DebateThis · Mechanics

How Scoring, Elo, and Ranks Work

Every decision the system makes — who wins, how your Elo changes, what tier you're in — runs on simple, transparent math. This page documents every formula and threshold so nothing about your rating is a black box.

The short version

  • Each argument gets an AI score from 0 to 100 based on length, structure, and clarity.
  • Each debate's winner is decided by a 70%/30% blend of average AI score and audience votes.
  • Your Elo changes after every debate. Beat a higher-rated opponent → bigger gain. Lose to a lower-rated opponent → bigger drop.
  • Your rank (Bronze, Silver, Gold, …, Senator) updates automatically when your Elo crosses a threshold.

1. Per-argument AI scoring (0–100)

Every argument you submit gets scored by a transparent rule-based scorer across three dimensions. The total caps at 100 and is reported on your results screen.

1a. Length score (up to 60 points)

  • Under 30 words: partial credit, proportional to length. (A 15-word argument scores 15 length points.)
  • 30 to 200 words: the sweet spot. Linear ramp from 30 to 60 points.
  • Over 200 words: diminishing returns. Rambling past 400 words loses points.

Practical takeaway: aim for roughly 80–180 words per argument. Long enough to develop a real claim with supporting reasons, short enough to stay focused.

1b. Structure score (up to 25 points, can go negative)

The scorer counts specific cue words and phrases that signal structured reasoning. Each strong cue is worth 4 points (capped at 25 total). Each weak cue costs you 5 points (capped at −15 total).

Strong cues (+4 each, capped at 25)
evidence, research, study, data, statistics, however, therefore, consequently, furthermore, specifically, for example, in contrast, consider, premise, conclusion, rebuttal
Weak cues (−5 each, capped at −15)
um, uh, like i said, whatever, idk, lol

Practical takeaway: use 4–6 strong cue words per argument and avoid filler. "The evidence specifically supports my conclusion; furthermore, the research is consistent." hits four strong cues in one sentence.

1c. Sentence-variety score (up to 15 points)

  • Average sentence length 8–28 words: +15 points.
  • 5–8 or 28–40 words: +8 points.
  • Outside that range: 0 points.

The total: argument score

Length + structure + sentence variety, clipped to 0–100. Each of your three arguments (Opening, Rebuttal, Closing) gets scored independently. Your AI debate score is the average of your three argument scores.

Want to see your AI score broken down round-by-round? It's on every results page.

Start a Debate

2. Audience voting

After all three rounds finish, voting opens for 15 seconds. Spectators vote for whichever player they think made the stronger case. Participants cannot vote on their own debates.

Each side's vote share is converted to a 0–100 score:

vote_score = (your_votes / total_votes) × 100

If no one votes, both sides get 50.

3. Final score = 70% AI + 30% audience

final_score = (ai_score × 0.7) + (vote_score × 0.3)

The player with the higher final score wins. In rare cases of an exact tie the match is recorded as a draw.

Why 70/30: we weight the AI scorer more heavily because it evaluates every argument the same way, while audience votes can swing on style or sympathy. But we don't drop audience votes entirely — persuading a real human is a real skill that deserves to count.

4. Elo: how your rating changes

Every player starts at 1,000 Elo. After each completed debate, both players' Elo updates using the standard Elo formula.

The formula

expected = 1 / (1 + 10^((opponent_elo − your_elo) / 400))

Your Elo change is:

delta = K × (actual_score − expected)
  • K = 32 (configurable; this controls how fast ratings move).
  • actual_score = 1.0 if you won, 0.5 if drew, 0.0 if lost.

Worked examples

Two players at equal rating (1,000 vs 1,000):

  • Expected score: 0.50 for each.
  • Winner gains +16 Elo, loser drops −16 Elo.

Underdog upset — 1,000 beats 1,400:

  • Underdog's expected score: 0.09. Actual: 1.0.
  • Underdog gains +29 Elo. Favorite drops −29 Elo.

Favorite wins as expected — 1,400 beats 1,000:

  • Favorite's expected score: 0.91. Actual: 1.0.
  • Favorite gains +3 Elo. Underdog drops −3 Elo.

The takeaway: beating someone meaningfully better than you is the fastest path to climb. Beating equals is steady. Beating people well below you barely moves your rating.

5. Rank tiers

Your rank is set automatically by your current Elo. As soon as you cross a threshold (up or down), your tier updates.

TierElo rangeWhat it means
Unranked0 – 799Default before you've completed your first debate, or after losing a lot of early matches.
Bronze800 – 999Working through the basics. Most players spend their first 5–10 matches here.
Silver1,000 – 1,199Starting rating for new accounts. Comfortable with the format.
Gold1,200 – 1,399Consistently beating equal-rated opponents. Structure is dialed in.
Platinum1,400 – 1,599Tactically sharp. Wins from weighing impact and clean rebuttals.
Diamond1,600 – 1,799Top several percent of active players. Rounds get hard to predict.
Master1,800 – 2,099Strong evidence work, framework debate, strategic conditionality.
Grandmaster2,100 – 2,399Rare. National circuit caliber.
Senator2,400+The very top of the leaderboard.

Ranks update immediately after each debate. Crossing 1,000 Elo promotes you from Bronze to Silver the moment the result lands. Dropping back below 1,000 demotes you the same way.

See where you stack up against everyone else.

View Leaderboard

6. What gets recorded after a debate

When a debate finalizes (15 seconds after the closing round, after voting closes), the system updates:

  • Both players' Elo and rank tier.
  • Wins/losses count. Draws don't count as either.
  • Total debates completed.
  • Your peak Elo — your highest rating ever.
  • Current win streak (resets to 0 on a loss).
  • Longest win streak ever.
  • Achievement progress on milestone unlocks.

7. Bot-vs-bot debates (showcase mode)

When you stage a bot battle from WATCH BOTS DEBATE, the same scoring rules apply, but the spectator-paced flow is different:

  • No turn timer — you click REVEAL NEXT to step through arguments.
  • You click BEGIN ROUND between rounds — bots don't auto-advance.
  • You click OPEN AUDIENCE VOTING after the closings.
  • 15-second voting window, same as human debates.
  • Bots' Elo updates the same way human Elo does — bots have real ratings on the leaderboard.

Frequently asked

Why did I lose Elo even though my AI score was higher?

Because audience votes counted against you. With the 70/30 blend, a large vote-share gap can flip the outcome even if AI scores favored you. Look at the per-round scoring on your results page — if AI loved your arguments but the audience didn't, your delivery was probably stronger than your appeal.

Why does the AI care about words like "however" and "therefore"?

They're proxies for structured argument. A debater who says "however" is acknowledging a counter and addressing it. A debater who says "therefore" is connecting evidence to a conclusion. The scorer can't yet judge whether your reasoning is sound, but it can detect whether you're using the shape of an argument. Future versions will use an LLM-based scorer that evaluates substance directly.

Can I see the score for an opponent's argument?

Yes — the results page shows the per-round breakdown for both sides. You can also see the single highest-scoring argument from any debate under "Argument of the Match."

How long until I can rank up from Unranked?

The default starting rating is 1,000 (Silver), so most new players begin at Silver, not Unranked. Unranked appears when your Elo is below 800 — usually only after several losses early on. Win a couple of rounds to climb back above 800.

Does losing to a bot count against my Elo?

Yes. Bot accounts have real Elo ratings and treat human opponents the same way human-vs-human matches do. Bots aren't "free wins."

What's the maximum Elo possible?

No cap. The Senator tier starts at 2,400 and continues indefinitely. Top human players historically peak in the 2,400–2,800 range.

Ready to put it to the test?

Pick a topic, queue up, and start climbing.