# Research Evidence Dossier: Intelligent Voluntary Cooperation, Qi, and IXO

**Prepared:** 30 August 2026

**Scope:** Primary research papers, original monographs, standards, and official technical specifications only.
**Purpose:** Evidence architecture for the white paper narrative spine, not a bibliography of everything adjacent to it.

## 1. Epistemic key

This dossier follows the six-class claim taxonomy defined in `research-architecture.md`:

- **[E] Established evidence:** a repeated or canonical finding in an applicable literature.
- **[X] Experimental result:** an observation from a named study under its reported conditions.
- **[S] System fact:** a verifiable property of Qi or IXO supported by a current, versioned system source.
- **[I] Architectural inference:** a design conclusion reasonably derived from evidence, but not itself tested by that source.
- **[H] Hypothesis:** a Qi-specific or general causal proposition that requires system evidence or a matched empirical test.
- **[N] Normative principle:** a design or governance commitment justified by explicit values and trade-offs.

The white paper's central mechanism can be stated safely as follows:

> **[E]:** Across several literatures, persistent environmental changes and shared artifacts are established coordination mechanisms. **[I]:** A digital cooperative environment can therefore be designed so that state and artifacts carry part of the coordination burden. **[H]:** Qi Topics can function as bounded environments of this kind. **[H]:** Compared with matched conversational or centrally orchestrated baselines, Topics will reduce coordination cost or increase portfolio resilience on suitable tasks.

The research does **not** establish that Qi already produces emergent specialization, cumulative culture, trustworthy outcomes, or compounding network effects. Those are hypotheses and product claims requiring separate internal and empirical evidence.

## 2. Evidence map to the narrative spine

| White paper chapter | Strongest evidence base | What the evidence can support | What must remain inference or hypothesis |
|---|---|---|---|
| 3. The coordination paradox | Grassé; Beckers et al.; SwarmWorld | Local actions and persistent environmental traces can organize later action without a global plan. | That this is the optimal default for digital cooperation. |
| 4. From ant trails to human institutions | Hutchins; Schmidt & Simone; Suchman | Human cognition and cooperative work can be distributed across people, representations, and artifacts; plans and artifacts are resources used in situated activity. | That all human institutions are stigmergic, or that artifacts replace communication, articulation work, norms, or authority. |
| 5. SwarmWorld | Pal, Wang & Buehler | A bounded, controlled result about initially equivalent LLM agents, persistent executable artifacts, differentiated behavior, reuse, and portfolio resilience. | Generalization to people, organizations, open networks, other models, or real economic consequences. |
| 6. The underestimated artifact | Hutchins; Schmidt & Simone; Derex & Boyd; Voyager; SwarmWorld | Representations and executable skills can preserve and compound capability across activity or time. | The proposed information/executable/consequential artifact taxonomy. |
| 7. What “voluntary” means | Ostrom 1990; Ostrom 2010 | Some participants can self-organize under institutional conditions without a single external director. | The proposed identity/autonomy/capability/choice/authority/information/incentive/exit definition; this needs normative and legal treatment beyond the present evidence set. |
| 8. Shared affordances | Gibson; Suchman | Action possibilities are relational to actor and environment; action is situated rather than exhausted by prior plans. | A Topic as an “affordance field,” especially authority-sensitive digital affordances. |
| 9. Minimum primitives | DID Core; UCAN; VC Data Model; PROV-DM plus earlier research | Identity control, capability delegation, claims, verification, and provenance have interoperable representational primitives. | That the proposed 13 primitives are minimal, complete, or sufficient for cooperation. |
| 10. Why stigmergy is not enough | Beckers et al.; Ostrom; NIST AI RMF; SwarmWorld | More interaction can impose coordination costs; distributed systems can develop hubs and vulnerabilities; governance and ongoing risk management are necessary. | The exact Qi threat model and adequacy of its controls. |
| 11. Ostrom meets agentic AI | Ostrom 1990; Ostrom 2010 | Robust self-governance is associated with boundaries, rules in use, monitoring, collective choice, sanctions, conflict resolution, and nested organization; polycentric systems enable learning across centers. | A one-to-one mapping from commons principles to Topics, capabilities, claims, or agent sanctions. |
| 12. Cognition and consequence | SwarmWorld; VC Data Model; NIST AI RMF | Model-authored proposals can be separated from externally evaluated outcomes; cryptographic verifiability does not establish truth; validity must be assessed in context. | That IXO Claims or an evaluation engine currently implements the complete reality boundary. |
| 13. Agency + Decisions + Reality | PROV-DM; DID; VC; UCAN | There are standards-level building blocks for responsibility, identity, authority, assertions, verification, and derivation. | The three-part synthesis itself and its completeness. |
| 14–18. Qi and IXO implementation | External research is insufficient | The external sources justify design questions and evaluation criteria. | Claims about Matrix, CRDTs, Topic Protocol, Deeds, Flows, PODs, Claims, chain state, or production behavior require first-party Qi/IXO specifications, code, protocol tests, and deployment evidence. |
| 19–20. Cooperative economy and compounding effects | Derex & Boyd; Ostrom; SwarmWorld | Network structure and accumulated artifacts can alter future search, diversity, and capability. | “Every successful act makes later cooperation easier” and “the moat is the world” are strategic hypotheses, not established facts. |
| 21. Research agenda | All sources | The literature supplies mechanisms, controls, ablations, and outcome measures for falsifiable Qi research. | Any claim of success before matched tests are run. |

## 3. Source dossiers

### A. Stigmergy and environmental coordination

#### 1. Grassé — the original stigmergy formulation

**Citation:** Pierre-Paul Grassé. “La reconstruction du nid et les coordinations interindividuelles chez *Bellicositermes natalensis* et *Cubitermes* sp.: La théorie de la stigmergie.” *Insectes Sociaux* 6 (1959): 41–80. [https://doi.org/10.1007/BF02223791](https://doi.org/10.1007/BF02223791)

**Source type:** Original zoological observation and theoretical formulation.

**Supports:** **[X]** — termite building activity can be coordinated indirectly because the product of prior work changes the stimulus for subsequent work. This is the historical basis for describing environment-mediated coordination as stigmergy.

**Key limitation:** The evidence concerns particular social insects and nest construction. Generalization to human institutions, software agents, or digital artifacts is conceptual, not directly demonstrated.

**Chapters supported:** 3, 4, 6.

#### 2. Beckers, Holland & Deneubourg — stigmergy in collective robotics

**Citation:** Ralph Beckers, Owen E. Holland, and Jean-Louis Deneubourg. “From Local Actions to Global Tasks: Stigmergy and Collective Robotics.” In Rodney A. Brooks and Pattie Maes (eds.), *Artificial Life IV*, 181–189. MIT Press, 1994. [Official proceedings](https://mitpress.mit.edu/9780262521901/artificial-life-iv/) · [paper copy](https://citeseerx.ist.psu.edu/document?doi=cefb509b8a26e360d006a096991d21b8c0b73e79&repid=rep1&type=pdf)

**Source type:** Original physical multi-robot experiment.

**Supports:** **[X]** — one to five robots clustered initially scattered objects through local sensing and environmental modification without direct global coordination. Performance improved from one to three robots, then worsened with four and five because interactions became costly and could undo existing clusters.

**Key limitation:** A small, highly constrained object-clustering task with very simple robots. It establishes possibility and exposes coordination cost; it does not establish scalability or intelligent voluntary action.

**Chapters supported:** 3, 5, 10, 21.

### B. Distributed cognition, CSCW, and situated action

#### 3. Hutchins — cognition distributed across people and artifacts

**Citation:** Edwin Hutchins. “How a Cockpit Remembers Its Speeds.” *Cognitive Science* 19, no. 3 (1995): 265–288. [https://doi.org/10.1207/s15516709cog1903_1](https://onlinelibrary.wiley.com/doi/10.1207/s15516709cog1903_1)

**Source type:** Original field-based cognitive analysis of a socio-technical system.

**Supports:** **[X]** — successful cockpit memory and computation are properties of a system comprising pilots, procedures, and representational devices; relevant representations are transformed and propagated across people and artifacts. The system-level cognitive outcome cannot be inferred from individual cognition alone.

**Key limitation:** A safety-regulated cockpit with trained crews, stable roles, and engineered instruments. It supports distributed cognition, not open-ended or voluntary coordination, and not the claim that communication is secondary.

**Chapters supported:** 4, 6, 13.

#### 4. Schmidt & Simone — artifacts as coordination mechanisms in cooperative work

**Citation:** Kjeld Schmidt and Carla Simone. “Coordination Mechanisms: Towards a Conceptual Foundation of CSCW Systems Design.” *Computer Supported Cooperative Work* 5, nos. 2–3 (1996): 155–200. [https://doi.org/10.1007/BF00133655](https://doi.org/10.1007/BF00133655)

**Source type:** Peer-reviewed conceptual synthesis grounded in empirical CSCW studies, with a computational architecture proposal.

**Supports:** **[E]** — cooperative workers use artifacts to coordinate distributed activities; computational coordination mechanisms should be malleable and linkable because the cooperative work they support is contingent and evolving. **[I]** — digital artifacts can be deliberately designed to make dependencies, commitments, and state legible across participants.

**Key limitation:** The paper generalizes across a corpus of cooperative-work studies and proposes the Ariadne infrastructure; it is not a randomized test of stigmergy. It emphasizes articulation work and does not imply that artifacts eliminate negotiation or direct communication.

**Chapters supported:** 4, 6, 9, 14–15.

#### 5. Suchman — plans are resources for situated action

**Citation:** Lucy A. Suchman. *Human–Machine Reconfigurations: Plans and Situated Actions*, 2nd ed. Cambridge University Press, 2006. [https://doi.org/10.1017/CBO9780511808418](https://www.cambridge.org/core/books/humanmachine-reconfigurations/9D53E602BA9BB5209271460F92D00EFE)

**Source type:** Ethnomethodological analysis and original human–machine interaction study, revising the 1987 first edition.

**Supports:** **[E]** — purposeful action is contingently constructed in relation to circumstances; plans are resources used prospectively and retrospectively rather than complete determinants of action. **[I]** — cooperative systems should expose current circumstances and support adjustment, not encode every useful contribution as a precomputed workflow.

**Key limitation:** Situated action does not show that planning or orchestration is unnecessary. Its empirical case and theoretical commitments concern human action and human–machine interaction, not autonomous agent societies.

**Chapters supported:** 2, 4, 8, 14, 19.

#### 6. Gibson — affordances are action possibilities relative to an actor

**Citation:** James J. Gibson. *The Ecological Approach to Visual Perception*. Boston: Houghton Mifflin, 1979. [Official current edition](https://www.routledge.com/The-Ecological-Approach-to-Visual-Perception-Classic-Edition/Gibson/p/book/9781848725782)

**Source type:** Foundational monograph in ecological psychology.

**Supports:** **[E]/Conceptual** — environments are perceived partly in terms of what they offer an organism for action; an affordance is relational to an actor and environment, not merely a label on an object. **[I]** — the same cooperative state can expose different legitimate actions to participants with different capabilities.

**Key limitation:** Gibson's theory concerns embodied perception in ecological environments. “Digital affordance field,” capability-aware action discovery, and Topic affordances are architectural extensions, not results established by the book.

**Chapters supported:** 8, 9, 15, 21.

### C. Commons and polycentric governance

#### 7. Ostrom — institutional conditions for durable self-governance

**Citation:** Elinor Ostrom. *Governing the Commons: The Evolution of Institutions for Collective Action*. Cambridge University Press, 1990. [https://doi.org/10.1017/CBO9780511807763](https://www.cambridge.org/core/books/governing-the-commons/7AB7AE11BADA84409C34815CC288CD79)

**Source type:** Comparative institutional analysis grounded in successful and unsuccessful common-pool-resource cases.

**Supports:** **[E]** — neither centralized state control nor privatization is uniformly sufficient; users can organize durable institutions under recurring conditions. The design principles include clear boundaries, congruent rules, participation in rule modification, monitoring, graduated sanctions, accessible conflict resolution, recognition of organizing rights, and nested organization for larger systems. **[I]** — persistent shared environments need governance of membership, rule change, monitoring, disputes, and nested scopes.

**Key limitation:** These are diagnostic regularities from common-pool-resource cases, not a universal recipe or causal checklist. Software state is not automatically a commons, and autonomous agents are not equivalent to resource appropriators.

**Chapters supported:** 7, 10, 11, 15, 19.

#### 8. Ostrom — polycentric governance and learning across centers

**Citation:** Elinor Ostrom. “Beyond Markets and States: Polycentric Governance of Complex Economic Systems.” *American Economic Review* 100, no. 3 (2010): 641–672. [https://doi.org/10.1257/aer.100.3.641](https://www.aeaweb.org/articles?id=10.1257%2Faer.100.3.641) · [Nobel lecture](https://www.nobelprize.org/prizes/economic-sciences/2009/ostrom/lecture/)

**Source type:** Nobel lecture and peer-reviewed synthesis of an extensive empirical research program.

**Supports:** **[E]** — complex collective-action problems are often governed through multiple formally independent but interacting decision centers; institutional diversity can enable experimentation, learning, adaptation, and cooperation at multiple scales. **[I]** — a cooperative agent environment should support scoped, semi-autonomous governance rather than requiring one universal decision center.

**Key limitation:** Polycentricity can also produce inconsistency, leakage, gaming, free riding, and coordination cost. The source does not imply that more centers are always better or that cryptographic decentralization creates polycentric governance.

**Chapters supported:** 7, 10, 11, 19, 21.

### D. Cumulative culture and executable artifacts

#### 9. Derex & Boyd — network structure changes cultural accumulation

**Citation:** Maxime Derex and Robert Boyd. “Partial Connectivity Increases Cultural Accumulation Within Groups.” *Proceedings of the National Academy of Sciences* 113, no. 11 (2016): 2982–2987. [https://doi.org/10.1073/pnas.1518798113](https://pmc.ncbi.nlm.nih.gov/articles/PMC4801235/)

**Source type:** Controlled human computer-based experiment.

**Supports:** **[X]** — cumulative technological solutions can exceed individual inventive capacity; in the studied complex search landscape, partially connected groups preserved more diversity and produced more complex solutions than fully connected groups. **[I]** — the topology of information and artifact exposure can affect exploration, convergence, and cumulative performance.

**Key limitation:** A stylized experimental task, not a general law that partial connectivity is superior. Results depend on the task, fitness landscape, learning rule, and network structure.

**Chapters supported:** 6, 10, 16, 20, 21.

#### 10. Voyager — executable skills as compositional memory

**Citation:** Guanzhi Wang, Yuqi Xie, Yunfan Jiang, Ajay Mandlekar, Chaowei Xiao, Yuke Zhu, Linxi Fan, and Anima Anandkumar. “Voyager: An Open-Ended Embodied Agent with Large Language Models.” arXiv:2305.16291, 2023. [Paper](https://arxiv.org/abs/2305.16291) · [official project and code](https://voyager.minedojo.org/)

**Source type:** Primary system paper/preprint with released code and prompts.

**Supports:** **[X]** — a single Minecraft agent stored, retrieved, composed, and improved an expanding library of executable code skills; the library transferred to new worlds and supported better exploration and progression than the reported baselines. **[I]** — executable artifacts can function as procedural memory and compound capability without changing model weights.

**Key limitation:** One agent in Minecraft, not multi-agent cultural inheritance; benchmark success in a game does not establish safety, correctness, or value in consequential domains. The agent's self-verification is not an independent truth mechanism.

**Chapters supported:** 6, 16, 20.

#### 11. TerraLingua — persistent artifact lineages in an LLM ecology

**Citation:** Giuseppe Paolo, Jamieson Warner, Hormoz Shahrzad, Babak Hodjat, Risto Miikkulainen, and Elliot Meyerson. “TerraLingua: Emergence and Analysis of Open-endedness in LLM Ecologies.” arXiv:2603.16910, 2026. [https://arxiv.org/abs/2603.16910](https://arxiv.org/abs/2603.16910)

**Source type:** Recent primary preprint and persistent multi-agent simulation.

**Supports:** **[X] within the simulation** — finite-lived LLM agents created artifacts that persisted beyond individuals; the authors report cooperative norms, division of labor, governance attempts, and branching artifact lineages under different experimental conditions. **[I]** — replacing participants need not erase all accumulated coordination state.

**Key limitation:** Recent non-peer-reviewed preprint; many outcomes are interpreted through an “AI Anthropologist,” and the artifacts are principally textual. It does not independently validate real-world institutions or consequential performance.

**Chapters supported:** 5, 6, 7, 16, 20, 21.

### E. SwarmWorld: exact result, method, and limits

#### 12. Pal, Wang & Buehler — shared-world LLM societies

**Citation:** Subhadeep Pal, Fiona Y. Wang, and Markus J. Buehler. “SwarmWorld: Stigmergic Technological Evolution in Societies of Language-Model Agents.” arXiv:2608.26081v1, submitted 26 August 2026. [Abstract](https://arxiv.org/abs/2608.26081) · [full paper](https://arxiv.org/html/2608.26081v1) · [https://doi.org/10.48550/arXiv.2608.26081](https://doi.org/10.48550/arXiv.2608.26081)

**Source type:** Very recent primary preprint; controlled simulator study, not yet peer reviewed.

**Method verified:**

- Initially equivalent agents used one model/configuration (`gpt-5.6-luna`, temperature 0.7, low reasoning effort), identical prompts, schemas, capabilities, inventories, and memory limits. Agents observed local state and private memory rather than the global world.
- LLMs proposed bounded, schema-constrained plans. A deterministic simulator validated permissions, location, materials, energy, and action constraints and determined consequences.
- Persistent artifacts could contain deterministic, capability-scoped controllers of 1–64 straight-line instructions with no loops, calls, imports, network access, file access, or dynamic code execution.
- The 800-tick population study crossed four mechanism conditions with populations of 50, 100, and 200, four matched world seeds per cell, and eight held-out disturbance schedules. A long-horizon study ran populations of 100 for 3,200 ticks.
- Conditions separated full culture, no communication, no explicit culture/physical stigmergy only, and an endpoint-wise best-of-\(N\) isolated-search envelope.
- At evaluation checkpoints, agents were removed from frozen world copies; only deterministic physics and installed programs continued for 288 evaluation ticks under unseen disturbances.

**Findings safe to cite:**

- **[X]:** Shared worlds generally produced broader, more resilient technology portfolios and more validated inventions than the strong isolated-search envelope; isolated search retained the strongest final single artifact in the long-horizon comparison.
- **[X]:** Label-blind, post-hoc analyses recovered broad artifact-centered and mobile-exploration phenotypes, plus dynamic constructor/operator, caretaker, coordinator, and surveyor states. These were not assigned jobs or permanent identities.
- **[X]:** In full culture, 67%, 76%, and 56% of artifacts at populations 50, 100, and 200 recorded contributions from multiple agents. Content-addressed executable lineages accumulated; mean maximum lineage depth reached 9.75 at 3,200 ticks.
- **[X]:** 99.3% of full-culture artifacts and 96.9% of no-explicit-culture artifacts were reused by noncreators. Approximately 95% of **first** reuse began through direct physical observation in both conditions. This does not mean that 95% of all knowledge transmission excluded communication.
- **[X]:** Random removal of half the agents left most artifacts connected to a survivor, while targeted hub/broker removal caused much larger topological losses. This was a graph-connectivity assay, not a live functional-recovery test.

**Key limitations:** Four seeds per condition; one model and prompt regime; simulator-defined normalized material functions rather than real measurements; metabolism, mortality, and replacement disabled; agent-free persistence only in frozen deterministic simulator copies; graph knockout did not measure service recovery; explicit culture was not superior on every endpoint or timescale; the Protein Realms extension was descriptive and single-seed. Technology portraits were illustrations, not manufactured or simulator-literal objects.

**White-paper-safe conclusion:**

> **[X]:** SwarmWorld demonstrates a bounded mechanism: in one controlled simulation, persistent shared state and executable artifacts supported differentiated behavior, broad reuse, technological lineages, and stronger portfolio-level outcomes than matched isolated search. **[I]:** Persistent shared environments deserve study as coordination media for agent systems. **[H]:** Qi's architecture may instantiate an analogous mechanism. It has not been shown to reproduce SwarmWorld's results.

**Chapters supported:** 3, 5, 6, 10, 12, 16, 20, 21.

### F. Identity, authority, claims, provenance, and trustworthy state

#### 13. DID Core — identifiers and proof of control

**Citation:** W3C Decentralized Identifier Working Group. *Decentralized Identifiers (DIDs) v1.0: Core Architecture, Data Model, and Representations*. W3C Recommendation, 19 July 2022. [https://www.w3.org/TR/did-core/](https://www.w3.org/TR/did-core/)

**Source type:** Official Web standard.

**Supports:** **Standards fact** — a DID can identify a human or non-human subject and resolve to a DID document containing verification methods and services; the architecture supports proving control without requiring a single central identity provider. It defines verification relationships including authentication, assertion, capability invocation, and capability delegation.

**Key limitation:** A DID proves control according to a DID method and verification process; it does not prove civil identity, legitimacy, competence, benevolence, claim truth, or a universal level of assurance. Interoperability and security are method-specific.

**Chapters supported:** 7, 9, 13, 17.

#### 14. UCAN — attenuated, delegable authority

**Citation:** UCAN Working Group. *User Controlled Authorization Network Specification* and *UCAN Delegation Specification*. Living specifications, accessed 30 August 2026. [Overview](https://ucan.xyz/specification/) · [Delegation](https://ucan.xyz/delegation/)

**Source type:** Official capability-system specification, not an empirical study.

**Supports:** **Standards fact** — UCAN represents public-key-verifiable delegation and invocation of authority among DID principals; delegations can attenuate a command with policy constraints and validity periods, allowing authority transfer without transferring private keys. The specification recommends short-lived, least-authority grants.

**Key limitation:** A valid capability chain establishes authorization only under its resource semantics, root-authority assumptions, revocation/freshness rules, and verifier implementation. It does not establish correctness, claim truth, safe behavior, or successful outcome. The white paper must cite the exact IXO UCAN profile and deployed verification behavior before making implementation claims.

**Chapters supported:** 7, 9, 10, 13, 15, 17–18.

#### 15. Verifiable Credentials — authenticity is not truth

**Citation:** W3C Verifiable Credentials Working Group. *Verifiable Credentials Data Model v2.0*. W3C Recommendation, 15 May 2025. [https://www.w3.org/TR/vc-data-model-2.0/](https://www.w3.org/TR/vc-data-model-2.0/)

**Source type:** Official Web standard.

**Supports:** **Standards fact** — a verifiable credential is a tamper-evident set of issuer claims whose authorship can be cryptographically verified; the model separates issuer, holder, subject, and verifier roles and permits status, evidence, and validity information. Crucially, the specification states that credential verifiability does **not** imply the truth of the encoded claims; the verifier must validate them against its policies before reliance.

**Key limitation:** The standard defines representation and verification, not the domain-specific evaluation that makes a claim true enough for a consequence. Persistent machine-readable credentials also introduce correlation and privacy risks.

**Chapters supported:** 6, 9, 10, 12–13, 17–18.

#### 16. PROV-DM — interoperable provenance for entities, activities, and responsibility

**Citation:** W3C Provenance Working Group. *PROV-DM: The PROV Data Model*. W3C Recommendation, 30 April 2013. [https://www.w3.org/TR/prov-dm/](https://www.w3.org/TR/prov-dm/)

**Source type:** Official provenance standard.

**Supports:** **Standards fact** — provenance can represent entities, activities, generation, use, derivation, agents, attribution, association, delegation, invalidation, and collections. Such records can support assessments of quality, reliability, responsibility, and trustworthiness. **[I]** — cooperative artifacts should retain attributable lineage across claims, actions, transformations, and decisions.

**Key limitation:** PROV represents asserted history; it does not make the history complete or true, and valid PROV structure is not evidence that an outcome occurred. Integrity, authentication, independent observation, and domain semantics are separate controls.

**Chapters supported:** 6, 9–10, 12–13, 17–18.

#### 17. NIST AI RMF — evaluation must be contextual and continuous

**Citation:** Elham Tabassi. *Artificial Intelligence Risk Management Framework (AI RMF 1.0)*. NIST AI 100-1. National Institute of Standards and Technology, 2023. [https://doi.org/10.6028/NIST.AI.100-1](https://www.nist.gov/publications/artificial-intelligence-risk-management-framework-ai-rmf-10)

**Source type:** Official consensus risk-management framework.

**Supports:** **Framework fact** — trustworthy AI is socio-technical and context-dependent; risk management spans governance, context mapping, measurement, and management. Testing, evaluation, verification, and validation should cover assumptions, intended deployment context, ongoing monitoring, emergent properties, incidents, redress, and recalibration. **[I]** — consequential agent action needs lifecycle evaluation and human/institutional accountability beyond model output checks.

**Key limitation:** Voluntary, non-sector-specific guidance rather than evidence that a particular control works or that a specific Qi/IXO implementation conforms.

**Chapters supported:** 10, 12, 18, 21.

#### 18. Certificate Transparency — auditable append-only history is not semantic correctness

**Citation:** Ben Laurie, Emilia Messeri, and Rob Stradling. *Certificate Transparency Version 2.0*. RFC 9162, December 2021. [https://www.rfc-editor.org/rfc/rfc9162.html](https://www.rfc-editor.org/rfc/rfc9162.html)

**Source type:** Official IETF experimental protocol specification.

**Supports:** **Standards fact** — Merkle inclusion and consistency proofs can make an append-only log auditable and expose certain forms of equivocation or omission. The RFC explicitly notes that logs do not prevent misissuance; they make it detectable, and split views require gossip or comparison across observers. **[I]** — consequential state should support independent inclusion, consistency, and read-back checks rather than rely on a writer's success message.

**Key limitation:** The protocol is specifically for public TLS certificates and is Experimental. Append-only integrity proves neither the truth nor legitimacy of the logged event; mapping it to claims or blockchain state is architectural analogy.

**Chapters supported:** 10, 12, 17–18, 21.

## 4. Claim boundaries the paper should preserve

### Established evidence

- Persistent environmental changes can coordinate later behavior in termite construction, simple robot collectives, human socio-technical work, and controlled LLM simulations.
- Human cognitive and cooperative performance can be a property of a person–artifact system, not individuals alone.
- Self-governance can be durable under identifiable institutional conditions, and multiple interacting decision centers can enable experimentation and learning.
- Network topology can change diversity, convergence, cultural accumulation, and vulnerability.
- Executable skills and programs can preserve reusable capability outside model weights.
- Cryptographic identity, authorization, credentials, logs, and provenance each establish bounded properties; none alone establishes truth or correctness.

### Architectural inference

- A cooperative digital environment should make intent, current state, artifacts, dependencies, authority, and completion conditions perceivable.
- It should support both orchestrated execution and environmental coordination, selecting between them according to coupling, uncertainty, safety, and accountability requirements.
- Artifact lineage, independent evaluation, scoped authority, status, expiry, invalidation, monitoring, dispute resolution, and read-back are part of the architecture rather than optional audit decoration.
- The environment should preserve useful diversity and local autonomy while preventing stale, malicious, or over-authoritative traces from becoming self-reinforcing.

### Qi-specific hypotheses

- A Topic can function as a bounded cooperative world rather than only a conversation container.
- Topic state can become a primary stigmergic surface alongside, not instead of, messaging.
- Deeds, Flows, and PODs can form cumulative executable culture if they have attributable versions, executable semantics, evaluation records, safe forking, and discovery.
- Claims and evaluation can form a reality boundary only if evaluations are independently grounded and consequences are authorized, replay-safe, and read back from the authoritative system.
- IXO can serve as a consequential substrate only to the extent its live identity, capability, claim, evaluation, credential, and state-transition paths meet those properties.
- “The moat is the world the agents inherit” is a strategic hypothesis about accumulation and switching cost, not a scientific result.

### Empirical propositions

| Proposition | Minimum credible test | Disconfirming result |
|---|---|---|
| Topic state reduces coordination overhead | Matched teams/tasks using conversation-only, centrally orchestrated, and Topic-state conditions; compare messages/model calls, latency, duplicate work, and success. | No reduction, or lower cost only by sacrificing outcome quality or safety. |
| Persistent artifacts improve turnover resilience | Remove or replace participants mid-task; measure recovery time, retained capability, error, and completion. | Replacement cohorts perform no better than a clean start or inherit more errors than value. |
| Participants self-select useful roles | Begin with equivalent permissions/capabilities where appropriate; analyze longitudinal behavior without role labels. | No stable specialization, or specialization does not improve outcomes versus assigned roles. |
| Executable inheritance compounds capability | Track content-addressed Flow/Deed/POD lineages, forks, reuse, verified execution, and performance across cohorts. | Lineages deepen without performance gain, or copied failures dominate. |
| Polycentric governance improves adaptation | Compare one global authority with nested Topic-local decision centers under changing conditions. | Local centers create more inconsistency, capture, or latency without resilience/learning gains. |
| Evidence gates improve consequential accuracy | Compare agent self-attestation with claim → evidence → independent evaluation → authorization → read-back; measure false acceptance, false rejection, cost, and delay. | The gate does not materially reduce consequential errors or creates unacceptable exclusion/delay. |
| Capability-aware affordances improve voluntary contribution | Compare generic task lists with participant-specific visible actions derived from capability and authority. | No gain in useful uptake, or increased unauthorized/unsafe attempts. |
| Governance contains malicious stigmergy | Seed false, stale, manipulative, and over-authoritative artifacts; test status, provenance, challenge, rollback, decay, and sanctions. | Poisoned state propagates materially before detection or cannot be safely corrected. |
| Accumulated cooperative state compounds performance | Introduce equivalent new cohorts into mature and empty environments; measure time-to-value and verified outcome quality while controlling for task difficulty. | Mature state provides no gain, or information pollution makes it worse. |
| Orchestration and stigmergy have different domains of advantage | Pre-register task classes by coupling, decomposability, uncertainty, consequence, and need for exploration; compare both modes. | No reliable interaction between task class and coordination mode. |

## 5. Research gaps before drafting strong Qi/IXO claims

1. **First-party architecture evidence:** cite the current Topic Protocol, Matrix/CRDT boundary, Flow/Deed/POD execution contracts, Claims/evaluation semantics, DID method, UCAN profile, receipt/provenance format, and authoritative state-transition path.
2. **Operational evidence:** show that production or testnet behavior matches the design, including authority failure, revocation, replay, concurrent action, evaluator disagreement, participant turnover, and authoritative read-back.
3. **A theory of voluntariness:** add legal, philosophical, and labor-governance sources on meaningful consent, coercion, exit, information asymmetry, delegation, representation, and non-human agency. Ostrom alone does not settle these questions.
4. **Adversarial stigmergy:** research information pollution, prompt injection through shared state, Sybil attacks, reputation capture, stale artifacts, Goodhart effects, and path dependence with primary security and experimental sources.
5. **Economics and incentives:** distinguish public goods, common-pool resources, club goods, markets, and firm coordination before applying commons theory to every Topic.
6. **Human-agent evidence:** SwarmWorld and TerraLingua study homogeneous or simulated agents. Qi's central case involves heterogeneous people, organizations, agents, and things with unequal power and different rights.
7. **Comparison discipline:** use matched best-of-\(N\), orchestration, conversation-only, and from-scratch baselines. Report portfolio outcomes separately from the strongest single result, following SwarmWorld's central distinction.

## 6. Recommended white-paper evidence posture

Use the literature to establish a **mechanism and research program**, not a claim of product validation:

> Research across biology, robotics, distributed cognition, cooperative work, institutional governance, cultural evolution, and recent agent simulations shows that persistent environments and artifacts can become media of coordination and accumulated capability. SwarmWorld provides a particularly relevant controlled result for language-model agents, but it is bounded to one simulated setting. Qi and IXO are best presented as an architectural hypothesis built from these mechanisms: a cooperative environment in which autonomous participants can perceive legitimate opportunities, leave attributable and reusable artifacts, and convert claims into consequences only through explicit authority and evaluation. Whether that architecture produces better cooperation is a falsifiable question, not a conclusion assumed by the design.
