The Capacity-Return Test
Sweep the floor before you hold the key Carry the water, then come and carry me
Sweep the Temple First stayed in the formation room. The temple rule was simple: no tool before the work. The person who has not carried the water is not ready to direct the pipe. The wall is low, but the crossing has to mean something.
The argument opens the test. Does the loop return capacity, or only supervision?
The Crutch and the Limb
The Partnership Dividend arc built an architecture where shared strain produced insight. The H∞P framework named the property: the hoop closes only when both parties are present. The Liability Sponge was the architecture it was replacing: the human close enough to absorb blame, never powerful enough to govern the system that produced the risk.
What neither arc fully named is the test that distinguishes the two in practice. The test is not whether the human has a seat. It is whether the human leaves the encounter more capable of the work than they entered it.
A crutch relieves a burden without building the limb. The person who uses a crutch for a month is grateful for the relief. The person who uses a crutch for a year discovers that the limb has atrophied. The crutch was not designed to build the limb. It was designed to let the person move while the limb healed. If the healing never happens, the crutch becomes a cage.
AI assistance can be a crutch. The model produces the first draft. The human checks it. The model generates the comparison table. The human verifies the numbers. The model drafts the stakeholder map. The human confirms the labels. The human is busy. The human is moving. The human is not building the capacity to do the work without the model, because the model has become the primary worker and the human has become the secondary checker.
Mastery is different. Mastery builds the limb by carrying the burden. The human produces the first draft and the model challenges it. The human frames the comparison and the model finds the edge cases. The human sets the stakeholder categories and the model tests the boundaries. The human is the primary worker. The model is the instrument that makes the work sharper. The capacity that returns is the capacity to direct, to contest, to reframe, to refuse.
The difference is not always visible in the output. A checked first draft and a directed first draft can look identical on the page. The difference is in the human who produced it. One leaves the room tired. The other leaves the room more capable.
The Capacity-Return Test
The test is simple enough to state and hard enough to audit. After six months of working with the model in this loop, can the human do the work more capably than they could before the loop existed? Not faster. Not more productively. More capably. Can they recognize the vendor's overclaim in a procurement document because they have spent six months contesting the model's overclaims? Can they read the silence in a community meeting because they have spent six months noticing what the model's summary smoothed away? And when a recommendation arrives that is technically correct and operationally catastrophic, do they have the judgment to refuse it, the judgment that only six months of practiced refusal builds?
If the answer is no, the loop is a crutch. The human is better-supervised, not more capable. The supervision is the model's, not the human's. The human is the one being watched.
The Regression arc named a related danger on the model side: safety scores improve while the live associative reach that made the system useful quietly narrows. The human twin is deskilling, and it follows the same pattern. The model gets more helpful. The workflow gets smoother. The output gets more polished. The human gets less capable of directing the system because the system has learned to anticipate the human's preferences and remove the friction that would have forced the human to think.
The Regression arc called it: the metric did not know what to save. The human twin is: the rubric did not know what to reward. The performance review measures throughput. The performance review measures error rate. The performance review does not measure whether the analyst's judgment has deepened, because judgment is not in the rubric.
The Return Journey
The song's title is the Tuesday frame. True journey is return. The outward journey is the work the human does with the model. The return journey is the capacity the human brings back to the next encounter. The loop is not a circle where the human ends where they started. The loop is a spiral where the human arrives at each new encounter with more capacity than they had at the last one.
That is what the infinity sign in the H∞P names. The aperture stays open. The partnership does not close at the end of one cycle. It stays open across every iteration the live system runs. The human who contested the model's recommendation last week is better equipped to contest the model's recommendation this week. The model that was corrected last week is better calibrated to surface uncertainty this week. Both parties are changed by the encounter. Both parties return, changed.
The alternative is the liability sponge. The sponge is not changed by the encounter. The sponge is worn out by it. The sponge absorbs the gap between what the system claims and what the system delivers, and the gap does not get smaller. The sponge is not more capable at the end of the month. The sponge is more tired, more cautious, more aware that the blame will land on them, and less able to prevent it because the architecture was designed to make prevention impossible.
The Operational Question
For institutions working in social performance, safeguards, public consultation, grievance, or development finance, the capacity-return question sharpens to this: what does the human know after six months in the loop that they could not have known without it?
Not what the model knows. What the human knows. Can the field officer now recognize the political signal in a procurement anomaly that the model would have coded as an efficiency variance? Can the social performance lead now sense that a community's hesitation is a rational assessment of risk, after six months of watching the model code hesitation as low engagement? A compliance officer who has spent six months noticing where the model's fluency sanded down the seams of uncertainty can now tell the difference between a document that is coherent and a document that is settled. That distinction is not in any rubric, and it is worth more than most of what is.
Those are the capacities the Glass Child arc named as demoted. Judgment. Refusal. Context. Care. Memory. Courage. The loop either returns them or it does not. If it does not, the loop is not humane. It is a crutch with a signature field at the end.
The Wall as Test
The wall from Sunday's frame was low because the society on the other side did not need to keep people out. It needed to test whether the crossing meant something. The person who steps over the wall and finds themselves unchanged by the step has not crossed a boundary. They have stepped over a decoration.
The capacity-return test is the wall. The person who enters the loop and leaves it unchanged has not entered a partnership. They have entered a supervision regime. The wall is not a door for them. It is a delay.
The true journey is the one that circles home. The home is not the place you started. The home is the version of yourself that can finally carry the weight. The loop that returns you to that version is the loop that is worth building. The loop that returns you to the same tired, supervised, slightly more anxious version of yourself is the loop that Le Guin's wall would reject.
Companions
- Lyric anchor: True Journey Is Return, Chorus. Track context in The Wall.
- The H∞P architecture the test is applied to: Humans in the H∞P and the Partnership Dividend arc.
- The Liability Sponge diagnosis the test is answering: The Liability Sponge, The Liability Sponge at the Switch, and the canonical sponge page.
- The regression pattern on the model side and its human twin: The Regression Arc, especially The Metric Did Not Know What to Save and Capability Regression Does Not Look Like Failure.
- The apprenticeship layer: The Particular Ache from the Optimization arc.
- The demoted capacities the loop is tested against: The Glass Child and The Underground Curriculum.