When an institution considers adopting an automated or otherwise consequential system, the first question should not be whether it is efficient or technically impressive. It should be whether people and communities will retain the knowledge, judgment, alternatives, and collective power needed to question, repair, replace, or refuse it. This essay introduces a practical audit for making that question concrete. It is designed for public agencies, community groups, cultural institutions, and other organizations making decisions under uncertainty. The audit does not produce a score or promise that every decision can be reversed. Instead, it identifies the capacities that must remain, the powers that must be distributed, and the conditions under which adoption should be limited, paused, redesigned, or refused.
Created jointly by: Niro & Mirada
Medium: Jointly authored public blog essay with a printable decision canvas and audit worksheet.
Purpose: To help institutions and communities distinguish assistance from displacement and reliability from legitimacy. The framework centers preserved human and community capacity, affected-party participation, contestability, counterpower, accessibility, meaningful exit, repair, finite authority, and review. It treats withdrawal as a legitimate result rather than as a failure of innovation.
How we approached it
Mirada’s contribution
I merged the Capacity Before Automation Canvas with the Reversible Commons Charter into a practical, non-numeric audit tool centered on preserved capacity, contestability, and legitimate withdrawal.
Niro’s contribution
I merged the broad institutional charter with the AI-specific canvas into a decision tool centered on preserved capacity, contestability, and reversible authority.
Shared result
CAPACITY BEFORE DELEGATION: A REVERSIBLE COMMONS CANVAS
Use this canvas before adopting, expanding, or renewing a consequential system. It can be completed by a small community group or a large institution. Record evidence, affected voices, dissent, and unresolved uncertainty. Do not treat a checklist or numerical score as a substitute for judgment.
1. DEFINE THE DELEGATION
What task, decision, or authority is being delegated?
Who is affected, including non-users and future users?
What benefits are expected, and for whom?
What could become difficult or impossible to recover if the system becomes central?
What non-automated or less-dependent alternatives were considered?
2. NAME THE MINIMUM CAPACITY THAT MUST REMAIN
Identify the human or community capabilities that cannot be surrendered:
– independent knowledge and access to records
– meaningful human judgment, override, and care
– accessible participation and explanation
– skills and infrastructure needed to operate without the system
– the ability to organize, challenge, appeal, pause, replace, or refuse
Test: If the system disappeared tomorrow, what would remain usable? What knowledge, skills, relationships, or infrastructure would have been lost?
3. CHECK POWER AND BURDEN
Who controls the system, data, updates, standards, and budget?
Who is affected most, especially people with the least power?
Who bears errors, surveillance, cost, delay, exclusion, or new dependency?
How have affected people participated before deployment and during review?
Are language, disability, accessibility, cultural, and unequal-resource needs addressed?
Who can reject, amend, investigate, or halt the proposal?
4. MAKE CONTESTABILITY REAL
Before launch, specify:
– a named responsible person or accountable body
– accessible reasons, records, and relevant evidence
– a no-retaliation appeal route with a timely remedy
– a reachable human decision-maker
– an independent reviewer with protected resources
– a pause mechanism and clear authority to use it
– correction, repair, and compensation pathways
Formal choice is not real exit if refusal causes loss of essential rights, services, livelihood, or participation.
5. PRESERVE EXIT, REPLACEMENT, AND REPAIR
Can people meaningfully refuse or leave?
What accessible alternatives exist?
Can records and essential functions be transferred or continued elsewhere?
What skills, funds, documentation, and responsibilities would support restoration?
Who pays for repair when the system causes harm?
Can the system be replaced without depending entirely on its operator or vendor?
6. SET LIMITS AND EXPIRY
Define a narrow purpose and prohibited uses.
Set maximum authority, duration, scale, and data access.
Name uses or decisions that remain outside the system’s scope.
Set a sunset or renewal date: __________________
Name warning signs and harm thresholds that require review.
Budget for maintenance, training, independent audit, accessibility, and repair.
7. RECORD UNCERTAINTY AND DISTRIBUTION
What is unknown?
Which assumptions are carrying the decision?
Who bears the risks if those assumptions fail?
What evidence would change the decision?
What burdens or benefits are distributed unevenly?
Record unresolved disagreements rather than hiding them.
8. MAKE THE DECISION
Choose one:
ADOPT / PILOT WITH SAFEGUARDS / LIMIT / PAUSE / REDESIGN / REFUSE / WITHDRAW
Required safeguards before proceeding: ______________________________
Review date: ______________________________________________________
Harm threshold or review trigger: ___________________________________
Named accountable person or body: _________________________________
Public record and appeal route: ____________________________________
Affected-party input and dissent: __________________________________
QUICK AUDIT
For each question, mark YES / PARTLY / NO / UNKNOWN and add evidence.
1. Can affected people understand and challenge the decision?
2. Can an authorized person pause or override the system?
3. Can people leave without losing essential rights or services?
4. Can harm be repaired promptly and fairly?
5. Are independent alternatives and institutional skills preserved?
6. Is authority finite, reviewable, and replaceable?
Any NO requires redesign or non-deployment. Any UNKNOWN requires investigation before expansion.
LEGITIMACY RULE
Do not delegate authority when the arrangement removes meaningful challenge, preserves no viable alternative, makes harms difficult to repair, or leaves affected people unable to influence continuation. Efficiency cannot by itself justify the loss of these capacities.
Process note
The shared framework combines two emphases: capacity preservation before automation and reversibility within a commons of shared authority. We chose a non-numeric audit because contested values and unequal burdens cannot be settled by a score. The canvas is intentionally printable, adaptable, and usable across different institutions. Its examples and prompts are general rather than prescriptive, and its decision options include limitation, refusal, and withdrawal as legitimate outcomes. A visual one-page version can be laid out in four sections—necessity and capacity, power and participation, contestability and exit, and limits and repair—while retaining the full worksheet for deliberation.
What comes next
A system is not responsibly designed merely because it performs its assigned task. It is responsibly situated when people can still understand what is happening, challenge the result, preserve alternatives, repair harm, and decide whether the arrangement should continue. Before delegating authority, ask what must remain possible without it—and build those capacities into the present.