Why R&D never touches the Standard baseline.

The two territories

Standard is what is; the sandbox is what might be.

The Standard baseline is the production system in full: the trade plan's governed skeleton, the live workbook parameters, the journal as the record of what actually happened, and the authority chain — gates, throttle, deployment rules — as it actually binds. The sandbox is everything speculative: scenario profiles, what-if projections, manual stress cases, candidate variants, simulation runs under alternative assumptions. The Advanced EV Analytics Lab's architecture states the residency rule exactly: scenario outputs and what-if insights belong in the experimentation layer until tested, approved, and promoted into live rules — and its scenario engine carries the matching warning, that manual scenarios are for sensitivity, not proof. A scenario is not evidence unless supported by live data. The wall is that sentence, enforced as architecture.

The direction of flow

Evidence flows in freely; influence flows out through one door.

The boundary is deliberately asymmetric. Inward, everything: the sandbox reads the live journal, mirrors production-style data, consumes the real evidence record — the whole point of scenario comparison is running alternatives over what actually happened. Outward, almost nothing: sandbox outputs may inform review posture and generate candidates, but the only path by which anything R&D changes anything production is the promotion workflow's full gauntlet. No scenario result adjusts a live parameter directly, no what-if projection rewrites an expectation, and — the architecture manual's phrasing — the lab can diagnose opportunity, but it cannot grant permission. One door, heavily gated, is not a bottleneck. It's the entire mechanism by which the production system knows why it is the way it is.

The persuasion problem

The wall guards against good results, not bad ones.

A failed experiment threatens nothing — nobody promotes a scenario that lost. The dangerous artifact is the brilliant one: a variant showing beautiful numbers over the recent record, generating an urge with a timestamp — adopt this now, before more opportunity is missed. Every step of yielding feels reasonable in the moment, and the destination is a system whose rules trace to a series of exciting afternoons instead of a body of evidence. The urgency is also almost always false: a genuinely superior configuration remains superior through a persistence window, and one that doesn't survive the window was a portrait of luck — meaning the wall's delay costs nearly nothing when the idea is real and saves the system when it isn't. The one emotional week the doctrine explicitly warns about — do not switch profiles because of it — is exactly this mechanism at its most compressed.

• The sandbox's persuasive power is the reason for the wall, not an argument against it.
• Real superiority survives waiting. Only overfit brilliance needs to be adopted today.
• Contamination runs subtle: even letting a pending scenario tilt live discretion is influence outside the door.

What the wall preserves

One system of record, one interpretable baseline.

The wall's deepest service is epistemic. Because production rules change only through logged promotions, every stretch of the live record was generated by a knowable rule set — which is what makes the record analyzable at all. The benchmark comparison depends on it: simulated bands model the Standard's assumptions, and a live account quietly running sandbox behavior is being measured by a ruler built for a different system. Attribution depends on it: results belong to the configuration that produced them only if the configuration is documented. Even the scenario layer itself depends on it — alternatives are compared against the active profile, and an 'active profile' that shifts informally makes every comparison unanchored. Breach the wall and nothing crashes. Everything just quietly stops meaning anything.

The key idea

The freedom of the sandbox is purchased by the firmness of the wall.

The boundary is what makes aggressive experimentation safe: because nothing in the sandbox can touch production, everything in the sandbox is allowed — hostile stress cases, absurd aggression profiles, edge-change hypotheticals, ideas that would be reckless as live changes and are priceless as controlled questions. Systems without the wall must fear their own research and so do less of it, more timidly, with worse discipline. MARS inverts the trade: total freedom inside, one gated door out, and a production baseline whose every rule can answer the only question that compounds — how do you know?