How the Combo Scores Work

The idea in one paragraph

Each of the 131 heroes is reduced to a measurable profile read from its eight skills: how much damage it can be expected to deal over a battle, what it heals, blocks, debuffs and controls, which battle events its skills create and react to, the range of every attack skill, and how strong its era's stat package is. Every possible three-hero formation is then assigned front, middle and back rows that its skill ranges actually support, filtered through composition checks (three damage dealers with nobody keeping them alive is not a formation), and graded out of 100 on eleven named checks. That score decides which lineups win a seat in their tier (the tier itself follows the heroes’ seasons, set out below), with a second step making sure the page is not a hundred rearrangements of the same three names; inside a tier the order comes from simulated battles, described below.

The eleven checks

These are the same audits a strong player runs before fielding a lineup, with points attached. The Combo Rater prints every check’s points for any trio; the notes on the ranked lists put the checks a lineup falls short on into words instead.

Those sum to 100. One deduction exists outside the table: a lineup that harms its own side (self-damage, friendly damage taken) loses up to 8 points for it.

Every list is ordered by battle

The scorecard finds and scores the lineups — but at the very top of a list its resolution runs out: the strongest thirty span three to nine points, and most sit about a tenth of a point from their nearest neighbour, gaps too small for the scorecard to separate honestly. So the head of each list is ordered by evidence of a different kind. After every regeneration, the top thirty fight each other round-robin on the site’s own battle simulator — six hundred battles per matchup, three hundred from each seat so neither side of the field is favoured, under the victory rule the game’s developers confirmed — and their ranks follow their records: most wins first, a tied pair settled by their own head-to-head, then by score. The record is drawn on each row as a bar with its count (27 of 29 in battle), so the claim the rank makes is one you can check, and even restage yourself in the Combo Simulator. Below the top thirty, every row in every tier fights a fixed panel of eight reference lineups drawn from across its category (a row that is itself on the panel fights the other seven) — the first, fourth, eighth, fifteenth, thirtieth, sixtieth, hundred-and-twentieth and two-hundred-and-fiftieth by score — from both seats, and the rows are ordered within their tier by that record (6 of 8 reference lineups), then by their mean win rate, then by score. This is why a row can show a higher score than the row above it anywhere on a list: the scorecard finds the lineups and fills each tier, and the battles order them.

Near the top of a list, not all eleven checks separate the scores. Range and timing are earned in full by every published lineup, and so is troop eligibility on the three single-type lists — they behave as qualification checks rather than as tie-breakers. So when two combos sit a point apart, the difference is coming from the remaining checks: damage output, trigger chains, season strength, a clear carry, protection, engine coherence, disruption and targeting. The Mixed list is the exception for troop eligibility: half of that check rides on team-mates of other types, so it never reaches full marks there and still separates lineups. The qualification checks bite on trios the lists never seat, which is where the Combo Rater shows them failing.

How much do those weights actually matter?

The point values above are the one genuinely arguable part of this model. Nothing in Rise of Castles says a trigger chain is worth 12 and targeting is worth 2. So rather than defend them, we measured how much they matter: every weight was moved up and down by a quarter, and every lineup was re-ranked by score.

The score order barely moves. Across all four lists and all eleven checks, the re-ranked order agreed with the original score order at a rank correlation of 0.99 or better, and at most three lineups entered or left any top ten. Season strength scored 0.995, clear carry 0.997, engine coherence 0.999. That was measured in August 2026, when the lists were still ordered by score; the order inside a tier now comes from battles, which the weights touch only as a final tie-break. The rankings are driven by what the heroes' skills do, not by the exact points attached to each check.

Individual lineups in the crowded middle of a list do move, by up to sixty places, and that is why no list is ordered by its score alone: when two lineups sit within a fraction of a point, the order between them is not a claim the scorecard can make, so battles settle it.

Expected damage, not printed damage

The single biggest difference between this scoring and reading skill descriptions: "45% chance to deal 410% damage" is treated as roughly 185% per round, not 410%. Every hero's skills carry activation chances, preparation rounds and cooldowns, and they differ enormously. A guaranteed pre-battle effect always happens; a combat skill rolls its chance every round; a skill that needs a round of preparation can be interrupted before it ever resolves. All damage is converted to expected damage over a full eight-round battle so a reliable 250% hero and an unreliable 500% hero are compared honestly.

Tiers: lineups are ranked within their hero economy, never across it

Rise of Castles has two hero economies. Season 0–4 heroes — from the Season 0 starting heroes every account begins with through Season 4 — complete with just 4 copies. X-season and SP heroes need 34 copies, and many are paid. No real account fields starting heroes beside an X29: whoever completed the X29 has outgrown their starters, and whoever is on starters cannot afford the X29. So this site never ranks those worlds against each other.

Every lineup is placed in a tier by every hero it requires you to own, which is what the chip on the row means: Current meta is X16-or-later, SP, or paid, all three; Mid progression is nothing newer than X15, with no SP and no paid hero (a developer pick that spans eras is shown there too, with its badge); Starter is free Season 0–4 heroes only — the best lineups anyone can field for free, ranked seriously rather than buried under lineups that need paid heroes. Scores are comparable within a tier; across tiers, the score gap mostly measures power creep, which is exactly what it should measure.

A paid hero counts as current whatever season it came from, and that is the point of the rule rather than an exception to it. The tier asks what your account has had to achieve, and reaching X29 takes years of play — which is what makes it an odd partner for a starter everybody was given. A paid hero proves nothing of the kind: you can buy it in your first week. So Theodora is a Season 4 hero that players at X29 still field, and the list says so. Whether such a lineup is any good is a separate question, and the season-strength check below still ages an old hero for the part of its kit that has been outclassed.

The same logic rejects a trio that spans those worlds on the free side. A Season 2 hero you were given, beside an X29, is a pairing no real roster fields, so it is not scored as Current or Starter. Adjacent bands inside Mid (an X03 beside an X15) are normal account progression and stay allowed.

One thing still ships when it spans: lineups recommended by the game’s own developers, always shown, always badged, always at whatever score they honestly earn — often mid-table. Today 168 rows put a Season 0–4 hero beside an X16-or-later or SP hero: 21 are developer picks, and in every one of the other 147 the Season 0–4 hero is a paid one, which is what makes the lineup Current meta.

Power creep is real, so it is measured

A skill multiplier only hits as hard as the stats behind it. An early-season hero carries a passive package around Might +40%; a current-season hero carries +65% on the squad, +40% on the legion, plus extras. Raw damage is therefore scaled by each hero's own stat chassis and by its era. Debuffs, control and mitigation are not aged — a −60% Might strip takes the same bite out of the enemy whoever applies it — which is why some veteran support heroes stay relevant while veteran damage dealers fade. That matches how high-level players actually field them.

The season-strength check follows the same rule, which it did not always do. It used to age every hero by release date alone, so a Season 1 hero who is 91% debuff was penalised for an obsolescence this model had already said did not apply to him — the page contradicting itself twelve points at a time. A hero now ages in proportion to the part of his contribution that decays: a pure carry is aged exactly as before, a pure support pays half the penalty, and most heroes sit somewhere between. Half, not none, because a support kit’s debuffs and mitigation are already left unaged where the model prices them; relieving them in full here as well would credit the same thing twice.

Paid heroes are marked, not penalised

Some heroes cost real money — SP heroes are the most expensive, followed by the other paid heroes, mostly X-season (plus Cleopatra VII, Octavius, and Theodora from the early seasons); the rest are free. Being paid does not change how hard a hero hits, so it never changes a score. It changes what you can field, so it is marked everywhere instead: every paid hero is labelled in the combo explanations ("Saladin (SP · paid)"), and any lineup whose three heroes are all obtainable without spending carries a budget note saying so.

Every ranked list also carries a “Free heroes only” filter with the count on its face, because a mark buried inside an expandable note is not the same as an answer. A large share of every list needs no paid hero at all — 552 of the 1,394 ranked lineups — and the best fully free lineups are Archer’s and Mixed’s Current meta #1. Filtering does not renumber anything: a free lineup at #12 really is the twelfth lineup in its tier, and calling it #1 would flatter you at the cost of the truth.

SP heroes are dated by the developers' own groupings

SP (premium) heroes carry no release season, but they are not all the same age — and pretending they are would overrate the old ones. Each SP hero's era is estimated from the dated heroes the game's developers field it with in their own recommended lineups: an SP hero the devs group with Season 1–3 heroes is priced as that generation, while one they group with X27+ is priced as current. This is official evidence, applied hero by hero, and it is why two SP heroes can sit far apart in the rankings.

It only applies when the evidence is good enough. A hero needs at least two developer lineups, and those lineups have to agree: if the seasons they span are too wide, the cohort is saying “this hero is fielded in every era”, not “this hero is old”. Ramses II spans S01 to X26. Where the evidence fails either test the hero takes a neutral value instead, and that is the common case — most SP heroes on these lists are dated neutrally, not from a cohort.

Synergy means a trigger chain, not three big numbers

When a note says "Saladin's prep-skips set off Lancelot's bonus effects," that is a literal claim about the two heroes' skill text: one hero creates a battle event (an extra basic attack, a dodge, a cast-chance boost, a prep skip) that the other hero's skills react to. Chains built on rare events score higher than chains built on events every hero produces anyway. A lineup with no chain at all is told so to its face: its note says the three work side by side rather than off each other.

Rows come from skill ranges

Your front row stands at distance 1–3 from the enemy rows, your middle at 2–4, your back at 3–5 — and every attack skill has a range. A formation is only shown with row assignments its ranges actually support. The standing prefers durable heroes and counter-engines (kits that feed on being attacked) in front, support in the middle, and the biggest ranged hitter in back — unless another legal standing scores higher. A tied score never moves the carry into the middle just because that permutation was tried first. When a hero has to give up attack skills to stand in a row, the scorecard says how many and which row. On the published lists this never happens — every shipped lineup has an arrangement that costs nothing — so you will only meet that sentence in the Combo Rater, on a trio of your own.

Most cards that offer a Footman: branch and a Cavalry: branch switch on the type the hero is fielded as, and only that branch ever acts. (Percival and Fortuneteller each carry one card that switches on the target’s type instead, where every branch can act in one battle.) The scorecard counts such a card — its damage and its buffs, debuffs and mitigation alike — only for the type the lineup fields, on every list including Mixed, and the Combo Simulator plays only that branch. Both branches on the same card would credit a squad with effects the game never lets it have: an Archer Oliver buffs himself and never strips the enemy, a Cavalry Oliver does the reverse.

Who gets a seat on the page

Scoring every legal trio is not the same as publishing every high score. Each tier keeps a fixed length (about 250 Current, 50 Mid, 30 Starter, plus any developer pick that missed the cut). The first seats in each tier are the highest scores, so a famous 85 cannot lose to a 64. The rest of the tier prefers lineups that share at most one hero with those headline cores — a new idea, not a remix of the same three names — then fills leftover seats by score. The second step only decides who is on the page; the order inside a tier comes from the battle records described above.

Developer picks

Lineups recommended inside the game by its developers are always shown in the ranked list, marked as developer picks, at whatever score they honestly earn today. Their scores are never inflated: the badge plus the scorecard tells you "the developers recommend this core, and here is what it is worth against the current roster."

The engine speaks the game's language

All of the above only works if every word in a skill description is understood exactly the way the game means it — Silence versus Disarm versus Suppression, Burning versus On Fire versus Flammable, Dodging (guaranteed) versus Evade (a chance), Fatal Blow's +100% versus Destructive Strike's 200%. The full vocabulary the engine reads — every skill type, attribute, control status, fire effect, mark and modifier family — is published as a complete Rise of Castles combat glossary, so you can check our definitions against the game's own text term by term.

Where the Simulator and the Rater fit

The Combo Simulator is a separate instrument built on the same foundation: it reads the same stored skill text as the ranking engine and fights any two formations 2,000 times on the server, on identical armies. The scores above never come from it — they are computed from the eleven checks — but the order inside every tier does: the battle records that order each list are fought on this engine. Its battle pacing is fitted to real, transcribed battle reports, but the game’s exact damage formula is unpublished, so its constants are deliberately uncalibrated. The troop-counter triangle’s built-in strength is never assumed either — the game does not publish that number. Every cross-type verdict is re-tested at hypothetical strengths from 0–40%, and the result says in plain words whether it would change. The counter effects individual heroes state on their own cards do act, because those are printed on the card.

The Combo Rater is the third instrument: it runs the same eleven checks on any three heroes, including combinations no published list contains, and sets that score beside the published list’s own scores. The number it prints is the number a ranked row would carry. It never publishes anything: which lineups make the lists, and in what order, stays with the offline pipeline described here.

What keeps this honest

Three outside sources are used to test the model rather than to set scores: real high-level lineup usage from the community, the in-game developer recommendations, and corrections from players who are deep in the game. The developer recommendations do two further jobs, both stated above: they are always listed, badged, at the score they earn (developer picks), and they date the SP heroes (SP dating). When the model and those sources disagree, the disagreement is investigated down to the specific skill text that caused it, and whichever side turns out to be wrong gets fixed. Several genuine bugs in earlier versions of this ranking were found exactly that way.

The model is deterministic: the same hero data always produces the same list. When a new hero ships, its skills are entered as written in the game, and the whole ranking — scores, rows, chains, notes, pair synergies — recomputes with no manual placement anywhere. If you believe a specific score line is wrong, that is exactly the kind of correction that improves this site: tell me which line and why.