When a sanctions package lands, a corridor closes, or an election shifts trade policy, institutions do not lack information. They lack signal. State outlets, opposition media, allied think tanks, and social amplification layers each supply a coherent story. The stories rarely disagree on every fact, but they disagree on causality, intent, magnitude, and probability. Risk committees, sovereign desks, and underwriting teams need to price exposure without importing someone else's political sugarcoating.

Earthian's answer is No Propaganda Geopolitics: a configurable inference mode in Geopolitics Axiom-0 that separates evidentiary claims from rhetorical framing, scores source independence, and returns scenario distributions institutions can defend in front of boards, regulators, and model validation.

Why propaganda is a risk variable, not a media problem

Propaganda is not only falsehood. More often it is selective truth: real events embedded in causal chains that serve a narrative. For financial institutions, the damage is operational:

• Mispriced sovereign and EM exposure when escalation probability is inferred from headline tone rather than capability and incentive structure
• Supply-chain false positives when boycott rhetoric is treated as enforced policy
• Insurance and credit committee drift when the same asset is assessed differently depending on which jurisdiction's briefing pack circulated last
• Model risk findings when geopolitical inputs cannot be traced to evidence distinct from advocacy

Legacy geopolitical platforms often aggregate country scores or analyst briefs. Those outputs inherit whichever narrative layer the analyst or data vendor preferred. Geopolitics Axiom-0 was built as an inference SLM: it reasons over mechanisms (sanctions transmission, trade rerouting, conflict diffusion, policy ratcheting) rather than summarizing press releases.

No Propaganda Geopolitics extends that architecture with an explicit narrative strip layer before scenario propagation runs.

What the feature does

No Propaganda Geopolitics is a toggle (API flag `propaganda_filter=strict` on Earthian Hub geopolitics endpoints) that instructs Geopolitics Axiom-0 to:

1. Ingest multi-side corpora for an event or region: official communiques, trade and customs statistics, satellite and AIS-derived movement proxies, sanctions registry deltas, central bank and finance ministry releases, and independent wire reporting, each tagged by source class and jurisdictional alignment
2. Decompose statements into claim graphs: atomic factual assertions ( troop movements, tariff rates, export license changes), attributed quotes, and causal assertions ("because," "in response to," "will lead to")
3. Score framing load: rhetorical density, unattributed causal edges, emotional intensifiers, and consistency drift across a source's own timeline
4. Strip or down-weight claims that fail cross-source corroboration thresholds or that exist only inside aligned narrative clusters on **both** sides of a conflict (state line and counter-state line)
5. Re-run scenario inference on the residual evidence graph, producing probability bands, financial translation vectors, and explainability traces without narrative padding

Institutions see what is evidenced, what is disputed, and what is pure framing, in separate lanes. Committees can accept or challenge each lane without untangling a polished brief.

Technical architecture inside Geopolitics Axiom-0

Geopolitics Axiom-0 is a domain-specialized small language model trained on international relations structure, trade topology, conflict epidemiology, sanctions law, and historical policy ratcheting. No Propaganda Geopolitics adds four technical modules on top of the base inference stack:

1. Source independence graph

Each document enters a source independence graph (SIG). Nodes are outlets, ministries, agencies, and individual brief authors; edges encode ownership, funding, jurisdictional alignment, and historical co-quoting. When two opposed narratives quote the same unattributed "regional officials," SIG collapses them to a single low-independence cluster. Claims that exist only inside that cluster are flagged narrative-local, not portfolio-grade.

2. Rhetorical feature extraction

A lightweight rhetorical head scores each sentence for:

• Causal assertion density without measurable predicate
• Dehumanization or moral absolutism markers (down-weighted, not censored; they predict mobilization risk but not balance-sheet fact)
• Temporal vagueness ("soon," "imminent") vs dated policy actions
• Cherry-picked baseline comparisons

These features do not delete text. They assign framing weights used in the next stage. A tariff announcement with high framing load but matching customs code changes retains the economic fact; the motivation story may drop to "unverified narrative."

3. Cross-side corroboration engine

For any contested theater (e.g., corridor closure, export ban, cyber attribution), the engine requires triangulation classes before a claim enters the evidence graph:

• Primary institutional record (law, regulation, filing)
• Physical or transactional proxy (trade volume, AIS gap, power flow, payment rail usage)
• Independent third geography reporting (outlet whose SIG independence score exceeds threshold from both aligned blocs)

Claims supported only by mutually opposing aligned sources are marked dual-propaganda: often useful for understanding mobilization, excluded from base-case loss propagation unless the institution opts in.

4. Scenario re-inference on stripped graph

Once the evidence graph is built, Geopolitics Axiom-0 runs standard scenario distribution modeling:

• Escalation and de-escalation branches with explicit triggers tied to evidenced policy levers
• Financial translation into FX, spread, commodity, and supply-chain nodes relevant to the user's portfolio or underwriting book
• Tail mass calibration from capability and incentive inference, not from headline intensity

Outputs include `evidence_coverage_ratio`, `narrative_stripped_claims[]`, and `dual_side_framing_summary` for audit. Model risk teams can replay which claims moved the median scenario.

Switching off propaganda from both sides

"No propaganda" means symmetric skepticism, not neutrality theater. The mode does not pick a moral winner. It applies the same corroboration rules to:

• State media and allied broadcasters
• Opposition exiles and diaspora advocacy networks
• Western government backgrounders and corporate strategic communications
• Social velocity clusters that amplify one-sided clips

An institution covering Taiwan strait exposure, Red Sea routing, or Eastern European energy corridors receives the same stripped evidence graph regardless of which capital's briefing cycle dominated the morning call. What changes is not the conclusion mandated but the epistemic hygiene of the inputs.

Where this shows up in Earthian Hub

No Propaganda Geopolitics is exposed through Earthian Hub alongside other Axiom workflows:

• REST / GraphQL: `/geopolitics/axiom/analyze` with `propaganda_filter` set to `strict`, `balanced`, or `off` (legacy narrative-inclusive mode for red-team comparison)
• Batch portfolio scans: sovereign and EM books rerun overnight with strict filter; diff reports highlight names whose risk rank moved purely because narrative claims were removed
• Committee mode: multiplayer sessions (see Earthian's Multiplayer AI for Risk Teams) where analysts challenge individual claims; stripped graph updates live as humans accept or reject evidence nodes
• API + Lucid / NatCat coordination: stripped geopolitical shock feeds propagate into Lucid Climate-0 and NatCat Lighthouse-0 when physical corridors and energy nodes are shared

Default for regulated institutions is strict in production; balanced retains low-weight narrative signals for early warning of mobilization and market sentiment decoupling.

Analytical outputs institutions actually use

Typical deliverables from a No Propaganda Geopolitics run:

Unlike a generic LLM asked to "be unbiased," every row traces to SIG nodes, corroboration class, and inference path. That is the bar for institutions that already govern climate and credit models under SR 11-7 and EU AI Act expectations.

Comparison to legacy geopolitical feeds

Who enables this first

Earthian partners use No Propaganda Geopolitics where narrative error is capital error:

• Asset managers with EM and frontier sovereign debt where official and opposition storylines diverge by hundreds of basis points of implied risk
• Global insurers underwriting political violence, trade credit, and marine war risks on the same corridors state media describes differently
• Banks with cross-border exposure and sanctions compliance obligations that require evidenced policy fact, not advocacy
• Public-sector and multilateral users preparing situation assessments that must survive inter-agency challenge

Governance and limitations

No Propaganda Geopolitics is not a claim to absolute truth. It is a method: downgrade single-source causal stories, elevate triangulated institutional and physical evidence, and show what remains. Unknowns stay explicit. Institutions can widen corroboration thresholds when they want conservative inclusion of contested claims.

Earthian documents filter settings, graph versions, and model hashes for replay. When a committee asks why a strait closure scenario dropped from 18% to 9%, the answer is a list of removed dual-propaganda claims, not a black-box reweight.

Bottom line

Geopolitical risk is too consequential to outsource to whichever narrative arrived last. Geopolitics Axiom-0 with No Propaganda Geopolitics gives institutions a switch: strip propaganda from both sides, keep the evidenced core, and price forward scenarios with the same inference rigor Earthian applies to climate, nat cat, and technology risk.

Facts without political sugarcoating are not a aesthetic preference. They are a model input requirement for any desk that must defend its numbers.

To evaluate No Propaganda Geopolitics on your sovereign book, trade corridor, or underwriting portfolio, contact [email protected].