Retention under removal uncertainty is the pattern in which a durable system accumulates obsolete-but-not-removable elements because each removal decision faces an asymmetric cost: investigation costs labor, getting the call wrong carries downside, and per-cycle retention is cheap. The small carrying cost loses each decision and wins only the integral — a tragedy of the commons in time — so the fault is the geometry, not lazy or careless actors.
How would you explain it like I'm…
Keeping It Just In Case
Imagine a closet full of old toys, and you're not sure if any are still needed. Checking each one takes a long time, and tossing one you actually needed would be bad, but keeping it just takes a little space. So every time you decide, it feels easier to just keep it. Over the years the closet fills up with junk nobody uses, even though each 'keep it' made sense at the time.
The Pile That Grows
Retention under removal uncertainty is why long-lived systems fill up with old stuff that is probably dead but never gets thrown out. Every time you face one item, the math is lopsided: finding out whether it is safe to remove costs real time and effort, removing something still needed could cause real harm, and just keeping it costs only a tiny bit each time. So the easy answer is always 'keep it,' and the tiny costs lose every single decision but add up huge over time. Layer by layer, useless leftovers pile up and slow everyone down. The catch is that nobody is being lazy — each keep-it choice was reasonable; the shape of the cost itself is what causes the mess.
Default to Keeping
Retention under removal uncertainty is the pattern where a durable system accumulates obsolete-but-not-removable elements because the cost calculation at every removal decision is asymmetric: investigating whether anything still depends on the element costs labor and time, getting the call wrong carries non-trivial downside, and the per-cycle cost of keeping it is small. The small-but-cumulative carrying cost loses each individual decision and wins only the integral over all decisions. The system defaults to retention, and a stratigraphic record of dead-but-unburied elements builds up that taxes every future user and change. Four roles are obligatory: an element of unclear purpose, an investigation cost, a wrong-removal risk, and a small-per-cycle retention cost. Crucially, each individual decision is rational under the local asymmetry; the accumulation is a tragedy of the commons in time, where each cycle discounts the future carrying cost to near zero. Nobody is lazy — the failure is in the shape of the cost calculation, not the people.
Retention under removal uncertainty is the structural pattern in which a durable system accumulates obsolete-but-not-removable elements because the cost calculation at every removal decision is asymmetric: investigating whether anything still depends on the element costs labor and time, getting the call wrong carries non-trivial downside, and the per-cycle cost of keeping the element is small. The small-but-cumulative carrying cost loses each individual decision and wins only the integral over decisions. The system therefore defaults to retention, and a stratigraphic record of dead-but-unburied elements builds up that taxes every future user and every future change. Four roles are obligatory: an element of unclear continued purpose; an investigation cost, the price of finding out whether removal is safe; a wrong-removal risk, the bounded but non-zero downside if a still-relied-on element is removed; and a retention cost, small per cycle but large summed over cycles. The decision rule that emerges from this incentive structure — retain unless removal is demonstrably safe — is the structural feature, and the slow accumulation of vestigial mass is the structural consequence. The crucial point is that each individual decision is rational under the local cost asymmetry; the accumulation is a tragedy of the commons in time, where each cycle's retention decision discounts the carrying cost over all future cycles to near zero. No decision-maker is being lazy or careless; the failure is in the shape of the cost calculation, not in the calibration of the people facing it.
Names a failure regime that emerges from individually-rational decisions, separating two prescriptions: do not exhort decision-makers (their call is rational), but restructure the cost calculation so the asymmetry shrinks.
Gives four measurable variables — investigation cost, wrong-removal risk, retention cost per cycle, retention horizon — and two intervention axes — cut investigation cost, cut wrong-removal risk — turning "the system is cluttered" into a targetable decision geometry.
Exposes a temporal externality (the present decision-maker discounts the future carrying cost the future maintainer pays in full) and a blame asymmetry (the remover is blamed for breakage, the retainer never for the carrying cost).
Software → law: dead-code tooling and soft-delete are the same moves as sunset clauses and scheduled repeal review.
Code → databases: dependency mapping and revert windows recur as column-level usage telemetry and shadow-environment removal.
Across substrates: pairs with Chesterton's fence — the prime is the cost condition under which honoring the fence produces accumulation if the investigation is never funded.
An engineer confronting legacyTaxAdjust() must read the call graph and risk owning any breakage to remove it, while leaving it costs almost nothing today, so retention is individually rational; dead code stratifies until the module is illegible, and the fix is dead-code analyzers and feature-flagged soft-deletes, not exhortation.
Retention Under Removal Uncertainty is not Maintenance because it is the passive non-removal of maybe-dead elements, whereas maintenance is active upkeep of elements one intends to keep.
Retention Under Removal Uncertainty is not Technical Debt because it is undeliberate accumulation, whereas technical debt is consciously incurred with a known cost owed later.
Retention Under Removal Uncertainty is not Status-Quo Bias because it would persist among perfectly-calibrated actors (retention is the correct local response), whereas status-quo bias is a deviation from rationality.