A team comp that works in Overwatch is not five meta heroes thrown together. It is a lineup where the roles cover each other, the archetypes fit the map, and the picks make each other more dangerous instead of fighting for the same space. Role balance gets you in the door. The archetype layer on top is what actually wins.
Role balance is the floor
Overwatch 5V5 is one tank, two damage, two supports. Every rule of thumb starts there. If your tank dies, the frontline folds. If neither damage hero can reach the enemy supports, fights feel hopeless. If your supports cannot keep the tank up, you lose the engagement before it starts. But every standard queue comp already satisfies role balance, so it is the floor, not the ceiling. What separates a comp that works from one that fights itself is the archetype on top.
Dive, Poke, and Brawl
Every 5V5 comp sits somewhere between three archetypes.
Dive is fast heroes who collapse on a target before the enemy can react. Tanks like Winston, D.Va, Doomfist, and Wrecking Ball; damage like Genji, Tracer, Sombra, and Soldier; supports like Lúcio, Kiriko, and Mercy. Dive wants flanks, high ground, and vertical maps.
Poke is long-range damage that farms ultimates from mid-range and forces the enemy out of position. Tanks like Sigma, Orisa, and Ramattra; damage like Widowmaker, Hanzo, Ashe, and Pharah; supports like Ana and Baptiste. Poke wants open maps with clean sightlines.
Brawl is high-health, close-range fighters who win the 50/50 through sustain and area denial. Tanks like Reinhardt, Junker Queen, Roadhog, and Mauga; damage like Mei, Reaper, and Torbjörn; supports like Brigitte, Moira, and Lifeweaver. Brawl wants chokes and tight indoor maps.
A comp that mixes archetypes too hard fights itself. A Winston wants to dive in and leave, a Widowmaker wants to hold a sightline and stay, an Ana wants to stand back and scope. Three agendas, no shared engagement window, and the team loses fights because they are never in the same one. The fix is archetype coherence. Pick heroes that all want the same kind of fight, then let the map decide which archetype to run.
Synergy versus counter
Here is the trade-off you cannot dodge. You cannot max synergy and counter at the same time.
Synergy is your team's internal cohesion. A Junker Queen and Mercy pairing means Mercy's damage boost pushes Queen's axe and shotgun breakpoints into one-shot range while Queen's pull clusters enemies for the boost to capitalise on. In recent community data that pair sits around 56.9% over roughly 6,000 tracked 5V5 matches, one of the cleaner synergy numbers in the game right now. Counter is about beating what the enemy picked, the Reaper-into-Winston "I saw your tank and swapped to shut it down" move.
The problem is that high-synergy and high-counter picks often point at different heroes. Swap to counter and you might break your team-up; stay synergised and you might walk into a bad matchup. The team builder's scoring takes this as given, so no hero wins on both dimensions at once and the ranking always reflects the trade.
How the team builder scores a lineup
The team builder blends three inputs into one win-chance number. Counter score is half of it, how each friendly hero does against the enemy lineup across all five matchups. Synergy score is another 30%, how each friendly hero does alongside the other four. Map performance is the last 20%, how each hero does on the chosen map, weighted by sample. Counter carries the most because counter-picking is the highest-leverage call in a match. Synergy matters but less, since a perfectly synergised comp still loses against the wrong enemy. Map is smallest because map effects are a weaker signal than matchups, and plenty of players do not optimise by map anyway.
Three synergy pairs from recent data
All from recent 5V5 community data, all ranks, all game types, and all liable to move with the patch.
Roadhog and Kiriko, around 56.0% over ~14,500 matches, the highest-sample pair of the three. Roadhog's problem is that he trades too much health and dies the moment a second cooldown lands on him. Kiriko's Suzu cleanses the crowd control, burst, and anti-heal he cannot escape on his own, and Teleport lets her reach him when he over-extends. One hero papers over the other's single worst weakness, and the huge sample means you can trust it.
Doomfist and Sojourn, around 56.3% over ~4,400 matches, a dive pairing. Doomfist pins a target long enough to kill it; Sojourn's railgun erases the low-health targets he has already chipped. He makes the opening, she converts it in the same two-second window. Comps built around the pair want dive-friendly maps with flank routes.
Junker Queen and Mercy, around 56.9% over ~6,000 matches, the top synergy of the three. Mercy's boost lifts Queen's axe and shotgun into one-shot territory against squishies, Queen's ultimate clusters targets for the boost to exploit, and Queen's bleed self-sustain lets Mercy stay in damage-boost mode longer. A brawl pairing that wants chokes and close-range maps.
When to break your comp
The team builder hands you a current best lineup, but locking a mathematically optimal comp and never adjusting is worse than a smart mid-match swap. If the enemy picks a hard counter to one of your damage heroes, taking a one-point comp-score hit to move into a better matchup is usually worth it. If they lock a full brawl comp on a poke map, your synergy number dips a little when you adjust but the map-weighted factor gives back more than you lost. The rule is that the builder's score is an average of a lot of real matches, and when a specific adjustment clearly helps against what is in front of you, the specific beats the average.
Before the match and during it
The team builder is the tool before you load in or during role queue, where you can see full-lineup predictions and swap suggestions. Once the match starts the Counterwatch app overlay takes over, updating synergy and counter reads in real time as enemy picks lock, with no alt-tabbing. For the scoring formula, the shrinkage, and how the terms are weighted, the methodology page has the detail.