most-missed question: “In which country does nodding your head up and down mean no?” Every player who saw it answered wrong. (It is Bulgaria.)

The method, in one paragraph: per-option answer counts from 28,891 production questions across 570 topics, aggregated per topic into an answer-weighted wrong rate (the sum of wrong answers divided by all answers, so heavily answered questions dominate a topic’s rate). A topic needs at least 5 answered questions and 100 total answers to rank, which 156 of 570 topics clear. Answer counts come from real duels and practice sessions; the data holds topic-level aggregates only, no user-level records.

Key takeaway: Trivia difficulty is empirical, not intuitive. The topics that feel hard (advanced science, mathematics) are not the ones people actually miss; cultural conventions, legal systems, and attribution traps are.

The full 570-topic index is published as an open dataset on Kaggle under CC BY 4.0, refreshed quarterly. Cite it as “LearnClash Trivia Difficulty Index, June 2026, Pluxia GmbH (learnclash.com)”.

Why LearnClash Uses Prime-Number Question Counts

Every LearnClash topic uses a prime-number question count: 37, 43, 47, 53, or 89. Never 40, 50, 90, or 100. This looks strange the first time you see it, but it’s a deliberate design choice grounded in how round numbers anchor completion behavior. Round targets create implicit quit points. Primes don’t.

Figure 8: Why LearnClash picks primes over round numbers. Round targets create implicit stopping cues; primes keep players engaged past arbitrary stopping points.

The operational reasoning:

• 37 questions covers narrow topics without padding. Five-point tier of bar-trivia granularity.
• 43 questions is the standard spoke depth. Hits most category-level topics.
• 47 questions is the “difficult” bucket for topics with a lot of edge cases.
• 53 questions is a deeper variant used when the domain has genuine depth.
• 89 questions is the pillar count. Reserved for broad-parent topics like geography or general knowledge.

Round numbers underperform in content-product contexts because they cue quitting. Listicle research shows odd-prime titles earn roughly 20 percent more clicks than even numbers (Outbrain, analyzed from 10 million headlines). BuzzFeed’s internal tests have pointed to 29 as a peak performer for the same reason. LearnClash carries that logic into the product layer. A player who’s answered 36 questions doesn’t think “I’m done at 40”. They’re one away from 37, which is a specific target. The next odd prime is 41 (too small an increase), then 43, which is where deeper topics land.

Key takeaway: Prime-count question lists are a product design choice, not a mathematical curiosity. They change completion behavior by removing the round-number quit cue.

The 89 count deserves a note. It’s the largest prime we use, and it shows up only on pillar topics where breadth is genuine. Geography has 89 questions because the space is huge: capitals, rivers, mountain ranges, colonial history, modern borders. General knowledge has 89 because it’s meant to cover every domain at a shallow depth. Using 100 here would be the obvious round-number choice, but the 11-question gap between 89 and 100 is enough to matter: a round 100 invites a clean stopping point, while a specific 89 points only at the finish. The companion piece on prime-number question counts walks through the behavioral research behind the choice.

What These Design Choices Mean for Learners

The design choices on this page are not academic curiosities. They map directly to three decisions you make as a LearnClash player: how hard to push, how often to practice, and when to trust that a question is really mastered.

Figure 9: Three practical takeaways from the LearnClash design.

Three diagnostics you can pull from the design:

If you are…	The design says	Action
New to LearnClash	K-factor 40 amplifies rating swings for the first 10 duels	Don’t read early ELO swings as skill signal
Stuck at a tier	The composite weights topic overlap at 50%, so breadth is what climbs	Practice wider, not harder
Focused on long-term retention	A mastered card takes around 97 cumulative days to exit the pool	Trust the slow curve; the Cepeda 10-20% rule applies

If you’re new, the “am I getting worse” feeling in the first week is almost always calibration, not skill. K-factor 40 produces wide rating swings for the first 10 duels before it drops to 20 and the noise flattens. If you’re stuck at a tier, the bottleneck is usually breadth: add new topics to your pool and play each a few times. The matchmaker then pairs you with higher-rated opponents on deeper topic overlap, which is where the climb happens.

For long-term retention, the 3-stage SRS path is slow on purpose. Cepeda’s 2006 meta-analysis across 184 distributed-practice studies found optimal gaps scale to 10-20 percent of the target retention period. Ninety days is the right interval for questions you want to keep for a year or more.

Did you know? Research on repeated testing in medical education (2026) showed physicians exposed to spaced retrieval performed at 58 percent versus 43 percent for control groups. The same mechanism drives LearnClash’s long-interval checks.

The Bottom Line

LearnClash’s design describes a competitive learning app with one job: keep duels inside the desirable-difficulty zone while the 3-stage SRS pushes questions toward mastery. The composite matchmaker’s balanced win band, the K-factor 40-then-20 calibration, the 7-day and 90-day SRS checkpoints, and the prime-number topic counts all pull in the same direction.

Figure 10: Four LearnClash design choices that reinforce each other. Remove any one and the loop breaks.

But the design only matters if you test it. The next duel is the fastest way. See our Kahoot vs Quizlet comparison for how the two biggest study apps handle retention, or does Quizlet actually have spaced repetition for how the free tier holds up against the 3-stage SRS above.

Key takeaway: ELO matching, 3-stage SRS, prime-count topics, and the short round are four levers pulling in the same direction. Pull any three without the fourth and the system stalls.

Deeper reads: the LearnClash ELO system explained, the full 3-stage SRS breakdown, the retrieval practice science behind our hard questions, and our nine-method study guide. Or browse the learning science cluster for every learning-science article. Duel me on memory psychology →

⚡ Start your 7-day Premium trial on LearnClash