G.7Sensitivities
Which guesses matter? I changed 30 inputs one at a time, each to its own low and high value, and left everything else at the central case. The cost of Years 0–3 turns most on heritage-worthy planning (the stalls that food guides and heritage records already name) counts and on a veto at Tekka; the endowment on outlets with the till dollar and opt-in at the till. Fundraising moves the endowment far more than any cost line moves the budget.
Two outputs are tracked: the four-year total, Years 0–3 at three centres ($19,683,808), and the corpus in Year 20 as a share of the $40M target in today's dollars (121%). Where the Monte Carlo (the model run thousands of times, its uncertain numbers drawn at random) already varies a parameter, its range is the one the code draws from, printed as the code writes it. Where the model has no range, 19 of the 30, the parameter moves 20% either way, and the table says so: that is the only rule this volume adds, and it is not a claim that 20% is the true uncertainty. The table keeps the six parameters that move the four-year total most and the six of the rest that move the corpus most; the download has all 30. All results are Derived. g7-sensitivity.csv
| Parameter | Central | Range | Total, low | Total, high | Corpus, low | Corpus, high |
|---|---|---|---|---|---|---|
| Outlets with the till dollar | 0%, 2%, 4%, 5% | Monte Carlo: [0.0, rng.uniform(0.01, 0.03), rng.uniform(0.02, 0.06), rng.triangular(0.02, 0.05, 0.1)] | – | – | 85% | 196% |
| Opt-in at the till | 8% | Monte Carlo: rng.triangular(0.03, 0.08, 0.15) | – | – | 85% | 187% |
| Diners who give | 2.0%, then 0.8% | Monte Carlo: rng.triangular(0.005, 0.02, 0.04); rng.triangular(0.002, 0.008, 0.015) | – | – | 102% | 144% |
| Return on the fixed-income core | 4.1% | ±20%: the model has no range | – | – | 102% | 143% |
| Other companies | $0.0M, $2.0M, $2.0M, $0.5M | Monte Carlo: rng.triangular(0.5, 2.0, 3.5) | – | – | 101% | 142% |
| A veto at Tekka | no | Monte Carlo: int(rng.random() >= 0.8) | −$3.23M | – | 158% | 121% |
| Heritage-worthy planning counts | 27, 8, 9 | ±20%: the model has no range | −$2.24M | +$2.29M | 134% | 121% |
| Heritage Plate scale | 1 | Monte Carlo: rng.triangular(0.25, 1.0, 2.0) | – | – | 110% | 137% |
| National core lines | $3,583,300 gross | ±20%: the model has no range | −$0.79M | +$0.79M | 135% | 108% |
| Apprentices a Fellowship | 2 | ±20%: the model has no range | −$1.12M | +$1.12M | 130% | 116% |
| Apprentice wage (the PWM path) | $50,684, $52,650, $52,650 | ±20%: the model has no range | −$1.01M | +$1.01M | 127% | 116% |
| Master Fellowship cost | $90,750 | ±20%: the model has no range | −$0.88M | +$0.88M | 126% | 116% |
Granted, one at a time is a narrow test: it misses what two guesses do together. The Monte Carlos of D.4, E.1 and B.9 move everything at once, and their results are in G.3 and G.4. Pledge rates, the draw (the share of the endowment spent each year) and the failure rate change only the endowment; the ledger's rates change both, because a dearer programme leaves less to join the corpus.