Familiar projects that improve the quality of an exercise of power while leaving the structure of that power untouched, and the questions that replace them.
The criterion
A project is low-leverage when it makes the institution better at exercising power instead of asking whether, how, and under what conditions it should exercise that power at all.
None of the projects in the catalog below is worthless in itself. Metric precision can matter. Explanation can matter. Participation can matter. Each becomes a sink when it substitutes for the structural question — when enormous effort improves the quality of an exercise of power while leaving the structure of that power untouched.
The recurring mistake takes one form:
Make the institution better at deciding, instead of asking whether, how, and under what conditions it should decide.
The catalog
Each card names the pattern and the question that restores leverage.
Effort flows into accuracy, coverage, and personalization as though enough fidelity will dissolve the political problem. It will not: even perfect prediction would not itself establish jurisdiction.
Replace with: Preserve standing for when representation fails — or when it succeeds but should not be authoritative.
The model of a person is not the person →Years can go into making the ledger more precise while the ledger's boundary stays untouched. Accurate accounting within a false boundary still produces a false account.
Replace with: Whose costs does the metric exclude, and who carries what it cannot see?
Falsified denominator →Dashboards, model cards, and disclosure notices explain the system while leaving it unchallengeable. Knowing why you were harmed is not being able to reverse it. Legibility is not standing.
Replace with: What state can the explanation change, and who can trigger that change?
Law VIII: observability without state transition is theater →A human clicking approve at the end of a machine-generated workflow, without time, information, authority, protection, or reversal, does not constitutionalize the system.
Replace with: Can anyone interrupt, override, investigate, and change the rule?
Meaningful control →Workshops, codes, and exhortations improve conduct while people remain dependent on somebody's goodwill. Better character is not a substitute for better structure.
Replace with: Which dependency can be redesigned so virtue is no longer the load-bearing part?
Dependence without standing →Citizen panels, employee feedback, and stakeholder workshops consume enormous energy while decision rules stay intact. If participants can force nothing, the engagement has absorbed the dissent without changing anything.
Replace with: What can participants force the institution to reconsider?
Corrective standing →A workflow that gets faster by pushing friction onto patients, claimants, contractors, or support staff has optimized burden transfer. Do not optimize a system until you know where its residuals go.
Replace with: Where do the residuals land, and does the ledger count them?
Absorption as concealment →Years negotiating principles — fairness, beneficence, responsibility — while deployments move into progressively more intimate domains. A technically excellent system performing an unjustified function is badly designed.
Replace with: Why is this institution entitled to make this inference or decision at all?
What does not convert →Repeated residuals get case-by-case repairs, better support, more patches. At some threshold, many exceptions are evidence against the rule or representation that produces them.
Replace with: Has the exception pattern been read as evidence about the architecture?
Corrective Learning evals →Whether the machine is agentic, aligned, or controllable matters less than the institutional relationship built around it. A perfectly obedient algorithm can carry extraordinary unanswerable power.
Replace with: Machine obedience is not human non-domination. What holds the operator answerable?
Democratic vs. coercive governability →One more dataset, one better model, one more sensor — imagined as finally removing the need for judgment, contestation, and exception handling. The mature goal is not complete representation.
Replace with: Does the system remain legitimate when it inevitably does not know?
Safe incompleteness →Two absences are deliberate. Nothing here says expertise is worthless — expertise stays authoritative within the jurisdiction it is qualified to answer, and loses only the conversion from knowing more to deciding. And nothing here says efficiency is worthless — the objection is to efficiency claims whose denominator excludes the people doing the absorbing. See also abstention, the capability of declining to decide, which most of these projects substitute for.
The permanent question
A project has restored leverage when getting someone wrong is no longer final: the exception can travel back into the rule that produced it, the challenge is affordable to the person who carries it, and the decision stays reopenable — which is what non-finality names. The machinery on this site exists to make that standard checkable rather than aspirational.
The underlying project — what domination means here, why no arrangement is exempt by construction, and the compact form of the whole discipline — is stated on the about page. The epistemology: the excluded case and consequential contradiction; the design stance: safe incompleteness.