Begin with the probability object
These tools answer five different questions that can all look like “pick something at random.” A team split partitions named people into blocks. A weighted choice assigns unequal interval lengths to labels. A date-and-time draw samples minute positions on one calendar interval. Dice odds add the probabilities of success counts across a finite pool. A streak experiment looks for adjacent successes over many generated sequences. Selecting the wrong object changes the rule, even if each button produces an unpredictable result.
For a workshop with ten attendees and tables of three, the relevant fact is the quotient and remainder: 10 = 3 × 3 + 1. A size-based split therefore produces three complete tables and one person left over. Calling that lone person a fourth “team” does not solve a seating constraint. Decide before the shuffle whether the facilitator joins, pairs are allowed, or the roster should be reduced. The draw allocates membership after that capacity policy exists.
Weights describe a rule before anybody wins
When Tea, Coffee, and Water receive weights 3, 5, and 2, their chances are 3/10, 5/10, and 2/10. Those are 30%, 50%, and 20%, not three votes for one label and five votes for another unless the entries really represent votes. Multiplying all weights by ten keeps the same distribution because 30/100 equals 3/10. A group should be able to say why Coffee receives the longer interval before its name appears; random selection cannot turn an unexplained advantage into an equal contest.
An equal picker is the better procedure when every eligible person gets one ticket. A weighted procedure is useful only when unequal tickets are intentional, visible, and tied to a stated source such as inventory units or published raffle entries. It cannot verify eligibility, resolve a dispute about the source list, or serve as a regulated prize-draw record.
Calendar ranges are made of eligible minutes
A range from 09:00 on September 1 to 17:00 on September 30 contains 42,241 eligible minutes when both endpoints count. Middle days contribute 1,440 minutes each; the opening and closing dates contribute fewer. That is correct when the policy is one continuous availability window, but it is not the same policy as giving every calendar date equal odds. Excluding nights, weekends, holidays, or booked appointments creates a set of separate slots, which this single-interval generator does not construct. Cross-border decisions also need a declared time zone, especially around daylight-saving changes.
Counts and runs require different mathematics
For eight d6 with success on 5 or 6, one die succeeds with probability 2/6 = 1/3 and the expected success count is 8/3. Asking for at least three successes is a binomial event: every possible count from three through eight contributes to its exact 53.18% probability. That value is fixed by the inputs. It is not an account of one table’s dice faces, and rerunning the calculator should not change it.
A five-success streak in one hundred 50% trials has another structure. Windows overlap: six consecutive successes contain two five-long windows. The streak tool therefore simulates complete sequences and estimates the share containing a qualifying run. With 10,000 experiments, two runs can differ by a few tenths of a percentage point from ordinary sampling noise. Independence and a constant success rate are assumptions, not conclusions about a tired player, a failing machine, or correlated weather observations.
Choose a procedure that can be explained afterward
For an informal game or classroom activity, write down the roster, weights, time endpoints, dice rule, or event model before pressing Run tool. Preserve the displayed result if other people need to inspect it. These browser calculations do not decide who is eligible, balance ability or access needs, schedule around real calendars, model special dice rules, or diagnose a real-world streak. They make a narrowly stated chance rule visible; responsibility for the surrounding rule remains with the people using it.
Five distinct outputs, five different records
The appropriate record follows the mathematical object. A team allocation needs the exact eligible roster, the requested block size, the remainder policy, and the shown groups. A weighted selection needs labels and weights because the winner alone cannot reveal whether the odds were 1/2 or 1/20. A time selection needs both endpoints and the time zone. A dice calculation needs dice count, sides, threshold, and required successes. A streak estimate needs the event count, the assumed rate, target run length, and number of experiments. Saving only the final name, timestamp, percentage, or group label loses the information required to reproduce or question the rule.
Those records are not paperwork for its own sake. They separate an input correction made before a draw from a redraw made after an unwelcome output. They also reveal when two people are arguing about different models: one may want every date equally likely while another wants every available minute equally likely; one may want an expected number of dice successes while another needs the chance of meeting a threshold. A browser result is easiest to defend when the population, event, and stopping rule were settled first.
Randomness does not repair a vague policy
Chance can resolve a tie only after the tie is defined. It cannot decide whether a late entrant counts, whether three raffle tickets are justified, whether a closed hour should be eligible, whether a game die rerolls, or whether observations really have a constant success rate. Adding random selection before those choices are made merely hides the policy inside the setup. For casual use, make the rule visible on the shared screen and accept the first valid result. For a decision that affects money, access, safety, employment, or legal rights, use the governing procedure and an appropriate auditable system rather than treating a local convenience tool as an authority.