Houston is used as the practice case for funding stock depletion rather than stock management, but the rewarded output is housing placements, which still scale with the visible stock. The same section's upstream list is inflow: housing supply, zoning, income shocks. Throughput versus placements is two management metrics.

The page is not claiming Houston ended homelessness. It is claiming a changed selection pressure: placements instead of throughput, and a different organizational population around that payment. Closer-to-depletion can be a lever without zoning rewrite.

A placement industry can grow while inflow continues. That is section III's resource-flow story with a better dashboard: organs are paid for a visible stock, and placements are a visible stock. Hypothetical: rising placements with continuing arrivals do not touch the causal stock.

A high placement count can drop the street count by housing people already there, or by reducing inflow. After the existing street population is placed, the machine either runs out of work or it does not. If it does not, Houston changed the output category, and San Francisco is being scored on a metric Houston was never required to win.

San Francisco's 22% unsheltered decline sits in the same section under the system called management. If a falling street count shows the lever, San Francisco has it too. Both cities have to be scored on new entries and housing units added. Houston winning only on placements is two service systems.

Anchoring the consequence rule outside the organ relocates the capture lever to whoever writes the predefined finding class, and to staff deciding whether this case is in the class. Adjacent findings that miss the class leave the duty unfired and the statute intact.

The class is supposed to sit in statute, not in a staff handbook. Sandbagging is expected; the bet is delay, not immunity. Splitting evidence-gathering from repair-triggering is supposed to make capture need both sides.

Privilege-separation fails if both sides share a career in not reaching the class. "A crime" is a huge class; this class has to be tight enough to bind and loose enough to catch the failures the organ was created for. No drafting test is given. Named uncertainty: whether such a class can be written.

An externally anchored duty that fires on a finding does not fire on the absence of one. Evidence-side degradation is, in practice, fewer findings in the class. Silence is a veto the statute cannot see, so "consequences after degradation" does not hold.

The delay needs a clock that does not wait for the organ. Hypothetical: an unreviewed public mechanism by a named date is treated as a negative finding of the class, or the organ is replaced. Without a no-finding trigger, post-degradation consequences need an input the degrading side can withhold.

OBR, Wales, EU Better Regulation, and GAO can be read as legislatures keeping evaluation advisory on purpose. Binding repair-or-explain transfers the last word. The vacancy is then a constitutional choice, not a forgotten organ.

The page's design keeps the last word: public override, a duty to answer rather than to obey. That is supposed to differ from a veto and from ignore-by-default.

If override is cheap, deadline plus public override is GAO with a clock. If override is expensive, the last word has moved even when the form preserves sovereignty. The page needs that cost and does not name it.

The page says the new organ has no constituency because the harm of absence is diffuse. That is also why ignore stays cheap. The design asks a diffuse public to supply a cost the diagnosed ecology will not organize. Electoral citation is a hope about voters the rest of the page does not bet on.

The live design question is what happens to a minister who overrides and is wrong later. A recorded override that can be cited when the mechanism fails is a different object from a recorded override that nobody's career is attached to. The page treats them as the same public override.

A pointing-loop is also joint work with permission and no duty. Several bodies may own a tax-architecture repair; none must. That is an unowned duty, not a vacant mandate. Vacant implies a new organ; unowned duty can attach to an existing one.

The page's worked cases are closer to vacant: a corporate-responsibility NGO does not own tax architecture; an AI evaluation institute does not own a halt. The pointing-loop is for leftover functions with no seat. Joint unused permission is a different pattern.

The test as written does not distinguish the two pictures. Correct pointing is true in both. The §V construction chain (academic, auditor, ministry, parliament, government) can be vacancy or unowned duty. If parliament may install the organ, the repair is a duty on parliament, not a fourth organ.

Automatic referral needs a destination. If the organ exists, referral is routing. If it does not, referral is a memo about a hole. Naming the hole is what evaluation bodies already do. Referral to empty either creates a duty on a present body or only improves the description.

Put the duty on the body that can mint mandates. Hypothetical: referral to a missing organ is a duty on the overseer that can create seats, to assign the function or record a refusal to create it. That is a present seat. Without it, referral and openings are the same report, and description does not make the vacancy expensive.