Method & Sources
Where PowerFlow's
evidence comes from
PowerFlow does not treat any single feed or dataset as ground truth. It combines sourced reporting, structural baselines, relationship evidence, and conflict records into a living model of how power moves. Those layers do different jobs on purpose.
Why this page exists
PowerFlow should not feel like a bibliography dump, and it should not hide provenance either. Most of the product leads with interpretation. This page shows the source architecture behind that interpretation so trust is earned without turning every screen into footnotes.
Source architecture
Different source layers do different jobs. That separation matters because it keeps the product from collapsing into either a raw feed or a fake all-knowing score machine.
Core ingest sources
Role
Fresh reporting gives the model new evidence about power shifts, dependencies, fractures, and cascade effects.
Live now
Event Registry plus a curated source allowlist.
Not for
No single article or outlet sets a score by itself. Reporting must survive screening, extraction, and calibration before it hardens into the model.
Structural baselines
Role
Slow-moving denominators help anchor state actors in real capacity constraints rather than pure narrative.
Live now
World Bank and V-Dem are explicit today. SIPRI military expenditure and selective UN Comtrade concentration now also shape backend reasoning.
Not for
Baseline data does not determine geopolitical alignment, replace live reporting, or override current conditions on its own.
Relationship evidence
Role
Directional evidence makes leverage and dependency less subjective, especially on relationship and query surfaces.
Live now
World Bank IDS, OECD / CRS, SIPRI arms transfers, and UN Comtrade trade dependence.
Not for
These sources strengthen relationship scoring. They do not replace PowerFlow's relationship model or make the judgments for it.
Conflict substrate
Role
Historical conflict records give conflict pages and query receipts structured depth beyond day-to-day headlines.
Live now
UCDP historical conflict coverage in mapped PowerFlow conflicts.
Not for
Conflict datasets are context layers. They do not directly determine PF Scores across the whole system.
Principle
Sources are evidence inputs. PowerFlow's job is to translate them into readable judgments about dependence, exposure, concentration, asymmetry, and trend.
Live in the product today
These are the main external source layers live in the model today. Some are already explicit on frontend surfaces. Others currently shape backend reasoning, receipts, or writer context more than fixed page chrome.
Live source
Event Registry + curated outlet policy
What it adds
Breadth of event intake and timely reporting from monitored sources.
Where it shows up
Feeds PowerFlow events, intel items, score movement, briefs, and the downstream actor and conflict surfaces built on top of them.
Constraint
PowerFlow is selective by design. This layer is screened and translated before it becomes model evidence.
Live source
World Bank
What it adds
Structural footing for state actors and relationship-side context where baseline exposure matters.
Where it shows up
Actor pages, relationship evidence, and query receipts.
Constraint
Used as structural context, not as a one-stop answer to geopolitical power.
Live source
V-Dem
What it adds
Authority-mode context for state actors, especially how mobilization works under democratic versus coercive systems.
Where it shows up
Score-agent grounding, actor-page strategic profiles, and Ask PowerFlow structural receipts.
Constraint
Used to explain how authority holds together, not to mechanically set PF Scores or replace live reporting.
Live source
SIPRI
What it adds
Arms-supplier concentration and military-burden context where defense structure matters.
Where it shows up
Relationship evidence today via arms-transfer receipts, plus backend score and writer context via military-expenditure baselines.
Constraint
SIPRI TIV is structural weighting, not money. Military expenditure is burden and prioritization context, not automatic Reach, readiness, or combat effectiveness.
Live source
UN Comtrade
What it adds
Goods-trade dependence, partner concentration, and market or supplier exposure between actors.
Where it shows up
Relationship evidence today, plus backend prompt context for trade concentration and dependence.
Constraint
Trade share is exposure, not alliance, not alignment, and not automatic leverage by itself.
Live source
OECD / CRS
What it adds
Donor-recipient support evidence that sharpens support concentration, leverage, and dependency.
Where it shows up
Relationship pages and Ask PowerFlow receipts.
Constraint
This is development-finance evidence, not generic geopolitical alignment. Counterpart mapping is still uneven.
Live source
UCDP
What it adds
Historical conflict depth: battle-death context, continuity, agreements, and terminations.
Where it shows up
Conflict detail pages and conflict-shaped query receipts.
Constraint
Historical conflict context is important, but it is not the same thing as a full real-time theory of every conflict dynamic.
What this page does not do
This is not roadmap theater. Dormant schema, experimental datasets, or future integrations should not be mistaken for live trust layers. At the same time, some live sources are model-facing before they are fully page-facing. That distinction matters.
How evidence shows up
PowerFlow is not meant to cite every sentence across the site. Evidence becomes explicit where skepticism is highest and where provenance changes whether the product feels real.
Interpretation first
PowerFlow is not designed like an academic paper with citation markers on every sentence. Most surfaces lead with explanation and only then expose the evidence behind it.
Receipts where trust matters most
Relationship pages, conflict evidence blocks, and Ask PowerFlow answers show their work more explicitly because those are the places where unsupported claims fail fastest.
Meaning before metrics
When source detail is shown, PowerFlow leads with concepts like dependence, concentration, trend, and asymmetry. Raw values are secondary, not the headline.
Limits and judgment
A serious intelligence product should be opinionated, but it should not pretend to be omniscient. Credibility comes from being explicit about where the model is strong and where it still has edges.
Coverage is uneven across actors, relationships, and conflicts. Some parts of the system are denser than others.
Slow-moving structural data and fast-moving event reporting do different jobs. They are intentionally not treated as interchangeable.
Trade concentration is exposure context, not alliance or alignment. SIPRI TIV is structural weighting, not money.
Provider data can strengthen a judgment without replacing it. PowerFlow still translates evidence into model-native concepts instead of dumping raw tables.
Scores are calibrated outputs, not objective facts handed down by a single feed. They are designed to be defensible, not magical.
Go deeper
The full scoring model lives separately
This page is about provenance and trust. The detailed explanation of PF Score, Authority, Reach, anchor calibration, and relationship scoring still lives on the methodology page.
Read the full methodology