A probability number for a geopolitical event answers one question while deferring another. It answers: given the current information environment, how likely is this outcome? What it defers is the question of which parts of that environment drove the reading, and how independently were they moving.
Source attribution closes that gap. When the probability that a regulatory review affecting a specific trade corridor proceeds to formal proceedings shifts from 44% to 57% over 36 hours, the shift itself is informative. The attribution behind it tells an analyst whether the evidence base was broad or narrow, correlated or independent, and therefore how defensible the reading is. Without attribution, a probability score is a number an analyst must accept on faith. With it, the number has a visible structure.
What Source Attribution Records
Attribution at the source level tracks, for each scoring event, the contribution of each source category to the resulting probability change. This is distinct from listing which sources a platform ingests. The question is not what the system has access to but which signals were actually responsible for a specific shift.
The accounting addresses three dimensions: which source categories contributed to the probability change; what each category's contribution was, expressed as points or proportion of the total shift; and how recent those contributing signals were. A signal from 12 hours ago and a signal from 9 days ago carrying identical content have different evidentiary status. Recency is part of what attribution records alongside the contribution values themselves.
Together, these three dimensions tell an analyst whether a probability reading reflects fresh intelligence across multiple structurally independent observation types, or concentrated activity in a single source category that may be internally correlated.
How Source Types Behave in Geopolitical Contexts
Geopolitical events generate coverage from structurally different observers, each with distinct biases and update patterns. Understanding attribution means understanding what each category is likely to see and when.
News wire coverage is fast but sometimes correlated. When multiple outlets pick up the same story from shared syndicated sources, the apparent signal volume exceeds the actual information content. Wire services also respond to public statements faster than underlying policy mechanics shift, which can produce probability volatility that reverses when the news cycle moves on.
Government document feeds, including regulatory publications, official filings, and parliamentary records, move more slowly but carry higher specificity. A government document signal typically reflects action at the procedural level rather than the discourse level. These signals tend to be durable because they correspond to institutional process, not public attention cycles.
Field intelligence, meaning observer-sourced and geographically specific inputs, has narrower coverage scope but may capture information not yet in public circulation. Field signals are structurally independent from wire coverage, which is precisely what makes them analytically valuable. When a field signal and a wire signal point in the same direction, they are doing so from genuinely different vantage points with different information access.
Policy preprints and analyst publications provide expert-consensus signals. When specialists begin publishing on near-term outcomes in a policy domain, it reflects the professional community's collective assessment of the direction of travel. This category moves slowly but correlates well with eventual outcomes in regulatory and political domains, particularly for events with long procedural lead times.
Concentration Risk and Correlated Signals
When attribution shows that 70% or more of a probability shift came from one source category, that concentration pattern carries analytic weight. It does not mean the shift is wrong. It means the evidence base is narrower than the probability change might suggest.
A 13-point shift distributed across news wires, government feeds, field intelligence, and analyst publications suggests that independent observers with different structural biases are updating in the same direction. That is meaningful convergence. The same 13-point shift driven 85% by news wire activity suggests a narrower story, possibly multiple outlets picking up a single narrative source without genuine independent confirmation.
Source concentration also relates to signal durability. News-wire-heavy shifts tend to be more volatile because wire coverage follows the attention cycle. A probability change driven primarily by wire activity may partially revert once the news cycle moves on, unless other source categories confirm the same direction. Knowing the concentration profile helps an analyst judge how stable the current reading is likely to be over the next several days before additional evidence accumulates.
Reading Convergence in Practice
A synthetic example makes this concrete. Consider an event defined as: a regional regulatory body initiates formal review proceedings affecting semiconductor component classification within 90 days. Initial probability: 44%, reflecting base rates for regulatory reviews of this type and region.
Over 36 hours, the probability moves to 57%. Attribution shows the following breakdown:
- Government document feeds: +6.1 points (48% of total shift)
- News wire coverage: +3.9 points (30% of total shift)
- Field intelligence: +2.7 points (21% of total shift)
- Analyst preprints: +0.3 points (2% of total shift)
The distribution changes the interpretation materially. Government document activity drove nearly half the shift, suggesting procedural movement within regulatory channels rather than a spike in public discourse. Wire coverage and field intelligence confirmed the same direction independently. Preprint activity was minimal, consistent with an event at an early stage before academic commentary accumulates.
An analyst presenting this reading would frame it differently than if the same 13-point shift had been 90% wire-driven: "The reading has moved to 57% primarily on government document activity confirmed by field intelligence. This appears to reflect procedural development rather than news cycle dynamics, and we expect it to be more durable." That framing is only possible with attribution. Without it, "57%" is a conclusion with no visible supporting structure.
Attribution as an Analytic Audit Trail
For corporate risk functions, probability estimates need to be defensible after the fact as well as in the moment. If a supply chain or financial exposure decision was made on a 57% probability reading, the risk function needs to be able to reconstruct why it held that position when asked by senior management or by a post-event review.
Attribution provides that reconstruction. Each probability reading carries a record of which source categories were contributing at that time and in what proportion. The post-event record is not just "we assessed 57%" but "we assessed 57%, driven primarily by government document activity and field intelligence convergence as of that date." One of these is auditable; the other is not.
This matters particularly for decisions that may face scrutiny after outcomes resolve. If the event occurs, did the organization have a defensible basis for its probability assessment? If it does not occur, can the organization explain what was driving the position and what changed? Attribution gives the risk function the material to answer both questions coherently, which is the baseline for treating probability estimates as institutional decisions rather than individual judgments.
What Attribution Does Not Tell You
Attribution records what moved the probability score. It does not certify that the sources which moved it were correct.
A source category can have strong historical calibration for one event type and weaker calibration for another. News wire coverage may be well-calibrated for corporate announcement events but less reliable for political transition events, where public discourse often lags the underlying decision process by weeks. Attribution shows contribution; calibration history shows track record. These are separate questions requiring separate data to answer.
The practical implication: when attribution shows that a shift was driven primarily by a source category with relatively lower calibration weight for the specific event type being scored, the analyst should treat the shift with appropriate skepticism even if the direction seems plausible on its face. Attribution is a starting point for analytic judgment, not a shortcut past it.
Used together, attribution and calibration weights give an analyst a richer picture than either provides alone: not just what moved the number, but how much to trust what moved it for this specific event category and time horizon. That combination is what makes a probability reading analytically defensible rather than merely numerically precise.