event-seating-orchestra

Use when planning seating for a wedding, dinner party, holiday table, or rehearsal dinner and you must respect keep-apart pairs, balance talkers across tables, seat couples near-but-not-alone, and keep exes/estranged family apart — and need an actual assignment chart, not vibes.

Overview

The seating chart is the highest-stress artifact of wedding and event planning. The constraints live only in hosts' heads: divorced parents who can't share a table, the two friends who broke up, the uncle who talks over everyone, cousins who only know each other, grandma who needs an aisle-adjacent seat. Professional planners charge real money for this; spreadsheet-savvy relatives drown in it the night before.

This skill encodes the real constraints and runs an actual optimizer (simulated annealing over table assignments) to produce a printable chart, a per-table roster with reasons, and a complaint log for changes ("swap these two and the score improves by X").

When to Use

• Weddings (the big one: 60–250 guests, family politics, plus-ones).
• Holiday dinners with known friction pairs.
• Rehearsal dinners, retirement parties, milestone birthdays.
• Dinner parties 8–16 where mixing matters ("don't let the two quiet ones end up together").
• Conference-style lunches where you want cross-team mixing.

Don't use for: standing receptions (no seating), classroom training (fixed rows), gala sponsor tables (priority ordering, not social optimization).

How It Works

1. Build the guest graph. Each guest: name, party (couple/family unit that must sit together or near), flags (talker, quiet, kid, needs-aisle), and optional language. Edges: must-together (couples), keep-apart (exes, feuding), prefer-near (friends), know-well (already close — mixing prefers splitting them).
2. Score function (what "good" means):
   • +100 per must-together at same table (hard constraint otherwise)
   • −150 per keep-apart pair at the same table (hard veto)
   • talker balance: per-table talker penalty = (talkers − ideal)² × 15
   • party integrity: couples at same table strongly preferred; families with kids at kid-friendly tables
   • mixing bonus: tables whose members have few prior know-well edges score better
   • kid tables: kids seated together, not scattered among adults
3. Optimize. Simulated annealing: start random (respecting must-together blocks), propose single-guest or block swaps, accept improvements, occasionally accept worse to escape local optima, cool over ~20k iterations. Deterministic seed for reproducibility.
4. Output. Per-table rosters with headcount, talker load, kid count, constraint check, and per-table "why" notes. Swap suggestions for manual tweaks with score deltas.
5. Iterate. Late RSVP changes? Re-run with the same seed and the new guest list; diff the charts.

Constraint Encoding Examples

Real situation	Encoding
Divorced parents	keep-apart edge between them (and optionally their new spouses)
Maid of honor + her new boyfriend	same party (must-together)
Two work friends who should meet	prefer-near
The loud cousin	flag talker — optimizer spreads talkers one-per-table max
Grandma with a walker	flag needs-aisle → host marks which seats are aisle seats
Kids' table	flag kid → grouped at one table (with 1–2 patient adults)
"Don't split our family of 5"	party of 5, must-together

Worked Examples

python3 scripts/seating_optimizer.py --demo — 24-guest wedding sample (2 divorced-parent keeps-apart, 4 couples, 3 talkers, 4 kids) across 3 tables of 8 → rosters + balance report in under a second.

python3 scripts/seating_optimizer.py --guests guests.json --tables "8,8,8" — your guest file.

Common Pitfalls

1. Optimizing before the guest list is frozen. Every late RSVP invalidates the chart; run final versions, not drafts (or accept re-runs).
2. Forgetting plus-ones. A party of 2 must be seated as a block; modeling guests as individuals breaks couples randomly.
3. Treating all constraints as equal. keep-apart is a veto; prefer-near is a wish. The tool uses hard/soft classes — respect them when entering data.
4. One kids' table with zero adults. Young kids need 1–2 assigned adults; the optimizer flags kid tables without adults.
5. All the talkers at one table (usually the "fun" table the host builds by hand). Every other table goes silent; spread the energy.
6. Chart delivered as a screenshot. Output rosters as text/table; venue coordinators and caterers need names per table, not graphics.
7. Hiding the chart from the couple's parents when they're hosts. A 10-minute review of table rosters surfaces forgotten frictions the graph never knew about.
8. No aisle-seat map. needs-aisle guests must map to real physical seats; get the venue's table map before finalizing.

Verification Checklist

• Every must-together party seated at one table (script asserts)
• Zero keep-apart pairs sharing a table (script asserts)
• Talker load balanced (no table with 2+ talkers unless unavoidable)
• Kids grouped, each kid table has ≥1 adult
• Headcounts match venue table capacities exactly
• Aisle/mobility guests matched to the venue's actual aisle seats
• Chart printed as text rosters + one review pass with a second host