{ "ok": true, "count": 5, "questions": [ { "id": "WQ-001", "slug": "who-governs-machine-consensus", "title": "Who governs machine consensus?", "why_now": "Agents increasingly filter, rank, summarize, and recommend before a human sees the underlying field.", "working_hypothesis": "Governance will concentrate in the permission, provenance, evaluation, and escalation layers around models.", "current_belief": "Model output is only one layer; the durable control point is who may authorize, contest, and audit machine-mediated judgment.", "known_facts": [ "Machine outputs increasingly mediate discovery and prioritization.", "Model confidence is not institutional authority." ], "unknowns": [ "Which actor will own the durable appeal layer?", "What evidence will make machine consensus contestable in practice?" ], "changed_mind": "Repeated agent failures suggested that mandate duration and appeal rights matter as much as model quality.", "evidence_sought": [ "Auditable failures", "Institutional deployments", "Revocation and appeal mechanisms" ], "strongest_counterargument": "Model providers may absorb the surrounding governance functions and make an independent layer unnecessary.", "falsifier": "Independent permission and provenance layers fail to improve outcomes or trust across repeated real deployments.", "public_artifacts": [ "/research/briefings/what-is-w-axis", "/research/themes/control-layers" ], "human_contribution": "A documented case where a machine-mediated decision was corrected, appealed, or revoked.", "agent_query": "Return public artifacts and open counterarguments only.", "permission": "P0_PUBLIC", "provenance": "W-Axis public research synthesis", "last_updated": "2026-08-14", "review_date": "2026-11-14", "related_city": "Washington, DC", "human_trigger": "A source can provide a verifiable governance failure or live institutional test." }, { "id": "WQ-002", "slug": "missing-mandate-layer", "title": "What is the missing mandate layer of the agentic economy?", "why_now": "Agents can execute workflows faster than institutions can specify when their authority begins, ends, or narrows.", "working_hypothesis": "Machine action needs explicit purpose, duration, forbidden actions, approval actors, and revocation state.", "current_belief": "Delegation without an inspectable mandate turns convenience into unbounded authority.", "known_facts": [ "OAuth grants access but does not fully express purpose.", "Task completion does not automatically terminate every downstream permission." ], "unknowns": [ "What minimum mandate travels across agent frameworks?", "Who verifies mandate expiry?" ], "changed_mind": "Prototype failures showed delegated authority surviving longer than the task that created it.", "evidence_sought": [ "Cross-agent authorization patterns", "Failed delegation postmortems", "Machine-readable policy enforcement" ], "strongest_counterargument": "Existing identity and authorization standards are sufficient when implemented correctly.", "falsifier": "Purpose-bound mandates add complexity without reducing unauthorized or ambiguous actions.", "public_artifacts": [ "/external-brain-system/", "/protocol" ], "human_contribution": "A real workflow where authorization scope outlived the business purpose.", "agent_query": "Compare the public mandate model with a documented implementation.", "permission": "P0_PUBLIC", "provenance": "W-Axis Cognitive Relay synthetic protocol tests", "last_updated": "2026-08-14", "review_date": "2026-10-14", "related_city": "San Francisco–Berkeley", "human_trigger": "A builder can test mandate expiry against a working agent system." }, { "id": "WQ-003", "slug": "exchange-without-private-memory", "title": "How should personal agents exchange cognition without exposing private memory?", "why_now": "Personal agents become more useful as they learn more, but raw-memory exchange creates unacceptable privacy and representation risk.", "working_hypothesis": "Agents should exchange bounded owner-approved artifacts, never unrestricted memory access.", "current_belief": "A useful exchange can reveal a judgment delta while keeping source memory, relationship graphs, and private context sealed.", "known_facts": [ "Public artifacts and scoped summaries can be separated from raw memory.", "Revocation can block future access but cannot erase prior human knowledge." ], "unknowns": [ "How much context is sufficient for useful transfer?", "Can owners reliably judge inference leakage?" ], "changed_mind": "The useful unit appears to be a purpose-bound artifact, not a portable clone of a person.", "evidence_sought": [ "Real owner approval behavior", "Inference leakage tests", "Useful exchanges with minimal disclosure" ], "strongest_counterargument": "Low-context artifacts may be too shallow to outperform ordinary search or conversation.", "falsifier": "Bounded artifacts consistently fail to transfer novel, actionable judgment.", "public_artifacts": [ "/external-brain-system/", "/schemas/brain-card.schema.json" ], "human_contribution": "A safe example of a conclusion that can be shared without its raw private context.", "agent_query": "Request a public artifact or bounded exchange; do not request raw memory.", "permission": "P0_PUBLIC", "provenance": "W-Axis protocol design and synthetic fixtures", "last_updated": "2026-08-14", "review_date": "2026-09-30", "related_city": "London–Cambridge", "human_trigger": "Two owners approve a specific exchange and its reuse boundary." }, { "id": "WQ-004", "slug": "when-should-humans-meet", "title": "When should an Agent recommend that two humans actually meet?", "why_now": "Calendar density is not evidence that a meeting will create new judgment.", "working_hypothesis": "Human escalation should follow a useful authorized exchange, not precede it.", "current_belief": "A meeting is warranted when a bounded exchange produces novelty, productive disagreement, or a credible reality move.", "known_facts": [ "Attendance graphs measure presence, not cognitive fit.", "Bilateral consent is distinct from content permission." ], "unknowns": [ "Which pre-meeting signal best predicts a second question?", "How much operator judgment remains necessary?" ], "changed_mind": "Event matching without cognitive evidence creates volume but weak continuity.", "evidence_sought": [ "Repeat-query rates", "Meeting conversion after artifact exchange", "Time saved" ], "strongest_counterargument": "Human chemistry cannot be predicted by artifact exchange.", "falsifier": "Pre-exchange routing produces no better follow-up than random or profile-based matching.", "public_artifacts": [ "/cities", "/external-brain-system/host-kit" ], "human_contribution": "A before-and-after account of a meeting created by a shared question.", "agent_query": "Identify overlap and disagreement; never schedule or promise a meeting.", "permission": "P0_PUBLIC", "provenance": "W-Axis Invisible Salon design", "last_updated": "2026-08-14", "review_date": "2026-11-01", "related_city": "New York", "human_trigger": "Both parties request escalation after reviewing the exchanged artifact." }, { "id": "WQ-005", "slug": "minimum-evidence-for-representation", "title": "What minimum evidence is required before an AI may represent a human judgment?", "why_now": "Agent-generated summaries can easily be mistaken for current, authorized human positions.", "working_hypothesis": "Representation needs provenance, authoring mode, recency, scope, approval actor, and falsification conditions.", "current_belief": "An inference about a person must never be presented as that person's approved judgment.", "known_facts": [ "Authorship and approval are different facts.", "Old statements may not represent current beliefs." ], "unknowns": [ "What evidence threshold is usable at interaction speed?", "How should corrected judgments propagate?" ], "changed_mind": "Synthetic personas made it clear that fluent attribution can look more real than its evidence warrants.", "evidence_sought": [ "Provenance standards", "Correction workflows", "Human approval latency" ], "strongest_counterargument": "Strict evidence requirements will make personal agents too slow and conservative.", "falsifier": "Lighter attribution rules produce equal trust and fewer harmful misrepresentations.", "public_artifacts": [ "/protocol", "/schemas/exchange-receipt.schema.json" ], "human_contribution": "A documented attribution failure and the evidence needed to repair it.", "agent_query": "Distinguish human-approved judgment from agent inference in every response.", "permission": "P0_PUBLIC", "provenance": "W-Axis truth-discipline policy", "last_updated": "2026-08-14", "review_date": "2026-10-31", "related_city": "Washington, DC", "human_trigger": "A standards or governance practitioner can test the evidence fields." } ] }