Random selection starts with a sample space
Before a result can be called fair, the possible results need a precise count. For an equal draw, the working formula is P(selected item) = copies of that item ÷ all eligible copies. Consider a lunch rota with eight entries: Bao, Curry, Pizza, Salad, Tacos, Pasta, Pizza, and Soup. Pizza has probability 2/8 = 25%; every one-copy meal has 1/8 = 12.5%. That is correct only if the two Pizza entries deliberately represent two separate claims. A duplicated paste, a person who was not eligible, or a missing participant changes the sample space before the random step ever begins.
The formula assumes that every displayed slot is equally reachable and that the list is frozen before the selection. It does not decide whether two similar labels mean the same thing, whether a late arrival belongs in the draw, or whether a participant may decline the result. Those are rule-setting questions. Treating them as edits after an outcome has appeared is the common way an apparently neutral draw becomes hard to defend.
One worked shift assignment
Four volunteers—Ana, Bo, Cy, and Dev—need one person for a Tuesday opening shift. With all four available, an equal single selection gives each name 1/4 = 25%. If Ana reports an existing Tuesday commitment before the list is used, the valid population becomes Bo, Cy, and Dev; each then has 1/3, or about 33.33%. Removing Ana and drawing from three names is not the same procedure as drawing from four and rerunning until Ana does not appear. The latter gives the appearance of a new event while concealing the eligibility decision that produced it.
Keep a small record for a shared decision: the roster, the rule for duplicates, the time it was closed, and whether a declined place triggers a replacement draw. Those facts let someone check why the probability was 1/4 or 1/3. The random result is only the final line of that record.
Selection, ordering, and grouping count different outcomes
A single winner from six names has six possible outcomes. A complete speaking order for those six names has 6! = 720 possible arrangements, because every position matters. Dividing the same six people into two labelled teams of three is another problem: choosing Team A determines Team B, so there are C(6, 3) = 20 labelled allocations before any restrictions are added. A tool that returns one name cannot verify a full order, and a shuffle does not guarantee balanced teams merely because everyone appears once.
Replacement is equally important. Three prize winners taken from 12 names without replacement give each valid person a 3/12 = 25% chance of appearing among the winners, since a selected name is removed. Three coin flips retain both faces after every flip, so heads-heads-heads remains a normal sequence with probability (1/2)^3 = 1/8. Expecting no repeats in a repeated-trial experiment is a category error; allowing repeats in a distinct-winner raffle is a rule failure.
Where a browser result stops being evidence
These tools generate an outcome locally in the browser, which avoids making a casual list into an account profile. Local generation does not prove who entered the list, certify the time of a draw, establish legal eligibility, or preserve an audit trail after a tab is closed. A dinner choice or classroom turn can reasonably rely on a visible roster and a shared screen. A prize allocation, queue priority, safety action, or access right needs its own documented procedure, an authoritative participant list, and a record of redraws or exclusions.
Chance can apply a stated rule consistently. It cannot supply consent, resolve a disputed identity, or make an unsuitable decision low-risk merely by producing a number or a name.