Nature’s Admonition

A decision can have a positive average payoff and still make the decision-maker poorer through repeated compounding. The consequence is a systematic preference for attractive arithmetic returns that damage the capital base from which every future return must be earned.

Enterprise problem: investors and operators need an accounting method that values deep loss according to its effect on what remains, not as one interchangeable line item in an average.

Whole-course context: Day 01 defined cost-effective protection as a falsifiable portfolio claim; today supplies the geometric wealth accounting needed to measure that claim.

Today’s slice: Chapter 2 moves from the Saint Petersburg paradox to Bernoulli's logarithmic map, the geometric average, and the merchant case in which costly insurance improves the whole.

End-of-day evidence: you will calculate arithmetic and geometric values for a risky shipment, identify the premium range that improves compounding, and reproduce the result in an interactive path experiment.

Still unsolved: the chapter values a known repeated payoff; it does not yet resolve the difference between an ensemble of parallel outcomes and the single path a real decision-maker experiences.

Arithmetic language hides the increasing harm of deep loss, so first-use terms must distinguish a payoff's average from its effect through time. The definitions below are the minimum vocabulary for reading the chapter's equations.

Term	Plain meaning
Expected value	Probability-weighted arithmetic average of possible outcomes
Geometric average	The nth root of the product of n positive wealth outcomes or total-return multipliers
Total-return multiplier	Ending wealth divided by starting wealth; a -20% return is 0.80
Log return	Logarithm of the total-return multiplier, which turns multiplication through time into addition
Concavity	A curve that bends downward, so equal-sized changes have unequal effects at different starting levels
Volatility drag	The gap by which variability lowers compound growth relative to arithmetic return
Recovery burden	Gain required after a loss to return to the starting level
Actuarial value	Expected claim payment minus premium on an arithmetic basis

The logarithm here is not a personality test. It is a map of multiplicative wealth: log(a × b) = log(a) + log(b). Averaging log wealth and then exponentiating gives the geometric average wealth outcome.

The Saint Petersburg paradox asks why a game with an enormous or unbounded arithmetic expectation can be worth surprisingly little to the person who must risk finite wealth. The answer is not that averages are useless; it is that the unit being averaged must match the process being lived.

Daniel Bernoulli proposed measuring each possible ending fortune relative to the fortune already held. In modern notation, the chapter's central map is:

geometric value = exp(Σ p_i × log(W_i))

Here W_i is a positive possible ending wealth and p_i is its probability. With equally likely cases, this is simply the geometric average of the ending wealth values. The map penalizes approaches toward zero increasingly because log(W) falls without bound as positive wealth approaches zero. A total loss cannot be repaired by later multiplication: once the capital multiplier is zero, every subsequent product remains zero.

This produces a normative rule—how a compounding decision should be evaluated—not a claim that people psychologically feel logarithmic utility. A decision-maker may dislike or enjoy risk for many reasons. The physical accounting remains: future gains apply to the capital that survived.

An infinite payoff can distract from the ordinary mechanism, so the chapter uses a finite, positively skewed teaching wager. The consequence is clear even without any infinity.

Take six equally likely gross winnings: $1, $2, $6, $22, $200, and $1,000,000. Their arithmetic average is $166,705.17. A person with $100,000 who pays $50,000 to play has possible ending fortunes of $50,001, $50,002, $50,006, $50,022, $50,200, and $1,050,000.

The arithmetic expected ending fortune is large because the rare million-dollar outcome dominates it. The geometric value is approximately:

($50,001 × $50,002 × $50,006 × $50,022 × $50,200 × $1,050,000)^(1/6) = $83,114

That is below the certain $100,000 starting fortune. Paying half of wealth for the game therefore fails the geometric criterion despite its spectacular arithmetic expectation. Reducing the stake changes every ending fortune and can make the wager acceptable; the value is personal to the scale of wealth at risk, not because money has a mysterious utility, but because the same dollar loss removes a different fraction of the future compounding base.

The first source diagram can seduce a reader into seeing six equally prominent dice while overlooking that one payoff supplies almost the entire arithmetic average. The consequence is a decision driven by visual frequency rather than payoff magnitude, so this lab preserves both the equal probabilities and the logarithmic separation among dollar outcomes.

Printed page 33, PDF page 51, labels the six fair-die payoffs as $1, $2, $6, $22, $200, and $1,000,000. Reset reproduces those values exactly. The horizontal axis is die face, and the vertical axis is payoff in dollars on a logarithmic scale; logarithmic means each equal vertical interval represents a multiplication rather than an equal dollar addition. Hatched square bars are the book benchmark. Outlined diamond bars are an illustrative scenario, so a learner can challenge the tail without overwriting the source.

The Face-six jackpot scale changes only the million-dollar outcome. The Face-five payoff scale changes only the $200 outcome. Both leave every face at probability 1/6. The controls answer a first-principles question: how sensitive is a reported average to the two tail values, before any claim is made about preferences or markets?

Use a falsification protocol rather than admiring the skew:

1. Reset and verify all six published payouts and the 1/6 unchanged probability readout.
2. Set the face-six scale to one quarter while leaving face five unchanged; record how much the arithmetic payoff falls.
3. Restore face six and vary face five. Compare its influence with the jackpot’s influence.
4. State the claim “the average is representative of a likely one-roll experience.” Reject it if five of six outcomes remain far below the average.
5. Repeat the interpretation without the logarithmic axis. If the first five bars become visually indistinguishable, record that linear presentation conceals structure even though it does not change the data.

The model assumes a fair independent die and guaranteed cash settlement. It does not estimate a real opportunity, make the million-dollar state achievable, or prove that rare upside is undesirable. It isolates the difference between probability mass and payoff magnitude.

Domain	Translation	Evidence to demand
Economics	A mean income or output figure may be dominated by a small upper tail	Median, quantiles, population weights, and who can actually receive the tail outcome
Startup	One imagined acquisition can dominate expected valuation	Base-rate distribution, dilution, runway in ordinary outcomes, and decision rights at exit
Business	One huge contract can dominate pipeline-weighted revenue	Concentration, close probability, delivery capacity, and cash collected rather than booked
Daily life	One dream outcome can dominate a career or education calculation	Acceptable ordinary paths, reversible steps, downside capacity, and nonfinancial value

A payoff table alone omits the finite capital used to buy the chance, so the same gamble can look harmless at one scale and destructive at another. The consequence is evaluating prizes without the fortune that remains after the stake is paid.

Printed page 34, PDF page 52, begins with $100,000, subtracts a $50,000 wager, and adds each published payoff. Reset therefore produces ending fortunes of $50,001, $50,002, $50,006, $50,022, $50,200, and $1,050,000. The horizontal axis remains die face; the vertical axis is ending wealth on a logarithmic scale. The compact state table below the plot preserves the source’s row-by-row accounting.

Starting wealth changes the capital base. Fraction wagered changes how much of that base is removed before the payoff arrives. The jackpot is held fixed, which exposes a scale effect: a nominal prize does not have the same geometric consequence for every starting fortune or stake.

Run the lab as a balance-sheet test:

1. Reset and reconcile every ending state using starting wealth − wager + payoff.
2. Confirm that the arithmetic ending wealth exceeds $100,000 while geometric ending wealth is below it.
3. Lower the wager fraction without changing the payoff. Find the first displayed setting at which geometric ending wealth exceeds starting wealth.
4. Double starting wealth while holding the wager fraction fixed. Explain why the fixed payoff tail now offsets a smaller portion of the capital at risk.
5. Falsify the claim “positive expected winnings imply an acceptable wager” whenever the combined ending-wealth criterion is below the no-wager alternative.

The reconstruction assumes the wager is paid once, payoffs settle immediately, wealth stays positive, and dollars are the only relevant state. It omits liquidity needs, taxes, utility, debt, repeatability, and uncertainty about the generator. The geometric comparison is accounting for multiplicative capital, not a universal theory of human welfare.

Domain	Translation	Evidence to demand
Economics	A subsidy, tariff, or guarantee changes agents with different starting buffers differently	Distribution of balance sheets before the policy and ending resources after each state
Startup	A fixed-cost bet consumes a different share of runway at different funding levels	Cash after failure, months to the next milestone, and financing available in that state
Business	A contract’s maximum upside is incomplete without working capital committed	State-by-state cash, collateral, cancellation liability, and delivery cost
Daily life	A fixed prize or expense must be judged relative to savings and obligations	Remaining emergency buffer, recovery time, and essential spending protected

The upper table on printed page 42 isolates the most fragile sizing choice: stake the entire capital base and let the six payoffs become the six ending fortunes. The consequence is an arithmetic average of $166,705 beside a geometric average rounded to only $61, a direct contradiction of the claim that one favorable mean describes typical compounded wealth.

This is the first of two distinct analytical tables on printed page 42, PDF page 60, and it therefore has its own lab. Reset uses $100,000 starting wealth and a 100% wager, producing $1, $2, $6, $22, $200, and $1,000,000 of ending wealth. The book prints arithmetic average $166,705 and geometric average $61; the lab keeps those rounded benchmarks visible beside its full-precision recalculation.

The Starting wealth control changes the finite base. The Wager fraction control is bounded between 50% and 100% so the experiment stays focused on aggressive sizing. The logarithmic vertical axis prevents the first five states from collapsing into a single pixel while marker shape and dash pattern distinguish the series without relying on color.

Test the table’s claim step by step:

1. Reset and verify that a full wager leaves no starting capital outside the game.
2. Compare arithmetic and geometric ending wealth and compute the ratio between them.
3. Reduce the wager in five-point increments. Identify the first setting that leaves a meaningful floor in five states.
4. Increase starting wealth while holding the payout schedule fixed. Observe whether the same full-wager rule becomes more or less defensible geometrically.
5. Reject any presentation that reports only $166,705 without the six ending states and the $61 geometric benchmark.

The source calculation gives every face equal probability and treats the table as a one-period valuation under Bernoulli’s logarithmic mapping. It does not claim a person will literally receive $61, nor does the lab simulate a market. The geometric value is the certain wealth equivalent under this specific multiplicative criterion.

Domain	Translation	Evidence to demand
Economics	National averages can rise while many balance sheets approach zero	Lower-tail wealth, insolvency transitions, and recovery access rather than GDP alone
Startup	An all-in launch can create a high expected valuation and almost no second attempt	Runway after each demand state, rollback ability, and financing dependence
Business	A single-project company can have positive expected project value but terminal downside	Covenant headroom, supplier liabilities, and continuation value after failure
Daily life	Staking all savings on one opportunity removes the reserve that makes future choices possible	Protected essentials, reversible commitment, and time needed to rebuild

The lower table on the same source page changes only sizing, yet it reverses the geometric verdict. The consequence is that a “good” or “bad” opportunity cannot be separated from the fraction of finite capital exposed to it.

Printed page 42, PDF page 60, starts again at $100,000 but wagers $10,000. Reset reproduces ending fortunes of $90,001, $90,002, $90,006, $90,022, $90,200, and $1,090,000. The book reports arithmetic average $256,705 and geometric average $136,445, both above the starting fortune. This lower table is analytically distinct from the full-wager table and is not merged with it in the course.

Starting wealth and Wager fraction recompute the six states. The fraction control spans 5% to 50%, allowing the learner to locate a region rather than memorize 10% as a universal answer. The underlying payoffs and probabilities remain fixed, so any change comes from exposure size alone.

Use the smaller wager to challenge overgeneralization:

1. Reset and reconcile the six published ending fortunes.
2. Compare the $136,445 geometric benchmark with the $61 benchmark from the separate full-wager lab.
3. Increase the fraction until geometric ending wealth first falls below starting wealth.
4. Reduce the fraction and note the opportunity cost: more capital survives ordinary states, but less participates in the jackpot.
5. Reject the claim “10% is optimal” because the source tests one wealth level, one payoff distribution, and one criterion; it establishes a counterexample to all-in sizing, not a universal allocation.

The model assumes divisible stakes, no transaction costs, no borrowing, fixed payoffs, and immediate settlement. Real projects may have minimum viable scale, path-dependent learning, changing opportunity sets, and losses beyond the stake. Those mechanisms can move or eliminate the apparent safe region.

Domain	Translation	Evidence to demand
Economics	Partial guarantees or capital buffers can preserve participation without removing all exposure	Marginal behavior, fiscal capacity, tail losses, and who bears residual risk
Startup	A staged pilot buys information while reserving runway for another iteration	Learning milestone, capped spend, conversion evidence, and remaining months of cash
Business	A limited order, regional rollout, or capped guarantee sizes a new opportunity	Unit economics at pilot scale, expansion triggers, and maximum contractual loss
Daily life	A bounded trial can preserve a job, savings, or support network while testing a new path	Review date, stop rule, reserve floor, and transferable learning

Sizing by intuition can still miss the exact point where the geometric criterion changes sign. The consequence is either overpaying for upside or rejecting a wager that a smaller stake would make acceptable.

Printed page 43, PDF page 61, plots fraction of starting wealth wagered on the horizontal axis and geometric expected ending wealth on the vertical axis. With $100,000 starting wealth and the published payoffs, the curve crosses the $100,000 starting-wealth line at the book’s rounded 37.7%, or $37,708. The reset lab recalculates the curve directly from the six states and displays the printed benchmark separately from the numerical root.

Starting wealth tests scale dependence because the payoff dollars remain fixed. Face-six jackpot scale tests tail dependence while preserving the other five payoffs and equal probabilities. The curve and crossing update together, preventing a control from changing a headline without changing its causal geometry.

Use the crossing as a falsifiable boundary:

1. Reset and locate the diamond marker where geometric ending wealth equals starting wealth.
2. Compare the numerical root with the printed 37.7% / $37,708 benchmark; attribute small differences to chart and display rounding.
3. Halve the jackpot while holding wealth fixed. Record the new fair fraction and reject any claim that fair value is invariant to the tail payoff.
4. Restore the jackpot and double starting wealth. Explain why a fixed prize supports a different fraction of a larger fortune.
5. State the decision rule before changing controls: above the crossing, reject the wager under this criterion; below it, the criterion alone does not establish suitability because omitted constraints may still dominate.

The root assumes a continuous wager fraction and strictly positive ending wealth. It is not a market price, expected profit forecast, or personalized recommendation. Probability error, inability to divide the stake, payout default, liquidity needs, and repeatability can all invalidate the boundary.

Domain	Translation	Evidence to demand
Economics	A policy intensity can cross from resilience-enhancing to balance-sheet damaging	Heterogeneous starting wealth, response curves, and fiscal or institutional capacity
Startup	A project budget has a maximum defensible share of runway under a declared state model	State payoffs, milestone probabilities, minimum build scale, and cash after failure
Business	Bid size or customer credit exposure has a geometric break-even boundary	Margin distribution, default recovery, concentration, and funding constraints
Daily life	A commitment can be scaled until its downside crosses a reserve floor	Savings, fixed obligations, reversibility, and personally unacceptable states

Insurance often appears overpriced when claim payments are compared only with premiums. The merchant example shows why that arithmetic comparison can miss a positive portfolio-level geometric effect.

A merchant starts each shipment cycle with 11,000 rubles: 3,000 in reserves and 8,000 committed to goods. If the voyage succeeds, the goods produce 10,000 rubles and ending wealth is 13,000. If ship and cargo are lost, only the 3,000 reserve remains. Assume success occurs on 95 of 100 comparable voyages and loss on 5.

An insurer charges 800 rubles to make the merchant whole on the cargo. On an arithmetic basis the policy is expensive: expected claims are 5% × 10,000 = 500, so the merchant's expected standalone insurance result is 500 - 800 = -300 rubles per voyage.

Now compare the whole:

Measure	Uninsured	Insured
Possible ending wealth	13,000 or 3,000	12,200
Arithmetic expected ending wealth	0.95 × 13,000 + 0.05 × 3,000 = 12,500	12,200
Geometric expected ending wealth	13,000^0.95 × 3,000^0.05 ≈ 12,081	12,200
Geometric return on 11,000	about 9.8%	about 10.9%

The policy lowers arithmetic expected wealth by 300 but raises geometric expected wealth by roughly 119. Both merchant and insurer can gain according to the metric relevant to each: the insurer keeps a positive arithmetic underwriting margin across many customers, while the merchant protects one recursively reinvested capital base.

The deep-loss asymmetry explains the difference. Losing from 11,000 to 3,000 is about a 73% loss. Returning from 3,000 to 11,000 requires about 267%, commonly rounded to 270% in the example. Equal percentage losses and gains do not cancel because their bases differ.

Arithmetic lines make the recurring premium look larger than the expected claim, while the logarithmic map exposes how much more damage the rare cargo loss does to the merchant’s finite base. The consequence is rejecting mutually beneficial insurance because the wrong dimension was summed.

Printed page 46, PDF page 64, contains a two-panel figure titled “Plunging into the Abyss.” The left panel shows the full logarithmic curve; the right panel enlarges the small premium movement near 12,000–13,000 rubles. Reset reconstructs the points described in the text: A is 13,000 rubles after a successful sale, B is an 800-ruble horizontal premium step, C is its smaller vertical log step, D is the horizontal 10,000-ruble cargo-loss step to 3,000, and E is the much larger vertical log plunge.

The horizontal axis is ending rubles and the vertical axis is the natural logarithm of ending rubles. Insurance premium changes the A → B → C path. Cargo loss changes the A → D → E path. The 5% loss frequency stays at the source benchmark, allowing the readouts to compare the insurer’s arithmetic margin with the merchant’s geometric effect.

Run the map as an accounting challenge:

1. Reset and identify all five labeled points before reading any summary metric.
2. Confirm that the 800-ruble premium exceeds the 500-ruble expected claim, producing a 300-ruble arithmetic underwriting margin.
3. Compare insured ending wealth of 12,200 rubles with uninsured geometric wealth of about 12,081 rubles.
4. Raise the premium until the geometric net effect becomes negative. This is the falsifier for “insurance is cost-effective” under the stated distribution.
5. Restore the premium and reduce cargo loss. Observe that a smaller deep-loss asymmetry makes the same premium harder to justify.
6. Increase cargo loss and inspect the vertical segment near the lower end of the log curve. Explain why equal horizontal ruble distances do not imply equal multiplicative damage.

The reconstruction follows the source text rather than digitizing pixels, so point coordinates are transparent calculations. It assumes 95 successful shipments out of 100, immediate full payment, a solvent insurer, no deductible, fixed cargo value, and identical reinvestment. Correlated piracy, delayed claims, premium changes, partial recovery, legal disputes, and a changing business would alter the result.

Domain	Translation	Evidence to demand
Economics	Risk pooling can create value for a finite household or firm even when premiums exceed expected claims	Pool independence, insurer capital, exclusions, distributional access, and systemic correlation
Startup	A reserve or rollback can lose money in the launch-success state and preserve another iteration after failure	Carrying cost, failure-state cash, recovery time, and whether the control shares the same failure mode
Business	Redundancy can be a negative standalone investment and a positive whole-system control	Incident severity, downtime avoided, common-mode risk, and tested failover performance
Daily life	Catastrophe insurance can have negative expected cash return and protect a nonreplaceable base	Coverage language, deductible, payout timing, insurer reliability, and uninsured remainder

Linear ledgers make an 800-ruble premium look large beside a 500-ruble expected claim, while the log map makes the avoided plunge toward low wealth visible. The consequence is a different ordering of what matters.

For a fractional loss L, the gain required to recover is:

recovery gain = L / (1 - L)

Loss	Wealth remaining	Gain needed to recover
10%	90%	11.1%
25%	75%	33.3%
50%	50%	100%
75%	25%	300%
90%	10%	900%

The curve is concave: as remaining wealth shrinks, the next unit lost does more damage to future possibility than the previous unit. That is the chapter's “nature's admonition.” It does not command avoidance of every gamble. It commands respect for the capital base and for protection whose arithmetic cost is smaller than its geometric benefit.

People often expect a percentage gain equal to a prior percentage loss to restore the starting point, but the gain is applied to a smaller base. The consequence is systematically underestimating the effort, time, or risk required to recover from deep drawdowns.

Printed page 50, PDF page 68, plots loss percentage on the horizontal axis and profit required to get back to even on the vertical axis. The source labels a -73% loss and an approximately +270% required recovery. Reset preserves that rounded benchmark. The solid circle curve is the exact identity gain = loss / (1 − loss); the dashed square line shows the false intuition that equal loss and gain percentages cancel.

Loss severity selects a point on the curve. Loss covered reduces the effective loss before applying the same recovery identity. Coverage is an illustrative overlay, not another curve transcribed from the book. Hatching, shapes, labels, and line styles keep the comparison legible without color.

Use the curve to falsify linear recovery stories:

1. Reset and calculate 0.73 / (1 − 0.73) = 2.7037, which the source rounds to 270%.
2. Move loss severity to 50% and verify that a 100% gain is needed.
3. Move it to 90% and verify that a 900% gain is needed; note the acceleration near total loss.
4. Restore 73%, then increase coverage. Confirm that the effective-loss marker moves left and down the same curve.
5. Reject the statement “we can make back the loss with an equal gain” for every nonzero loss.
6. Reject a proposed control if its own cost or failure mode creates a larger effective loss than it removes.

The curve is an algebraic identity for positive capital, not an empirical forecast of how quickly a market or business will recover. It assumes no withdrawals, contributions, leverage, taxes, path-dependent cash flows, or time value. A total loss is excluded because recovery from a zero multiplicative base is undefined without new external capital.

Domain	Translation	Evidence to demand
Economics	A recession can destroy productive capacity that headline growth must rebuild from a smaller base	Level of output and employment, not only growth rate; scarring, entry, and external support
Startup	A 70% runway loss requires far more than a 70% improvement in burn efficiency to restore months remaining	Cash level, fixed obligations, financing probability, and time to the next milestone
Business	Revenue, trust, or capacity lost in an incident may require disproportionate acquisition and repair	Customer cohort recovery, margin, remediation cost, and operational bottlenecks
Daily life	Debt, depleted savings, injury, or burnout can reduce the base available for recovery	Remaining capacity, external support, compounding costs, and a realistic recovery horizon

The logarithm can remain an abstract formula until its steep approach toward zero is visible. The consequence is treating a deep loss as one more equal subtraction rather than a plunge that permanently changes the multiplicative path.

Printed page 55, PDF page 73, pairs a graph labeled Y = LOG(X) with a historical Rhine Falls engraving. The graph is analytical; the engraving is an editorial metaphor and contains no data to digitize. The lab reconstructs the mathematical panel as y = ln(x), the natural logarithm, and honestly omits the decorative image.

The horizontal axis is a wealth ratio, where 1x is a reference capital level, and the vertical axis is its natural log. Starting wealth ratio moves the before point along the curve. Loss from that level moves the after point to a smaller ratio. The dashed triangle segment makes the logarithmic drop explicit.

Run the falls experiment from first principles:

1. Reset at a 4x starting ratio and 50% loss. Record the before point, after point, and log change.
2. Change the starting ratio but keep the loss at 50%. Confirm that the log change remains ln(0.5) even though the dollar-like horizontal distance changes.
3. Hold the start fixed and increase the loss toward 90%. Observe the rapidly steepening vertical plunge.
4. Compare two successive 50% losses with one 75% loss: both leave one quarter of the starting base and therefore have the same total log change.
5. Falsify the claim “the chart predicts returns.” Its curve is a mathematical transformation of wealth ratios; it contains no historical frequency or forecast.

The reconstruction chooses natural log because the chapter’s formulas use exponentials and logs as inverse mappings; another log base rescales the vertical axis without changing ordering or concavity. The visual applies to strictly positive ratios. It does not model negative equity, external capital injections, liabilities, or preferences over nonfinancial outcomes.

Domain	Translation	Evidence to demand
Economics	Log changes compare proportional movement across different-sized economies or balance sheets	Positive base, consistent measurement, distributional effects, and level recovery
Startup	Burn and setbacks compound through the runway that remains	Cash trajectory, milestone timing, financing constraints, and irreversible commitments
Business	Repeated percentage impacts multiply rather than add on a fixed ledger	Beginning and ending levels, shared dependencies, and restoration of operating capacity
Daily life	Savings, debt capacity, health, and time can have path-dependent proportional changes	A defined positive base, hard constraints, external support, and nonfinancial limits

A static calculation can conceal how the same loss repeats through a finite capital base. This lab lets the learner vary the merchant's exposure and watch the arithmetic and geometric scoreboards diverge.

The horizontal axis is repeated shipment or venture number and the vertical axis is compounded wealth. The ensemble projection, exposed time-growth path, and covered time-growth path use redundant line styles and marker shapes. Loss severity changes the uninsured capital impairment while the loss frequency remains fixed at 5%. Coverage scales both the claim and its proportional share of the 800-ruble full premium. Readouts compare arithmetic return, exposed and covered geometric growth, paid premium, recovery burden, and covered ending wealth. The starting buffer, horizon, success payoff, frequency, and contract terms stay fixed.

Run the experiment as a controlled accounting test:

1. Reset to the merchant-style baseline and record the premium, loss frequency, loss severity, arithmetic insurance value, and both geometric rates.
2. Increase loss severity while frequency remains fixed. Confirm whether the geometric value of coverage changes faster than the arithmetic projection suggests.
3. Restore severity and reduce coverage one step at a time. Because premium and claim scale together in this lab, identify where partial coverage ceases to protect enough of the deep loss.
4. Compare zero and full coverage at the minimum and maximum loss severities. Explain why the same contract fraction has different geometric value as the capital impairment changes.
5. State which omitted change—frequency, premium loading, buffer, exclusions, or counterparty failure—would most plausibly reverse the displayed ordering.

Interpret the break-even premium only inside the displayed generator. The model assumes independent voyages, a stable loss frequency, immediate certain claims, frictionless reinvestment, and no insurer default. Real contracts add deductibles, exclusions, inflation, correlated losses, legal delay, counterparty capital, changing exposure, and uncertain probabilities. The lab demonstrates geometric accounting; it neither prices insurance nor recommends a product.

An economy loses productive capacity when each household or firm must self-fund every rare loss. The consequence is either excessive caution or repeated ruin.

Risk pooling turns many individual N = 1 exposures into an insurer's broader book of claims. A farmer can plant, a merchant can ship, and a business can build because a bounded premium replaces a potentially capital-destroying loss. The arrangement can be mutually beneficial even when expected premiums exceed expected claims, because the insured and insurer face different capital bases and aggregation opportunities.

This does not justify any premium. Market power, adverse selection, moral hazard, correlated catastrophe, and insurer insolvency can erase the gain. The economic test remains whether the contract's full cost is smaller than the damage it removes from the insured's compounding path.

A startup's headline valuation is irrelevant if one loss removes the cash needed for the next product cycle. The consequence of a deep setback is not just its amount; it is the experiments, hires, and customer learning that can no longer occur.

Suppose a company has £1 million and risks £700,000 on a single launch. A failed launch leaves £300,000, requiring a 233% gain merely to restore the original base. Spending £80,000 on staged rollout, contractual exit rights, security review, and rollback capacity may look like negative expected value when the launch succeeds. Its geometric value lies in limiting the failed state enough to preserve another attempt.

Model success and failure as ending cash states, not as isolated project return. Then test whether the mitigation improves the geometric path after its recurring cost. The goal is not to insure ordinary product learning away; it is to prevent one experiment from ending the learning process.

Operational controls are often cut because their direct return is negative and incidents are rare. The consequence is a false economy when one severe event damages revenue, trust, capacity, and financing at the same time.

A tested rollback process consumes engineering time every release. A spare production line incurs carrying cost. Cyber insurance and offline backups require recurring expenditure. Evaluate each against the whole state after failure: cash lost, days unavailable, customers retained, regulatory obligations, recovery cost, and whether normal operations can resume.

An 800-unit control that prevents an occasional 10,000-unit capital impairment can be rational even if its arithmetic expected claim is only 500. But the control must actually pay in the named state. A backup connected to the same compromised credentials or an insurer that excludes the relevant event offers a story, not the required payoff.

Household decisions become misleading when each expense is treated as an equal subtraction from an annual average. The consequence of a deep loss is the removal of choices and the addition of costly recovery mechanisms.

An emergency repair paid from a buffer may be a £2,000 loss. The same repair funded by high-cost debt can create interest, missed bills, and reduced ability to handle the next shock. Insurance for a catastrophic loss may have a negative expected standalone return and still protect housing, income, or caregiving capacity.

The framework does not say to insure everything. Small, frequent losses may be cheaper to self-fund; policies may contain gaps; cash has opportunity cost. List the remaining wealth or capacity after the loss, the gain or time required to recover, and the premium required to cap it. Choose with the whole path visible.

Geometric wealth accounting is exact for multiplicative capital, but applying it carelessly can still produce false confidence. The consequence is a precise answer to the wrong objective or an estimated distribution.

The geometric mean requires positive wealth multipliers; zero is absorbing and negative wealth needs a different model. Probabilities may be unknown, returns may not be independent, and repeated opportunities may change after each outcome. Consumption, liabilities, withdrawals, taxes, and cash flows can make order matter even when a simple product is commutative. Organizations also optimize safety, mission, law, and human welfare, not wealth alone.

Do not rename logarithmic compounding “risk aversion” and assume it describes everyone's preferences. Do not infer a market distribution from the dice. Use the map to value damage to a stated capital base, then stress the inputs and preserve the decision's nonfinancial constraints.

Source attribution can become misleading when exact transcription and new teaching controls are blended together. The consequence is false precision, so every lab keeps its printed/PDF page, book benchmark, and illustrative scenario visibly separate.

The chapter's historical arc, finite Saint Petersburg wager, Bernoulli logarithmic formulation, merchant shipment, and arithmetic-cost-versus-geometric-effect distinction are paraphrased and recalculated from Chapter 2 of Mark Spitznagel's Safe Haven (Wiley, 2021), PDF pages 47–74. The complete analytical inventory is: payoff profile (printed 33/PDF 51), half-wealth ending table (34/52), full-wager table (42/60), ten-percent-wager table (42/60), fair-value curve (43/61), merchant logarithmic map (46/64), recovery-gain curve (50/68), and logarithmic curve/Rhine Falls comparison (55/73). The two page-42 tables are distinct items because each reports its own six ending states and arithmetic/geometric summaries.

Exact readable source values are preserved at Reset and identified as book benchmarks. Curves derived directly from printed formulas are recalculated rather than pixel-digitized. Adjustable overlays, cross-domain applications, protocols, annotations, and dynamic domains are original teaching reconstructions. The chapter-opening die, ship drawings, and Rhine Falls engraving are editorial illustrations rather than datasets; only the mathematical panel of the Rhine Falls comparison is modeled. No lab estimates a market distribution or recommends a transaction.

Chapter 2 changes the unit of account from isolated profit to surviving capital. Once returns compound, deep losses carry an increasing recovery burden that an arithmetic average does not display.

• The geometric average tracks multiplicative wealth through time.
• Logarithms turn multiplied wealth ratios into additive terms and expose the cost of approaching zero.
• Positive arithmetic expectation can coexist with negative compound growth.
• Insurance can lose money alone and still raise the geometric growth of the whole.
• The premium is not automatically justified; coverage, dependence, and counterparty performance matter.
• A deep loss is costly because it shrinks the base on which every later gain operates.

Mastery means being able to reproduce the merchant result and state where it stops applying. Complete each calculation without appealing to a label such as “conservative.”

• [ ] I can convert a percentage return into a total-return multiplier.
• [ ] I can calculate an arithmetic expected ending wealth.
• [ ] I can calculate a geometric expected ending wealth from probabilities and logs.
• [ ] I can derive the recovery gain after a fractional loss.
• [ ] I can explain why the merchant and insurer can both gain in their relevant frames.
• [ ] I can find the premium at which protection stops improving compound growth in the lab.
• [ ] I can name at least three omitted mechanisms that could reverse the result.
• [ ] I can explain why geometric accounting is not personalized advice.