Luck and location explain the ten lowest earners. One machine is leaking.
Replacing the bottom ten would cost $220,000 and change little. Nine of them earned what their bets predict, and nine sit in Back Forty, the floor's quietest zone. The one real fault is M-1158, which pays out far too often. Every figure below follows the filters above.
Daily coin-in for the selection
The July plan at a glance
The promotion runs only if it passes four go/no-go checks on July 4. Otherwise the plan spends $1,200 and keeps $98,800.
Ranked fairly, the bottom ten are mostly unlucky or badly placed.
Raw house win punishes machines with little play, short time on the floor, or bad luck. The luck score compares each machine's actual win with what its bets predicted, in standard deviations. The hit-rate score checks how often it pays anything against the published game sheet. The location index compares its coin-in with the same game elsewhere on the same days.
The manager's bottom ten, re-examined
Lowest ten by raw house win within the current filters. Click a row to open that machine below.
Luck scores against pure chance
Expected vs. actual house win per machine
Points on the dashed line earned exactly what their bets predicted. Low points near the origin had little play, not bad luck.
Demand is steady and weekly. Where a machine sits matters more than which machine it is.
Saturdays bring in about 60% more than Tuesdays, and citywide events add more. Back Forty machines earn about 0.65× what the same game earns elsewhere. When its main walkway closed for renovation, that zone fell further while the rest of the floor picked up the play.
Floor map: earnings vs. the same game elsewhere
Zone earnings, adjusted for game
1.00 = the same game's floor average on the same days.
Natural experiment: the Back Forty walkway closure
Average coin-in by weekday
Events in the period
Citywide events come from public calendars. Property promotions are the casino's own.
Real-world context
The challenge ships real public tables for context: Las Vegas visitation from the LVCVA and statewide slot results from the Missouri Gaming Commission. Their publishers have not cleared redistribution, so this public dashboard does not republish their figures. Neither was an input to the forecast: the LVCVA series was published after 2025, and both describe other populations, not this casino's guests.
Repairs take about three hours. What varies is how often machines break and how long they wait.
Ridgeline machines fault several times as often per machine as Halcyon's. Waits are a queue: with two technicians on shift a new fault is picked up at once unless two others are open. Short-staffed shifts in May doubled the play lost while waiting.
Wagers lost to fault downtime, by month
Valued at each machine's own average coin-in for that hour of day.
Faults per machine per month, by manufacturer
Follows the date and zone filters.
Wait for a technician, by staffing and queue
All six months, floor-wide.
Technician staffing, planned vs. worked
Average headcount per shift. Shifts are named in the roster's own terms.
Card holders lost about what the games predict. Comps got much cheaper in points on May 1.
These figures cover loyalty-card play only, which is about two-thirds of all coin-in. The top tenth of card holders account for about two-thirds of expected loss. On May 1 the points price of every comp fell about 53% for every tier, with no measurable gain in play.
Wagered, paid back, and net for the selection
Net = payout − wagers. Expected loss (theo) is what the games' payout rates predict for those bets.
How concentrated are results?
Points needed per $1 of comps, by tier
All guests, full period.
Food and drink cash, carded vs. uncarded
Uncarded guests cannot use comps, yet their cash fell more after May. The drop is a summer dining slowdown, not comps replacing cash.
A weekday-and-calendar forecast beats both baselines on a time-based holdout.
Trained on January–April and scored on May–June. Event and holiday adjustments use only events announced by April 30, so nothing from the holdout leaks in. The frozen forecast is also the yardstick for measuring the July promotion.
Holdout: actual daily coin-in vs. forecasts frozen on April 30
Error on the holdout
MAE = average absolute daily miss in dollars. Bias = average of actual − forecast; negative means the forecast ran high.
How the model changes a decision
- It is the counterfactual for the promotion: actual coin-in minus the frozen forecast gives the lift.
- Using only data before the event, the February tournament produced $566K–$884K of net extra wagers, against a $158K break-even for a $15,000 package.
- On 20 ordinary weekends the same method finds deviations from −$300K to +$212K, so a July lift above $212K stands out from all of them.
Placebo test: forecast deviation on ordinary weekends vs. the tournament
Included: every Friday–Sunday from January 1 to June 30 with no citywide event, casino promotion or federal holiday. The tournament ranks first of 21, a rank p-value of about 0.05; with 20 comparison weekends that is the strongest claim the test supports.
Inspect the one leak. Run one tournament if it passes its checks. Keep the rest.
$16,200 committed, $83,800 kept. If any promotion check fails on July 4, spend $1,200 and keep $98,800. Each action links to the evidence behind it.
Inspect and reconcile M-1158
- Hits 38.57% of 674,454 spins vs 31.54% on the other 47 Desert Gold machines, every month.
- Buys information: paytable, software, meters and jackpot accounting. Any repair is separate and unscoped.
- If a correction follows, about $200 a day of expected house win returns, so it pays back in about six days.
- Success: a documented reconciliation, then a three-day hit rate inside 30.6%–32.4%.
One tournament-style promotion weekend
- February's tournament: $566K–$884K net extra wagers against a pre-event baseline. Break-even: $158K ($211K with $5K of extra costs).
- Measured against the frozen forecast for July 18–27, including the following week for displaced play.
- Counts as clearly above noise if the weekend alone exceeds $212K.
- A second weekend would have to be ordered before the first runs, so it is not funded.
Go/no-go rules, checked by July 4
- February lift under the stricter baseline is at least twice break-even ($317K). Currently $566K: passes.
- The package covers prizes, marketing and staffing; costs outside it are $5,000 or less.
- No citywide event or holiday on July 18–20, so the forecast measures the weekend cleanly.
- Open to all guests, with no extra offers aimed at the heaviest-losing players.
If any rule fails: inspection only, keep $98,800.
Commitment dates
Promotion downside scenarios (July net, expected house win)
Expected house win = extra wagers × 9.5%, minus the $15,000 package. Not realized win, which also depends on luck, and not full contribution after comp and food costs.
Plans compared in 10,000 simulated Julys
Conditional on assumed input ranges; not observed frequencies. The two-weekend plan scores higher only because we assumed the second weekend earns 50–100% of the first.
Not funded
- Replace the bottom ten: nine are within normal luck and nine sit in Back Forty. Buying ten costs $220,000.
- Move machines or add Back Forty signs: the walkway closure shows that removing access shifts play to other zones. It does not show what signs would add.
- Extra technician: waits cost about $4,000 a month in expected house win against an $11,520 package.
- Ridgeline maintenance: slightly negative within July (−$2.7K simulated); evidence is one machine. Better as an August test.
- Floor-wide comp change: the May 1 cut raised comp cost about $1,400 a month with no measurable play gain, but the evidence is before-and-after only.