A library beats a blank page, especially when the question is hostile.

The problem with ad-hoc

A blank test surface returns the operator's own agenda.

Given an empty configuration and no prompts, an operator tests the question currently on their mind — usually a variation on something recent, and usually in the direction they suspect is favourable. Nothing about that is dishonest, and it produces a research record with a systematic hole in it: the adverse conditions nobody was thinking about go untested, indefinitely, because nothing prompts them. Hostile scenarios in particular are almost never generated spontaneously. They have to be sitting there already, as items in a list, waiting to be run.

Saved slots

A configuration you can return to is a configuration you can compare against.

The scenario slots hold recorded parameter states — save writes the current configuration, load recalls it. The value is not convenience but comparability: a saved baseline can be reloaded exactly six months later and re-run, which turns a historical result into a re-testable claim rather than a screenshot. This is the same property the run registry provides for outputs, applied to inputs. Between the two, a result from last quarter can be regenerated on demand, and a regenerable result is the only kind that can be argued with. The slots also solve a smaller problem worth naming: an operator mid-investigation often needs to step away from a configuration to check something else, and without a save that departure destroys the setup. Cheap state capture is what allows a research session to branch without the operator hoarding a workbook they are afraid to touch.

Presets and stress packs

Adverse conditions arrive pre-written so nobody has to volunteer for them.

The prebuilt presets carry per-preset assumption columns spanning conservative through aggressive, plus explicitly hostile ones — a high-fee configuration and a trend-collapse configuration. The stress library goes further, applying coordinated parameter overrides in a single action so that a fee doubling, a volatility expansion, or a regime collapse can be imposed without hand-editing a dozen tabs. The design intent is to remove friction specifically from the unpleasant tests. Where testing adversity is laborious it will happen rarely, and it will happen least in the periods when the account is doing well — which is exactly when the information would be cheapest to act on. The coordination matters as much as the convenience: a hand-assembled adverse scenario tends to be adverse in one dimension while the others stay benign, which is not what hostile conditions look like. Real stress arrives correlated, and a pack that moves fees, volatility and trend behaviour together is testing something the piecemeal version never reaches.

The rejected archive

Negative knowledge is knowledge, and most processes throw it away.

Rejected scenarios remain in the library with their verdicts attached. This is the least intuitive of the four categories and arguably the most valuable, because a research programme without a rejection record repeats itself. The same appealing idea recurs every eighteen months, gets re-tested at real cost, and fails for the same reason nobody wrote down. The archive also fixes the denominator problem: a library showing nine rejections behind one promotion tells a reviewer something important about how selective the process actually is, and a library showing only the promotion tells them nothing.

How a library changes the review

Standing tests run on cadence rather than on inspiration.

Once the adverse configurations exist as library items, they can be scheduled. A quarterly re-run of the stress packs against current calibration answers a question no ad-hoc process asks: has the system's tolerance for hostile conditions changed as its evidence base has? That question has no natural prompt, arises from no live event, and is genuinely important — a configuration that survived a fee doubling last year may not survive one now if the measured edge has thinned. The library is what makes the question askable without anyone having to think of it.

• Hostile presets exist so adverse testing does not depend on the operator's mood.
• Saved slots make a historical result regenerable, not merely remembered.
• The rejection archive prevents re-testing settled ground and reveals the process's true selectivity.

The key idea

What gets tested is decided by what is easy to test.

Every research process is shaped less by its standards than by its friction, and a blank configuration surface makes the comfortable questions cheap and the uncomfortable ones expensive. A standing library inverts that: the adverse conditions are one selection away, the historical baselines are reloadable, and the failures are still on the shelf with their reasons attached. None of it changes what a scenario contract means. It changes which scenarios ever get written.