Team composition theory, counter-picking, ban strategy and reading the meta — grounded in win-rate data from millions of ranked matches.
Method over memorization: every claim links to the live table it comes from.
A coherent Dota 2 team composition is one where all five heroes serve the same game plan: it mixes physical and magical damage, has at least one reliable initiation and one save, aligns its power spikes around the same timing window, survives its lanes, and can state its win condition in one sentence. Five individually strong heroes that fail any of these checks lose to five average heroes that pass all of them.
Most winning drafts collapse into one of four plans. Knowing which one you are drafting toward — and which one the enemy is — tells you what to pick and ban next.
Two heroes "synergize" when one reliably creates the situation the other needs: a setup stun into a channeled ultimate, a mana battery under a spell-spammer, twin power spikes that hit at the same minute. That shows up directly in the data — some duos win far above what either hero manages alone. The strongest same-team pairs on patch 7.41e right now:
Browse every duo on the hero synergy hub — each pair links to a page with its win rate history and per-rank breakdown.
Counter-pick by targeting the enemy's win condition, not their heroes one by one: pick what invalidates their plan (mana burn into spell-reliant cores, constant pushing into greedy scaling drafts, saves into pick-off lineups). A direct hero counter is worth taking when the matchup edge is large and the hero still fits your own composition — otherwise it is a trap.
Most hero-versus-hero edges are worth a few percentage points; playing a hero badly costs far more. The practical rule: take the counter when the edge is large and you can actually play the hero; otherwise take your comfort pick and counter with itemization and movement instead. This matters more at lower ranks, where execution variance dwarfs draft edges.
A hero that wins the lane can still lose the game — and vice versa. When you read a matchup table, ask when the edge applies: laning-phase counters (harass, kill threat) fade after minute 15; structural counters (mana burn vs. mana dependence, break vs. passive-reliant heroes) last all game. Structural counters are the ones worth spending a pick on.
Bans should remove the enemy's most probable good outcome, and what that is changes by bracket. In Herald–Guardian and Crusader–Archon, ban the simple pub-stompers that over-perform at that bracket — the per-rank tier lists show exactly which heroes those are, and they differ sharply from the global list. At high ranks, ban the patch's contested meta answers and your opponents' signature heroes instead.
The largest one-sided matchups on patch 7.41e illustrate the ceiling:
Edges that size are worth drafting around; the long tail of 51–53% matchups is not. Look up any hero on the counters hub before you lock.
Read the meta on two axes — win rate and pick rate — and watch the direction of movement, not just the level. High pick + high win marks the patch's contested answers (first-phase pick or ban material); low pick + high win marks niche stompers you can often get for free; a rising win-rate trend catches buffs and item interactions before the tier lists fully reprice them.
Pro drafts optimize around coordination that pubs don't have: heroes that need four teammates to follow up under-perform in ranked and over-perform on stage. Copy pro drafts for ideas, but validate against ranked data at your bracket before adopting them. The TI 2026 pick/ban tracker vs. the ranked tier list makes the gap visible side by side.
The first two weeks after a balance patch are where win rates move fastest — and where stale knowledge costs the most. The patch 7.41e winners & losers page tracks exactly which changes actually moved numbers, as the data arrives.
Prepare drafts the way you would prepare any decision under uncertainty: know your own hero pool cold, scout what the opponent is likely to do, rehearse the pick/ban sequence against realistic responses, and pressure-test the final composition with a model that tells you the honest probability — not the flattering one.
Honesty note: no draft tool sees execution. Our models are optimized for calibration — when they say 60%, drafts like that win about 60% of the time (54.93% accuracy on 758,807 held-out ranked matches; full receipts on the methodology and benchmark pages). Use the probability to compare options, not to blame the draft after a loss.
Win-rate figures on this page are from the patch 7.41e ranked snapshot, updated Aug 28, 2026.