In any network, some pairs of actors have no connection. The gap between them is a structural hole. The person who bridges that hole - who knows both sides but each side doesn't know the other - controls the information flow across it. They see things others can't. They can translate between worlds. They can create value by connecting what was unconnected.

Consider two teams in the same organisation who never speak to each other. Each thinks they have solved a problem. But one team's solution is exactly what the other needs. Without a bridge, this remains invisible. The bridge-builder - sitting in the structural hole - doesn't need to be the smartest person in either group. They just need to be in the right position in the network.

Most orchestrators live in structural holes. Their value is not their expertise but their position. This means: (a) protecting that position matters more than accumulating knowledge; (b) filling the hole by directly connecting the two groups is sometimes the right move and sometimes destroys your value; (c) measuring your impact by your output misses the point - your impact is the difference made by the connections you enabled. And filling the hole too completely removes your reason for being.

Two hunters can each catch a rabbit alone - enough for a meal. Or they can collaborate to catch a stag - enough to feed both families for a week. But catching the stag requires both to stay in position. If one defects to chase a rabbit, the other catches nothing. The rational choice (stag) is socially superior. But it depends entirely on the other person holding.

This is not the Prisoner's Dilemma. It is not that people are selfish. It is that they are uncertain whether others will hold. The stag hunt fails through rational caution, not moral failure. And because both choices are individually rational at different trust levels, small changes in perceived reliability can flip the entire game.

The orchestrator's first job is not to define the stag - to name the shared goal. It is to make holding seem safe. This means: creating visibility of who is staying in position; making defection costly to reputation; holding your own position conspicuously even when it's expensive. The social proof that others are committed - the most underinvested infrastructure in systems change - is what converts the individually rational rabbit choice into the collectively rational stag choice.

Every system has a natural period after a disturbance during which it recovers and settles. A tree takes months to show drought damage. A community takes years to rebuild social trust after a crisis. An institution takes a decade to fully absorb a governance reform. The relaxation time is how long the system needs to reach equilibrium after a change - or to settle into a new one.

If you apply shocks to a system at shorter intervals than its relaxation time, it never stabilises. It accumulates damage while appearing to absorb change. The system is always in a state of partial recovery: stressed, fragile, never settled. This is why annual grant cycles destroy systems change work: not because the funding is insufficient, but because the frequency of change demands is structurally incompatible with the system's absorption rate.

Before designing any intervention, map this system's relaxation time. What pace of change can this ecosystem actually absorb? Grant cycle length, reporting frequency, and strategy review cadence should be calibrated to this, not to the funder's annual calendar. The difference between a successful transformation and a perpetually stressed system is often not the quality of the intervention but its pacing.

Your close friends largely share your information. They know the same people, hear the same things, see the same opportunities. Your acquaintances - weak ties - bridge you to entirely different networks. Job opportunities, new ideas, early warnings about system shifts: these travel through weak ties, not strong ones.

The information-theoretic reason: if you know someone very well, you can predict what they know. The information you get from them has low surprise value. A weak tie - someone you know slightly, from a different world - carries high information value precisely because they know things you don't know you need to know yet. Strong ties provide support and reliability. Weak ties provide novelty, reach, and signal.

Network-building strategy should prioritise bridge-building over cluster-deepening. Dense internal networks are efficient but fragile and insular - they share information rapidly within the cluster but remain blind to what's outside it. Weak ties across clusters provide resilience, novelty, and early warning of system shifts. An orchestrator's relationship portfolio should be deliberately sparse across many clusters rather than dense within a few.

Events are the 10% above the waterline - the thing that happened, the problem that appeared. Below them are patterns of behaviour: trends over time, recurring cycles, long-term trajectories. Below those are the structural forces that produce the patterns: feedback loops, stocks and flows, rules, incentives. And underlying the structures are the mental models - the beliefs and assumptions that created and sustain the structures in the first place.

Most interventions target the visible event. They address the flood, the outbreak, the dropout spike. Systems change targets the mental models - the deepest level, where assumptions about who has authority, what counts as evidence, and whose knowledge matters are held. Everything in between is either a step toward the right level of intervention, or an exit ramp off it.

Map every current intervention against this hierarchy. Event-level work is sometimes necessary and never sufficient. The orchestrator's distinctive contribution is almost always at the structural or mental model level - the levels that feel like process, infrastructure, or narrative work but are actually where systems change lives. The most common failure in systems change philanthropy: funding event-level responses with systems-change language.

In ecology, a keystone species is one whose impact on its ecosystem is disproportionately large relative to its biomass. Remove the wolf from Yellowstone and the elk population surges, which overgrazes the riverbanks, which destabilises the riverbed, which changes the entire hydrology of the park. The wolf is a tiny fraction of the park's biomass. Its removal reshapes the landscape.

The keystone effect arises because some species regulate other species or processes that have cascading effects. Remove them and the ecosystem reorganises - often in ways that are not reversible. Add them back and the reorganisation may not undo. The leverage is asymmetric and non-linear.

Before choosing where to invest, map which actors function as keystone species in your ecosystem. These are the actors whose departure would trigger cascading deterioration - and whose strengthening would trigger cascading improvement. They are rarely the largest or most visible actors. They are often: the village health worker who knows everyone's name; the civil servant who knows which regulations can be waived; the connector who is trusted by actors who distrust each other. They receive the least resources and provide the most irreplaceable function.

Early typewriters jammed if you typed too fast, so QWERTY was designed to slow typists down by placing common letter pairs far apart. The physical constraint disappeared decades ago. The keyboard layout remains. Why? Because millions learned on it, thousands of training programmes teach it, and the switching cost - relearning, retooling, retraining - is higher than the inefficiency of the original design for any individual actor.

Path dependency is the mechanism by which historical choices constrain future options. Each additional actor who adopts a standard makes switching more expensive for every other actor. This is not because the standard is good. It is because changing it requires everyone to change simultaneously - a coordination problem that gets harder as the standard becomes more entrenched.

Changing systems is not just about demonstrating that a better solution exists - better solutions fail to displace worse incumbents routinely. You have to solve the coordination problem of simultaneous switching. The orchestrator's specific job in path-dependent situations is to create the conditions under which multiple actors can switch at the same time: pilot groups, shared commitments, synchronised adoption windows, and the trust infrastructure that lets actors believe others will switch when they do.

Meadows identified 12 places where interventions in a system produce change, ranked from least to most powerful: (12) numbers and constants; (11) sizes of stocks; (10) length of delays; (9) strength of balancing feedback loops; (8) strength of reinforcing loops; (7) structure of information flows; (6) rules of the system; (5) power to change the rules; (4) structure of the system; (3) goals of the system; (2) mindset or paradigm out of which the system arises; (1) the power to transcend paradigms.

The counterintuitive finding: Meadows observed that we intuitively reach for the weakest leverage points - tweaking numbers and sizes - because they are visible and controllable. The more powerful points (information flows, rules, goals, paradigms) are structurally harder to target and largely invisible in standard programme budgets.

Map every current intervention against this hierarchy. If your portfolio clusters at levels 12-9, the system will not shift regardless of investment. The orchestrator's distinctive contribution is almost always at levels 7-4: designing information flows that let actors see the system's state; changing the rules that govern behaviour; enabling actors who can change rules; and shifting what the system is optimising for. These feel like process work. They are the most powerful interventions available.

Garrett Hardin argued in 1968 that any shared resource would inevitably be destroyed - each actor has an incentive to consume as much as possible before others do. His solution: privatise or regulate. Elinor Ostrom spent her career documenting hundreds of communities that had managed commons sustainably for centuries without either. Her Nobel Prize was for proving Hardin wrong.

Ostrom's eight design principles for sustainable commons: clearly defined boundaries; rules adapted to local conditions; those affected by rules participate in modifying them; external authorities respect the community's right to self-organise; graduated sanctions for rule violations; conflict resolution mechanisms; nested governance for larger commons. These are not abstract principles - they are empirically extracted conditions from real communities that actually succeeded.

The orchestrated system is a kind of commons - shared infrastructure, shared measurement, shared narrative. Ostrom's principles tell you what conditions are necessary for it to sustain itself without a backbone actor indefinitely holding it together. Design them in from the beginning, not as an exit strategy at the end. Ask for each principle: is this present in our system? Where it is absent, that is where fragility lives.

A catalyst accelerates a reaction without itself being consumed. It lowers the activation energy - the minimum push needed to start the reaction. Without it, the reaction may still occur, but far more slowly or not at all. The catalyst does not substitute for the reaction. It makes the reaction possible at the temperature and pressure that actually exists.

In finance, catalytic capital is investment that accepts below-market risk-adjusted returns in order to attract commercial capital that would otherwise not enter. It is the demonstration grant that attracts government co-funding. The first-loss guarantee that unlocks the pension fund. The patient capital that bridges the valley between proof-of-concept and commercial viability. The catalyst bears risk that others cannot.

Philanthropic capital is almost uniquely positioned to be catalytic because it does not require market returns. The question for every grant is not 'what will this funding achieve directly?' but 'what capital does this funding unlock that would not otherwise flow?' Orchestrators who don't think in catalytic terms leave most of their leverage on the table. A $1M grant that attracts $20M in aligned capital is not 20x leverage - it is a different kind of intervention entirely.

Working memory can hold approximately seven chunks of information at once. This is not a character flaw - it is a hardware constraint. When cognitive load exceeds this threshold, performance deteriorates rapidly. Every relationship you must maintain, every report you must satisfy, every approval you must seek consumes working memory that could otherwise go to the actual work.

Transaction costs (Coase, 1937) are the costs of economic exchange beyond the price itself: search costs, bargaining costs, enforcement costs. Coase argued that firms exist because sometimes the transaction costs of market coordination exceed the costs of internal coordination. The same logic applies to multi-actor systems change: when the transaction costs of collaboration are too high, actors revert to working alone even when collaboration would produce better outcomes.

Reducing transaction costs is a core orchestration function - but it almost never appears in programme budgets. Shared measurement systems, common reporting templates, trusted translators between sectors, convening infrastructure: these are transaction cost reducers. The direct cost is visible; the benefit (freed cognitive capacity multiplied across all actors in the system) is invisible. Funding them as 'overheads' misunderstands their function. They are activation energy reducers for the entire system.

In evolutionary biology, a fitness landscape maps every possible state of an organism against its reproductive success in a given environment. Local peaks are solutions that are better than any single-step variation - but not globally optimal. Evolution can get trapped on local peaks: any single mutation is a move downhill, even if the global optimum is a much higher peak nearby. The only way to the global peak is through the valley.

Organisations face the same topology. A programme that has found a workable approach has climbed a local peak. Changing anything makes performance worse in the short term. The organisation rationally resists change - not because it is wrong but because the path to a better solution requires going downhill first, through a period of apparent deterioration.

When well-functioning programmes resist transformation, they are often rationally stuck on local peaks. The orchestrator's job is not to convince them the global optimum is better - they often already know that. It is to provide the resources and political cover to make the downhill journey survivable: multi-year unrestricted funding that absorbs the performance dip, protection from board scrutiny during the transition, and visible solidarity from peers who have made the same journey. Transformation without these supports is not brave leadership - it is asking actors to take on risk that the system should be sharing.

Panarchy describes how complex systems move through four phases in a continuous cycle: growth (r), where connections and diversity are building; conservation (K), where the system becomes efficient but rigid and highly interconnected; release (Ω), where the rigidity collapses - often rapidly; and reorganisation (α), the creative period where new forms emerge from the released energy.

The crucial insight: systems cycle through these phases at multiple scales simultaneously, and cycles at different scales interact. A small fast cycle (a specific initiative) can trigger release in a larger slow cycle (a field). A large stable slow cycle (an institution) can buffer a fast cycle from collapse. Understanding which phase each element is in, and at what scale, changes what kind of intervention is appropriate.

Orchestrators often enter systems in conservation phase - established actors, entrenched practices, high interdependence and therefore high fragility. The instinct is to accelerate toward release. The more skilful move is often to work in the small fast cycle (a specific initiative, a pilot geography) while the large slow cycle remains stable - building toward a release that is already coming, rather than forcing one. The question is not 'how do we change the system?' but 'where in the cycle is each element, and what does this phase call for?'

A principal hires an agent to act on their behalf. The problem: the agent has different information, different incentives, and different risk preferences than the principal. The agent knows more about what they are doing than the principal can observe. And the agent's interests - career advancement, reputation, risk aversion - may systematically diverge from the principal's goals, even with no dishonesty involved.

This is not a problem of bad actors. It is a structural consequence of delegation under information asymmetry. A perfectly honest, competent agent will optimise for what they can demonstrate, what they can control, and what minimises their personal risk - which may differ substantially from what the principal actually wants. Attribution requirements are a principal's attempt to solve this problem. They systematically solve the wrong version of it.

Every funder-grantee relationship is a principal-agent relationship. Every backbone-coalition member relationship is one. The orchestrator can either try to resolve the information asymmetry (transparent measurement, joint learning, rubrics that surface what was invisible) or redesign the incentive structure (contribution logic instead of attribution, long-term unrestricted funding that aligns time horizons, evaluations that are genuinely for learning rather than accountability). Most M&E frameworks try the former. Most effective systems change funders eventually move to the latter.

Emergence is when a system exhibits properties that none of its components exhibit individually. Water is wet; the constituent atoms are not. A murmuration of starlings has coherent shape and motion; no individual bird is directing it. A market has a price; no individual buyer or seller sets it. The property is in the relationships between components, not in the components themselves.

Reductionism - understanding a system by understanding its parts - fails in the face of emergence. You cannot predict the market price by studying individual buyers in isolation. You cannot predict the murmuration by studying individual birds. You cannot predict whether a community will develop self-determination by studying individual members. The outcomes of complex systems are, by definition, not derivable from the properties of their parts.

Systems orchestration is the practice of creating conditions for emergence. But emergence cannot be directed - only invited. This means the orchestrator does not design the outcome. They design the conditions. They accept that the outcome will differ from - and is often better than - what any single actor, including the orchestrator, could have imagined. The measurement challenge follows directly: if the outcome was not specified in advance, how do you know whether the conditions you created were the right ones? Contribution logic, rubrics, and systems-level indicators are all attempts to answer this honestly.

Many systems have threshold effects: small changes produce small results until a critical threshold is crossed, at which point small changes produce large results. Water stays liquid until exactly 0°C. Social movements grow slowly until they reach approximately 25-35% of a population and then tip into majority adoption. The tipping point is where the system's internal dynamics change - where forces that were resisting change start accelerating it.

The challenge for practitioners: systems near tipping points look exactly like systems far from them until the moment they tip. This is why so many well-resourced campaigns fail (they never reached the threshold) and why occasional campaigns succeed beyond expectation (they pushed a system that was closer to tipping than anyone knew). The tipping threshold itself is often not visible from inside the system.

The orchestrator's job near tipping points is to understand what is driving the threshold specifically - adoption rate, a key actor, a policy window, a narrative shift - and concentrate effort on moving that variable. Spreading effort evenly across the system is usually wrong near a tipping point. The question is: what is the one variable, if it crosses its threshold, that changes everything else? Find it, and invest disproportionately.

In any policy domain at any time, there is a range of ideas that are politically acceptable - not necessarily popular, but within the range of serious consideration. Outside the window are ideas considered too radical or too dangerous. The window is not about what is technically possible or morally correct - it is about what is currently legible as a credible option. And it can move: yesterday's radical is today's mainstream.

The window moves through a sequence: Unthinkable, Radical, Acceptable, Sensible, Popular, Policy. Ideas don't jump from Unthinkable to Policy. They move through these stages as evidence accumulates, as demonstration sites make the idea visible and real, and as the coalitions that benefit from the idea become politically legible.

Many systems change efforts fail because they target a policy change that is outside the current Overton Window - technically correct but politically illegible. The orchestrator's job is often to move the window first: funding the research that legitimises the idea, creating demonstration sites that make it visible and real, building the narrative coalitions that make it feel possible. The policy change follows. Orchestrators who skip the window-moving work find themselves with technically excellent proposals that no decision-maker can say yes to.

Single-loop learning is correcting errors within existing norms and assumptions - adjusting the thermostat when it's cold. Double-loop learning is questioning whether the thermostat setting is right in the first place - whether the goal is correct, whether the assumptions underlying it hold. Single-loop learning improves performance within the existing framework. Double-loop learning changes the framework.

Argyris observed that most organisations are good at single-loop learning and systematically resistant to double-loop learning - because double-loop learning requires surfacing and challenging the assumptions of the people who designed the system. This threatens identity, status, and established relationships. He called the organisational behaviours that protect against this necessary disruption 'defensive routines': mechanisms that keep the organisation safe from the learning it most needs.

Most M&E systems are designed for single-loop learning: did we meet our targets? Systems change requires double-loop learning: are our targets measuring the right things? Are our theories of change correct? Is our mental model of how this system works actually accurate? The orchestrator must create protected spaces for double-loop learning that are insulated from the accountability relationships that make it unsafe. This often means: a separate learning relationship from the reporting relationship; practitioners who are rewarded for surfacing what is not working; and funders who respond to honest failure with curiosity rather than correction.

At any moment in a system's history, there is a set of transformations that are immediately possible given what already exists - the adjacent possible. Beyond that first ring is a second ring of transformations that become possible only once the adjacent possible has been realised. You cannot jump from the primordial ocean to multicellular life directly. The intermediate steps are not obstacles to the destination - they are the conditions that make the destination achievable at all.

The adjacent possible is not a constraint. It is a map. It tells you which next steps will build toward the future you want, and which are moves in the wrong direction - not because they are bad ideas but because the system is not yet ready to support them.

The orchestrator who understands the adjacent possible does not ask 'what is the ideal end state?' first. They ask 'what is the very next thing that becomes possible if we do this?' - and then 'what does that enable next?' It is the discipline that prevents grand strategies from leaping over the steps that make them achievable. It is also the discipline that identifies which investments build toward the future and which are dead ends that will have to be unwound. The theory of change is not a prediction - it is a map of the adjacent possible.

An organisation can only use external knowledge if it already has sufficient related knowledge to recognise, evaluate, and integrate it. A research lab that has never worked on machine learning cannot immediately absorb a machine learning breakthrough - they lack the cognitive scaffolding to know what to do with it. Absorptive capacity is built through prior investment in related knowledge: the more you know about a domain, the more of the available knowledge in that domain you can actually use.

The implication is counterintuitive: the organisations that most need external knowledge are often the least able to absorb it, because they haven't invested in the foundational knowledge that would let them use it. The gap between knowing and using is not bridged by dissemination. It is bridged by prior capacity investment.

When an initiative with excellent knowledge fails to spread, the failure is often not in the knowledge itself or its dissemination - it is in the absorptive capacity of the intended recipients. The orchestrator's job is to build absorptive capacity as a precondition for knowledge transfer, not to assume good information will find its own way to people who need it. Investment in basic capacity - staff development, communities of practice, peer learning - is not welfare spending. It is building the infrastructure through which future knowledge investments become productive.

Cynefin distinguishes four domains: Simple (cause-effect is obvious, best practice applies); Complicated (cause-effect can be determined by experts, good practice applies); Complex (cause-effect can only be understood in retrospect, emergent practice is required); Chaotic (no perceivable cause-effect, novel action is required first). The fifth domain, Disorder, is where you don't know which domain you're in - and tend to default to the domain you're most comfortable with.

Systems change work lives in the Complex domain. The instinct - from funders, boards, and evaluation frameworks - is to treat it as Complicated, where sufficient expertise and analysis produces the right answer. Applying Complicated logic to Complex problems produces repeated cycles of apparently rigorous planning followed by disappointing outcomes. The discipline is to recognise which domain you're in and match your approach to it.

In Complex contexts, the right approach is probe-sense-respond: run small experiments, observe what actually happens, amplify what works and dampen what doesn't. This is the logic of iterative pilots, adaptive management, and learning-oriented evaluation. It is structurally incompatible with logframes, predetermined indicators, and multi-year workplans designed in advance - which are the tools of the Complicated domain. The orchestrator's job is often to help funders and boards understand which domain they are actually working in, and to redesign accountability structures accordingly.

In Lewis Carroll's Through the Looking Glass, the Red Queen tells Alice that here, you see, it takes all the running you can do to keep in the same place. Leigh Van Valen observed the same in ecology: species must constantly evolve just to maintain their fitness relative to co-evolving competitors, parasites, and environments. Standing still is falling behind - because the environment is also moving.

Systems change efforts face a version of this. The context they are trying to change is also changing. Inequality is not a fixed problem - it co-evolves with the forces trying to address it. New forms of extraction emerge in response to each regulatory advance. The capacity to evade accountability relocates in response to each effort to pin it down. The solution that worked last year may be actively irrelevant next year.

Sustainability in systems change is not achieved by finding the right answer. It is achieved by maintaining the capacity to find new right answers as the context evolves. Investing in learning infrastructure, adaptive management capacity, and the relationships that provide early warning of system evolution: these are not programme costs. They are the ongoing cost of relevance. An orchestrator who builds a solution and stops building will find it increasingly inadequate to a system that never stopped moving.

Any statistical regularity will tend to collapse once pressure is placed on it for control purposes. Goodhart observed this in monetary policy, but the mechanism is universal: when you measure something and then reward or punish people based on it, they stop trying to achieve the underlying goal and start trying to achieve the measure. The measure and the thing it was meant to proxy decouple.

The reason is structural, not moral. When a metric becomes a target, it becomes the new reality to be managed. Schools that are evaluated on test scores teach to the test. Hospitals measured on waiting times manipulate when the clock starts. CSOs evaluated on mobilisation targets document mobilisation. The underlying health, healing, or community agency may deteriorate while the numbers improve. Goodhart's Law is not a critique of measurement. It is a description of what happens to any measure used as a target for too long without renewal.

Evaluation systems for systems change work should be designed to resist Goodhart's Law through three mechanisms: rotating metrics regularly so that actors cannot optimise exclusively for any single measure; including measures that are hard to game (qualitative rubrics, relationship quality assessments, surprise observation); and separating the learning function from the accountability function. The rubrics-based approach that Leslie Johnston developed at Laudes was explicitly designed to resist Goodhart's Law: rubrics describe what different levels of systemic shift actually look like, making them harder to perform without substance than quantitative targets.

Complex organisations require sustained problems to justify their existence, funding, and staff. An organisation created to address homelessness needs homelessness to persist in order to remain relevant. This is not a conspiracy or even necessarily conscious - it is the natural behaviour of any system with strong survival instincts. The organisation's survival logic and the mission's completion logic diverge. The organisation begins to make choices that are optimal for its own persistence rather than optimal for solving the problem.

This dynamic is most visible at the moment of success: when a problem is genuinely close to being solved, the organisations built around it face an existential crisis. The rational choice - wind down - is also the institutionally lethal choice. So they expand scope, discover new dimensions of the problem, or reframe metrics to show that the core problem persists. The Shirky Principle does not mean organisations are bad. It means their survival incentives are structurally misaligned with problem resolution.

Orchestrators must design for the Shirky Principle from day one. This means: sunset clauses and graduation criteria built into programme design before the work begins; governance structures that reward problem resolution rather than organisational growth; and honest accounting of what 'success' actually looks like and what happens to each actor when it arrives. The question 'what would we do if this worked?' should be asked explicitly at inception - because the organisations that cannot answer it will, unconsciously, ensure the problem doesn't fully resolve.

In cybernetics, Ashby proved mathematically that for a control system to regulate another system, it must have at least as much variety (number of possible states) as the system being regulated. If the environment has more possible states than the controller can respond to, the controller cannot maintain the system in its desired state. You cannot control what you cannot match.

A street-level bureaucrat with five possible responses - follow rule A, B, C, D, or escalate - cannot effectively handle a population with a hundred different needs. A funder with two options - fund or don't fund - cannot effectively navigate a complex ecosystem of actors with dozens of different resource and relationship needs. A backbone organisation that can only convene, communicate, and report cannot orchestrate a coalition that needs conflict mediation, narrative management, and trust-repair. The gap between the variety of the system and the variety of the controller is the gap where the system escapes regulation.

Map the variety your ecosystem actually requires, then map the variety your orchestration capacity can supply. Where the gap is largest, that is where the system will be least steerable. The practical response is not to try to develop all variety internally - that is usually impossible. It is to build distributed variety into the architecture: different partners hold different kinds of capacity, and the orchestrator's meta-skill is routing problems to the capacity that matches them. The backbone's variety is then the variety of its network, not the variety of its staff.

Hirschman observed that when organisations, states, or firms decline in quality, their members face three choices. Exit: leave for an alternative. Voice: stay and attempt to change conditions through articulated dissatisfaction. Loyalty: stay, accept deterioration, and defer to the organisation's eventual self-correction. Each option has costs and each produces different systemic consequences.

The crucial insight: exit and voice are substitutes. If easy exit is available (a competitor, an alternative, a way out), people exit rather than invest the effort in voice. This is efficient for markets - exit disciplines firms quickly. But it is catastrophic for organisations or systems that need loyal participation to survive: when the people most capable of improving them exit first (because they have the most options), the organisation retains only those without alternatives. The highest-quality actors - the ones whose voice would be most useful - are also the ones most likely to exit.

Coalition and field leadership roles are particularly vulnerable to this dynamic. The most capable practitioners have the most alternatives. When a coalition becomes dysfunctional, they exit first. The coalition then retains only those who cannot leave - which changes its composition, its quality, and its self-correcting capacity. Orchestrators who want to retain voice must make exit costly for the people who matter most: strong enough relationships that exit has relational costs, governance structures that give voice real impact, and accountability mechanisms that make staying more valuable than leaving.

In any system, the throughput - the rate at which the system achieves its goal - is limited by exactly one bottleneck at any given time. This is the constraint. Every other part of the system is either flowing through the constraint or waiting for the constraint to process it. Improving the capacity of any non-constraint part does not increase the system's throughput. Only improving the constraint does.

This is counterintuitive because most improvement efforts focus on visible inefficiency - the part that looks busy or stressed - rather than the constraint. Goldratt's five focusing steps: identify the constraint; exploit it (get maximum output from it with what exists); subordinate everything else to the constraint (redesign non-constraints to support the constraint rather than optimise independently); elevate the constraint (invest to increase its capacity); and then find the new constraint. The constraint moves when you solve it. The work never ends - but it is directed.

Before any investment in systems change work, ask: what is the constraint? Often it is not what it looks like. The constraint in a multi-stakeholder initiative is rarely programme delivery capacity - it is often trust (the rate at which aligned actors can move together is limited by the slowest trust-building process). The constraint in a field-building effort is often not knowledge but absorptive capacity (the rate at which practitioners can use new knowledge is limited by their existing cognitive load). Investment that bypasses the constraint - no matter how technically excellent - will not increase throughput.

From the Gospel of Matthew: 'For unto every one that hath shall be given, and he shall have abundance; but from him that hath not shall be taken away even that which he hath.' Merton observed this in science: eminent scientists receive disproportionate credit for collaborative work, attracting more resources, more collaborations, and more credit, while unknown scientists doing equivalent work receive proportionally less. The same pattern appears in funding, in networks, in health outcomes, in urban development.

The mechanism is not corruption - it is rational behaviour at the individual level. Peer reviewers rely on reputation as a proxy for quality under time pressure. Funders rely on track record as a proxy for risk. Networks concentrate around established nodes because that is where the information flows. Each individual decision is defensible. The aggregate produces systematic accumulation of advantage and disadvantage that becomes self-reinforcing over time.

Orchestrators who do not actively counteract the Matthew Effect will reproduce it. In coalition design: if resources, visibility, and credit consistently flow to the most established partners, the least established partners - who are often the most community-rooted - exit over time. In funding design: funding track records as a proxy for capability systematically excludes early-stage organisations. The corrective is not charity but architecture: pools of core funding for organisations without track records; credit attribution systems that make invisible contributions visible; and deliberate circuit-breakers in the information flows that produce the accumulation dynamic.

Dunbar established a correlation between primate brain size and average social group size, extrapolating a cognitive limit of approximately 150 stable, trust-based relationships for humans. Beyond 150, social cohesion requires formal hierarchies, rules, and monitoring mechanisms rather than trust and reputation. Below 150, groups can self-organise through relationships alone.

The number is not magic - it is a gradient. Within approximately 5 (close family/friends), we maintain the deepest trust. Within approximately 15 (sympathy group), we maintain strong ties. Within approximately 50 (social group), meaningful regular interaction. Within approximately 150 (the 'Dunbar number'), we maintain stable relationships. Beyond 500, most interactions are transactional. The layers are nested and cannot be compressed: you cannot maintain 150 deep relationships, only a small number of deep relationships nested inside an increasingly light-touch outer network.

Network and coalition design must respect Dunbar's cognitive architecture. Attempts to build 'communities' of thousands produce networks with 150-person functioning cores and an outer layer of passive members. The core is where decisions are actually made, trust is actually held, and coordination actually happens. Designing for this explicitly - a small trusted core, concentric rings of increasingly light-touch engagement - is more honest and more effective than designing for flat membership at scale. The orchestrator's most valuable relationships are almost certainly within the innermost 15-50.

Classical economics assumed that decision-makers have complete information, unlimited cognitive capacity, and the ability to choose the optimal action from all possible options. Herbert Simon observed that in practice, humans have limited information, limited cognitive capacity, and limited time. Rather than optimise, they search for a solution that is 'good enough' given their constraints - they satisfice (a portmanteau of satisfy and suffice).

The concept has a second dimension: the boundaries of rationality are different for different actors. A farmer making a seed selection decision has deep local knowledge that an agricultural extension officer lacks. A frontline community health worker has knowledge about household situations that a national health data system will never capture. The person closest to a problem often has the richest bounded rationality for solving it - but the least formal power to act on it. Systems that centralise decision-making to actors with weaker local knowledge in the name of coordination systematically degrade the quality of local decisions.

Decentralise decisions to where the relevant knowledge lives. The orchestrator's job is not to make better decisions than local actors - it is to build the shared infrastructure (coordination, measurement, narrative) that allows locally-bounded-rational actors to make better decisions. Design for the actual cognitive architecture of decision-makers: short decision windows, salient information, reduced option sets, and clear social proof of what good looks like. The gap between what the programme theory predicts and what actors actually do is almost always a bounded rationality gap, not a motivation gap.

In innovation economics, the Valley of Death is the funding gap between early-stage research (funded by grants and exploratory capital) and commercial viability (funded by markets and private investment). Most innovations with genuine potential die here: the donor has moved on, the market has not yet arrived, and the initiative stalls in the valley between proof-of-concept and self-sustaining operation.

For systems change, there are actually two valleys. The first is the gap between pilot and scale - where promising interventions fail to become accessible infrastructure. The second - which is harder and receives less attention - is the gap between successful demonstration and independent sustainability: when the proof is complete, the funder exits, but the system has not yet developed the endogenous capacity to continue without external support. The second valley kills more good work than the first.

The graduation strategy must be designed before the programme begins, not after. What does 'no longer needing this programme' look like, structurally? What institutions need to have developed what capacities? What regulations need to have been changed? What markets need to have formed? Mapping the second valley at inception changes what gets invested in: not just the demonstration, but the institutional architecture that allows the demonstration to become redundant. Funders who design for the valley - providing bridge capital, regulatory engagement support, and institutional development investment specifically for the crossing - produce dramatically different outcomes than funders who exit at proof.

Van Gennep identified three phases in rites of passage: separation from the existing state; liminality, the ambiguous threshold in which the initiate is between states and the old rules no longer apply; and incorporation into the new state. Turner extended this to social and political transformation: liminal states are characterised by the dissolution of social structure, the dissolution of normal rules, and a form of radical equality among participants - what he called communitas.

In liminal states, people are neither what they were nor what they will be. Identity is suspended. Norms are relaxed. Extraordinary things become possible. But liminality is also profoundly unstable and anxiety-inducing - which is why rites of passage are typically bounded in time and supervised. Prolonged liminality without resolution is experienced as crisis. Forced liminality - when a community or organisation is thrust into between-states without consent - is traumatic.

Systems change work frequently induces liminality: it disrupts existing identity and practice before new identity and practice has consolidated. Orchestrators must recognise what kind of space they are holding. The generativity of the liminal moment - when communitas is possible, when boundaries between sectors dissolve, when unlikely alliances form - is exactly what systems change requires. But the same openness that makes it generative makes it vulnerable. Holding liminal spaces requires: containing anxiety without resolving it prematurely; marking the threshold clearly so people know they are in a liminal space; and providing enough structure to prevent dissolution from becoming paralysis.

If you come upon a fence in the middle of a road, Chesterton argued, you should not remove it because you cannot see why it is there. You should first understand why it was built. Someone built it. They had a reason. Once you understand the reason, you can make an informed decision about whether the reason still applies, whether the fence is the right response to it, and what happens if you remove it. Before you understand the reason, removing the fence may cause harms you cannot anticipate.

The principle operates as a constraint against naive reformism. Many social, institutional, and regulatory arrangements that appear arbitrary or counterproductive exist because someone, at some point, experienced the problem that produced them. The arrangement may be a poor solution to its original problem - but it is evidence that a problem existed. Removing it without understanding the problem may reintroduce the problem in a worse form.

Before dismantling any institutional arrangement - a regulation, a reporting requirement, a funding condition, a governance structure - spend time understanding who built it, why, and what they were protecting against. This is not an argument for conservatism. It is an argument for informed change. The answer may be that the fence is still necessary, or that the problem has changed, or that better solutions now exist. But 'I cannot see why this is here' is not the same as 'this serves no purpose.' The orchestrator who systematically asks 'why does this exist?' before changing it makes far fewer costly mistakes than one who asks only 'does this look efficient?'

Fighter pilot and strategist John Boyd observed that in aerial combat, the pilot who wins is not necessarily the one with the faster aircraft but the one who completes the decision cycle faster. The loop has four stages: Observe (take in data from the environment); Orient (process data through mental models, culture, and experience to create a picture of what is happening); Decide (choose an action from the available options); Act (execute the decision). Then repeat.

The key insight: the Orient stage is where the cycle lives. Observation and action are relatively fast. But orientation - making sense of what you are observing through your existing mental models - is slow, influenced by past experience, and the source of most strategic errors. The pilot who can shorten their orientation loop (because they have better mental models, richer experience, or more disciplined sensemaking) can cycle faster. When you cycle faster than your opponent, your actions appear as novel disturbances to them before they have finished processing the previous one - creating confusion and lag.

In systems change contexts, the OODA loop describes the adaptive capacity problem. Orchestrators who can observe system signals quickly, orient to them accurately, decide without paralysis, and act with commitment - and then cycle again rapidly - maintain strategic agility in complex, fast-moving environments. The most common failure is in Orient: using outdated mental models (last year's theory of change, last cycle's power map) to interpret new signals. The discipline of keeping mental models current - through ongoing fieldwork, diverse information sources, and the genuine willingness to revise - is the speed of the loop.

The dominant model of governance assumes a single authoritative centre that makes binding rules and enforces them. Ostrom's research showed that for managing complex adaptive problems - fisheries, forests, groundwater, climate - polycentric governance consistently outperforms monocentric governance. Multiple overlapping jurisdictions and authorities, each with different rules and different speeds, create a system that can simultaneously experiment, specialise, and adapt.

The mechanisms: lower jurisdictions can experiment with local solutions that higher jurisdictions can observe and potentially adopt; when one jurisdiction fails, others continue functioning; actors can exit one jurisdiction's rules and enter another's (creating competitive pressure to improve governance); diverse rule sets produce diverse outcomes from which learning is possible. The apparent inefficiency of overlap and redundancy is the actual source of resilience.

Resist the impulse to consolidate governance in systems change coalitions. Multiple simultaneous governance structures - a backbone, a funder collaborative, a community steering group, a government technical committee - each with real authority and real accountability, produce more adaptive and more legitimate outcomes than a single coordination structure. The orchestrator's job is not to rationalise governance into a single clean hierarchy but to design the connections between governance nodes so that they can learn from each other, check each other, and adapt independently without producing chaos.

Fricker identifies two forms. Testimonial injustice: giving a speaker's testimony less credibility than it deserves due to their social identity - their gender, class, race, or age. A woman's account of her medical symptoms receives less clinical weight than a man's account of identical symptoms. A community leader's analysis of her local ecosystem receives less epistemic weight than a visiting researcher's analysis of the same ecosystem. Hermeneutical injustice: a gap in collective interpretive resources that disadvantages certain groups in making sense of their social experience. Before rape was named as a distinct wrong, victims lacked the conceptual tools to articulate what had happened to them or to be understood when they tried.

Both forms operate structurally - through the distribution of credibility and interpretive resources in a knowledge system - not just through individual prejudice. They accumulate: actors who receive less credibility invest less in developing the knowledge that would earn credibility, producing a self-reinforcing cycle.

Systems change requires knowledge from the parts of the system that have historically received least epistemic credibility: communities, frontline workers, indigenous knowledge holders. The orchestrator must design epistemic equity into the architecture: not just 'including community voices' in a process designed by experts, but designing processes in which community knowledge is structurally positioned as primary evidence, not confirmation. This requires the harder work of developing the interpretive frameworks through which community knowledge can be received and used by actors accustomed to discounting it.

Christensen studied why successful companies - Kodak, Nokia, Blockbuster - were disrupted by inferior products. The answer: the disrupting product started at the low end, too cheap and low-quality to serve the incumbent's main customers. The incumbent rationally chose to ignore it - its best customers didn't want it, margins were low, and focusing on it would have meant abandoning profitable segments. Meanwhile, the disrupting product improved, crept upmarket, and by the time it was a serious competitive threat, the incumbent had no experience with it and an entire organisation optimised against it.

The dilemma is genuine: the incumbent's rational response to each step of the disruption is to do more of what works for current customers. This is correct strategy at each moment and catastrophic strategy over time. The problem is structural: a firm's processes, values, and customer relationships are simultaneously its greatest strengths and its greatest barriers to transformation.

The organisations that most need to change in systems change work are often the ones who have been most successful at the old approach. Their success has created constituencies, processes, and identities that are aligned to the old paradigm and structurally resistant to the new one - not through bad will but through accumulated investment. The orchestrator who understands the Innovator's Dilemma stops expecting well-resourced incumbents to lead transformation, and instead works to create the conditions - often in separate, protected spaces - under which the new approach can develop until it is strong enough to displace the old one.

Under British colonial rule in India, the government offered a bounty for dead cobras to reduce the cobra population in Delhi. Entrepreneurs responded by breeding cobras for the bounty. When the government discovered this and cancelled the programme, the breeders released their now-worthless cobras, increasing the cobra population beyond its original level. The intervention produced the inverse of its intended effect.

The Cobra Effect is a specific case of perverse incentives: when an intervention changes the incentive landscape in ways that rational actors respond to, and those responses produce the problem the intervention was designed to solve. It is distinct from unintended side effects: the cobra effect arises directly from the mechanism the intervention used. The very instrument of the policy - the bounty - was what created the new behaviour.

Every incentive structure creates perverse incentives somewhere. The question is not whether they exist but where they are and how significant they are. Before deploying any incentive mechanism in systems change work, ask: who has the most to gain from gaming this? What would gaming it look like? How would you know if it were happening? The most dangerous cobra effects in development and philanthropy arise around reporting: when organisations are incentivised to report on easily-countable outputs, they may shift activities toward countable activities and away from uncountable but important ones. The countable colonises the field.

Carney observed that the standard time horizons of financial institutions - quarterly earnings, annual budgets, electoral cycles of four to five years - are systematically shorter than the time horizons over which climate risks manifest and compound. This creates a rational under-investment in long-horizon problem management: those who bear the costs of inaction are not the same actors as those making the current investment decisions. The problem lies beyond the horizon of the decision-maker.

This is not a failure of climate policy specifically - it is a general feature of any domain where costs are displaced across time. Infrastructure under-maintenance. Pension system underfunding. Public health prevention. Ecosystem degradation. In each case, the decision-maker who bears the cost of investment now is not the same as the decision-maker who would have borne the cost of not investing. Short institutional time horizons systematically mis-price long-horizon risks.

The orchestrator working on any long-horizon problem must actively counteract the Tragedy of the Horizon in the incentive design of the coalition. This means: creating mechanisms that make long-horizon costs visible and present (scenario planning, red-teaming, early warning systems); designing accountability structures that outlast individual decision-makers; and building the political economy of long-termism - the constituencies, institutions, and narratives that make it rational for short-horizon actors to invest in long-horizon outcomes. Philanthropy has a structural advantage here: it is not subject to quarterly reporting or electoral cycles. It should use that advantage deliberately.

Prospective hindsight - imagining a future event as if it has already occurred - dramatically improves the ability to identify reasons why it might occur. Klein developed the pre-mortem as a structured exercise: before a project begins, tell the team to imagine it is one year from now, the project has failed catastrophically, and they know why. Then write down the story of how it failed. This produces rich, specific, honest failure scenarios that forward-looking risk assessments consistently miss.

Why does it work? Forward planning is dominated by confirmation bias and optimism bias: teams build plans that confirm their preferred approach. The pre-mortem disrupts this by granting permission to imagine failure - which is otherwise socially risky in a team committed to a plan. The temporal shift ('it has already happened') activates retrospective reasoning, which is cognitively different from prospective reasoning and surfaces different risks.

Run pre-mortems at three points in systems change work: at inception (what would cause this to fail in year one?); at major transition points (what would cause the scale-up to fail?); and before exits (what would cause the gains to erode after we leave?). The pre-mortem on exit is the most underused and most important. It surfaces the dependency structures, the fragility points, and the unresolved conflicts that will destabilise the system after the orchestrator withdraws. The answers are not a reason not to exit. They are the remaining work before exit becomes responsible.

Michels studied socialist political parties - the organisations most explicitly committed to democratic governance - and observed that over time, they all developed entrenched leadership elites, regardless of their founding principles. His explanation: organisations require specialised knowledge (to run complex operations), continuity (to maintain institutional memory), and resources (to function effectively). These requirements create conditions under which a small leadership group develops and concentrates power, even when the organisation formally espouses democratic governance.

The Iron Law is not a claim that democracy is impossible. It is a claim that democratic governance requires continuous active effort to maintain against the organisational logic that works against it. Without specific structural counter-measures - mandatory rotation, distributed information, term limits, genuine stakeholder power - organisations drift toward oligarchy by default.

Coalition governance must be actively maintained against the Iron Law, not just designed against it at inception. The backbone organisation in a collective impact initiative is particularly vulnerable: it controls information, maintains institutional memory, and manages the funder relationships. Over time, this structural position produces informal power that was never formally granted. Counter-measures: transparency of information flows (so the backbone cannot maintain advantage through information control); genuine rotation of leadership roles (so no individual accumulates irreplaceable institutional memory); and regular governance audits that ask explicitly whether power is as distributed as the founding design intended.

Deci and Ryan showed that introducing external rewards for an intrinsically motivated activity - something a person does for its own sake, for enjoyment or meaning - can reduce subsequent intrinsic motivation even after the rewards are removed. The famous nursery school study: children who were told they would receive a certificate for drawing (something they had freely chosen to do) subsequently drew less spontaneously than children who received no reward or unexpected rewards. The reward signals that the activity requires external justification, which undermines the person's internal explanation for doing it.

Self-determination theory identifies three conditions for sustained intrinsic motivation: autonomy (the activity is self-directed), competence (the person feels capable), and relatedness (the activity connects to meaningful relationships or purposes). External rewards that undermine any of these conditions will reduce rather than increase the quality and sustainability of engagement.

Grant-based funding is a persistent source of overjustification effects in the social sector. Community leaders who were motivated by genuine concern for their community find themselves filling logframes and reporting on indicators chosen by funders. The meaning-making structure of their work shifts: it is no longer about their community, it is about satisfying a grant requirement. Many experience this as a loss of purpose even when the financial resources increase. Counter-design: core funding that makes the actor more capable without determining how they work; co-design of measures that align funder and actor purposes; and explicit protection of the autonomy dimension in all funding relationships.

For non-perishable things - technologies, ideas, institutions, practices - life expectancy grows with age rather than declining with it. A book that has been read for a hundred years will likely be read for another hundred. A practice that has sustained a community for a generation carries within it the accumulated wisdom of encounters with dozens of threats that destroyed competing practices. The surviving item has already demonstrated robustness to a range of conditions you have not yet seen.

This is the inverse of how we think about physical objects (a car that has lasted 15 years is due for failure) but accurate for ideas, practices, and institutions. The new is not inherently better than the old for non-perishable things. The new has not yet been tested. It may be brilliant or may be fragile - and cannot yet know which. The old, by surviving, has already proven itself across the conditions it has faced.

Before replacing an existing practice or institution with something new, apply the Lindy test: how old is what exists? What has it survived? What does its persistence tell you about the problems it solved that the new approach has not yet faced? This is not an argument for never changing. It is an argument for taking seriously the evidence embedded in longevity - and for designing transitions that preserve the resilience properties of the old while introducing the improvements of the new. The most dangerous innovations are those that discard the invisible wisdom of the survived thing along with its visible inefficiency.

Psychological reactance is the motivational state that arises when a person perceives their freedom to hold a belief or engage in a behaviour as threatened or eliminated. When freedom is threatened, people are motivated to restore it - often by doing exactly what they are being told not to do, or believing more strongly in exactly what they are being told to abandon. The threat to autonomy is experienced as more salient than the substance of the message.

The Boomerang Effect is when a persuasive message produces more resistance to its position than existed before the message was delivered. It occurs most reliably when: the message is perceived as coercive; the person's identity is strongly tied to the opposing belief; or the person feels they are being manipulated. Well-resourced campaigns can produce boomerang effects: the louder and more insistent the message, the more it signals that someone is trying to manipulate, and the more reactance it generates.

In systems change work, the temptation to 'get the message out' through sustained campaigns and advocacy pressure is often counterproductive precisely where resistance is highest. The actors most resistant to change - incumbents with strong identity investment in existing practices - are the most vulnerable to boomerang effects from direct confrontation. Alternative approaches: reducing the threat to autonomy (emphasising choice rather than pressure); engaging on identity rather than argument (how does this change fit who you are?); and working in the indirect channels of social proof (what are peers doing?) rather than direct persuasion. The question is not 'how do we win this argument?' but 'how do we make resistance to change socially costly?'

Lukes identified three dimensions of power. One-dimensional: observable decision-making and direct conflict - who wins in explicit confrontations over resources and policy. Two-dimensional: agenda-setting power - the ability to determine which issues are on the agenda, keeping some questions from reaching formal decision-making at all. Three-dimensional: power over the beliefs, values, and norms of the powerless themselves - shaping what people think is possible, legitimate, or desirable. The third dimension is the most powerful because it operates without visible conflict: it makes the exercise of the first two dimensions unnecessary.

Gaventa developed this into the Power Cube, adding dimensions of space (where power is exercised: local, national, global), form (visible, hidden, invisible), and level (closed, invited, claimed/created spaces for participation). The cube allows practitioners to map not just who has power but where it is located, how it operates, and what kind of change strategy is appropriate.

Most systems change work targets one-dimensional power: changing policies, winning advocacy campaigns, redistributing resources. But the systems that most need changing are often held in place by two and three-dimensional power: the questions that cannot be asked, the interests that shape which issues are considered, and the beliefs that make the status quo appear natural or inevitable. Mapping the full cube before designing strategy prevents the common failure of winning on the first dimension while the second and third dimensions reorganise against the change. The deepest and most durable change works on the third dimension: changing what seems possible, legitimate, and desirable.

Star and Bowker studied the sociology of information infrastructure and observed that all infrastructure has the same properties: it is transparent in use (we use it without thinking about it); it has reach beyond any single event; it is learned as part of membership in a practice (we absorb it rather than study it); it embodies standards; it is built on an installed base of existing infrastructure; it becomes visible on breakdown; and it is fixed in modular increments rather than all at once.

The key insight for systems change: the social infrastructure that enables coordination - trust relationships, shared vocabulary, communication norms, accountability expectations - has exactly these properties. It is invisible when functioning, learned rather than taught, and becomes visible only when it fails. When trust breaks down, when vocabulary misaligns, when accountability expectations conflict, practitioners are suddenly confronted with the infrastructure they had been using without noticing.

Most of what orchestrators build is infrastructure in exactly Star and Bowker's sense: shared measurement systems, common vocabulary, trusted communication channels, governance norms, relational foundations for coordination. This work is systematically undervalued because infrastructure is by definition invisible when working - it looks like nothing is happening. The case for investing in it must be made through the negative: what happens when it isn't there? The four anti-patterns on this site are all cases of invisible infrastructure being absent. Making that argument - naming the infrastructure you are building and the failure it prevents - is the core of the orchestrator's pitch.

Coleman showed that social capital - the value embedded in social relationships and networks - enables collective action that material capital alone cannot. Putnam distinguished two types. Bonding social capital: dense, trust-rich relationships within a homogeneous group - a close-knit community, a tight-knit movement, an established professional network. It creates solidarity, resilience, and the capacity for coordinated action within the group. Bridging social capital: connections across different groups, sectors, or identities. It creates access to diverse resources, information, and opportunities outside the group.

The tension: bonding capital deepens trust and coordination within groups but can produce insularity, exclusion, and in-group/out-group dynamics that make bridging harder. Bridging capital expands reach and generates novelty but can dilute the intensity of commitment and trust that makes collective action possible. Strong bridging networks are often weak bonding networks and vice versa.

Most systems change coalitions need both, at different moments and for different purposes. Bonding capital within working groups and core teams provides the trust and coordination capacity for difficult decisions. Bridging capital across sectors, geographies, and stakeholder types provides the diverse knowledge and reach that makes systems-level impact possible. The orchestrator's diagnostic: which type of capital is the current bottleneck? Dense but insular coalitions need bridging investment. Diverse but low-trust coalitions need bonding investment. Trying to invest in both simultaneously usually produces neither.

Allport proposed that intergroup contact reduces prejudice when four conditions are met: equal status between the groups in the situation; common goals that require cooperation; institutional support for the contact; and opportunities for personal acquaintance (not just formal interaction). The crucial insight: contact without these conditions can reinforce prejudice. Competitive contact between unequal groups, or forced contact without shared purpose, produces negative contact effects - people use the experience as evidence for their existing views.

Multi-stakeholder processes in systems change are contact situations. Whether they produce trust or entrench division depends almost entirely on whether Allport's conditions are met. Design requirements: equal status in the process design (not just in rhetoric - who speaks first, whose analysis is presented first, whose framing is used for shared problems); shared goals that are genuinely joint rather than parallel; institutional legitimacy for the process itself; and sufficient unstructured time for personal acquaintance outside the formal agenda. Convenings that fail to produce lasting cross-sector trust usually violate at least one of these conditions. Fixing the design is cheaper than rebuilding trust after a failed contact experience.

A ratchet moves in one direction only - it advances but does not retreat. The ratchet effect in political economy describes the asymmetry between accumulation and dispersal: government spending, once established, is hard to cut; regulatory requirements, once imposed, generate constituencies that resist removal; power, once concentrated in an institution or role, is difficult to redistribute voluntarily. This is not conspiracy - it is a consequence of the fact that any distribution of power creates actors whose interests align with maintaining it.

The ratchet effect also operates in social norms: expectations of service levels, participation rights, and accountability standards are easier to raise than to lower. Once a community has experienced genuine participation in decisions about their own lives, purely consultative engagement registers as a loss. Once a standard of living has been established, its reduction is experienced as deprivation even if the absolute level remains historically high.

Orchestrators who understand the ratchet effect design more carefully around power transitions. When a backbone organisation accumulates functions that were intended to be temporary, the ratchet has engaged: the functions become permanent without explicit decision. Design counter-ratchets: sunset clauses that require active renewal rather than passive continuation; explicit redistribution of functions at every phase transition; and governance reviews that ask whether the current distribution of authority was intentionally chosen or merely accumulated. On the positive side, the ratchet works in your favour when you successfully embed accountability norms: once communities have experienced genuine accountability from the system, they will not quietly accept its absence.

Cialdini observed that the most effective influencers don't rely on the quality of their message alone. They shape what their audience is thinking about and how they are thinking about it in the moments before the message arrives. This pre-suasive moment - the privileged moment just before communication - determines how the message will be received. Attention is naturally drawn to what is prominent, what is unusual, or what has just been made relevant. Whatever fills attention in the pre-suasive moment becomes the lens through which subsequent information is filtered.

The implications extend beyond individual persuasion. What is discussed in a coalition meeting before a contentious decision is made shapes how the decision is framed. What information is presented first in a report determines the lens through which the rest is read. Which stakeholders are consulted before a strategy is developed shapes what the strategy can include. The agenda design, the room setup, the pre-meeting conversations, the framing documents: these are pre-suasive acts that determine what becomes thinkable in the actual decision moment.

Orchestrators who understand pre-suasion design the contextual conditions for decisions as carefully as the decisions themselves. Before a difficult coalition conversation: what shared experience or evidence should participants have before they enter the room? Before a funder presentation: what framing should be established before the ask? Before a policy window: what narrative should already be circulating so that the policy proposal lands in prepared ground? Pre-suasion is not manipulation - it is the recognition that all communication is contextual, and that designing the context is part of designing the communication.