AI macro currency strength & forward outlook

cMind ship một AI-assisted, math-deterministic macro currency-strength engine. Nó ranks một configurable universe của currencies — 8 majors cộng emerging-market và exotic currencies — by current fundamental strength, và projects một forward directional outlook cho mọi pair over a chosen horizon (1M / 3M / 6M / 12M). Mọi rank, mọi pair bias và mọi số được compute bởi pure deterministic math trong domain core; LLM chỉ gathers các forward-looking inputs mà data can't publish và explains kết quả in plain English. Nó không bao giờ invent một rank, một direction hoặc một số.

Honest limitation. Fundamentals predict medium-to-long-term value tốt và short-term value poorly. Treat cái này như một positioning / confluence filter, không phải a short-term timing signal. Readings gần high-impact releases (NFP/CPI/central-bank) noisy. Not financial advice.

How it works

1. Current fundamentals đến từ Economic Calendar, không phải LLM. Hard numbers — policy rates, CPI vs target, GDP, employment, trade balance — và their surprise z-scores được sourced point-in-time từ economic calendar module (FRED/BLS/BEA/ECB and central-bank schedules). Một historical snapshot không bao giờ leaks look-ahead.
2. LLM gathers chỉ những gì calendar không thể publish — per currency: forward trajectory (expected policy-rate path in bp, inflation-trend-vs-target, growth momentum) và a geopolitical outlook (risk-on/off, tariffs, fiscal/debt, elections), cộng any EM/exotic current figures calendar lacks. Strict JSON, tier-aware validation, web search on.
3. Domain computes ranking và forward matrix deterministically. Mỗi driver scored as a within-tier z-score (vì vậy một 50%-inflation exotic không bao giờ distorts majors), winsorized, weight-summed into a composite, và ranked strongest→weakest với stable ISO tie-break. Forward layer carries each composite along its trajectory — projected = current + horizonScale · Σ trajectoryDriver·weight — và maps each pair's projected differential to a directional bias (▲ appreciate / ▬ neutral / ▼ depreciate) với conviction.
4. LLM explains ranking và top pair calls in plain language.

The drivers

Driver	Effect on strength	Notes
Policy rate & trajectory	Higher / hawkish ⇒ stronger	Highest weight; central-bank divergence drives biggest gaps.
Inflation (CPI vs target)	Above target ⇒ weaker	Scored inversely (purchasing-power drag).
GDP growth	Higher relative growth ⇒ stronger	Differential vs the panel.
Employment	Stronger labour ⇒ stronger	Feeds the policy path.
Trade balance / current account	Surplus ⇒ stronger	Structural demand.
Policy stance	Hawkish ⇒ stronger	Primary long-term driver.
Surprise momentum	Recent beats ⇒ stronger	From calendar's surprise z-scores.
Geopolitical / risk	Risk-off ⇒ safe havens (USD/JPY/CHF) stronger	Bounded forward risk delta.
Real yield / carry (EM/exotic)	Positive real rate ⇒ stronger	Dominant EM driver in calm regimes.
External vulnerability (EM/exotic)	Deficits / low reserves / USD debt ⇒ weaker	Structural depreciation pressure.
Terms of trade (commodity exporters)	Rising export prices ⇒ stronger	BRL, ZAR, CLP, NOK, AUD, CAD.
Political / institutional risk (EM/exotic)	Instability ⇒ weaker	Wider dead-band, capped conviction.

Tiered universe (majors + EM + exotics)

Universe là deployment-configurable (App:CurrencyStrength:Universe) — adding a currency là config, không phải code. Mỗi currency mang a tier (Major / EmergingMarket / Exotic) that tunes weighting, dead-band width và conviction cap:

• Majors — USD, EUR, GBP, JPY, AUD, NZD, CAD, CHF (rate-level led).
• Emerging markets — CNH, INR, BRL, MXN, ZAR, KRW, SGD, PLN (+ Scandi NOK/SEK); carry + risk + external-vulnerability weighted up, medium confidence.
• Exotics — TRY, HUF, CZK, cộng USD-pegged HKD/SAR; low confidence, wider dead-band, capped conviction. Pegged / heavily-managed currencies (HKD, SAR, CNH) được flagged, their trajectory down-weighted, và their pair outlook clamped toward Neutral vì vậy một peg không bao giờ được read as a free-floating signal.

Vì official EM/exotic stats lower-frequency, revised và sometimes opaque, AI-gathered figures carry a per-tier confidence shown as reliability badge.

Graceful degradation

Calendar	AI	Result
✅	✅	Full ranking + forward projection + narrative (CalendarAndAi).
✅	❌	Calendar-only current ranking, no forward projection (CalendarOnly).
❌	✅	AI-gathered current figures + forward, lower confidence (AiOnly).
❌	❌	No snapshot — widget hides và page shows empty state.

App runs unchanged either way. AI gated on AI key; calendar leg respects its own white-label gate + runtime toggle.

Using it

• Enable AI (Settings → AI) và turn on widget từ your own dashboard Customize dialog ("Currency strength" — opt-in, hidden by default). Widget shows top strong/weak currencies và top 3M pair call; nó links to full page.
• Full page — /ai/currency-strength: horizon selector (1M/3M/6M/12M), tier filter (All/Majors/EM/Exotics), current ranking, forward forecast, pair-outlook matrix (bias + conviction, pegged/low-confidence flagged), và AI narrative. Press Refresh now (owner) to regenerate. Background worker (App:CurrencyStrength:RefreshEnabled, default true) refreshes on a schedule nên page populated out of the box; deployment hoặc owner turns it off (hoặc disables AI / economic-calendar feature, mà refresher honours by degrading to no snapshot).

Programmatic access

Một shared read model (ICurrencyStrengthQuery) reachable three ways:

• In-app AI — injected directly (in-process) vào AI features.
• MCP — currency_strength tool (params horizon, tier) cho AI clients/agents.
• cBot REST — GET /api/market/v1/currency-strength/{latest,history,pair/{base}/{quote}}, secured by cùng CalendarJwt machinery như calendar cBot API với added market:read scope. Một cBot registers API client với market:read, exchanges its id + secret for a short-lived JWT at POST /api/calendar/v1/token, và calls endpoints với a Bearer token. Không second JWT scheme, không second secret — a leaked token is read-only, market-scoped, short-lived và revocable.

Xem calendar cBot API cho token flow và copy-paste sample.