Sortition
A governance function is established. The executive board commissions the platform organization to propose a list of suitable members. The platform lead nominates three senior engineers from the platform team, two architects from consuming teams who have historically been cooperative, and one external consultant who advised on the platform’s initial design. The list is approved without modification.
The governance body is now populated by people selected for proximity to the platform, not for independence from it. The negative criterion established earlier in this series (governance legitimation must not depend on the platform’s selection) is violated in the first act. No one notices, because the selection process looked like competence-based appointment. It was. Competence was the criterion. Independence was not.
Three mechanisms and their failures
The question is how the governance function is populated in a way that satisfies the negative criterion: its legitimation must not depend on the platform’s selection.
Three conventional mechanisms are available.
Appointment. A senior leader selects governance members based on expertise and judgment. The failure mode is reproduction: whoever appoints selects people whose judgment aligns with their own. The opening vignette is this mechanism in action. Appointment satisfies the competence requirement. It violates the independence requirement.
Election. The engineering population votes for governance representatives. The failure mode is campaigning: election selects for people who are good at being elected, not for people who are good at governing. In an engineering organization, election tends to produce candidates with high visibility (conference speakers, internal advocates, prolific communicators) rather than candidates with the judgment and distance required for rule-setting. Election also creates the wrong incentive: accountability to the electorate’s current preferences, when the function must sometimes impose obligations the electorate would prefer to avoid. A governance body elected by the teams it governs has a structural interest in popularity. It will tend to avoid mandates that are costly to teams, even when those mandates serve organizational legibility or options preservation. Election satisfies the independence requirement (the platform does not choose). It undermines the function’s ability to impose legitimate obligations.
Co-optation. The existing governance body selects its own successors. The failure mode is closure: the body reproduces itself. Over time, it converges on a type. Dissenting perspectives are filtered out through the natural preference for colleagues whose judgment resembles one’s own. The body becomes a guild: internally coherent, externally opaque, accountable to its own standards rather than to the population it governs. Co-optation satisfies neither independence nor representativeness. The body’s composition depends on the body’s preferences; it reflects itself, not the governed.
Each mechanism fails the negative criterion in its own way.
What sortition is
Random selection from a defined pool, with limited terms and mandatory cooldown before re-eligibility.
The pool: all engineers above a seniority threshold (senior engineer and above, or equivalent), excluding current members of the platform team and current members of the audit function. The exclusions enforce the negative criterion directly: no one whose current role creates a structural interest in the governance body’s decisions can serve on it.
The term: fixed duration (twelve to eighteen months), staggered so that the body never turns over entirely at once. Continuity through overlap, not through permanence.
The cooldown: a minimum interval (equal to the term length) before a former member becomes eligible again. This prevents consolidation through repeated service. Drawn members may decline; the mechanism accepts the self-selection this introduces as less distorting than any selection-based alternative.
The mechanism: drawn, not chosen. Judgment requires distance from the power it legitimates. Selection by the platform produces alignment with the platform. Selection by the engineering population produces accountability to current preferences. Self-selection produces continuity. A governance body drawn at random from a qualified pool is structurally independent of the selection mechanism. Its members still have positions, loyalties, and preferences. What they lack is a debt to whoever put them there.
What sortition produces
Four structural properties distinguish sortition from the three conventional mechanisms.
It interrupts selection logic. No actor chooses who governs. The platform cannot select allies. The engineering population cannot elect advocates. The body cannot perpetuate itself. The absence of selection is the mechanism by which the negative criterion is satisfied: governance legitimation does not depend on any actor’s choice.
It produces heterogeneity. A randomly drawn body reflects the composition of the pool rather than the preferences of a selector. Over successive terms, the governance function is populated by engineers from different teams, different technical backgrounds, different organizational tenures, different relationships to the platform. No single perspective dominates across terms. This matters because platform governance decisions distribute costs unevenly: a mandate that benefits backend teams may burden frontend teams, a mandate that serves security may constrain velocity. A heterogeneous body is more likely to surface these asymmetries before approval than a body selected for alignment.
It prevents consolidation. Limited terms with cooldown mean that no individual accumulates governance power over time. The body’s composition changes predictably. Institutional memory is carried by process (documented mandates, ADRs, precedent) rather than by personnel. The onboarding cost is high; the capture resistance justifies it.
It produces legitimacy through representativeness. The governed accept the authority of a body that could, in principle, include any senior engineer among them. The lottery is legitimate not because it selects the best, but because it selects without preference. The engineer whose team bears the cost of a mandate can accept the mandate’s authority more readily when the body that approved it was drawn from the same population, with the same probability of inclusion. This is statistical legitimacy: the body is a sample of the governed, not a selection by the governing.
What sortition does not solve
Sortition is not a complete governance mechanism. It addresses the population question. It does not address competence, process, or correction.
Expertise must be purchased, not assumed. A randomly drawn governance body has no guaranteed technical depth in any specific domain. When a mandate concerns container orchestration, the body may contain no container specialist. The structural answer is advisory input: the platform team, domain experts, and affected consuming teams provide technical briefings. The governance body decides; it does not need to be expert in every domain it governs. The structure is that of a jury: it prevents expertise from collapsing into authority.
The consuming teams’ role here is structural, not merely consultative. The exception process is their procedural voice: a channel to challenge mandates case by case. Sortition extends this from procedural to constitutional: consuming teams are not only challengers of individual mandates but the population from which the governance body is drawn. They provide testimony when mandates affect them. They challenge mandates through the exception process. And they know that next term, they may sit on the body that decides. The governed become the governing because any of them might be drawn next. The asymmetry between rule-setters and cost-bearers persists, but becomes rotational.
Majority composition of the pool matters. If the engineering population is homogeneous (same background, same training, same organizational socialization), random selection reproduces that homogeneity. Sortition reflects the pool faithfully. The pool definition itself is a governance decision: who sets the seniority threshold, who maintains the eligibility list, whether promotion processes that shape the pool are themselves biased. Sortition displaces the selection problem from the governance body to the pool boundary. It does not eliminate it.
Productive friction is not guaranteed. A governance body that happens to draw members who all agree may produce consensus without deliberation. The mechanism does not ensure disagreement. It ensures that agreement, when it occurs, arises from deliberation rather than selection pressure.
No built-in correction mechanism. A governance body that makes a bad decision has no internal mechanism to reverse it. Correction is external: audit reviews mandate outcomes, the executive board retains dissolution authority, the exception process surfaces where mandates fail. Sortition produces the body. The rules around it (mandate justification, outcome review, lifecycle dissolution) constrain what the body can do.
What the model requires
Sortition does not function in isolation. It requires the machinery described in the preceding text: mandate justification (naming the job, cost-bearers, exception path, and falsification condition), independent audit of mandate outcomes, documented decisions that survive personnel turnover, and executive board authority as boundary condition. Without these constraints, sortition produces authority without accountability.
Three tests for whether a sortition implementation satisfies the negative criterion:
Can the platform influence who is drawn? If yes, the criterion is violated.
Can any actor accumulate governance tenure beyond a single term without returning to the pool? If yes, consolidation is possible and the mechanism has failed.
Can the body issue mandates without documented justification subject to independent review? If yes, sortition has produced authority without accountability, which is the failure mode it was designed to prevent.
A governance body that passes all three tests is not guaranteed to govern well. It is guaranteed to govern without structural capture. The three conventional mechanisms each fail the negative criterion. Sortition satisfies it without introducing the structural failures of the alternatives. Hybrid mechanisms exist (shortlists generated by lottery, stratified sampling, rotating nomination rights); they are evaluated by the same criterion. The label is irrelevant. The criterion is whether the population mechanism reintroduces the dependency the architecture exists to prevent.