Evidence Without Morality

Chapter Five: Testing the World We Live In

The argument so far has been structural and theoretical. Constraint stacking and probability explain how failure can emerge without moral cause. But theory requires testing. The remaining question is whether these dynamics describe the world as it actually exists.

To answer this, the analysis turns to simulation — not to predict individual outcomes, but to test whether equal effort produces converging results under realistic conditions.

Simulation is the right instrument here for a specific reason: the real world never runs the controlled experiment. In observed data, effort and conditions vary together — the person with the thin margin is also tired, also stressed, also short on time — and every correlation can therefore be argued in either direction forever. This is precisely the ambiguity that keeps the character debate alive. A model breaks the deadlock by doing what no dataset can: holding one variable rigidly constant while the other runs. If equal effort produces unequal outcomes inside a machine where effort is equal by construction, then unequal effort cannot be the necessary explanation for the same pattern outside the machine. The model does not prove what the world is doing. It proves what the world does not need in order to do it.

The model is built to be fair to the opposing view. It assumes rational behavior throughout. Agents work when work is available. Bills are paid when possible. No one sabotages themselves. There is no laziness parameter. There is no moral sorting of any kind.

The environment, however, includes ordinary risk calibrated to U.S. conditions: job loss, medical expense, transportation failure, rent obligations. These events occur at realistic frequencies. Nothing is extreme. Everything is ordinary.

A population of 1,500 agents is run across five simulated years — 260 weekly steps — and the entire simulation is repeated 120 times to ensure results are not artifacts of a particular random sequence.

The results are consistent and striking.

When no one has a stability advantage, roughly 96 out of 100 agents experience housing failure over five years. When a small group (5%) begins with better buffers and lower shock exposure, the rate drops modestly to 93. When that group grows to 10%, it drops to 90.

Stability helps. But the magnitude is small relative to the baseline. Most outcomes are still decided by timing, shock size, and how much room existed when something went wrong.

Two clarifications prevent two predictable misreadings. First, the 96% figure is not a forecast of actual eviction rates. The model deliberately strips away the stabilizers that real households reach for — family transfers, informal loans, doubling up, public assistance, charity. It isolates the raw mechanics of margin against shock. Real-world failure rates are lower than the model's because real people are constantly rescued by buffers external to their own effort — which is not a rebuttal of the argument but its strongest confirmation. What stands between most households and the model's baseline is not their diligence, which the model already granted them at maximum. It is other people's margin.

Second, the objection that the model is rigged — that its parameters were chosen to produce the result — has a straightforward answer: the parameters are public, listed in the appendix, and each one errs toward generosity. Employment recovers faster than typical for the wage tier. Shocks arrive at conservative frequencies. No shock exceeds the ordinary catalog of a working life. Anyone who believes different parameters would rescue the sorting story is invited to say which parameter, and to notice that rescuing it requires assuming either implausibly large savings or implausibly benign luck — that is, requires assuming away the very conditions the argument is about.

The result is not collapse. It is churn. Over time, different individuals occupy failure states, but the overall distribution remains steady. The simulation does not invent this pattern. It reflects the mechanics described in the previous chapters.

Churn is also why the pattern is so hard to see from inside. A stable failure rate with rotating membership means that most people who fail eventually recover partially, and most people who are stable have a failure somewhere in their past or future. Everyone's biography therefore contains apparent evidence for the effort story — the year they worked hard and things improved — and the aggregate, where the real signal lives, is visible from no one's kitchen table. The system's statistics and its testimonies permanently disagree, and testimony is more vivid. This is not a failure of honesty. It is a failure of vantage.

There is no moral gradient explaining loss. Failure persists because the system requires continuous clearance of thresholds.

The most significant aspect of these results is what they do not show. There is no dramatic separation between the diligent and the negligent. There is no virtue gradient. What appears is a steady failure rate that persists regardless of how individual behavior is specified — because the failure is not produced by individual behavior. It is produced by the conditions.

Chapter Six: Why a 10% Failure Rate Is the Signal, Not the Reassurance

A common reaction to simulation results showing minority failure is relief. If only 10 or 20 percent fail, the system appears fundamentally fair. Most people make it. The few who do not must have done something differently.

This reaction misunderstands what the number represents.

In constraint-based systems, a persistent minority failure rate is not an error signal. It is a stability signal. Systems that eliminate failure entirely lose the disciplinary pressure that keeps behavior in line. Systems that allow failure to dominate lose the legitimacy that keeps people participating. A stable failure band — somewhere in the range of 10 to 20 percent — is evidence that the system is operating within its functional range.

The number matters more than the narrative attached to it. Ten percent failure is not evidence of insufficient effort by those 10 percent. It is evidence of managed exposure. The system is calibrated to produce that level of failure, and it does so reliably.

Over time, the identities of those who fail change. The proportion does not. This churn is essential to how the system maintains legitimacy: because failure is dispersed across populations and generations, it remains explainable through individual stories. Each case can be moralized. The aggregate pattern stays invisible.

Compare this with how the same logic is accepted without controversy in other domains. Every airline schedule is built to a target rate of delay; every insurer to a target rate of claims; every lender to a target rate of default. When a bank prices loans expecting 4% to default, no one describes the eventual defaulters as a moral discovery — the 4% was in the spreadsheet before any borrower signed. The system knew, in advance and in aggregate, that some of its participants would fail, and priced their failure into its own stability. The identical structure, operating on housing and employment, is described differently: there the pre-scheduled percentage is discussed as though it were a census of the unworthy. The only difference between the two descriptions is that one system publishes its expected loss rate and the other calls its expected loss rate a moral outcome.

If effort were the deciding factor, failure rates would decline as norms and work ethic improved. They do not. They stabilize.

This is the most quietly damning trend line available. Over recent decades, by any measurable proxy — hours worked, credentials earned, second jobs held — exertion at the lower half of the income distribution has risen. If the sorting story were true, the failure band should have narrowed in proportion. It held. Effort went up; the percentage cleared did not. A test that returns the same grade regardless of the studying is not measuring the studying.

The issue is not why some people fail. The issue is why the system requires some people to fail in order to continue.

Once this is understood, debates about motivation and personal responsibility lose much of their explanatory force. The system is not sorting people by effort. It is regulating pressure — maintaining a failure band sufficient to discipline behavior without provoking mass withdrawal.