Introducing ranked AI judging

How a debate
becomes a number

The most emotional screen in moggle is a scoreline. So we built the scoring to be something rare for AI: deterministic and auditable. The same debate always produces the same score, and every single point traces back to a line you can read.

The one idea

The code decides, the AI only perceives

The scoring engine is built on one split. The AI model is only ever asked small, checkable questions about a piece of text: “Is this a claim or a rebuttal?” “Did this response meet that objection, did it hold, dent, or break?” It never assigns a score and never says who is winning.

Everything that decides the numberis plain arithmetic, with no model and no randomness. That's what makes the result reproducible and fair.

Same debate in, same score out, every time. And because the math is fixed, every point on the scorecard is one you can check by hand.

The pipeline

From six speeches to a scoreline

1
Read the speeches
The judge splits each speech into moves: new claims, rebuttals (attacks on the other side's claims), and defenses (answers to attacks on your own).
2
Build the argument graph
Those moves become a map. Claims are points on the board, attacks and defenses are the lines between them. From here on, only the graph is scored, never the raw transcript.
3
Weigh each claim
Every claim is tagged Framing (it reframes the whole motion) or Substantive (a single supporting argument).
4
Resolve the clashes
Where a claim was attacked and answered, the judge decides whether the defense Held, was Dented, or Broke. A claim that's attacked and never answered counts as dropped, but only after a second check confirms nothing later in the debate quietly answered it.
5
Do the math
Pure arithmetic turns the finished graph into points, a winner, and the 0 to 100 scoreline.

Fate

What happens to a claim

Every claim ends in exactly one fate, and that fate decides how much of its value it keeps:

What happened	Fate	Keeps
Never attacked	Uncontested	100%
Attacked, defense held	Survived	60%
Attacked, defense dented	Weakened	30%
Attacked and dropped or broken	Defeated	25%

The scoring table

What a claim is worth

A claim's points are its base weight(how much it's worth at full strength) times its survival (how much it kept). A framing claim is worth more than any single argument, because it reshapes the whole debate.

Base weight

Framing claim: 30Substantive claim: 20

Survival kept

Uncontested 100% · Survived 60%Weakened 30% · Defeated 25%

Put together, those are the actual point values:

Fate	Substantive	Framing
Uncontested	20	30
Survived	12	18
Weakened	6	9
Defeated	5	8

The 25% floor for a defeated claim is deliberate. An argument you actually made and then lost still earns something, because it was an on-topic attempt. Zero is reserved for arguments that were never made at all.

Rebuttal credit

Attacking well is worth points too

Points come from claims. So a side that argues almost entirely by rebutting, picking apart the opponent's case, would score nothing no matter how well they did it. That isn't fair, so it isn't how moggle works.

A rebuttal earns its side a share of the damage it did. The harder the target fell, the more it's worth, and even forcing your opponent to spend a defense earns a little:

Your rebuttal…	Earns (vs a substantive claim)
Broke the claim	8
Dented the claim	7
Was answered, but forced a defense	4

A real example. One early match ended 12-0: one side made a single claim, the other argued purely by rebutting it. With rebuttal credit, the rebutting side is no longer shut out. The exchange reads 75-25, and that credit shows up as its own line on the scorecard.

The headline

From raw points to 0 to 100

Each side's points add up to a raw total. The headline scoreline is simply each side's share of the total, scaled to 100:

Raw points	Scoreline
12 to 4	75-25
15 to 57	21-79
nobody scored	50-50

The ledgerstays in raw points, which is the checkable breakdown, while the headline is the share. They don't add up to the same number on purpose, the same way a final score and a point-by-point card are two different views of one game.

The call

Winner, too close, and stable

The winner is decided on raw totals, so the headline can never disagree with who won. A verdict is flagged too close when the two sides land within 6 points of each other on the 0 to 100 scale, and stable when the margin is comfortable.

The chart

Debate shape & the key moment

The momentum chart replays the debate one speech at a time and plots who was ahead after each, so you can see exactly where the match turned.

The key momentis the single biggest swing on that line, usually the rebuttal that moved it the most. It's pulled out on your result screen with the exact line that did it.

Fairness

The rules we protect

Deterministic

Same debate, same score, every time. No randomness anywhere in the math.

Auditable

Every point is a line on the ledger. The headline is a stated transform of visible totals.

Blind resolution

The judgment that most affects the score sees no names, no sides, no motion. Bias has nothing to grab.

The model never scores

The AI classifies and perceives; code does all the arithmetic. The part that writes your feedback can't touch the numbers.

Engagement is never zero

Argue on-topic and you always score something. Zero is reserved for arguments never made.

Curious what the format itself looks like? Start with what a match is →