Losing streaks have a distribution, and you were given it.

The unit that hurts

Operators do not experience percentages. They experience counts.

A drawdown expressed as a depth is an abstraction reconstructed after the fact. What is actually lived is a sequence: this trade lost, the next lost, the one after that lost. The count is concrete, cumulative, and arrives with a running commentary about what it implies. This is why depth-only preparation leaves a gap — an operator can be entirely comfortable with an adverse depth number and still be dismantled by the arithmetic that produces it, because the depth was rehearsed and the sequence was not. The simulation models clustering explicitly, and the resulting streak statistics are the missing half of the preparation.

Clustering is modelled, not assumed away

Losses arrive together more often than independence would predict.

The simulation does not treat outcomes as cleanly independent draws. Loss and win clustering are modelled directly, correlation across branches, sessions and direction is converted into an effective concurrent-risk multiplier, and shock weeks are injected for crisis-path modelling. This matters because clustering is precisely the property that makes streaks feel meaningful — a run of losses that arrives in one week, across correlated positions, in a single hostile session, feels like a regime rather than a sequence. The model producing the streak distribution has that behaviour inside it, so the frequencies it reports are frequencies for a world where losses do bunch.

The worst-period tables

Worst cycle, worst week, worst month — each with its own distribution.

Alongside consecutive-loss counts, the simulation reports the damage concentrated into single periods: the worst cycle, the worst week, the worst month across governed paths, and how severely negative cycles cluster. These are the numbers that answer the question an operator actually asks during a bad week, which is not how deep can this get but is this week abnormal. Usually it is not, and having the answer available in advance converts the week from a referendum on the system into a data point with a known position. The account's worst week having a modelled distribution is, on its own, one of the more calming facts in the stack.

What a streak does and does not license

A run of losses is not evidence about the edge.

The critical inference to block is that a long losing run constitutes information about whether the system still works. It does not, on its own, because the distribution says healthy paths produce those runs at known rates — which means observing one is consistent with the system being entirely intact. Edge-change has its own evidence threshold requiring persistence across review periods, statistical degradation through the analytics gates, elimination of mundane causes, and structural corroboration. A streak is not a shortcut through any of those, and the fact that it is the most emotionally compelling evidence available is exactly why the threshold exists.

The preparation this enables

Rehearse the count, not only the depth.

The practical output is a second rehearsal alongside the tail-depth one. At the modelled worst streak, what does the gate state become? How many cycles does the run span at the current rhythm? What does the deployment look like at that point in the ladder, and what does the recovery arithmetic require? Written answers, produced in calm, are what the operator consults during the real thing. The depth rehearsal prepares for the balance; the streak rehearsal prepares for the sequence — and the sequence is what is happening in the moment the decision to abandon the system gets made.

• Read the streak distribution alongside the depth bands; they prepare for different ordeals.
• Clustering is in the model, so the frequencies already account for losses arriving together.
• A streak inside the distribution authorises nothing — the edge-change threshold is unmoved by it.

The key idea

Being shown the worst run in advance is what makes meeting it survivable.

The distribution's least pleasant entries are its most protective ones, and the streak statistics are the entries closest to lived experience. An operator who knows that a run of a given length is met by a substantial share of healthy paths has that fact available at the moment it is most needed and least discoverable. The run still hurts. It just stops being an argument, and a losing streak that cannot argue is a losing streak the system survives.