Metodologia (v1.2.0)

ReliefSignal agrega informações publicamente disponíveis em registros estruturados com timestamps e referências à fonte. É não político, não militar e não operacional.

Governance: events may be auto-flagged for human review based on confidence, ambiguity, severity, and source restrictions. Rejected events are excluded from scoring.

Language safeguard: summaries are normalized to neutral tone using a rule-based filter before storage.

Modelo de Risco (V1)

O score de risco (0–100) é um composto ponderado:

0.35 MilitaryIntensity
0.15 AirspaceImpact
0.15 MaritimeImpact
0.20 CommoditySignal
0.15 DiplomaticEscalation

A V1 é conservadora: reporta sinais observados e drivers com base em fontes públicas. Não prevê resultados.

Live-read audit path

Use this page together with the first-fold trust path and score explainability surfaces: the dashboard tells you what is elevated, while methodology tells you how that elevation is formed and what its limits are.

Benchmark proof package

This methodology page is part of ReliefSignal's public proof package: a reference-grade system should be able to explain not just its outputs, but the benchmark it is trying to satisfy — speed of understanding, transparency, stress usability, continuity support and public accountability.

Benchmark claims under inspection

Five-second understanding
Public transparency and method
Continuity and fallback awareness
Benchmark accountability

Inspection path

Read the public benchmark defense layer first
Read the live dashboard first fold
Inspect the methodology
Inspect source provenance
Inspect corrections

Benchmark proof matrix

The About page now provides a claim → surface → inspection matrix so users can verify benchmark claims through explicit public routes.

Public benchmark contract

Methodology is part of the public benchmark contract: if ReliefSignal claims benchmark ambition, this page must help users inspect how that ambition is operationalized and where its limits remain.

Trust-impacting revision path

Use methodology together with corrections and methodology changelog whenever a trust-impacting change may alter how users read confidence, provenance, thresholds or trustworthiness.

Corrections Methodology changelog

Public revision-accountability contract

Methodology should help users understand not only how outputs are formed, but also how trust-impacting revisions, corrections and methodological changes are exposed over time.

Usefulness under uncertainty — inspection checklist

Methodology should not stand alone as theory. It should help users verify that the live product remains useful under uncertainty, degraded attention and compressed reading conditions.

1. Front-door stress scan
Check whether world → local reading remains understandable in one compressed pass.
Open this proof surface
2. Pocket brief under degraded attention
Verify that the compact brief still explains what changed, why it matters and what to check next.
Open this proof surface
3. Methodology under pressure
Check whether the product still explains how it forms outputs and where its limits remain.
Open this proof surface
4. Corrections and trust path
Verify that uncertainty, corrections and trust-impacting revisions remain inspectable.
Open this proof surface

Global watchlist model (v1.2.0)

Watchlists are configurable by environment (not hardcoded in product logic).
Each watchlist has explicit scope (ISO3 + term matcher) and weighted priority boost.
Review queue priority combines severity, confidence risk, freshness, and watchlist weights.

Notification methodology

User-controlled preferences: channel, destination, language, severity, watchlists, delivery mode, quiet hours.
Dispatch respects quiet-hours and minimum severity before any alert is considered.
Dedupe + cooldown logic prevents repeated spam for equivalent alert payloads.
Dispatch logging keeps auditable records of sent/skipped/failed decisions.

Data sources transparency

See the complete public source directory and links used by the ingestion pipeline.

Tier policy: A=official/IGO/government, B=established media, C=auxiliary weak signals.

Official social channels may be monitored as weak signals, but scoring prioritizes corroborated web/official sources.

Safeguard: weak-signal-only evidence is capped and cannot alone drive the highest risk tier in Telegram local summaries.

View all sources

Frequência de atualização

Eventos são ingeridos de forma programada. Scores são calculados e armazenados como valores atuais e snapshots diários.

Ver changelog de metodologia

Limitações

A cobertura depende da disponibilidade pública e do alcance de idiomas.
Não é um produto de orientação oficial de autoridades.
Sem precisão tática: apenas coropléticos; modelo de localização grosseiro reforçado no banco.