Saved draws · Seed portability

Same seed, different teams: what to save before switching tools

Published: October 28, 2026

Save the accepted assignment, not just its seed. In our browser comparison, the same eight labels, two groups and seed 40 produced different teammates in two generators. Keep the original tool, ordered roster, settings and exported result together when you need to recreate a draw.

We ran this check on October 28, 2026, using invented labels P01 through P08 in that order. Manage My Batch used Number of groups set to two, seed 40, and empty keep-together and keep-apart fields.

GroupMixer used one session, two groups of capacity four, no constraints and its Auto configuration with deterministic seed 40. Its two default objectives remained enabled. These are different implementations, not a test of an agreed shared algorithm.

What actually changed between the tools?

Manage My Batch's downloaded CSV placed P02, P08, P04 and P03 in Group 1. Group 2 contained P05, P06, P07 and P01.

GroupMixer's downloaded schedule placed P01, P02, P05 and P08 in Group 1, with P03, P04, P06 and P07 in Group 2. Both files contained the eight labels once, in groups of four.

P01 and P02 shared a group in GroupMixer but were separated in Manage My Batch. Swapping the labels Group 1 and Group 2 could not repair that mismatch.

Comparing the actual files yielded twelve unordered teammate pairs in each output, four shared pairs and eight pairs present only in each output. The replay check prints those sets. The difference survived removing group names and sorting members for comparison.

Did each tool reproduce its own result?

In this session, yes. Clicking Generate Groups again in Manage My Batch retained the same exported assignment text.

A second GroupMixer run with unchanged settings completed with effective seed 40 and the same labeled memberships.

Its raw assignment-row order differed; the labeled memberships did not. Both GroupMixer run records reported stopping at the optimal score. The original records are retained; no predicted shuffle replaced them.

What should travel with a saved draw?

Send the accepted exported assignment when recipients need those exact teams. For a reproduction attempt, also retain the tool address, ordered input, group settings, constraints, seed and any available version information. Check memberships after regenerating. NumPy's compatibility policy illustrates how much more specific reproducibility conditions can be than the seed alone; it does not describe either tested website's internals.

These observations cover two tools and one synthetic roster, not every seed or later website version. We did not establish which internal algorithm caused the cross-tool difference. Our own Random Team Split form has no user-entered seed control; save its CSV or TXT result instead of expecting these seed values to work here.