A community’s power is not revealed by the number of meetings it convenes or the brilliance of its prototypes. It is revealed when an affected person can say stop, suspend a decision and obtain a response without paying for it through lost standing, time or safety. That possibility turns the future into a common practice: it requires authority to be distributed, evidence to be preserved and repair to be prepared before conflict arrives.

Many participatory initiatives promise to listen while reserving the final decision for an executive team, a supplier or a small technical group. A consultation gathers opinions and a workshop produces colourful notes, although the timetable, budget and definition of success are already closed. In that setting, the community provides information without gaining capacity. The process may feel courteous while leaving intact the power relationship it claimed to transform.

Saying stop does not mean granting an arbitrary veto over every change. It means defining a legitimate interruption when harm, significant uncertainty or a previously agreed breach appears. A stop must have known thresholds, a person or group authorised to trigger it, a review period and public conditions for restarting. Without those elements, the promise becomes decorative; with them, prudence becomes shared infrastructure.

The issue matters especially when a community rehearses technologies, rules or services that it does not yet understand well. A pilot distributes benefits, but it also distributes risks and invisible labour. Those receiving a benefit usually appear in the presentation; those absorbing errors, learning procedures or handling complaints rarely occupy the same slide. Mature governance asks from the beginning who can halt the trial and what will happen to the person who identifies a problem.

Saying stop reveals the architecture of power#

Every collective decision contains several distinct capacities: proposing, funding, implementing, evaluating, stopping and repairing. An institution can open the first capacity to participation while retaining the other five. Attendance and comment counts therefore cannot measure shared power. A useful map identifies who controls each verb, what information supports that action and to whom the decision maker must provide an explanation.

Every collective decision contains several distinct capacities: proposing, funding, implementing, evaluating, stopping and repairing.

The power to stop feels uncomfortable because it exposes dependency. If only a supplier can switch off a system, the community does not govern that system. If management can ignore an alert without justification, an oversight committee does not truly supervise. If a family must prove irreversible harm before it is heard, the complaint channel arrives too late. A formal distribution of roles is no substitute for effective and accessible authority.

There is also a difference between pausing and abandoning. A pause protects time for understanding, reducing exposure and bringing in missing voices. It does not prejudge the outcome or turn every doubt into failure. On the contrary, it enables a project to survive its mistakes without hiding them. When the team knows that interruption has a procedure and boundaries, it can report early, before a small incident becomes an institutional crisis.

A right to stop needs protection from retaliation. An employee, student or volunteer is unlikely to warn about a risk if shifts, opportunities or recognition may be lost. Safeguards should include proportionate confidentiality, a record of the alert, an explicit ban on punishment and an independent route when the direct manager is part of the problem. Power appears not only in the word stop, but in the possibility of uttering it without becoming isolated.

Case 1: a school halts automated attention scoring#

A secondary school decides to test an application that estimates student attention during digital activities. The supplier presents simple charts, and the leadership team hopes to identify difficulties earlier. For two weeks, teachers receive scores for each class. Several figures look useful, but one tutor notices that the system penalises pauses associated with screen readers, language changes and unstable connections.

The initial agreement had created a small safeguards group representing students, families, counselling, teachers and data protection. Any member could request a seventy-two-hour pause after documenting a systematic difference or a risk to rights. The tutor triggers the mechanism with five anonymised examples. The application stops producing new scores while the group reviews its purpose, not merely its accuracy.

The inquiry uncovers a problem deeper than a technical error. Nobody had demonstrated that cursor movement or response time validly described attention. In addition, scores had begun to influence assessment conversations even though the pilot prohibited that use. The group hears students with different needs, compares records and demands information about limitations from the supplier. School leaders acknowledge that they had confused the availability of data with educational legitimacy.

The school does not restore the classification feature. It retains only aggregated indicators of access failures, deletes individual scores and communicates the decision to the whole community. It also offers a review to students mentioned during the trial. The outcome is not framed as a rejection of innovation, but as institutional learning: the stop signal revealed that the original question was poorly framed and prevented an unfair practice from becoming established.

The case leaves four lessons. The threshold had been defined before the dispute; the reporter did not need to prove harmful intent; the review included those living with the consequences; and the exit covered repair, not only shutdown. Without those conditions, the same observation would have been reduced to a technical ticket handled privately by the company.

Case 2: a neighbourhood pauses a heat adaptation plan#

A local authority and several neighbourhood associations trial a plan to adapt streets and climate shelters. A model combines temperature, vegetation, average age and distance from facilities to prioritise investment. The first proposal directs shade and fountains towards high-footfall areas. At an assembly, carers from a housing complex point out that the map omits daily routes to a health centre and several bus stops without benches.

The participatory process had given a neighbourhood panel the power to freeze up to ten per cent of the budget before construction contracts were awarded. The panel uses that capacity and requests accompanied walks at three different times. The decision does not stop the entire plan; it protects a meaningful portion while checking whether missing data have erased needs. The authority publishes the reason for the pause and adjusts the procurement timetable.

The walks show that average temperature does not describe actual exposure. Older people wait beside heat-radiating surfaces, adults pushing buggies choose longer accessible routes, and the official shelter closes during part of the afternoon. Administrative data were accurate, but the unit of analysis was inadequate. Situated experience contributes precision that the model could not obtain from an office.

The panel agrees on benches, temporary shade, longer opening hours and a revised priority rule. A later evaluation measures pedestrian-level temperature, waiting time and use by time band, as well as the number of interventions. The frozen budget is released when the corrections are contracted. The pause demonstrates that a partial boundary with a duration and purpose can improve public action without paralysing it.

The case also avoids romanticising participation. Carers contributed hours for which other participants received a salary. The new protocol provides compensation, childcare and accessible materials. If a community must provide unpaid labour to correct professional decisions, the supposed sharing of power remains unequal. Authority to stop requires resources if it is to be exercised by people with the least available time.

Case 3: a cooperative challenges algorithmic shift allocation#

A delivery cooperative introduces a system that proposes shifts according to demand, past availability and performance. Its aim is to reduce calls and balance workload. After the first month, several members notice that rejecting an urgent shift appears to reduce future offers. The supplier denies any direct penalty but cannot explain how every model variable interacts.

The internal rules require any mechanism with a material effect on income to have a comprehensible explanation and a manual alternative. An elected committee can suspend it by simple majority for fifteen days. Faced with the initial evidence, the committee halts automated allocation, preserves necessary records and temporarily returns to a published rotation. Nobody loses priority for filing the complaint.

The audit finds an indirect correlation: accepting urgent work improved an availability measure that later carried weight in the recommendation. The system contained no instruction labelled punishment, but it produced that effect for members with caring responsibilities or dependence on public transport. The cooperative analyses income distribution, interviews affected groups and tests scenarios before redefining the rule.

The solution separates availability from performance, limits permitted variables and allows every proposal to be challenged through a specific explanation. For three months, the committee receives a weekly disparities report and retains authority to stop the system again. The supplier accepts a clause covering technical access and response deadlines. The cooperative learns that purchasing a tool does not transfer responsibility for work organised through that tool.

This third case shows the importance of a safe fallback state. The organisation could return to a familiar rotation because it had preserved and documented it. A system that removes every alternative before proving reliability turns a stop into an existential threat. Rehearsing the future requires a temporary capacity to operate without the new arrangement.

How to design stopping authority that works#

The first component is scope. The agreement should state what may be paused: a recommendation, data collection, a payment, procurement or the whole project. The greater the possible harm, the more direct the power of interruption should be. A scheduled review may suffice for low-risk activities; effects on rights, income, health or service access demand a rapid response.

The second component is the threshold. Some signals are objective, such as a disparity above an agreed value, a security incident or data used beyond their stated purpose. Others require judgement, such as an incomprehensible explanation or a pattern of humiliation. A protocol can combine both without pretending to mathematical certainty. It should accept reasonable indications and place the burden of investigation on the organisation, not the affected person.

The third component is the trigger’s legitimacy. Authority can sit with several figures to avoid a single point of silence: operational staff, user representatives, protection officers, ethics committees or public officials. An outside channel becomes necessary where conflicts of interest exist. No design is neutral if it requires permission from the very party that benefits when a project continues.

The fourth component is the state maintained during a pause. Protection of people, custody of evidence, minimum service continuity and communication must be defined. Switching off without an alternative can shift harm to other groups; continuing operation while debate proceeds can make harm irreversible. Preparation identifies a safe mode, limits new cases and retains only the information essential for review.

The fifth component is the exit. Restarting requires verifiable criteria, accountable roles and a public explanation. Cancellation requires contracts, data and repair to be managed. Modification requires the hypothesis to be tested again. Every decision should leave a record of evidence, disagreement and commitments. Institutional memory prevents the same risk from returning when personnel change.

Rehearsing the future without turning people into test material#

A legitimate trial begins with a concrete question and a limited scale. It declares what knowledge is missing, how long the test will last and which groups may bear unequal consequences. It does not call a permanent deployment a pilot or use novelty to reduce safeguards. Uncertainty justifies greater care because experience is not yet sufficient to estimate every effect.

Participation should produce observable decisions. It may determine criteria, exclude uses, reserve budgets, select indicators or stop stages. The convening authority explains which issues remain outside the process and why. That honesty is more democratic than an unlimited invitation whose real influence cannot be traced. Feedback shows which contribution changed which decision and which proposal was not accepted.

Learning needs diversity and legible conflict. An average can conceal that one group’s benefit depends on another group’s sacrifice. Results are therefore disaggregated where relevant and accompanied by testimony, withdrawals, complaints and care costs. Dissenting voices are not used as decoration; they receive a reasoned response and can trigger review when they reveal an unforeseen risk.

Repair is prepared before harm. The team identifies how it will correct a record, return a payment, restore access, compensate time or provide support. It also assigns budget and authority. A channel unable to change an outcome merely administers frustration. Trust does not come from a promise of perfection, but from evidence that a practical route exists when a decision fails.

Evidence, review and public accountability#

Evaluation should examine at least four dimensions: outcome, distribution, experience and capacity for correction. Outcome asks whether the intervention achieves its purpose. Distribution observes who receives benefits and risks. Experience examines dignity, understanding and burden. Correction measures how long an alert takes to alter the course. A project may improve an average while failing on the other three dimensions.

Every review needs a date and an expiry point. Social conditions, suppliers and uses change. An authorisation granted for a test should not become indefinite permission through inertia. The schedule may include early reviews, an independent assessment and an explicit final decision. If evidence is missing, continuation is not presumed; it must be justified again.

Publication should protect privacy without erasing responsibility. It can explain purpose, rules, measures, incidents, pauses and repairs through aggregated data while documenting who decided. It should also acknowledge uncertainty and disagreement. Communication that only celebrates success obstructs learning and penalises the person who warns. Useful transparency enables questions to be asked and responses to be followed.

Shared power does not remove leadership; it makes leadership accountable. Managers retain duties, but they accept limits, listen to distributed knowledge and answer for decisions. The community does not replace specialists, although it can challenge assumptions and set values. When someone can say stop and the system knows what to do next, the future ceases to be a spectacle presented to citizens and becomes a practice that citizens can govern.

Institutional sources for further reading#

These sources converge on an operational idea: participation is more than speaking, and oversight is more than watching. A community gains power when it has information, resources, authority and remedies capable of changing the course of a decision. Saying stop then becomes a form of public care and a condition for rehearsing futures that can belong to everyone affected.