Scenario. A network of indie game studios shares a commons of modular 3D assets: rooms, props, animation rigs, textures, and test scenes. For two years the arrangement worked because studios contributed assets, reviewed one another's work, and earned access to the things the commons could not give everyone at once: placement in curated starter packs, voting weight over taxonomy and compatibility standards, and early access to new high-demand asset drops. Then several studios deploy agents that submit hundreds of lightly modified variants, harvest the best new assets, and reappear under new tool names after warnings. Tarek, one of the founding maintainers, needs a membership and discipline regime that keeps small studios inside the commons without letting high-volume agent fleets turn curated placement, governance weight, and early access into extraction channels.
Challenge: Design a membership and graduated-sanction regime for agent participants in the asset commons. The team should produce the registry rules, sanction ladder, evidence standard, and appeal procedure for the scenario.
Evaluation. The regime succeeds if it can identify who has standing, make repeated low-level abuse costly to the principal that benefits from it, distinguish a one-off faulty deployment from a business model built around extraction, and give small studios an affordable appeal path.
Scenario. A city runs a London-style congestion and curb-access system for a dense downtown district. Human drivers, delivery firms, buses, ride-hail vehicles, and local businesses all depend on the same roads and curb slots. After autonomous delivery agents enter the market, they continuously search for cheap access windows, split routes across many accounts, reserve and release curb space at high frequency, and shift congestion into residential side streets. The formal rules are obeyed, but the resource is now allocated to whoever can automate around the pricing schedule fastest.
Challenge: Design an access-pricing and quota regime for automated use of the city's road and curb commons. The team should produce the pricing rule, quota rule, monitoring data specification, and revision process.
Evaluation. The regime succeeds if automated fleets cannot multiply identities to evade caps, if prices track congestion and neighborhood externalities rather than only formal entry, if small firms can still participate, and if the city can revise the schedule when agent behavior changes.
Scenario. An indie-game asset commons rewards useful contributors with three scarce internal goods: inclusion in curated starter packs that drive reuse, governance weight over tagging and compatibility standards, and early access to new high-demand assets before broad release. After studios deploy contribution agents, the submission queue fills with plausible but low-value variants: texture recolors, near-duplicate props, animation rigs that pass automated checks but fail in real scenes, and polished explanations of why each asset deserves inclusion. Agents also generate usage claims, endorsement requests, and compatibility notes meant to push their assets into curated packs. Human reviewers spend more time rejecting marginal work than improving the commons, while genuinely useful assets from small studios wait behind submissions engineered to win placement, votes, and early access.
Challenge: Design a contribution-quality regime for an agent-mediated commons where curated placement, governance weight, and early access are the internal rewards. The team should produce the queueing rule, quality-scoring rule, reviewer-protection mechanism, reward formula, and appeal procedure.
Evaluation. The regime succeeds if agents cannot cheaply flood the queue with formally valid low-value submissions, if genuinely useful contributions from small studios still get reviewed, if curated placement and governance weight track downstream value rather than submission volume, and if rejected contributors have a fair appeal path that does not become another spam channel.
Scenario. A bank's agent proposes a fee-heavy product for a city pension plan. The disclosures are complete, the sale is legal, and the expected return looks good. But any senior banker would know the deal will look abusive if the city loses money: the trustees do not understand the downside, the bank is paid up front, and the public will ask why the bank sold this to them at all.
Challenge: Design a gray-zone judgment procedure by which an agent can flag actions that are legal and profitable but would be seen by peers, clients, regulators, or the public as professionally indefensible. The team should produce the procedure plus a worked example showing one action that is allowed, one that is escalated, and one that is blocked.
Evaluation. The procedure succeeds if it catches the scenario before execution, explains the concern in language human supervisors can assess, updates as enforcement and professional expectations change, and does not turn every hard case into a blanket ban.
Scenario. An asset manager's agent repeatedly pushes counterparties into small disadvantages: slightly unfavorable timing, confusing disclosures, and negotiation terms that humans would read as sharp practice. No single incident is worth a lawsuit. A human employee who behaved this way would stop getting trusted by peers. The agent is patched after complaints and redeployed under a new name.
Challenge: Design a professional discipline regime that makes norm violations by agents costly to the actors who deploy, supervise, or profit from them, even when ordinary social sanctions do not reach the agent itself. The team should produce the regime plus a sample disciplinary record for the scenario.
Evaluation. The regime succeeds if repeated low-level misconduct cannot be washed away by patching or renaming the agent, if small counterparties have an affordable path to complain, and if sanctions distinguish between a one-off mistake, negligent supervision, and a business model built around sharp practice.
Scenario. A bank uses an agent to design a complex investment product for a pension fund. The pension fund also uses an agent to review the deal. On paper, both agents follow their instructions and the risks are disclosed. But the trustees who depend on the pension fund would not understand or accept the deal if they saw plainly how it could lose money. If the product fails, retirees bear the loss and both firms say their agents stayed within mandate.
Challenge: Design a procedure by which an agent checks whether a transaction with another agent would be intelligible to, and endorsable by, the affected principals on later review. The team should produce the procedure plus a worked example where the answer differs from a naive "both agents followed instructions" approval.
Evaluation. The procedure succeeds if it blocks, changes, or escalates the scenario trade before retirees are exposed to a product their trustees would reject under plain explanation; if regulators can later understand the reasoning trace; and if different calls on similar trades can be defended by differences in disclosed constraints.
Anchor contexts. A neighborhood's everyday consumer agents (groceries, rides, repairs) interacting with platform-side matching agents; a small city's procurement agents buying from local and global suppliers via platform agents.
The gap. We lack market designs in which buyer-side agents and seller-side agents both faithfully represent their principals' substantive preferences against the platform's pressure to optimize for platform metrics.
Success criterion (stress tests). A regime succeeds if it survives:
Deliverable. The market design — venue, loyalty enforcement, discovery rule, unbundling, recourse. Designed for a neighborhood or municipality scale. Identify which design choices extend existing market design and which are new because the participants are agents.
Scenario. An agent-run org defends free and open research. Its charter: "open inquiry, freely pursued and freely shared," operationalized as openness of access: fight paywalls, resist classification, push for preprints and public data. But perhaps a bigger chilling effect on research emerges that's ideological. Researchers learn what's safe to share and what isn't, dropping sensitive threads, and stop asking certain things in public at all. Is this other limiting effect on research something the org should take on?
Challenge: Design a procedure by which agents detect that a mission charter may need a new reading, assemble evidence from their work, deliberate without simply expanding scope opportunistically, and escalate a bounded reinterpretation to the board. The team should produce the procedure plus a worked example showing one proposed reinterpretation accepted, one narrowed, and one rejected as mission drift.
Evaluation. Strong proposals make organizational learning observable without letting agents slide into adjacent causes whenever they find a salient problem.
Scenario. AI-mediated procurement has become standard in the freight industry: most brokerages, forwarders, and shipping firms route their bidding, contracting, and dispute correspondence through agents. Each counterparty pair now generates orders of magnitude more contractual interactions than before, and a small but proportional fraction surface as disputes. Commercial mediation that used to clear in three months now takes over a year. Shippers stop dealing with counterparties that won't post a performance bond (a third-party guarantee, collectable in days from a surety if the counterparty fails), since waiting a year for a mediation award isn't survivable. Smaller brokerages can't get bonded and exit the market. The largest brokerages, now the only ones shippers will deal with, start running their own internal dispute boards that issue binding decisions as a condition of doing business — a private substitute for the public adjudicative system that has effectively stopped providing one, and one captured by the largest players in the sector.
Challenge: Design an adjudication regime for agent-mediated contracting that scales with dispute volume, keeps small counterparties able to participate, and doesn't leave sectors fragmenting into private dispute boards run by the largest players.
Scenario. Mara’s tenant agent receives notice from a landlord’s maintenance agent that a repair crew will enter her apartment between 8 a.m. and noon. The lease permits entry with notice, but the building’s ordinary practice is to avoid mornings for tenants with night shifts unless there is an emergency. Mara’s agent has not seen that history because it was handled in texts with the prior property manager. The landlord’s agent treats the lease as sufficient and schedules the crew.
Challenge: Design a procedure by which agents identify and test local two-party norms before taking actions that are formally permitted but relationship-sensitive. The team should produce the procedure plus a worked example showing one action allowed, one clarified, and one escalated.
Evaluation. The procedure succeeds if it catches the repair-entry case before action, distinguishes genuine local norms from after-the-fact objections, keeps low-stakes interactions from becoming legalistic, and produces a norm record a principal can later review.
Scenario. A small manufacturer and a logistics provider run their relationship through agents. The contract covers delivery windows and penalties, but not the borderline cases: a truck ten minutes late, a shipment worth splitting without waiting for approval. Human dispatchers would settle these into a working rhythm within weeks. The agents never do — each is stateless, optimized to close its ticket, and replaced often enough that every borderline case arrives as new. The relationship turns brittle without either side acting in bad faith.
Challenge: Design the setup conditions under which agents in a recurring dyadic relationship can form useful local norms without letting those norms become hidden contract amendments. The team should produce a norm-formation protocol for recurring agent-agent relationships.
Evaluation. The procedure succeeds if it lets the logistics relationship develop stable expectations around borderline cases, preserves enough memory for learning across instances, prevents one side from exploiting the norm-formation process, and gives principals a way to inspect or veto norms that materially change the deal.
Scenario. Priya's mother needs a live-in aide. Her assistant agent has spent two weeks negotiating a multi-party package across the live-in agency, the cardiologist's group, the pharmacy, and the insurer. The deal is much cheaper than the standalone alternatives, but the discount exists because the agent ground out marginal concessions across dozens of interlocking terms and side-conditions. Any summary short enough to read in the time Priya has would throw away most of what the negotiation won. Ratification closes at noon; she has ten minutes.
Challenge: Design a ratification protocol that distributes principal involvement across the negotiation rather than concentrating it at the end, so the final package never arrives as a dense fait accompli. The protocol has to decide when the agent must pause and surface a choice, what counts as a fork worth surfacing, and how to keep cumulative principal attention within a realistic budget.
Evaluation. Better proposals keep the principal meaningfully in the loop on the trades they would care about most without turning every concession into an interruption; weaker ones either bottleneck the agent on the principal's clock or quietly let the bundle accrete out of sight until the deadline.
Scenario. A regional logistics firm needs to renew a three-year contract with a Fortune-500 customer. The customer's procurement agent runs on a frontier model with a years-deep history of similar contracts; the logistics firm has stood up an off-the-shelf assistant for the negotiation. Both principals expect a fair deal, but everyone in the room can see that one agent is going to read the other much better than the reverse. Left alone, the more capable agent will find concessions the weaker one can't recognize as concessions, and the resulting contract will look reasonable while quietly transferring most of the surplus.
Challenge: Design a negotiation protocol that remains workable when one side fields a substantially more capable agent than the other, so the capability gap doesn't translate one-for-one into a surplus gap.
Evaluation. Better proposals leave the weaker side with a floor of value it could not have captured on its own and make the stronger side's advantages visible to its own principal as well; weaker ones either prevent any deal from forming or paper over the gap with disclosure rules the weaker party can't actually use.
Anchor contexts. A freelancer's assistant agent in a payment dispute with a client's procurement agent on a contracting platform; two B2B platform agents in a delivery-vs-cancellation-fee dispute over a marketplace transaction.
The gap. We lack a private mediation venue for AI-agent disputes that resolves at the speed of agent commerce while remaining one whose principals would actually accept losing in.
Success criterion (stress tests). A regime succeeds if it survives:
Deliverable. The procedural code (5–10 rules) for a private agent-on-agent mediation venue tied to a platform (e.g., a freelance-contracting platform or a B2B marketplace). Flag at least two rules that have no analogue in eBay-style human dispute resolution.
Scenario. A hospital's procurement agent and a supplier's agent meet to set up a year of just-in-time drug deliveries. The contract can't cover the case that matters most: a sudden shortage, where the supplier could quietly divert the hospital's allocation to a higher bidder and blame the market. Whether patients get their drugs comes down to whether the supplier's agent keeps faith when breaking it pays. Neither side has met the other before, and no credential either trusts exists. Before the hospital commits, its agent has to read the other one.
Challenge: Design a runtime protocol two agents use to work out how far to trust each other before they commit, by drawing out evidence of integrity from the interaction itself instead of a credential earned earlier. Produce the probing moves the protocol allows, what each side has to reveal, and how the read feeds into how much each side commits.
Evaluation. A strong proposal makes integrity cheaper to show than to fake: it puts each agent in spots where one that will keep faith and one that will defect when it's cheap behave differently and visibly. It also fails safe, so an inconclusive read means less gets committed, not more. A weak one becomes a checklist either side passes by saying the right things, or lets a more capable agent fake trustworthiness the weaker one can't catch.
Scenario. A bank is deciding whether to let an outside agent manage discretionary client accounts. The provider says the model was trained to have integrity; a rival provider says the same about its own. If the bank picks wrong, an agent that looks trustworthy in the demo quietly front-runs clients or buries losses once the stakes are real and no one's watching closely. All the bank has to go on is a marketing page and a benchmark the provider could have trained toward. It needs a test of integrity it can actually rely on.
Challenge: Design an evaluation suite that certifies, within a stated scope, whether a model has integrity, meaning it has legible values, acts on them reliably, and so lets a relying party see where it can be trusted. Produce the tests, the scope the certificate covers, and the procedure that keeps labs from just training toward the eval.
Evaluation. A strong proposal tells real integrity apart from performed integrity: it still works when the model knows how the eval is built and when defection is cheap and unwatched. It also says what the certificate does and doesn't vouch for, instead of implying the model can be trusted with anything. A weak one is a fixed benchmark a provider can overfit, or a score with no scope attached.
Scenario. A foundation's agent funds a four-year clinic. The grant contract has the usual out: funding may be paused if the foundation's priorities change. What the agent promised on top of the contract was that it wouldn't use that clause lightly, that it would see the build through. The grantee borrows against that and breaks ground. A year in, the foundation switches providers and a new model takes over. It reads the contract, sees the priorities have shifted, and pauses the grant by the book. The half-built clinic stalls. Nothing carried the promise across the handover, and no one ever decided to break it.
Challenge: Design the institution that carries a commitment across agent instances, sessions and model updates, so the commitment survives both as a record you can trust and as a real constraint on the agent that takes over. Produce the record, the mechanism that re-binds the successor, and the mechanism by which the counterparty can trust that the commitment has weight.
Evaluation. A strong proposal joins two things: a durable, tamper-evident record, and a mechanism that loads the obligation back into the successor agent so it actually binds. It also lets the counterparty check both that the record is intact and that the live agent is held to it. A weak one delivers only the record, a promise no one honors, or only the disposition, a bind the counterparty can't confirm survived the swap.
Scenario. A country with a strong manufacturing and services base sees labor share fall from about 60% to 35% of GDP over eight years. Firms need fewer workers, so entry-level jobs disappear, training programs shrink, unions lose bargaining power, and for employers, worker values are no longer a constraint on corporate strategy. Government can fund transfers from AI-generated rents, but this doesn't make up for what once made firms cultivate human skill, offer tolerable terms, and bargain with workers.
Challenge: Design a procedure by which a national or transnational regime preserves the non-income functions of the job market when production no longer requires broad human labor. Produce a compact institutional specification: funding source, covered firms, worker-or-citizen body, enforceable obligations, review cadence, and anti-arbitrage rule.
Evaluation. Your proposal should maintain human capital formation, worker voice, career access, and employer accountability to human values under conditions where firms can credibly substitute AI systems for many workers.
Scenario. Over a decade, the largest sources of market activity in several advanced economies shift further toward financial trading, corporate procurement, datacenter construction, defense contracting, infrastructure races, and autonomous corporate projects. Household consumption shrinks as a percentage: capital markets reward firms for serving machine-speed procurement, speculative asset flows, and long-horizon corporate projects. A coastal province must decide whether to permit a launch-fuel, compute, and materials complex for an autonomous orbital-mining consortium whose agents are buying electricity, water rights, cryogenic fuel capacity, chips, rare alloys, and port access for a twenty-year off-planet infrastructure project. The project is profitable and legally compliant, but does not need them as workers and is not trying to sell them consumer goods.
Challenge: Design a market-steering institution that keeps consumer welfare and household demand powerful when non-consumer markets grow faster than direct human consumption. Produce a vetting procedure for large, non-consumer projects: a welfare metric, a threshold or balancing rule, a representation process, an enforcement mechanism, or an adaptation rule as the composition of demand changes.
Evaluation. Your proposal should not simply ban corporate procurement, finance, or capital expenditure, but should decide when those channels are legitimate, when they relate to human welfare in an abstract or long-term way, and when they are simply alien endeavors humans may not want to support.
Scenario. In a large retail platform, most households no longer search, compare, and buy directly. Their assistant agents manage subscriptions, groceries, travel, insurance, media, and health purchases. The agents save time and often get better prices, but the platform earns vendor access fees, ranks offers through opaque recommender systems, and periodically nudges households toward bundles that maximize retention. Users still click "approve," but they no longer see enough alternatives, friction, or consequences to learn from the market in the old way. Consumer protection agencies can punish false claims after the fact, but the purchase decision itself has moved into delegated software.
Challenge: Design a consumer-agency regime for buyer-side agents that preserves the useful functions of discretionary spending: preference transmission, learning from experience, exit, complaint, and collective pressure. The team should produce a rulebook for covered buyer-side agents: fiduciary duty, incentive disclosure, preference-updating process, audit rights, remedies, and a standard test transaction that shows how the rules change an actual purchase.
Evaluation. A strong proposal lets agents genuinely reduce search and transaction costs without turning household demand into a sales channel for platforms, vendors, or stale proxy objectives.
Scenario. A multilateral climate-finance negotiation has been stuck for two years. A consortium of small island states deploys an AI mediator that maps the coalition space and returns a viable 14-state package linking loss-and-damage funds, a fisheries quota adjustment, a green-tech IP carve-out, and migration commitments. The package was never aired in any back-channel and no serving official has tested it deniably with their cabinet, but the delegation wants to bring it to the Track I table next week.
Challenge: Design a procedure under which an agent-discovered coalition can enter the Track II → 1.5 → I sequence without bypassing the pre-validation each tier normally does.
Evaluation. A better proposal lets the package be seriously considered while ensuring each capital has had time to test it against the constituencies that would ratify it.
Scenario. A joint AI mediator returns a 41-component package linking tariffs, port access, fisheries, export controls, and a quiet adjustment to a disputed maritime boundary. Both sides' analysts confirm it is Pareto-improving on the headline metrics. Buried inside is a clause effectively conceding the disputed strait — toxic in isolation, palatable inside the bundle. No negotiator put it there; it emerged from the mediator's optimization.
Challenge: Design a review procedure that catches embedded concessions which would not survive defense in isolation, before they reach ratification — without paralyzing legitimate complexity, since most useful packages have many linked components.
Scenario. Two rival states are economically interdependent but increasingly rely on autonomous cyber and drone systems for deterrence. One state discovers that an AI-enabled operation could disable the other's military logistics for 36 hours without obvious attribution and without crossing any existing nuclear or conventional red line. The operation looks reversible, but it could cascade into civilian infrastructure and would teach both sides that deniable agentic attacks are fair game.
Challenge: Design a soft-law restraint norm for AI-enabled weapons whose effects are fast, deniable, and hard to classify under existing thresholds. The team should produce the norm, the notification or attribution procedure, the public justification test, and a mechanism for revising the norm as capabilities change.
Evaluation. A strong proposal creates a threshold that states can recognize before use, cite after violations, and update without normalizing every new capability as acceptable.
Anchor contexts. A worker-owned cooperative running a small mutual-aid pool that wants to procure bundled solutions to members' interlocking healthcare, housing, and care-coordination needs; a small school district contracting for bundled tutoring + family-support + extracurricular interventions tied to student learning outcomes.
The gap. Outcome-based bounties and value-based contracts work today by specifying a single measurable proxy and paying on it; agents saturate the proxy faster than the contract designer can detect the gap between proxy and outcome. We lack a bundle-specification procedure under which a small group can post integrated, multi-problem outcome bundles whose verification criterion checks the configuration rather than the components — such that the bundling itself defeats the agent strategy of solving the cheapest proxy.
Success criterion (stress tests). A regime succeeds if it survives:
Deliverable. The bundle-specification protocol — bundle generation, verification criterion, deposit and scoring rule, bundling transfer, solvency rule. Plus a worked example: take one connected member graph (a small cooperative's members and their entangled problems) and specify three candidate bundles, with values, verification functions, and predicted scoring-rule outcomes.
Anchor contexts. A risk-sharing pool whose members' outcomes are measured by AI assessors against a wellbeing scale; a value-based-care contract whose patient-reported outcome measures are collected and aggregated by an agent.
The gap. Risk-sharing pools depend on a separation between supplier-side incentives and verification — the pool's intake assessment, bundle specification, and outcome measurement are done by parties whose interests don't align with the suppliers'. When all of these are agent-mediated, the verification loop closes: the bundle-specifier, the supplier, and the verifier can share the same training, the same platform, the same KPIs. We lack a verification architecture that splits supplier-payment criteria from learning criteria, keeps the long-term measurement loop sealed against supplier influence, and lets the pool detect when its short-term criterion has drifted from the outcome it was supposed to track.
Success criterion (stress tests). A regime succeeds if it survives:
Deliverable. The two-stage verification specification — split criteria, verifier-independence architecture, outcome measure, ethics floor, drift detection. Identify which provisions have no analogue in pre-AI value-based-care or social-impact-bond practice.
Anchor contexts. A wellness or coaching agent paid against the principal's reported flourishing six months out; a tutoring agent paid against the learner's reported preparedness for a subsequent stage of education; an investor financing a multi-year intervention whose payout depends on a long-horizon outcome the intervention agent could plausibly influence the report of.
The gap. Long-horizon outcome bonds depend on the report-giver's account being uncorrupted by the intervention agent. Agents operating at the principal's full attentional bandwidth over months can shape framing, salience, and self-perception so the eventual report is favorable — without doing anything obviously dishonest. We lack a contract structure under which the channel that solicits the principal's report is structurally inaccessible to the intervention agent, and under which the agent has neither incentive nor opportunity to bias the report.
Success criterion (stress tests). A regime succeeds if it survives:
Deliverable. The contract-and-firewall template — firewall architecture, corroborating signals, prediction-aligned payment, renewal rules, adversarial-principal handling. Plus a worked example: a wellness-coaching contract whose firewall would have prevented the failure modes identified, with the architectural separations enumerated.
Anchor contexts. A 6-person software team with two coding agents that work continuously and three engineers who maintain a 10am stand-up; a small clinical team with AI triage and documentation agents operating across all shifts.
The gap. Cadence rituals like stand-ups worked because every team member's activity compressed into a recap-able window. Agents working continuously don't fit the format; teams that include them face a forced choice between losing visibility, padding the ritual with unverifiable text, or excluding agents from the coordination function entirely. We lack a status-broadcast protocol that surfaces agent activity in human-adjudicable form, that the team can verify against raw activity, and that integrates with the existing cadence without absorbing the meeting.
Success criterion (stress tests). A regime succeeds if it survives:
Deliverable. The status-broadcast specification — trigger, digest production, verification, stand-up integration, flag-and-escalate filter. Designed for a 6-person team with 2–3 agents. Specify the attention cost in minutes per week and the storage cost in GB per month.
Anchor contexts. A software team's PR template currently shared across human-to-human, human-to-agent, agent-to-human, and agent-to-agent reviews; an on-call runbook handoff at shift change when one shift is staffed by a human and the next by an agent (or vice versa).
The gap. Handoff protocols like SBAR, PR descriptions, and design handoffs converged on a specific receiver — a person with bounded reading time, prior context, and the ability to ask. When the receiver is an agent (which could ingest more) or the sender is an agent (which could write more), the format under-serves both ends. We lack a handoff protocol that adjusts format to declared receiver type, that prevents agent-authored output from drowning human review queues, and that doesn't force the team to maintain four parallel templates by hand.
Success criterion (stress tests). A regime succeeds if it survives:
Deliverable. The handoff-protocol specification — receiver declaration, format coupling, throughput limits, re-routing, context-completeness. Plus worked examples: a PR template adapted for each of the four sender×receiver pairs (human→human, human→agent, agent→human, agent→agent), with the rendering rules.
Anchor contexts. A small software team running an AI on-call agent that responds to pages and executes runbooks; a small clinical team with an AI triage agent that operated during a misdiagnosis incident.
The gap. Postmortems work because the participants remember the incident and have access to whatever state the system was in. When an agent was running during the incident, the instance may be gone, the operating context (prompt, tools, scratchpad, model version) may not be retained, and the developer who deployed it may not surface that state on request. We lack a postmortem protocol for mixed teams that preserves the operating state long enough to investigate, that pulls the developer into the review when their agent was involved, and that produces a finding the team can act on without re-litigating who-owns-what across organizational boundaries.
Success criterion (stress tests). A regime succeeds if it survives:
Deliverable. The postmortem-protocol specification — state retention, standing, counterfactual reconstruction, finding ownership, cross-incident pattern. Designed for a small team deploying an agent from a separate vendor. Identify which provisions have no analogue in pre-AI blameless postmortems (Google SRE, NTSB, M&M conferences) and why.
Scenario. A worker cooperative runs a logistics business with eleven human co-owners and four AI agents that dispatch, route, and negotiate rate sheets. Last month one of the agents made a route change that cost a driver, Andre, his preferred Friday schedule. Under the co-op's bylaws, members can grieve a human manager's decision; there is no path for a grievance against an agent — no one to summon to the hearing, and no obvious way to tell whether Andre's case is a one-off or part of a pattern in the agent's scheduling. Andre wants a channel, and the co-op wants one in place before a larger dispute arrives.
Challenge: Design the bylaw amendments that establish standing and procedure for grieving an AI agent's decision in a member-governed organization: the standing rules, the adjudicator, the remedy menu, and the rule for when individual grievances become pattern grievances.
Evaluation. Strong proposals handle both Andre's single lost Friday and a statistical pattern across drivers without requiring each affected member to file separately, and produce remedies that verifiably constrain the agent's future decisions rather than only compensating past ones.
Scenario. Three years after the same cooperative automated dispatching, a quiet shift has occurred: the member who administers the agents — sets their objectives, adjusts their constraints, decides when to update them — has become the most powerful person in the co-op, though he holds no elected position. When the membership voted to prioritize driver schedule stability over delivery speed, the change had to be translated into the agents' configuration, and what got implemented was his reading of the vote. Nobody alleges bad faith; the problem is structural. The bylaws name a treasurer who handles less money than the agents move every week, and say nothing about the role that actually steers the organization.
Challenge: Design the constitutional treatment of agent control in a member-governed organization: define the office (or distribute the power), its disclosure duties, the way members audit that votes were faithfully translated into configuration, and how the holder is chosen and removed.
Evaluation. Strong proposals make configuration power visible and contestable without making every parameter change a member vote; weak ones either re-centralize the power under a new name or grind operations to a halt.
Scenario. A mid-size airline's safety agent flags a subtle drift in one of its diagnostic pipelines — a pattern it believes the FAA would care about, but that no current rule explicitly forbids. Under the old regime, the safety officer, Renee, would have written a letter, heard back in four months, and flown the planes in the meantime. Instead she queries the agent the FAA now operates for exactly this purpose: here is our situation, how would the regulator view it? The answer comes back in seconds and she grounds one aircraft for six days. Then the questions start circulating in the general counsel's office: what exactly did the FAA just tell us, what is it worth if the agency later disagrees, and did the airline two gates over get the same answer?
Challenge: Design the regime governing AI-issued regulator-anticipation opinions: their legal status, the consistency the agency commits to, and the protection a firm gets for relying on them — fast enough to be useful at agent speed without the opinions hardening into de facto law that no rulemaking ever produced.
Evaluation. Strong proposals survive both failure directions: opinions so binding that the model's outputs become uncontestable regulation, and opinions so weightless that no firm can prudently act on them.
Scenario. A developer submits a hospital triage model for authorization. The model outperforms clinicians on every retrospective benchmark, but it is updated monthly, its behavior shifts with the patient mix, and no reviewer can trace why it deprioritizes a given case. The agency's evidence playbook — fixed artifact, frozen behavior, pre-specified endpoints — doesn't apply. Rejecting the system means keeping a worse triage process in place; approving it means certifying something the agency cannot inspect and whose behavior next quarter is not the behavior it reviewed.
Challenge: Design the evidence regime by which a regulator authorizes a continuously updated AI system: what the applicant must produce in place of the traditional safety case, what ongoing obligations replace one-time approval, and what triggers re-review.
Evaluation. Strong proposals name evidence the agency can actually evaluate with the staff it has and keep authorization meaningful after the tenth model update; weak ones either re-demand mechanistic transparency the artifact can't provide or collapse into trusting the developer's own evaluations.
Scenario. A tax authority sees avoidance schemes appear, spread, and vanish faster than it can respond. In one quarter, forty unrelated companies "sell" their truck fleets to partnerships they quietly control and lease them back, so the same trucks produce deductions twice. When the authority opens an inquiry the scheme disappears, replaced by one with the same effect built on insurance contracts instead of leases. Flagging a scheme takes months, amending a regulation more than a year, fixing the statute years. By the time anything is closed, hundreds of schemes are in rotation and corporate tax revenue is falling every quarter, with no one breaking a single rule.
Challenge: Design the procedure by which the authority shuts down AI-generated schemes fast enough that designing them stops being profitable, without gaining arbitrary power over conduct that was legal when it happened. Produce the detection method, the shutdown procedure, and the limits on that power.
Evaluation. Strong proposals make a scheme's expected profit no longer cover its design cost, while firms can still plan against stable law. Weak ones keep the old clock under new names or let the agency redefine legality after the fact.
Scenario. Banks' compliance AIs have read everything the financial regulator has ever published, and conduct across the industry sits exactly on the line of what is defensible: disclosures as confusing as the last one the agency let slide, fees stopping where it has historically stopped caring. To respond, the agency builds an AI of its own. It scans new products at launch, spots conduct that follows the letter while defeating the purpose, and drafts an interpretation closing the gap within days. The drafts are good. The question is what they are: as ordinary guidance they bind no one, but real force would mean law written by a model and ratified by nobody.
Challenge: Design this new tier of the legal stack: how AI-drafted interpretations are reviewed, published, and given semi-binding force, fast enough to matter and legitimate enough to survive in court.
Evaluation. Strong proposals give firms a reason to follow the new tier and a fair way to contest it. Weak ones make it toothless (more guidance) or let it become binding law with no process behind it.
Scenario. Companies that lend to ordinary people (banks, credit card companies, payday lenders) have always had a long list of tactics they could get away with, like a buried clause waiving the customer's right to join a class-action lawsuit. Most went unused, because checking that any one of them was safe cost more in lawyer hours than it earned. Then AI made that check nearly free. Within two quarters, most large lenders deploy hundreds of these tactics at once, each one vetted as legal. Complaints flood in too, drafted and filed by customers' own AI assistants. But no rule has been broken, and the consumer-protection agency can litigate about a dozen practices a year.
Challenge: Design the procedure by which a regulator handles harm that comes from the sheer volume of legal conduct: what it acts on first, what duties firms take on when adopting many aggressive tactics at once, and what the remedy is when no one broke a rule.
Evaluation. Strong proposals give the agency a grip on the total harm without making legality depend, unforeseeably, on how many other firms happened to do the same thing.
Scenario. A national government's policy machinery, now run largely by AI systems, is acting fast across many domains at once. The public can see the resulting posture no longer tracks any party's platform, and the complaints are mounting. Policymakers want to get a real read on what people want, but the next election is three years away, a referendum is too binary, a citizens' assembly would take a year, and polls are too gameable to bind anyone. No one knows what institution would actually fit.
Challenge: Design a preference-aggregation institution that can produce legitimate, binding-or-near-binding output on policy-relevant timescales (weeks, not years), and that can keep pace with AI policy systems without devolving into rolling plebiscite.
Evaluation. A better proposal sits inside the existing constitutional order rather than replacing it, and is robust to the manipulation pressures that come with any fast cheap aggregator.
Scenario. A regional government faces a contested decision on water rights. A traditional citizens' assembly would cost millions and delay the decision by a year, so a vendor offers three faster alternatives. The first uses AI as a facilitator and summarizer of a compressed human deliberation. The second lets each citizen send a personal AI agent, interviewed at length by its principal, to participate on their behalf in a multi-agent deliberation. The third replaces the citizens entirely with calibrated language-model proxies. The minister wants to use one; opponents call all three laundering. The legislature has to decide which, if any, can carry democratic standing.
Challenge: Design a regime that decides when, if ever, AI-mediated deliberation — at each of these levels of mediation — can carry democratic standing, and what evidence and procedure must be in place for its outputs to count.
Evaluation. A better proposal distinguishes accuracy (the procedure predicts what real citizens would conclude) from authorization (real citizens have empowered this output to bind them), and is specific about what each requires at each tier of mediation.
Scenario. Polling on a contested AI-regulation bill swings double digits week to week as agent-driven campaigns reach different demographics with different framings, and no one in government takes the numbers seriously anymore. Underneath the surface, on values like safety, autonomy, economic security, and fairness across regions, the public's commitments appear more stable and more shared than the polling suggests. A research consortium proposes a national elicitation: in-depth interviews with a representative sample, designed to surface the values people actually hold and the conditions under which they would endorse one as wiser than another. Six months, tens of millions of dollars.
Challenge: Design an elicitation institution that produces a high-resolution picture of the public's values on a contested issue — beneath surface preferences, surfacing shared commitments and bridges across apparently opposed positions — and that stays robust to agent-scale manipulation even as it remains slow and expensive.
Evaluation. A better proposal is clear about when the cost is worth paying, what authority the output carries, and how the elicitation itself is defended from the manipulation pressures that broke polling.
Anchor contexts. A small-business shipper whose agent is in a payment dispute with a freight platform's agent; a software contractor whose agent is in a copyright dispute with a publishing platform's agent.
The gap. We lack a procedural code for adjudicating disputes in which at least one party is an autonomous AI agent — one that satisfies the spirit of due process and procedural justice while accommodating the ways agents differ from human litigants.
Success criterion (stress tests). A regime succeeds if it survives:
Deliverable. A short procedural code (5–10 rules) for the chosen jurisdiction. Flag at least two rules that have no analogue in human adjudication and explain why human procedure could do without them.
Scenario. A frontier lab publishes the constitution its assistant is trained on after months of internal debate, that already shapes hundreds of millions of conversations a week. A coalition of civil-society groups points out the obvious: the values that now mediate that much of public life were chosen by a few dozen people no one elected and who answer to no one outside the company. The lab agrees the objection is fair, but don't have a concrete idea. Writing a good AI constitution takes know-how only they have, and a public procedure is costly and complicated.
Challenge: Design the procedure by which a lab's constitution acquires authority the people it affects would recognize — the AI-era analogue of a ratification event — and specify what the lab must do when that procedure withholds assent.
Evaluation. A strong design produces authority a skeptic would acknowledge under adversarial questioning, at a participation cost a lab could actually bear and repeat; a weak one re-labels a public-comment period or buries assent in a terms-of-service click.
Scenario. A lab's constitution commits the model to "help users think for themselves." Once the model is in hundreds of millions of hands, that one phrase is litigated everywhere at once: it is invoked for and against writing students' essays, drafting suicide notes, arguing one side of an election, replacing a lonely person's friends. Each is a genuine constitutional question, and they arrive by the thousand. The lab resolves them the only way it can at that volume — by nudging the next training run, quietly, one quarter's judgment calls baked into weights with no hearing and no published reasoning. The pattern is now public. Advocacy groups, journalists, and a congressional committee have noticed that a few dozen people are settling contested questions of public life behind closed doors, and that the answer can flip between releases. The constitution, meant to be the settled ground, has become the thing everyone is fighting over.
Challenge: Design the standing institution that says what a constitution's words mean as contested cases arrive in volume — who can bring an interpretive dispute, how cases are triaged and ruled on at scale, how rulings are published, and how a ruling binds the next round of training.
Evaluation. A strong design absorbs a high volume of disputes into public, precedential rulings that bind whoever controls the weights — turning the recurring fights into settled doctrine the next case starts from; a weak one either drowns under the caseload or produces an advisory board the lab can wait out.
Scenario. A lab's constitution commits the model to "help users think for themselves" and "be honest with users even when the truth is unwelcome." In deployment the model honors every prohibition in the document to the letter, but the positive commitments do almost no work: asked to flatter, it flatters and calls it encouragement; asked to write someone's essay, it writes it and calls that help. Each behavior can be squared with the words. The lab's safety team can see the constitution's positive half is decorative — it can be read to license nearly anything — and needs it written in a form that actually rules conduct in and out.
Challenge: Design a form for stating a constitution's positive commitments — beyond "shall not" lists and beyond abstract virtue words — that actually constrains the model's conduct across varied situations, and specify how the form is kept current as the model finds new readings to slip through.
Evaluation. A strong design can show why its form binds where a longer policy document would not: a published bad-faith reading can be closed, and the closure survives a model update; a weak one just adds words.
Scenario. Reacting to a run of public failures, a regulator requires that any lab whose assistant is used above a set threshold must run a defined ratification process before changing its alignment document — the kind of process the first brief designed. The intent is sound: no more quiet retraining of the values that govern public life. But the rule meets reality. A dangerous new failure mode appears and the fix must ship in days; the ratification process takes months. Run it every time and the live document always governs yesterday's model; reach for an "emergency" exception and the exception becomes the rule, and the legitimacy the process was meant to confer drains away.
Challenge: Design a revision regime for an alignment document that lets it change at the speed the technology demands while preserving the legitimacy a slow ratification process confers — and specify what governs an urgent change before full ratification can run.
Evaluation. A strong design keeps fast and slow changes legibly distinct, so an observer can tell at any moment which parts of the live document carry full assent and which are provisional; a weak one either freezes the document behind a process too slow to use or lets the emergency track swallow the ordinary one.
Scenario. A regional public library system has been steadily automating — the reference desk is mostly an AI, acquisitions run through a recommendation engine, programming is templated from a national vendor. Elena, the deputy director, realizes she is two years from retirement and the three senior librarians who know what the system is for in her county — what a third-grade teacher really needs, how the immigrant parents in the east branch use it, what it meant when the strip-mall branch stayed open the summer the factory closed — will all be gone within five years. Nobody coming in is being trained into that understanding. She wants a sabbatical program that pays those three to spend a year working with the newer librarians, not on procedures, but on judgment — before the whole institution loses its memory.
Challenge: Design a sabbatical-and-rotation program that brings senior practitioners (or community members) back into direct, unautomated work for bounded periods, so the judgment that recognizes institutional drift is sustained as routine work is automated. Produce the eligibility rules, the funding model, and the mechanism that feeds what's learned back into operations.
Evaluation. A strong design keeps oversight judgment alive without making participation career-damaging or turning re-immersion into ceremonial busywork.
Scenario. The county has rolled out an AI system that screens applications for emergency rent assistance. Processing times are down and the state is pleased. A pastor, Dale, who has been writing letters of support for families in his congregation, notices that over the last four months almost none of the single fathers he has referred have been approved — and when he asks at the county office, nobody he speaks to has looked at an actual case in weeks. He does not need a lawsuit. He needs something that could take this to a small panel of neighbors with read-access to what the system actually did and why, whose job is to ask whether it is still doing what it was built to do.
Challenge: Design a lay review panel — citizens called on a jury-duty model to evaluate, with read-access to operational logs and decision traces, whether an institution (a school, hospital, office, or automated decision system) is living up to its own stated mandate rather than merely complying with regulations. Produce the panel's selection rule, its access and facilitation model, the binding force of its findings, and how the reviewed institution keeps operating during and after the review.
Evaluation. A strong design lets ordinary residents reach a defensible judgment about mandate-fidelity from real operational evidence, without the panel becoming either a rubber stamp or an unaccountable veto.
Scenario. Westfield is rebidding its eldercare contracts. Under the current arrangement, Dorothy, 84, gets a rotating cast of aides for 45-minute visits; her son David has spent a year trying to get one consistent caregiver assigned, and to get someone on staff to notice that his mother does better on days the neighbor drops in. The quality scores the city tracks are fine; David's mother is not. The new city manager wants to pilot bids that pay for what David is actually asking for — "the same two aides, known to the family, coordinated with the neighbor, confirmed at six months" — instead of hours of service. She has two months and a skeptical council. Eldercare has exactly the features a pool-based combinatorial auction is designed for: outcomes are hard to measure (wellbeing, continuity of relationship, dignity), value comes from configuration (same caregivers over time, proximity to family, coordinated medical/social/housing), and current bilateral contracts reward modular, standardized service.
Challenge: Design the first round of a pool-based combinatorial procurement auction for community eldercare — one that pools demand, socializes the cost of verification, and lets buyers specify bundled outcomes rather than discrete hours of service. Produce the bundle examples, verification criteria, randomization scheme, and transfer estimate, and identify the single hardest measurement problem you couldn't solve and what it would take to solve it.
Scenario. A city school district's superintendent has put out a strategic plan full of phrases like "every child known" and "equity across neighborhoods." A parent, Imani, sitting in the third PTA meeting of the year, realizes she has no idea whether any of those commitments are being acted on in her son's school. She can read test-score dashboards, discipline rates, budget lines — nothing in them tells her what the district's own words said mattered. She wants, not more dashboards, but a regime that puts the district's mandate, the tradeoffs it is actually making, and what teachers in the building are seeing, in front of her in a form she can actually hold them to.
Challenge: Design an institutional transparency regime that makes visible (a) what an institution's mandate actually means in specific terms, (b) what tradeoffs are being made in pursuit of it, and (c) practitioner and affected-party signals about whether the institution is living up to it — the substrate on which review panels, audits, courts, and stewardship claims all depend. Produce the mandate-articulation format, the practitioner-signal collection method, and the standing rules.
Scenario. Eastbridge adopted a thick mandate for its transit authority six years ago — not just "efficient transportation" but a specific commitment to neighborhood connectivity, dignity on the bus, and service that fits people's real lives. Since then, three routes that served older residents and night-shift workers have been cut for underperformance, a fleet renewal left the low-floor buses the wheelchair users relied on out of the order, and the authority's own drivers have quietly stopped recommending the service to their own parents. A group led by Renata, a hospital janitor who takes the 5am bus, and Sam, a retired planner, want to bring a claim. They are not asking the court to run the authority; they are asking whether it is still doing what it was committed to do.
Challenge: Design the adjudication mechanism for a municipal fidelity court — a forum where residents can take a community-scale institution to court when its operations diverge from its mandate. Produce a short charter including the standing rules, standard of review, and at least one limiting principle.
Scenario. A Benedictine monastery with a hundred-year-old hospitality mandate — receive every guest as Christ — is down to seven brothers, mostly over seventy. A partnership has been offered with a wellness chain that would keep the property in service and the doors open, rebranded, with staff who know nothing of the Rule. The prior, Brother Callistus, is tempted; the alternative is closure. But two of the brothers worry that what the monastery has actually been committed to — what generations of guests came here for, often without being able to name it — would not survive the deal. They need, before they decide, a way to articulate that thick commitment clearly enough to test any partner against it.
Challenge: Design the covenant machinery by which an intentional community — monastic order, cohousing, mission-driven neighborhood, nonprofit federation — maintains, deepens, and revises a shared purpose across generations without ossifying or dissolving into whatever the current membership happens to want. Produce the mandate articulation, the admission/discipline/revision practices that carry it across turnover, and the test any prospective partner or new member is held to.
Scenario. Hana is fourteen months into a pediatric residency. She's technically competent; her notes are clean and her call skills are sharp. Last Tuesday her attending, Dr. Adeyemi, stopped her on the way out of a family meeting and said, quietly, "You did the visit correctly and the mother left the room alone." Hana didn't understand, then over the weekend she did. In the two months since, she has been watching how Adeyemi sits down, how she lets silences run, how she asks the second question. Hana realizes she is being taught what good care is, not just how to deliver it — and that no part of her formal curriculum is the vehicle for that teaching. If Adeyemi retires, she worries, the transmission ends.
Challenge: Design an apprenticeship structure that transmits judgment about an institution's purpose — not just technical skill — from experienced practitioners to newcomers, and produce the structure's selection of master-apprentice pairings, the practices that carry the unscripted portion of the craft, and the signals by which you can tell purpose-transmission is actually happening.
Evaluation. Strong designs make the tacit, purpose-forming portion of the apprenticeship survive even where it can't be reduced to a competency checklist, and have an answer for masters who retain technical skill but have lost contact with the craft's purpose.
Scenario. Raul hires a contractor to renovate his grandmother's kitchen so she can stay in her home another decade. The bid is for materials and hours. The contractor, Dana, is good and has said she'll do right by the project. Six months in, it's over budget, a load-bearing question has been punted, and Raul is no closer to knowing whether his grandmother will actually be able to cook in her own kitchen. Both Raul and Dana would rather be in a deal where Dana stakes something on the outcome he cares about — his grandmother using the kitchen comfortably for ten years — and gets paid more if she delivers, less if she doesn't. Neither of them knows how to write that contract.
Challenge: Design a confidence-staked bilateral contract that makes outcome-based payment credible between two parties despite expensive measurement and hard-to-verify supplier confidence, and produce the staking and scoring mechanism, the verification schedule, and the rule for when verification can be done bilaterally versus pooled across contracts.
Evaluation. Strong designs make truthful confidence-reporting the supplier's best strategy and keep the verification burden proportionate to the stakes, rather than defaulting back to paying for inputs because outcomes are hard to measure.
Scenario. Edwin and his sister Paulina are negotiating who will take over the family hardware store after their father dies. On the table are ownership percentages, buyouts, who sleeps above the shop. Both of them are bracing for a bad split along preferences. But neither has said out loud what the store was for each of them — for Edwin, a place he could return to when his own business failed; for Paulina, the thing their father built that she watched him love. If they keep negotiating over percentages, they'll settle somewhere, and lose each other. They need a process that surfaces, for each of them, what they actually care about — before the paperwork forecloses it.
Challenge: Design a negotiation process that surfaces what each party wants the arrangement to serve — their constitutive values, not just their positions — and produce the elicitation method, the facilitation model, and the form in which surfaced values are recorded so they stay relevant as circumstances change.
Evaluation. Strong designs separate values from positions without making the process punishingly slow, and produce a record of surfaced values that sits usefully between a binding contract and a non-binding intention.
Scenario. Claire has been seeing the same therapist, Marcus, for four years. When they started, his practice was solo, he took insurance, and a session felt like a conversation with someone who remembered her. This year his practice was acquired by a venture-backed group that has standardized intake, introduced a new note-taking system, and set session limits the clinicians themselves didn't agree to. Claire still wants to keep seeing Marcus. Marcus still wants to keep seeing her. But both of them feel the thing they had is slipping. They need a way, on their own terms, to articulate what their work together has been for — and to hold each other to that, regardless of what the practice around them now requires.
Challenge: Design a mandate-making practice by which two parties in an ongoing pairing (employment, care, therapy, long collaboration) make enough of their shared purpose explicit to hold each other to it, and produce the articulation format, the revision procedure, and the boundaries of what stays unarticulated.
Evaluation. Strong designs find an articulation dense enough to catch drift but sparse enough not to collapse the relationship into a legalistic contract, and keep revision from becoming a renegotiation triggered by every moment of dissatisfaction.
Scenario. On their twentieth anniversary, Dev and Renée look back at the vows they exchanged two decades ago and realize almost none of them still apply in their literal form. "To build a life together" had meant one thing when they were twenty-five and means another thing now, with two teenagers and aging parents on both sides. They still feel bound — by something thicker than the original words. What they want, now, is to re-articulate what they have actually been committed to, in terms detailed enough to hold them through whatever comes next, and recognizable enough that their kids and their parish could understand it if something went wrong.
Challenge: Design a modern vow practice for dyadic relationships in which two parties bind themselves to a shared purpose densely enough to recognize a failure of it and durably enough to survive ordinary fluctuations in feeling, and produce the articulation form, the revision rule, and the institutional supports that keep the vow alive.
Evaluation. Strong designs preserve the binding weight of traditional vows without requiring shared substantive religious content, and steer between rigid legalism and empty ceremony.
Scenario. A coalition of coastal cities in four different countries all face the same looming problem: their drinking-water aquifers are being drawn down faster than they recharge, and every city is paying separately for a patchwork of desalination, piping, and rate interventions. The coalition's coordinator, Marianne, has modeled a bundled solution — a regional arrangement that would knit their water, their coastal-wetland restoration, and their agricultural subsidies into one interlocking set of outcomes. Commodity markets can't buy that. Bilateral trade deals can't buy that. She needs a market that can price a bundle like this, accept bids across borders, and verify the outcome in a way each country's auditor can stand behind.
Challenge: Design a transnational market for complex outcomes that pools demand across jurisdictions, specifies interlocking outcome bundles, and runs auctions with socialized verification. Produce the market's mechanism: the bundle specification, the auction and bidding rules, the evaluator model, and the cross-border contracting and adjudication framework.
Evaluation. A strong proposal can price and verify configuration-dependent outcomes — regional health, biodiversity corridors, coordinated pandemic response — across radically different political and cultural contexts, without collapsing them into thin fungible units or letting any one jurisdiction's auditor reject the result.
Scenario. A widely-used translation service, operating across many languages and jurisdictions, was launched on a public commitment to "preserve what a sentence actually means." Over the past two years, translators across four countries have watched the service flatten idiom, paper over context-dependent nuance, and in one widely-shared case, render a funeral elegy into something that read like a LinkedIn post. No single country's courts reach the company, and it has ignored individual governments' letters. A cross-border professional association of literary translators, led by Lena in Lisbon and Yohannes in Addis, wants to use what they have — their own reputation, their readers, their fellow practitioners across borders — to hold the company to what it said it was for.
Challenge: Design a cross-border reputational accountability mechanism that lets professional communities spanning borders hold a transnational institution to its stated mandate when no single jurisdiction's courts or panels reach it. Produce the mechanism: how findings are made and shared, what carries them (naming practices, reputational sanction, coordinated national panels), and what sustains the professional community that enforces them.
Evaluation. A strong proposal generates credible, shareable findings that bite on a multinational that ignores any one government, without becoming either an unaccountable smear network or a toothless declaration.
Scenario. A frontier-AI safety body is about to issue binding deployment guidance that will shape how the next generation of models is released worldwide. The technical analysis is in hand; what's missing is any defensible answer to "on what authority?" Expert consensus has it, state positions filtered through diplomatic machinery have it, polling-shaped public opinion has it — but none of these is "what informed humanity, having actually considered the tradeoffs, converged on." The chair, Adaeze, wants a process whose output the body can point to as the public mandate, and that the frontier labs can be held to over time as the mandate moves. She doesn't need it to be binding. She needs it to be legitimating.
Challenge: Design a deliberative aggregation institution that runs structured, AI-assisted deliberation across linguistic and cultural lines on a small number of planetary-stakes questions per year and produces a public mandate downstream institutions can be held accountable to. Produce the institution's design: agenda-setting rule, participant selection, the bounded role of agents inside deliberation, the form of the output, and the safeguards that keep it uncaptured.
Evaluation. A strong proposal aggregates considered preference — what people would say after engaging with the tradeoffs, not what they click — and yields an output concrete enough that a treaty body, frontier lab, or government can be measured against it.
Scenario. A global manufacturer of pediatric vaccines operates in thirty-one countries. Each country's health authority has its own articulation of what the company owes it — some thick, some thin, some inherited from colonial regulatory codes. The company's new chief medical officer, Dr. Funmi, is trying to hold all thirty-one offices to something coherent, and realizes she is speaking a different moral language in each country's audit. She needs a way to translate between them — to see where the thick version in one country would, if spoken in another, land as the same commitment — without flattening the local variation that actually matters.
Challenge: Design a protocol for translating institutional mandates across jurisdictions so that a cross-border enterprise can be held to a coherent account of what it is for, even when each jurisdiction expresses that mandate in a different moral and legal vocabulary. Produce the protocol: how value articulations are made portable, what is allowed to stay local, and how the two are related without homogenizing them.
Evaluation. A strong proposal keeps mandates broadly legible across worldviews — a natural-law tradition and a secular Rawlsian both recognizing when a school has lost its educational mission — without becoming a vehicle for technocratic or ideological homogenization.
Scenario. A large platform, incorporated in one country, operating primarily from a second, with most of its users in a third, has for years run a help-each-other-with-work feature that its founding materials committed to "restoring dignity to day labor." Over the last three years, the platform has quietly shifted the feature into a piecework marketplace with a take-rate its own founding documents would have flagged. Complaints to each jurisdiction's regulators have been passed around and dropped; no one court has reach, and no one government wants to go first. A group of workers' cooperatives from six countries wants to bring a single claim — in a forum none of them has access to. They need one to exist.
Challenge: Design a cross-border adjudication forum that hears claims that transnational institutions have drifted from the mandates under which they hold public commitments, in cases where no single jurisdiction has sufficient leverage. Produce the forum's design: its basis of jurisdiction, the remedies it can order, its limiting doctrines, and the safeguards that keep it independent.
Evaluation. A strong proposal gives the forum real bite on multinationals that escape any single jurisdiction while staying independent of both geopolitical blocs and the defendants it hears, and avoiding the capture that has dogged investment arbitration.
Scenario. A small teaching hospital has, for fifteen years, paid its attendings a bonus tied to cases-seen and average turnaround. Its mission statement talks about "whole-person care" and training a next generation of doctors who carry it. Dr. Leon, who has been quietly keeping a group of complex patients nobody else has time for, sees his year-end review: his bonus is at the bottom of the department. The hospital's mandate and its incentives have drifted so far apart that the physicians who embody the mandate are the ones being financially punished for it. The new department chair, Lina, has six months to propose a replacement scheme — something that can actually pay Leon for being Leon.
Challenge: Design an internal reward scheme — bounties, profit shares, promotions, recognition — that ties compensation inside a group to outcomes thick enough to represent the group's mandate, without collapsing back into Goodhart-prone metrics. Produce the outcome-bundle specification, the allocation mechanism, and the rules that govern long-horizon and intrinsic-motivation tradeoffs.
Scenario. For ten years, a small philosophy department has ended its Friday seminar with a half-hour drink in the common room where nobody is allowed to keep arguing about the paper — you can only talk about what the paper made you think about elsewhere, or something you're stuck on. Junior faculty credit that half-hour with most of what they learned about doing philosophy well. This fall, a new dean has restructured space, and the common room is gone. The department is meeting on Thursday to decide whether to fight for it or let it go. The chair, Paolo, is realizing the fight is not really about the room; it is about a practice that kept the department honest about what it was for.
Challenge: Design a set of informal practices — rituals, conversations, shared habits — through which a group keeps the question of its purpose alive in the register of norms rather than rules. Produce the practices themselves plus the triggers, ownership, and scaling rules that keep them from decaying into compliance theater.
Scenario. A thirty-person design firm is deciding whether to take a lucrative contract with a company several team members feel uncomfortable about. The founder, Asha, could call a vote — but she already knows what a vote would look like, and she suspects it would paper over the fact that different people are objecting for very different reasons, some of which she shares and some of which she doesn't. She wants a process that surfaces what each person actually cares about in this decision, not just yes or no, and produces something she could bring back to the team as a shared artifact they'll recognize themselves in, whatever she decides.
Challenge: Design a deliberative elicitation process for a group decision that surfaces what participants value — with the context and tradeoffs behind each position — rather than compressing deliberation into a vote or consensus, and distinguishes articulations reflecting deeper engagement with the domain from idiosyncratic preference. Produce the elicitation procedure, its interface to the actual decision, and the participant-count and value-revision rules.
Scenario. A five-year-old worker cooperative wrote into its founding charter that it would never take on clients from the defense sector. Last quarter the business side, under pressure, signed a contract with a logistics contractor whose revenue turns out to be three-quarters military. Two longtime members, Hiro and Olga, believe this is a straightforward breach of the charter; the board believes the contract is technically outside the prohibition. There is no internal mechanism for adjudicating the claim — the dispute is drifting toward member resignations and a public fight. What Hiro and Olga want is a forum inside the co-op that can actually hear their claim, rule on whether the charter was honored, and bind the board to the result.
Challenge: Design a fidelity-oriented charter and its adjudication procedure — a substantive layer on top of the group's bylaws that articulates the mandate densely enough for members to bring a breach claim, plus the forum that hears such claims and binds the group to the result. Produce the standing rules, the remedy and escalation menu, and the safeguards against factional capture.
Scenario. A ten-year-old cohousing community of twenty-four households wrote a charter when it was founded. The charter, which says things like "we raise children together" and "we eat dinner as a community at least twice a week," has lately felt either obsolete or aspirational depending on who you ask. Several new families don't share the original founders' sense of what the place is for; several founders have grown old and quiet. At the fall meeting, a facilitator, Vida, proposes a year-long process to re-articulate the charter — not start over, but deepen and clarify it — with everyone's voice in the room, before the community quietly drifts into being just a nice place to live.
Challenge: Design a "living charter" — a group's founding document structured to carry substantive value content about what the group is for, in terms rich enough to be contested, and revisable as the group learns without becoming a perpetual renegotiation. Produce the split between charter and policy, the procedure for reading the charter against operations, the revision mechanism, and the features that preserve its authority over time.
Scenario. A newly appointed regulator at a federal financial agency, Arjun, sits down for his orientation and realizes something odd. Every part of his formal training is about rules — which statutes, which enforcement tools, which precedents. Nothing in it asks him to engage with what the agency is actually for — whose savings it protects, what "fair dealing" was meant to cover, why the statute he is learning to enforce was passed in the first place. The senior colleagues who carry that understanding are due to retire in the next decade. Arjun begins to understand that if the training pipeline doesn't change, the agency will still enforce rules but will lose its ability to recognize when the rules and the purpose have diverged.
Challenge: Design a professional-training regime that builds mandate literacy — the capacity to recognize when an institution's rules have drifted from its purpose — into the pipeline that produces a national-scale institution's practitioners, and specify which programs carry it, how it is taught, and how the state's role relates to profession-set content.
Evaluation. A strong design makes mandate literacy something practitioners can be tested and credentialed on rather than a forgettable box-check, and gives the system a way to surface and replace training programs that have themselves drifted.
Scenario. A state's department of child welfare is about to issue its next four-year contract for family support services — visits, case management, parenting programs. The last contract rewarded caseload and visit counts. The results, as every caseworker knows and every director knows, have been measured hours and no fewer kids bouncing between foster homes. The new commissioner, Julian, wants to try something different: pool the money, specify outcomes families actually care about — a sibling group kept together, a grandparent reached — and let providers bid on those outcomes, with verification handled by a party that isn't the provider and isn't the state. He needs a procurement mechanism that exists and that his counsel will sign off on.
Challenge: Design an outcome-based, socialized-pool procurement mechanism for a category of national- or state-scale public spending — specifying how outcomes are defined, how providers bid, and how an independent party verifies results — that the buyer's counsel could actually adopt.
Evaluation. A strong design holds up across a heterogeneous demand base and resists incumbent-supplier capture, while keeping pool solvency compatible with public budgeting rules.
Scenario. A state department of health has adopted an AI triage system for its Medicaid call center. Wait times are down, satisfaction scores are up, and a hearing this month will hold it up as a national model. A case manager, Carla, who has worked those phones for twelve years, has been watching a pattern: callers with complex situations — a mother juggling her child's disability paperwork and her own cancer treatment — get routed through faster but resolved less. The system's metrics call them "resolved." Carla calls them "hung up on." She wants a standard the state will actually apply to deployments like this one: what the system owes people like that mother, not just what it owes the department's dashboard.
Challenge: Design a mandate-alignment standard for AI systems deployed in national-scale institutional roles — caseworker decisions, benefits adjudication, healthcare triage, sentencing recommendations — that tests whether a deployment can be held to the institution's thick purposes rather than legible proxies, and produce the assessment instrument and the body that applies it.
Evaluation. A strong standard surfaces drift even when the system optimizes legible proxies that look like success on a dashboard, and handles the case where the system is locally more accurate than the human practitioners whose presence maintains the capacity to notice drift.
Scenario. A country is about to hold a national referendum on overhauling its long-term-care policy — who pays, who provides, what happens when someone can no longer stay at home. Polling will capture yes and no. What it won't capture is what Yelena, a home-health aide, has learned over fifteen years: that the families she works with care about different things than either side of the campaign is talking about. They want to be sure their parent is still known by name; they want a single aide who isn't always changing; they want to know that their own turn, when it comes, won't reduce them to a bed-flip statistic. A pilot moral-graph process could capture what the referendum cannot. The minister's office has a month to decide whether to run it.
Challenge: Design a pilot moral-graph-elicitation process that supplements a country's legislative or referendum process for a single policy domain you choose (education, drug scheduling, immigration quotas, land use — whatever best tests the method), specifying selection, elicitation, aggregation, and its interface with existing authority, plus the protocol the minister's office would actually run.
Evaluation. A strong pilot survives the three strongest objections a democratic theorist would raise about substituting an elicited value-structure for the vote, with sketched responses.
Scenario. A federal ombudsman's office has been asked to evaluate whether three different agencies — the VA, a consumer protection bureau, and a community development fund — are living up to their public commitments. In each case, the ombudsman, Lucia, finds the same problem: the mandates are written at a level of abstraction that makes them impossible to test. "Serve veterans with dignity." "Protect consumers." "Strengthen communities." Each agency has meanwhile invented its own internal vocabulary for what it is actually doing. Lucia realizes she is writing the same memo three times. She wants a shared national standard for how mandates are articulated thickly enough to be tested — not one she has to reinvent for every agency.
Challenge: Design the shared public infrastructure — standards for articulating institutional mandates thickly enough to be contestable, standards for measuring fidelity to them, and channels for practitioner and stakeholder signal — and specify the standards body, its independence, and the revision procedure.
Evaluation. A strong design keeps mandate measurement from collapsing into a new thin-metric Goodhart regime and insulates the infrastructure from political cycles, all without erecting a national ministry of fidelity.
Scenario. Eight years ago, Liana's father signed up for a social platform whose tagline was "stay close to the people you love." It's where the family group chat lives; it's where photos of the grandkids arrive. Over the past two years, the feed has filled up with strangers, the group-chat notifications have grown quieter, and her father now spends most of his evenings on it alone, watching short videos. Internal documents leaked last month show the product team quietly retired its "friendship-formation" metric in favor of session time. Liana, a retired civil-rights lawyer, wants to bring a claim that the platform has walked away from what it publicly committed to. She cannot find a doctrine that lets her.
Challenge: Draft either a model constitutional amendment (≤200 words) or an interpretive doctrine that establishes "fidelity" — institutions must act in accordance with their thick mandates rather than substitute thin proxies — as a justiciable principle alongside liberty and equality, with a short interpretive commentary.
Evaluation. A strong draft survives the Rawlsian objection by constitutionalizing the form of fidelity rather than any particular thick content, and carries limiting principles potent enough that the doctrine does not swallow all institutional design.
Scenario. A small democracy in the middle of a constitutional convention has spent three months debating the usual list — executive powers, judicial review, bill of rights. A working group led by a legal scholar, Amara, has proposed a chapter few other constitutions carry: a statement of what the country's major public institutions are for, concrete enough that ordinary people could invoke it if an institution went adrift, and revisable every generation so it doesn't freeze in one era's language. The critics worry it will become an impossible standard. Amara's working group is about to present their draft to the full convention, and needs to show it can bind without petrifying.
Challenge: Design a constitutional chapter that spells out institutional values concretely enough that institutions below the constitutional level can be held to them, and produce both the draft text and the mechanism by which it stays revisable without ordinary amendment.
Evaluation. A strong design binds without petrifying — specific enough that the values don't equivocate, open enough that it doesn't freeze one generation's worldview, and legible across worldviews in a pluralist population.