{"path":"research/carriers.md","content":"# Carriers: The Institutions That Could Bear Deliberus Into Society\n\n**Date**: 2026-08-20 · **Type**: enumeration + ranking, founder-prompted, following [fractal-priors-for-success.md](fractal-priors-for-success.md) condition (c): the historical winners among legibility technologies spread through *institutional carriers whose own legitimacy depended on adoption*. This doc names the plausible carriers, tiered by how much their legitimacy depends on what the reasoning layer provides, how concretely they could embed it, and how reachable they are from the project's current position. **Sources**: corpus + live web verification this session (Full Fact AI's tooling and US expansion, EFCSN's Integrity Base timeline, Meaning Alignment's platform pivot, the MArgE paper) + training-data knowledge of the stable orgs, flagged where not re-verified. Named-person outreach strategy stays in `.private/` per the born-private rule; this page is org-level analysis.\n\n## §0 What makes something a carrier\n\nNot every ally is a carrier. A carrier is an institution that would *embed and propagate* the reasoning layer because its own function or legitimacy depends on what the layer provides — the way universities carried printing, states carried statistics, and auditors carried double-entry. Test: **if the carrier adopts, does its standing improve in its own world — and does adoption by it put the layer in front of people who never chose it?** Both must be yes.\n\n## Tier 1 — Scientific publishing and review: the drowning carrier\n\nLegitimacy-dependence: **maximal and acute** — the slop flood threatens the thing these institutions exist to guarantee ([ai-slop-and-the-reasoning-layer.md](ai-slop-and-the-reasoning-layer.md)). Candidates: **Cambridge University Press & Assessment** (its academic-publishing head publicly calling for \"radical change\"); **eLife** (already unbundled assessment from accept/reject — the institutional temperament closest to ours); **arXiv** (Ginsparg's team actively fighting slop, already reaching for provenance gates); **Center for Open Science / OSF** (Nosek — Registered Reports' inventors, methods-reform carriers, whose own preprint server closed under slop); **PLOS**; the integrity layer (**COPE**, **STM Integrity Hub**, **Retraction Watch/Crossref**); and the detection vendors (**Clear Skies**, **Pangram**) whose product class the article itself says may fail as models improve — making them natural customers for the structural complement to detection. Entry point already identified: pre-submission self-review, which needs no institutional buy-in ([peer-review-and-the-reasoning-layer.md](peer-review-and-the-reasoning-layer.md)).\n\n## Tier 2 — The fact-checking claim infrastructure: carriers already speaking claim\n\nThe discovery of this sweep: this world is **already building shared claim infrastructure**, at production scale, right now. **Full Fact AI** (London): claim-type classifiers, claim matching, ~a third of a million sentences daily across 40+ organizations and 30 countries, expanding to US newsrooms before the 2026 midterms, and building an **AI Trust Benchmark**. **EFCSN's Integrity Base**: a shared European database of fact-checks, claims and appearances — auto-ingested, AI-annotated, human-reviewed — expected live autumn 2026. **IFCN/Poynter** as the network hub. And critically, the existing standard: **ClaimReview (schema.org)** — 130,000+ fact-checks already published as structured claim markup consumed by search engines. These organizations hold claims WITHOUT the reasoning layer: verdicts and matches, no argument structure, no strength semantics, no decomposition. **The interop move is obvious and cheap**: the graph can read and emit ClaimReview, making Deliberus the reasoning layer on top of infrastructure fact-checkers already trust. Their legitimacy-dependence is high (drowning in volume, funding pressure, needing differentiation from raw AI), their claim-native culture removes the biggest translation barrier, and their networks are the distribution.\n\n## Tier 3 — Deliberative democracy and the plurality movement\n\nLegitimacy-dependence: high — these practitioners sell deliberation *quality* and need instruments that certify it. Platform-carriers: **Polis / the Computational Democracy Project** (the bridging lineage the corpus already builds on), **Decidim** (Barcelona; participatory-democracy infrastructure across hundreds of municipalities), **vTaiwan / g0v and the Plurality orbit** (Audrey Tang, Glen Weyl / RadicalxChange — the community most explicitly seeking tools of this class), **Community Notes** (the production proof of the cross-tribal incentive), **Deliberati** (Jonathan Warden's argument-based ranking work — a close neighbor). Practitioner-carriers: the **OECD deliberative-wave** network, **newDemocracy**, **Involve**, **MASS LBP**, **Missions Publiques**, the **Danish Board of Technology** lineage — citizens'-assembly organizers whose facilitation work is exactly the co-present modality the workshop layer targets. Sweden-specific and workshop-native: **Digidem Lab** (Gothenburg, runs Decidim in Sweden) and the **folkbildning tradition** (studieförbunden — a century-old Swedish institution whose entire premise is structured collective learning circles; \"discuss through Deliberus\" is a studiecirkel with a map).\n\n## Tier 4 — The AI-safety and epistemics ecosystem\n\nLegitimacy-dependence: high on the funding side (they need demonstrable epistemic infrastructure to point at), symbiotic on the research side. **LawZero** (the documented honesty-substrate convergence), the **Collective Intelligence Project** (Siddarth/Huang — mission-adjacent to the point of overlap), the **Meaning Alignment Institute** (verified this session: their OpenAI-funded moral-graph work is being turned into a standalone deliberation platform — the values-elicitation sibling of our values-bedrock claim), the **DeepMind Habermas-machine team** (whose own steering result creates their need for the non-steering complement), the forecasting world (**Metaculus**, **Manifold**, **QURI/Ozzie Gooen** — already in the founder's documented orbit), and the EA/rationalist forum infrastructure (**LessWrong/Lightcone**, the **EA Forum**) as the natural beachhead communities the outreach frontier already names. The academic wing matters more than it looks: **computational argumentation is productizing** — the MArgE paper (Toni's group, Imperial College) builds multi-LLM claim verification on *quantitative bipolar argumentation with gradual semantics*, i.e., our exact formalism family, published August 2025 — which makes that community both a validator of the foundations and a future consumer/contributor pool.\n\n## Tier 5 — Education: the scaffolding carrier\n\nLegitimacy-dependence: moderate but structural — critical-thinking education has always lacked an instrument that makes reasoning structure *visible and practicable*. **Kialo Edu** (already in classrooms at scale — competitor and proof-of-demand in one), **ThinkerAnalytix** (argument-mapping pedagogy, the van Gelder lineage whose deliberate-practice evidence is the field's best), the **Philosophy for Children** movement, debate leagues and **Ethics Bowl**, IB Theory of Knowledge / AP Seminar curricula, university writing-and-reasoning centers. The adult-development lens ([fractal-priors-for-success.md](fractal-priors-for-success.md) §7) says this tier is where the scaffolding bet gets tested a generation early: education is where societies already accept that cognitive operations are teachable with support. Carrier caution from the corpus: education is also a well-documented *ceiling* (tools that live only in classrooms stay there — [adoption-problem.md](adoption-problem.md) §5).\n\n## Tier 6 — Evidence synthesis, audit, and policy\n\nLegitimacy-dependence: high and chronic — these institutions ARE the L5 condition (audited anyway; structuring now is cheaper than justifying later). **Cochrane** and **Campbell** (systematic review is drowning in synthesis burden; living reviews are our staleness problem with a budget), the **GRADE working group** (evidence-weighting formalism — the closest existing practice to methodology-weighted strength), the **What Works** centres, parliamentary technology assessment (**STOA**, national TA bodies), regulatory impact-assessment units, and the evaluation offices of major foundations. Slowest-moving tier, highest institutional durability once entered.\n\n## Tier 7 — The open-web standards commons\n\nThe infrastructure-studies lens says the win condition is becoming a *layer others embed* — this tier is where layers get ratified. **NLnet/NGI** (funder AND standards-culture carrier; precedent in hand), **Knowledge Pixels / the nanopublication network** (interop peer for the fact layer), **Wikimedia/Wikidata** (the largest structured-claims commons on earth; the Abstract Wikipedia lineage is the most kindred large-scale project alive), the **Crossref/ORCID/DataCite** persistent-identifier layer (claims-with-resolvable-IDs is their paradigm, one level down), **ClaimReview** again as the schema bridge, and the indie neighbors now appearing (**FactHarbor**, **Reality Check** — the class is being born around us, which is the assembly-theory assimilation prediction §8 arriving on schedule).\n\n## §8 Ranking, and the three doors to knock first\n\nScoring the tiers on legitimacy-dependence × embedding capacity × reachability-from-here: **(1) the fact-checking claim infrastructure** — claim-native culture, live shared-database projects with 2026 timelines, a cheap concrete interop offer (ClaimReview in/out), and acute need; **(2) scientific publishing/preprint integrity** — maximal need, an entry point requiring nobody's permission, and the strongest why-now story; **(3) the Swedish civic layer** (Digidem, folkbildning, the deliberative practitioners) — smallest, but workshop-native, local, and the only tier where the co-present modality the corpus just ratified can be exercised this season. The plurality/AI-epistemics orbit (Tiers 3–4) is where the *idea* travels fastest and funding lives; the standards commons (Tier 7) is where the endgame (layer-not-destination) is decided; education (Tier 5) is the decade bet. A carrier plan is not an outreach plan — per the dogfooding-phase rule, nothing here licenses \"come try it\" promotion; it maps who to *watch, interoperate with, and be findable by* until the friends-round evidence exists.\n\n## Cross-references\n\n[fractal-priors-for-success.md](fractal-priors-for-success.md) (condition (c)) · [societal-integration-outlook.md](societal-integration-outlook.md) (allies ranked by force — carriers are the subset with embedding capacity) · [ai-slop-and-the-reasoning-layer.md](ai-slop-and-the-reasoning-layer.md) + [peer-review-and-the-reasoning-layer.md](peer-review-and-the-reasoning-layer.md) (Tier 1's need) · [competitive-landscape.md](../competitive-landscape.md) (the rival/peer boundary) · [science-reasoning-infrastructure.md](science-reasoning-infrastructure.md) (Tier 7's lineage) · [islands-of-coherence.md](islands-of-coherence.md) (the workshop layer Tier 3 serves) · [incentives-analysis.md](incentives-analysis.md) (L-conditions the tiers instantiate)\n"}