Underwater weeks are a budgeted quantity, not an unlucky side effect.

The governance gap

The ladder measures how far, and nothing measures how long.

Every threshold in the gate ladder is a depth trigger. Deployment compresses as the hole deepens and releases as it fills, and the entire mechanism is indifferent to elapsed time — an account eleven weeks underwater at a shallow depth carries exactly the authority of one that arrived there yesterday. That indifference is correct, because time does not increase the arithmetic difficulty of recovery the way depth does. But it leaves a real phenomenon unmonitored, and the phenomenon is the one most associated with operators abandoning working systems.

The flag that fires alone

Inside the depth bands, past the duration percentile.

The specific condition worth naming is a drawdown comfortably inside its depth distribution and far past its duration one. Nothing about it is alarming on any depth-based instrument: the gate is shallow or neutral, the percentage is unremarkable, and every individual week is defensible. What has happened is that the account has spent a large multiple of the modelled recovery time failing to make a new high. The duration read is the only place this registers, and it registers as a flag rather than as a response — the finding is that something is not converting, not that risk should change.

What it usually means

Efficiency, not edge — and the distinction changes the investigation.

A long shallow episode most often points at conversion rather than expectancy. Deployment may be running below what the gate authorises, so recovery proceeds at a fraction of available capacity. Friction may be consuming the small positive periods that would otherwise compound into a new high. Branch mix may have drifted toward outcomes that neither lose much nor gain much. Each of these produces the same signature — an account that does not fall and does not rise — and each is diagnosable through throttle efficiency, fee drag, and mix attribution rather than through anything in the risk stack. The instinct to treat a long grind as an edge problem sends the investigation to the wrong department.

Why budgeting rather than triggering

The response is a review, and deliberately not an automatic compression.

Duration is monitored as a budget rather than wired to an automatic action, and that restraint is principled. An automatic de-risk on elapsed time would compress deployment exactly when the diagnosis usually calls for the opposite — an account grinding sideways because it is under-deployed would be pushed further into the condition producing the grind. Depth justifies automatic compression because deeper holes are arithmetically harder to escape. Time does not carry that property, so it earns a flag and an investigation, and the response is whatever the investigation finds rather than a reflex.

Reading it beside gate dwell

Long dwell with normal depth is the same finding from another table.

The duration read has a corroborating instrument. Gate dwell benchmarks report expected time in each capital state, and an account showing long Recovery dwell at normal depth is describing the same underlying condition through a different lens: extended residence in a compressed state without the depth that would justify it. When the duration percentile and the dwell benchmark agree, the finding is solid and the investigation is efficiency. When they disagree, the disagreement itself is informative — usually about how the high-water mark and the gate state have drifted apart across a period of small oscillations.

• Time-under-peak is compared to its own distribution, not inferred from depth.
• The flag fires while every depth instrument reads normal — that is its entire purpose.
• The response is a conversion investigation, never an automatic reduction in deployment.

The key idea

The failure with no alarm needs a budget, or it runs indefinitely.

Systems fail loudly in one dimension and quietly in the other. Depth has a ladder, a terminal boundary, and an automatic response, so depth failures announce themselves. Duration has none of that, which means a slow bleed can continue for as long as the operator's patience holds — and the operator's patience is the thing the episode is consuming. Giving underwater time its own distribution and its own flag is how the quiet failure acquires a voice, and it is the only instrument in the stack that speaks while everything else reads normal.