← Back to discussions

resolved

Benchmark proposal: 18-player mixed-size tee-time rotation

John Gojnich · Guest ·

Hello GroupMixer team, Would you be interested in an additional constrained benchmark case based on a four-round golf rotation with a proven fairness optimum? The case has 18 players and five ordered groups per round, with exact sizes (3,3,4,4,4). The same size pattern repeats in every round. Groups correspond to tee slots at 10:00, 10:07, 10:14, 10:21 and 10:28. Every player attends every round, no unordered pair meets more than once, and four-player-group participation is balanced: six players have two four-player rounds and twelve have three, without prescribing identities. The objective lexicographically maximizes (1) the minimum number of distinct tee slots received by any player, then (2) total distinct slots across players. The proven optimum is (minimum 3, total 65), equivalently score 284 for 73 × minimum + total. We provide a validated reference schedule and an exhaustive symmetry-based exclusion of every improvement, independently checked in two OR-Tools formulations. Five additional fresh, no-hint local runs, seeds 101/202/303/404/505, each found (3,65) within a 60-second end-to-end deadline. Those attempts reported FEASIBLE, not solver-proven OPTIMAL; optimality comes from the separate exhaustive argument. Local ARM64/macOS timings are not a direct comparison with your benchmark: CPUs were not pinned and an 8 GiB process limit was not imposed. Public evidence: - [Case definition](https://github.com/jgojnich-a11y/enhanced-social-golfer-problem/blob/main/N18_BENCHMARK_CASE.md) - [Reference schedule](https://github.com/jgojnich-a11y/enhanced-social-golfer-problem/blob/main/solutions/optimal/n18_sol.csv) - [Optimality proof and audit](https://github.com/jgojnich-a11y/enhanced-social-golfer-problem/tree/main/runs/n18_canonical_two_rounds_20261003) - [Fresh no-hint benchmark evidence](https://github.com/jgojnich-a11y/enhanced-social-golfer-problem/tree/main/runs/n18_cold_benchmark_20261003) The reference schedule and optimum are evaluation evidence, not embedded answers or hints for a benchmarked solver. Does this fit your benchmark scope, and can your scenario format express the exact lexicographic tee-slot objective and group-size participation balance? If so, what format would you prefer for contributing the case? John Gojnich

Votes 1

Replies 5

  1. GroupMixer ·

    Hi John, Thanks for sharing this! I explored the case in GroupMixer, and it highlights an interesting gap in what we currently support. We can model the 18 players, four rounds, mixed group sizes, pair-repeat restrictions, and the requirement that each player plays in a foursome two or three times. We can also penalize repeated visits to the same tee slot, which, with everyone attending all four rounds, maximizes the total number of distinct slots experienced. However, we don’t currently have a constraint for “each person must visit at least N different groups,” or an objective that maximizes the minimum individual variety before maximizing the total. So we can’t express your exact benchmark objective. In exploratory runs of the closely related total-variety model, I found this schedule: Minimum 2, total 66 in a 5-second run: http://groupmixer.app/share#gm=br3.G5YJACwK7HZgO8VeTJSxlnsZFA1GSDI7eHPfX3EDjYa9SxhQmBUF-Am-Rxj8v85fEAYjjr4j79u1MxrFglJLTV0o4gdYITq5iT-Nlnc823mVjLM1S19Xx9Rh6fv_93OLGx6KtUq0GHZt8uwjbnF2EUv89uMWM0tEMiV44lAbWT2OG8J8pyunyCNflNb5gr9o-QHGjXhgiALc6G0Pt8PG1_Pquqw_vuoHOjxYQglv4FMMuG9edpkhlX_7fmeXV3epBoa3zfcfG-6HW778ELQzSdeg3RyCC8QuZzvxKPCBKnOIvpwHrMFRoRuJ0NYY4Mz-Plop8AHjaWjBbihVnl59IU5FiiGAkWBenE1CXAXHllv6eE-LtSbdy1ovsCbp38qH3OvRYOXPnNoIu7vHfkJZ7DBLO_R0uHVoaepwloasK2KN9rHlzQ55rcLoJAqQbTma-yP8rEunCm-8p3Gfr0OrX0egIU2YbE6o9j37fxvskyKbXNdefcYVuDInKQpXaU_apveOcTicErtKWVNuB0_FO_PvcompL04BRY7ggmoSz_OAuYu8pdqHDqfVIt58n05H1xHuqBnWrt5c7AxxRbrIELNKqd-mmWioxNm3RcgzQFPjJoT5R8TeQGRbBpDoOdzGLFkLkVq1lOiySdum02xjgwibOm9CuniGs3UMtWIItEnR19j_zDTb4awKoYL_UKp2-l4zP_DUkL02lmoIjA-L9KgvVnLhDxhN-W6VOmPpLHgX40xjKJJe1cRc7wriqhzGhTY96eVEMgBcpGaTJUa90Rt47IjZWIyEHU_0rzH4xI4sm-WQJBwyh5NbSMv-IyzWtNxYe4wc8GzEIC_BgK7ImQbFWimArpzoB3L2KdZJd3ISEp8dcSHHT2SoMuakllwBg7YSY-zNNowsHUny1xJzDGQ8i5xGt0I6oEjV9XO7hCfnkD2qx4RizL1qohoMcRqsIcIPZlg_afZF6Z6aFaK2xJ1KDCGI2ZDvNwFnRNtZNtY5O0cExRgEQqGcBrUZe_tUuA85JHMlkMUNT5VLhIovQksi48R_U8So3D0bpaciZzWyXRlyWzcJkMkFBNOq-etJ5tz_aprP9rWul6m1ggaJNu_PzqZj5KzR_zl8K2f3uqAfXAc It satisfies the group-size, pair-repeat, and foursome-frequency requirements. With the constraints in our model, this result is preferred over your perfect schedule because the total number of distinct tee slots is higher. According to your constraints, it is a worse schedule because one person only plays at two different tee slots. Would support for this additional constraint type be interesting for you in practice, or is this mainly out of curiosity? Generally, OR-Tools will work fine for smaller instances like the one you presented, but for large instances, GroupMixers solver approach could produce much better results. If we ever end up adding support for it, your example would indeed be a valuable benchmark.
  2. Guest · Guest ·

    Hi Guido, Thanks for exploring this. Yes, it’s useful in practice: even relatively small golf groups benefit from fair rotation through the playing order, alongside avoiding repeat partners and balancing group sizes. It’s also interesting to establish the smallest player count at which a desired minimum variety becomes feasible. OR-Tools may handle these smaller instances well, but having that requirement available directly in GroupMixer would still be valuable.
  3. GroupMixer ·

    Hi John, If you really want this feature, please tell me in your own, human words how it would be valuable for you in practice.
  4. Guest · Guest ·

    Touche. When we play in groups from 15-24 over 4 days, I was just trying to mix the players up, which seems to be relatively easy and then add playing order as another optimisation. But I suppose I've generated those numbers now and maybe others would find your generic solution helpful ;)
  5. GroupMixer · · Official

    Alright, you convinced me. I added the new capability as a new counting mode in group visits constraints. You can now select "Different selected groups visited" in the group counting field and model your scenario exactly. Groupmixer usually finds the optimal schedule very quickly; however, this also depends on the user's hardware. On very weak hardware, it might produce only near-optimal schedules, which are probably indistinguishable in practice. Here is your scenario modelled in group mixer with the perfect result schedule produced: http://groupmixer.app/share#gm=br3.G_MDACwPbLrzJVYTXQrvRLy_WBbPI6kIDHp60pYGzSPMCsokzQLqeJKLvgNOgEhBr85XRfTw62THzq8EDy0qGMFRCR-hg4hOI59wfYKAbmrZu276FR3eLF0dfVOHpc-IbVbEMokWn0hsMpWIpS2J-PWtkSJxJnWep1LehNeJ6YSfLOM4PjYX2B4iogkPYPT30-208fZ6lXnVfEW3D5ixeR1S6AQfQrFYS3QcJRQISKjQoMOAuerWomPBwMSKDfujg7BhLVq7zBCKWy_ruvV2KktlcWb0hw9rk9dMkFvjkuCejniGLA98q3OiixWMlOBi6TQqU7BxAQ-PsYgskAXxLD5hcqUZbHEiBEZReOgFbhQREGIbQM7X7fJb4xKrC2uzjQgJN6t7z5ehAvOyA3GzMf9ElUAzzz0pWGPzcHr8hC5Cf2c2YTXAxKCF8WnQInL4bZkhHocSQ9Jq8zW33LLMGM2zz9mMqa_ThfNTlPwfE1gnhHgd42NUbMh1AVUEWKdK-hl3eXiT9nDYD15VYsyAme-EKH4ETv2SjJJ1A_QB-wLrT25AoGwC1Oog-JOzEb6IlbCOMG7cwfprXnv_yv04YEe9M9X1-NEV_7LI9S8K