Purpose
This project is a forecasting record, not simply a geopolitical dashboard. It tracks Iran's political stability, regional escalation, maritime trade, foreign-exchange and fiscal stress, energy stress, external support, and Iran-linked Houthi capabilities using public information.
Observation, inference and forecast
Observation is a reported or measured fact attributed to a source. Inference interprets observations and should carry a confidence judgment where consequential. Forecast is a forward-looking probability attached to a defined event and horizon. Forecast probabilities are analytical judgments, not published statistics or market prices.
Early-warning states
Green — Normal: affirmative evidence indicates no unusual movement toward the specified failure mechanism. Yellow — Watch: deterioration is evident but the threshold sequence has not advanced far. Orange — Warning: multiple corroborating indicators show substantial movement toward the threshold. Red — Critical: the failure mechanism is operating or appears close to becoming self-reinforcing. Gray — Insufficient data: public evidence is inadequate; gray must never be interpreted as green.
Direction is separate: ↑ deteriorating, → broadly stable, ↓ improving. Evidence coverage is separately rated High, Medium, or Low.
Forecast definitions
Sustained major regional escalation within 90 days: renewed sustained major interstate strikes or material geographic widening beyond episodic contained exchanges.
Regime-ending transition within 12 months: loss or replacement of the Islamic Republic's governing institutional order.
FX crisis: disorderly parallel-market depreciation or a foreign-currency/payment shortage severe enough to require broad emergency controls or materially disrupt normal imports and payments.
Hanke hyperinflation: inflation exceeding 50% per month for at least 30 consecutive days. Monthly reports may use exchange-rate-implied inflation as an additional early-warning signal.
Major restructuring: substantial redistribution of governing power within Islamic Republic institutions without ending that institutional system.
Probability, confidence and the uncertainty rule
Probability answers “how likely is the event?” Confidence answers “how strong and complete is the evidence supporting that probability?” They are not interchangeable.
Explicit uncertainty rule: first set the point forecast. Then estimate the 10th and 90th percentiles of the point probabilities that independent, well-informed analysts using the same public information, event definition and horizon would reasonably give. The displayed judgmental uncertainty range is that reasonable-analyst interval. It is intended to contain roughly 80% of such analysts’ point estimates. It is not a statistical confidence interval, Bayesian credible interval, sampling-error interval, or a probability that the “true probability” lies inside the range.
Revision dates and source dates
The revision date is the forecast publication date on which evidence changed the assessment. The source/event date records when the underlying event occurred or the supporting source was published. These can differ when evidence is incorporated after publication.
Revision taxonomy
Every material revision should be classified as New event, Newly incorporated evidence, Model reassessment, or Correction. A correction never silently rewrites forecast history.
Diagnostic pathways
The displayed chains are diagnostic pathways, not deterministic causal stages. Stages can be skipped, reversed, occur simultaneously, or be reached through mechanisms outside the displayed pathway. Three mechanisms receive disproportionate weight: FX/fiscal: depreciation → intervention failure → monetary substitution → shortening price/wage intervals → fiscal monetization. Energy/real economy: rationing → industrial/freight disruption → essential-goods shortages → public-sector arrears → failure of priority allocations. Regime/coercion: economic protest → coordinated national protest → elite fracture → coercive refusal → defections.
Leading, coincident and lagging indicators
Payment rejection, parallel FX, tanker flows and security-force compensation can be leading indicators. Shortages and coordinated protests are often coincident indicators. CPI and GDP frequently confirm deterioration after it is already visible elsewhere.
Absence indicators
Important non-events constrain forecasts: no verified security-force defections, no nationwide bazaar closure, no failure of military fuel priority, no high-frequency economy-wide wage indexation, and no cutoff by a systemically important Chinese bank are examples. Absence is used only where observation coverage is adequate.
Original-indicator protocols
The Iran–China oil-credit haircut is a model estimate, not an observed contract price. Its central estimate is the sum of bounded delivery, sanctions/payment, counterparty, basis, illiquidity/enforcement, and political-tail components. Component anchors and caps are published with the current estimate. Houthi sustainability is separated into Iranian matériel support, Iranian technical/financial support, and Houthi independent/local capability; local capability alone does not count as continued Iranian support.
Source provenance
Priority goes to primary documents, official data, reputable international reporting and clearly identified specialist sources. Claims from governments, belligerents or opposition groups are treated as claims unless corroborated. Derived quantities such as the China oil-credit haircut are labeled model estimates.
Resolution and corrections
Scored forecasts require an event definition and horizon sufficiently explicit for independent coders to resolve. Legacy scenario partitions are retained for continuity but are not treated as ideal scored events where categories overlap conceptually. Corrections preserve the original historical record and display the correction date, corrected claim, reason, and any forecast effect.
Limitations
Iranian economic, military and political data are incomplete, delayed, sometimes contradictory and strategically manipulated. Open-source visibility is asymmetric. Missing observations are not treated as benign conditions.
Prospective scoring and retrospective decompositions
Only forecasts issued prospectively are eligible for Brier scores and calibration. A forecast must be explicitly marked scored=true and must have been created no later than its forecast publication date. A probability introduced later to decompose or reinterpret an older forecast is labeled Retrospective methodological decomposition — not scored. Such estimates remain visible for methodological transparency but never enter forecast-performance statistics.
The site separately records the assessment date, the evidence cutoff for that assessment, and any later methodology/page revision date. A later methodological revision does not alter the information set attributed to the original assessment.
Protocol versioning
Protocol v1.0 freezes the permanent core scored set, exact resolution rules, evidence-cutoff policy, warning transition rules, evidence-reliability rubric, China-haircut anchors and overlap correction, Houthi-support definition, and calibration policy. Substantive changes require a new numbered protocol with a prospective effective date. Historical forecasts remain governed by the rules and wording under which they were issued; later operationalizations are identified rather than silently back-applied.
Resolution, missing outcomes and dependence
Every prospective core forecast receives an immutable forecast ID, opening timestamp, and exact resolution deadline. If public evidence cannot reliably determine Yes or No at resolution, the outcome is coded indeterminate and excluded from scoring under the frozen rule; the reason remains public. Ambiguous outcomes require a written resolution note tied to the forecast ID.
Weekly forecasts with overlapping horizons are serially dependent. Per-forecast Brier scores remain useful, but overlapping forecasts are not treated as independent observations when interpreting calibration, uncertainty, or comparative performance.
Evidence persistence and revision attribution
Evidence receives persistent IDs. Evidence carried into later weeks is marked carried forward rather than recreated as a new observation. Material evidence is classified as raising, lowering, or ambiguous and as minor, moderate, or major under the frozen rubric. Beginning with Protocol v1 forecasts, probability revisions record approximate signed percentage-point contributions that reconcile the previous point forecast to the new point forecast. These contributions audit judgment; they are not causal estimates.
Protocol-development register
Methodological criticisms and research ideas that would alter frozen v1 are recorded in a non-operative candidate register rather than silently changing v1. These include dependence-aware performance aggregation, benchmark forecasts, overlapping-horizon resolution automation, richer resolution records, evidence-storage architecture, equivalent-haircut terminology and uncertainty decomposition, Houthi stock-flow accounting, warning-state validation, endogenous missingness, source-independence measurement, and prospective-validation maturity. None applies unless adopted prospectively in a later numbered protocol.
Publication validation
Executable horizon convention: 90-day forecasts use 90 local calendar days at the same local clock time; 12-month forecasts use the same local calendar date/time one year later (February 29 maps to February 28). The UTC offset may therefore change with daylight-saving time. The validator computes and checks these deadlines.
Validation distinguishes errors from warnings. Known post-cutoff source publication, timestamp-order violations, wrong deadlines, arithmetic failures, malformed core records, and unauthorized historical mutation are errors. Missing precise source-publication time is a warning and remains explicitly unknown rather than being invented.
Repository validation checks the machine-readable history on every change. Protocol-v1 records are checked for the four core forecasts, IDs, probability/range bounds, nine warning categories and allowed states, evidence audit fields, haircut arithmetic when present, and signed revision-attribution reconciliation. A formal JSON Schema supplies an additional structural contract. Historical immutability is also checked by the Sunday publication procedure before a new record is committed.
Forecast ledger and evaluation
The machine-readable ledger is designed to retain forecast date, creation/provenance status, event definition, horizon, probability, uncertainty, evidence confidence, revision type and source information. As forecasts resolve, resolution date and outcome can be added. This permits Brier scores and calibration curves without hindsight contamination.