What Speaker Battles Taught Us About Feedback
By Dara Halloran · Community & Practice
A Speaker Battle is simple. Two members speak on the same prompt, one after the other, and both are scored on the same five axes: clarity, tonality, structure, energy and intellect.
There are two kinds. In an audience battle, often run inside a live session, the room votes: every audience member scores both speakers on each axis from one to five, and can add a short note. In a live battle, two members meet one to one on video with timed turns, a set amount of preparation time followed by a set amount of speaking time, and an AI judge scores both turns on the same five axes.
We built battles to make practice competitive enough to be memorable. What they turned into was a running lesson in how feedback should be shaped, mostly through things we had to change.
Comparison beats evaluation
Ask someone "how was that talk" and you get politeness. Ask them "which of these two, and on what" and you get a real answer.
This is not a fact about any particular group of people being conflict-averse. It is a fact about how judgement works. Absolute ratings are unstable, because people do not carry a calibrated internal scale for "how good was that". Comparison is easier. As far back as 1927, the psychologist L. L. Thurstone set out a model, the law of comparative judgement, showing how consistent measurements can be built from nothing more than repeated judgements of which of two things has more of a quality.1 Paired comparison has been a standard tool for measuring hard-to-define qualities ever since.
A battle puts that structure in front of the room. The voter still gives each speaker a score, but they give both scores on the same axes in the same sitting, straight after hearing both. The question in their head is "which of these two was clearer", and that is a question people can answer.
The practical consequence for anyone practising anywhere: stop asking "was that good". Ask "was that better than last time, and on what".
The criteria have to match what you are trying to improve
Battle votes used to ask about four criteria: tonality, body language, narrative and an overall score. It looked reasonable, and it had two problems.
The overall score was a summary of the other three, so counting it alongside them counted the same impression twice. And the four criteria only fed three of the five axes the rest of Exprea measures. Clarity and intellect, whether the point was clear and whether the thinking was strong, got no human input from battles at all.
So the ballot now asks about the same five axes as everything else. Battles voted on before the switch are still shown with the questions the room was actually asked, because a finished battle is a record of what people answered, not something to rewrite.
The lesson generalises. If the feedback form asks about polish and the thing you are working on is the argument, the form will steer everyone, voters and speakers alike, towards polish.
Short notes get written
The note a voter can leave is optional and capped at 280 characters. The cap is deliberate. One useful sentence beats a paragraph nobody writes, and a short box makes it clear that one sentence is the job.
A good note has a shape, and it is the same shape that makes any feedback usable:
- One moment that worked: a specific line or beat, rather than a general quality
- One moment that lost you: the point where attention dropped, again specific
- One thing to try next time: an action you can take, where an adjective would give you nothing
"Be more confident" is a wish. "You looked at the floor for the whole second story, so try picking one person for that bit" is feedback, because the speaker knows what to do next time. There is more on this shape in feedback someone can actually use.
Notes are written by people, so they can also be reported, the same way a comment can.
Fairness is part of the feedback
A score only teaches you something if you trust it. Several rules exist for that reason alone.
The two speakers cannot vote in their own battle, and neither can the member who set it up. Each person votes once. In a battle that belongs to a session, only people registered for that session can vote, so the scores come from people who were in the room.
In live battles, the order of speakers used to follow whoever sent the challenge. That looked neutral and was not. The second speaker hears the first speaker's whole turn before giving their own, which means more information and more thinking time. Speaker order is now a coin toss.
Live battles also let a speaker buy an extra thirty seconds on their own turn. It costs a little XP and takes a small deduction from their structure score, because running over is itself a structural choice. The clock is part of the craft, and the scoring says so.
Losing is the useful half
Winning tells you the overall impression worked. It does not tell you why.
Losing, with five axis scores and a few one-line notes attached, tells you where it did not. A lower clarity score next to a higher energy score is a different message from the reverse, and both are more useful than a single verdict.
That is the argument for the whole format. The room exists to give you information you cannot get by rehearsing alone in a kitchen, more than to rank you, and information often arrives most clearly attached to a result you did not want.
If you lose a battle, read the axis scores before you read the result. Pick the lowest one, find the note that explains it, and make that the single thing you change next time.
References
- Thurstone, L. L. (1927). A law of comparative judgment. Psychological Review, 34(4), 273–286. https://doi.org/10.1037/h0070288