Which tiers actually earn their risk.

The question

Authorization and conversion are different verdicts.

A tier can be perfectly authorized — evidence strong, gate permissive, allocator resolved by the book — and still convert poorly, if the conditions that earn high-tier authorization aren't the conditions where the edge actually pays best. Attribution separates the two verdicts by slicing results per tier: trades and cycles taken at each rung, the R they produced, and the risk they consumed producing it. The benchmark's tier-usage and deployment distributions supply the expected shape — where governed futures spend their deployment time and what each zone contributes — and the live slices get read against that expectation. The output isn't a grade on the ladder; it's a map of where the ladder's authorizations are being converted and where they're leaking.

The zone expectations

Each zone of the ladder has a different job — and a different test.

Attribution reads differently per zone because the zones exist for different reasons. T3–T4 is the workhorse middle where a healthy allocator spends most of its time; it should carry the bulk of both deployment and return, and its conversion efficiency is effectively the system's baseline. T5 is earned acceleration — gate-permitting, evidence-stacked — and its test is whether the acceleration actually accelerates: per unit of deployed pool, T5 cycles should out-earn the workhorse zone, or the 'earning' of the tier isn't being paid out. T6–T7 are rare by design; their sample will be thin for a long time, and the honest read is directional. T1–T2 are survival tiers whose job is rhythm and containment, not return — attributing 'weak performance' to them is a category error; their test is whether they kept cycles small and the operator in motion during defense.

The signature failures

Three leak patterns attribution reliably catches.

The first is inverted conversion: high tiers deploying meaningfully more risk per cycle while returning no more R than the workhorse zone — the acceleration is consuming its own advantage, usually a sign that the evidence conditions triggering high tiers aren't the market conditions where the edge widens. The second is workhorse erosion: T3–T4 conversion drifting below its own history while the exciting tiers look fine — dangerous precisely because the middle carries most of the deployment, so a small leak there outweighs a large one at T7. The third is defensive bleed: survival tiers producing outsized losses relative to their tiny pools, which almost always traces to execution quality degrading under pressure rather than to the tiers themselves — the sizing shrank but the discipline shrank faster.

• Judge every tier against the benchmark's expectation for that tier — never against the flat average of all tiers.
• Weight leaks by deployment share: a 10% efficiency leak in the workhorse zone outranks a dramatic one in a tier used twice a quarter.
• Thin-sample zones get directional notes, not verdicts. T6–T7 conclusions ripen in years, not months.

What attribution feeds

Findings route to diagnosis and the sandbox — never to live ladder surgery.

Attribution is an evidence producer, not an authority. A confirmed conversion leak doesn't rewrite pool rows or re-space tiers on the spot; it generates a diagnostic question — is the leak execution, conditions, or structure? — and, if structural, a candidate change that enters the scenario pipeline like every other candidate: formulated with expected behavior, simulated, held against live evidence through a persistence window, and promoted only on a documented verdict. The ladder's parameters are production rules, and production rules change through the promotion workflow, not through one quarter's attribution table. What attribution does immediately affect is attention: it tells the weekly review where to look, which is most of what an early-warning instrument is for.

The key idea

A ladder that isn't attributed is a ladder on faith.

Governance answers 'was this deployment allowed?' — attribution answers 'was it worth it?' — and a system running only the first question is trusting that authorization and conversion coincide, which is exactly the kind of assumption MARS exists to retire. Sliced per tier, read against benchmark shape, weighted by deployment, and routed through the sandbox, attribution closes the loop between the ladder's design and the ladder's results. The rungs that earn their risk get documented. The rungs that don't get investigated. Nothing gets assumed.