How rooms and speaking order are drawn

1. Why speaking order matters

A judge in a preliminary round of nine students listens to roughly ninety minutes of continuous performance, then produces a ranking. Memory and attention are not constant across those ninety minutes. The literature on sequentially judged contests gives us good reason to take this seriously. Bruine de Bruin (2005) found systematic serial-position effects in jury evaluations at the Eurovision Song Contest and in figure skating, where later performers received higher marks. Glejser and Heyndels (2001) documented the same pattern across decades of results from the Queen Elisabeth music competition, a setting where performance order is drawn by lot. Page and Page (2010) confirmed it at scale in televised singing competitions.

We want to be careful about what we claim. These studies cover music juries, skating panels, and television audiences. Nobody has measured the size of order effects in high school forensics specifically, and we do not pretend to know it. The direction of the published evidence is consistent, though, and the cost of taking it seriously is small. A student who draws the first speaking slot in all three preliminary rounds has been treated differently from one who closes every round, whatever the true magnitude turns out to be.

Our position is practical rather than theoretical. We do not attempt to correct judge scores for order, which would require knowing the bias precisely. We make the draw itself equitable, so that no competitor systematically inherits the same seat.

2. What we protect first

Scheduling objectives conflict. A draw that perfectly balances speaking positions might place two teammates in the same room. When objectives collide, the engine resolves them in a fixed order:

1. School separation. Competitors from the same school are kept apart whenever the rooms allow it. Schools that ask to be linked (for example, a high school and its middle school) are kept apart like teammates. This is the strongest rule in the draw.
2. Opponent variety. Across the preliminary rounds, each competitor should meet as many different opponents as possible. Meeting the same opponent three times is penalized far more heavily than meeting two opponents twice.
3. Nobody speaks last twice. In some events going last is a real advantage, so no competitor takes the final slot of a room in more than one round, whatever size the rooms are, and nobody who went last earlier in the day goes last in a final.
4. Position non-duplication. A competitor should not draw the same speaking slot in two rounds.
5. Tier balance. We divide each room into an early tier (slots one and two), a late tier (the final two slots in rooms of six or more), and a middle. Each competitor’s positions should spread across the three tiers.

The ordering reflects competitive judgment, not measurement. Michigan’s rules and long tab-room practice treat school separation as non-negotiable; the remaining four are ranked by how directly they touch competitive outcomes.

3. How it works

The engine scores any candidate schedule with a single penalty number. Each school collision costs 1,000,000 points. Each repeated head-to-head matchup costs 10,000, squared as repetition accumulates. Each extra time a competitor speaks last costs 5,000. Each duplicated speaking position costs 1,000, and each unit of tier imbalance costs 100. The wide spacing between weights means a lower priority can never outvote a higher one: no quantity of beautifully balanced speaking orders is worth a single same-school pairing. The specific values are design choices that encode the priority ranking. They are not derived from data, and we say so plainly.

Finding the schedule with the lowest penalty is a combinatorial search problem; for a field of sixty entries across three rounds, the space of possible schedules is astronomically large. We use simulated annealing, a standard optimization technique. The engine starts from a random draw, then repeatedly proposes small changes: swap two competitors between rooms, or swap two speaking slots within a room. Changes that lower the penalty are always kept. Changes that raise it are sometimes kept early in the run and almost never kept late. This willingness to briefly accept worse schedules is what lets the search escape locally good but globally poor arrangements. A run takes thirty thousand proposals, which completes in well under a second on our infrastructure.

The output assigns every competitor a room and an explicit speaking position for every preliminary round, and both are written to the tournament record together. Elimination rounds are excluded by design: semifinal and final seating follows the break, which is a separate part of the tab engine.

4. When the rooms can’t be perfect

Sometimes the arithmetic does not work. If a school enters eight students in an event sectioned into six rooms, at least two of its students must share a room. No algorithm changes that.

What an algorithm can do is refuse to hide it. When the best schedule the search can find still contains a same-school pairing, the tournament editor is told exactly how many pairings were unavoidable, in the same breath as the result. The tab staff can then decide what to do with that information: accept the draw, add a section, or talk to the coach. We consider the honest report as important as the optimization itself.

5. What this can’t fix

Three are worth stating. First, the bias literature we lean on comes from adjacent domains. Until someone runs the study on forensics ballots, the size of order effects here is an open question, and we would genuinely like to see that study done. Second, simulated annealing is stochastic: two runs on the same field produce different schedules, both near-optimal but not identical, and the engine makes no claim of finding the unique best draw. Third, equitable scheduling addresses one source of unfairness among many. Judge assignment, room acoustics, and time of day all carry their own effects, and a balanced draw does not absolve a tournament of attending to them.

The implementation, including the penalty weights and the test suite, ships with the platform. Tab staff who want to verify the behavior described here can do so against the tournament record itself: every speaking order the engine produces is visible on the published schematic.