Global Risk Index architecture & proof system
The Global Risk Index (GRI) is a deterministic weighted-intensity index of qualifying Geomacro risk evidence after event classification and current-contract story assignment. It is designed so a published score can be inspected through its inputs, concentration controls, contribution ledger, change attribution and integrity hashes.
System contract
One current calculation path
Sources
Evidence + observation time
Classification
Domain · severity · confidence · provenance
Story assignment
Underlying-development correlation
GRI engine
Source cap + story cap + aggregation
Proof
Contribution ledger + hashes
Publish
Verified immutable snapshot
Observed evidence
Model-produced inputs
Deterministic aggregate
Definition
What GRI measures — and what it does not
GRI is
- A 0–100 weighted intensity index of qualifying Geomacro risk evidence.
- Deterministic after event classification and current story assignment.
- Confidence- and recency-weighted with source and story concentration controls.
- Published with coverage, evidence counts, confidence and verification context.
GRI is not
- A prediction-market probability.
- A census of every event in the world.
- A claim that missing evidence means zero global risk.
- A predictive-performance claim without a separately preserved validation result.
Current domains
Three equal base-weight scoring domains
Geopolitics
Base weight 1/3
Macro
Base weight 1/3
Rare earth / critical minerals
Base weight 1/3
If one domain has no eligible evidence, it is excluded instead of receiving a synthetic zero. Remaining active weights are renormalized and coverage is disclosed separately. Crypto may exist elsewhere in Geomacro's broader data/technical architecture, but it is not a current GRI v1.2 scoring domain.
Eligibility
Canonical evidence window and provenance
Current GRI observations must satisfy the supported-domain, severity, confidence, observation-time, classification-provenance and story-assignment contracts. The trailing window is 72 hours and the exponential recency half-life is 24 hours.
The canonical observation time represents when Geomacro knew the observation. Publisher time remains provenance and cannot backdate a historical snapshot.
Concentration control
Source cap, then story cap
1. Source cap
Within a domain, one stable source receives at most 1.0 total evidence weight. Its eligible observations share that budget in proportion to raw weight.
2. Story cap
After source capping, observations assigned to the same underlying development share one story budget based on the strongest constituent source total, capped at 1.0.
Repeated publication from one source or multiple publishers therefore cannot multiply one underlying development into unlimited independent evidence weight.
Aggregation
From effective evidence to the global score
01
Domain score
Σ(severity × effectiveEventWeight) / Σ effectiveEventWeight
02
Active weights
Renormalize the 1/3 base weights only across domains with eligible evidence.
03
Global score
GRI raw = Σ(active normalized weight × domain score); display = round(raw).
The published snapshot retains higher-precision raw values in addition to the integer display score.
Change attribution
Every material move should reconcile
An observation can change contribution because it was added, removed, rescored or reweighted by recency, source concentration, story concentration or active-domain normalization. The proof preserves the effective weights used by each snapshot and records reconciliation/change residuals under the current numeric contract.
Proof package
Published score → contribution ledger → evidence
Evidence / classification proof
- Source and observation provenance.
- Severity and confidence.
- Classification provider/model/version/prompt/input provenance.
- Current story assignment and story-correlation provenance.
Aggregate / integrity proof
- Raw, source-capped, story-capped and effective evidence weights.
- Domain scores, normalized weights and exact contribution points.
- Methodology, input, evidence, calculation and proof hashes.
- Disposition/change integrity fields required by the current public contract.
- Reconciliation and change residuals.
Publication integrity
Draft → verify → immutable publish
1. Draft
Build a candidate snapshot that is not yet authoritative.
2. Proof
Persist contribution, evidence and integrity material required by the contract.
3. Verify
Check contract versions, hashes and reconciliation before publication.
4. Publish
Expose only a qualifying immutable published snapshot through the public read model.
A correction should become a new publication or methodology version rather than a silent rewrite of historical proof.
Validation boundary
Proof is not the same as predictive validation
Deterministic proof demonstrates how a score was calculated. It does not by itself prove that the index predicts markets, losses or future geopolitical events. External benchmark, historical replay and out-of-sample validation should remain separate from the production score calculation and should be claimed only when a preserved run supports the claim.
Versioning
Material numeric changes require a new version
Versioned contract
- Scoring domain set or base weights.
- Lookback window or recency half-life.
- Source/story concentration semantics.
- Eligibility and timestamp rules.
- Missing-domain normalization.
- Rounding or contribution/change-attribution semantics.
Public verification model
Product surfaces can stay readable while the verification path exposes the methodology, contribution ledger, evidence context, hashes and change attribution needed to inspect a published GRI.