Inspiration
What it does
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
What's next for Gilded Ace Floor Review
What it does
An interactive dashboard and analysis for The Gilded Ace, a fictional 300-machine casino (TAMU Datathon 2026, Dashboard Royale). It asks whether the ten machines that earned the least in January to June 2025 are weak, unlucky, badly placed or leaking, and where a $100,000 budget should go for the 30 days from July 1.
Short answer: do not replace the bottom ten. Spend $16,200 to inspect one machine (M-1158) and run one slot-tournament weekend (July 18 to 20). Keep $83,800.
The dashboard is one self-contained HTML file with the data inlined (docs/index.html in the attached ZIP); open it in any browser. The Claims & Evidence appendix is docs/claims.html and also the Evidence view inside the dashboard. The AI-use disclosure is submission/ai_disclosure.md.
How we built it
DuckDB SQL (pipeline/sql/) and Python analyses (pipeline/analyses.py) build every number from the challenge files; the build stops unless the derived tables reconcile with the published totals, and scripts/check_numbers.py rebuilds every figure the write-up quotes. Models: a bank x game model, a ridge forecast with rolling-origin validation against a weekday baseline, natural-experiment estimates with block bootstrap, and luck and hit-rate z-scores.
Write-up
Machines (Q1)
Nine of the ten lowest earners sit in Back Forty, where every game earns 30% less per hour than average with the walkway open; eight run low-edge games (theo hold 8% or less); two left the floor; one arrived in May. All ten covered their costs. Luck is ordinary (z-score spread 1.07, within chance); two need checking: M-1158 pays on 38.6% of spins against a configured 31.5% (z +124), and Comet Cash holds 7.69%, not 8.00%, over 24 machines (z -3.46).
Demand to comps (Q2-Q5)
Daily coin-in fell 6% from February to June, 2.9% on event-free days, like city visitors. After eight Center Pit machines arrived, the floor rose 2.6% for four weeks, city visitors 1.8%; over eight, 0.6% (-1.7% to 2.5%): 13% of their play looked new. The Back Forty walkway closure cut that zone 31.5% against the rest, but not measurably the floor. Downtime cost at most $98,886 of theo; grave-shift repairs waited twice as long. Tracked players lost $15.59M of $163.8M wagered, close to theo; the 1% who lost most account for 17.8% of losses. Comps cost $190,669; on May 1 points needed per comp dollar halved, adding about $75 a day.
Models
The bank x game model, fitted on January to April, predicted May to June as well as each machine's history (MAE $1.07 against $1.06). A ridge forecast of coin-in prices the tournament; frozen on May 1, its pre-selected trend variant lost to the weekday baseline (MAE $71,152 against $68,699). Without the trend it scored $46,303 and, picked on May alone, still won June ($55,901 against $76,281).
July plan (Q6): spend $16,200, keep $83,800
Inspect M-1158 ($1,200): $2,900 expected, $5,800 if a setting is wrong. Success: hit rate back to spec within a day; else ask maker Ridgeline.
Run one slot-tournament weekend, July 18 to 20 ($15,000): $46,600 expected ($14,600 to $78,600); break-even lift 3.3%; past promotions lifted weekends 3.2% and 17.5%. Repeat it only if the weekend beat its neighbours by 6.3% and, with July 25 to 27, beat twice July 11 to 13 by $158,600 of coin-in.
Not funded: replacing the ten ($33,800); two in-place swaps gained nothing measurable. Moving three Back Forty machines to busier slots ($5,400) needs 26% of their play to be new. Free: query Comet Cash's paytable, the comp change and Ridgeline's fault rate.
Built With
- duckdb
- html
- javascript
- python
- sql
