Collect
Retrieve each registered BRH or IHSI public dataset on an independent schedule.
Intelligence
METHODOLOGY & GOVERNANCE
Every value is traceable.
The platform preserves the BRH or IHSI evidence behind each published observation and separates automated processing from human judgment.
DATA PIPELINE
One controlled path to publication
Preserve
Store the exact HTML or PDF, final source URL, collection time, media type, and SHA-256 digest.
Validate
Normalize source-labelled values and check period semantics, plausible ranges, and series-specific changes.
Publish
Write accepted or warning observations as immutable revisions and expose them through the generic API.
PUBLICATION RULES
Automated checks, cautious outcomes
Plausible rangeSeries-specific hard bounds; invalid values are quarantined
Change thresholdSeries-specific comparison with the previous published period
DuplicateSame indicator, period, and value produces no new revision
RevisionA changed value for the same indicator and period creates a new immutable version
AnomalySuspicious candidates are quarantined and never auto-published
ANALYSIS METHODOLOGY
Transparent transformations, common periods
The workspace aggregates each series according to its economic meaning, aligns only periods shared by every series, and computes statistics from the displayed transformed values.
Common frequencyA selected frequency is never higher than the least frequent source series; no interpolation is used.
AggregationRates default to averages, stocks and indexes to period-end values, and flows to sums, as documented per indicator.
AlignmentOnly common economic periods are retained. Missing values are never filled, interpolated, or replaced with zero.
TransformationsLevel, changes, index normalization, and logarithms run after frequency alignment using explicit formulas.
CorrelationPearson and lagged correlations describe exploratory co-movement, not causal or predictive relationships.
FORECASTING METHODOLOGY
Prediction requires evidence beyond fit
Forecasts are experimental analytical outputs. Models must earn selection through out-of-sample walk-forward testing against simple baselines.
TargetsUSD/HTG source observations and monthly published inflation only; annual indicators are excluded.
Information timingSource publication date is used first, then first-seen and collection timestamps. Future information is never admitted at a historical origin.
BacktestingExpanding-window origins evaluate each supported horizon separately. Random train/test splits are not used.
BaselinesPersistence, drift, and monthly seasonal persistence are evaluated before ARIMA, ETS, or autoregression.
SelectionMASE is primary, RMSE secondary, and simpler models win near ties. A weak model is disclosed rather than promoted.
UncertaintyNative 80% and 95% prediction intervals are shown where supported; unavailable intervals are never fabricated.
RevisionsCurrent data require latest-revision evaluation, which may overstate performance until historical vintages accumulate.
LimitationsStructural breaks and unprecedented shocks can make historical forecast performance unrepresentative.
SOURCE & STATUS
BRH + IHSI
The dashboard does not estimate missing values, recalculate source-published inflation, or publish quarantined candidates. Exact source archives are downloadable from the Data Explorer.
AuthorityBRH · IHSI
SeriesExchange rate, CPI, inflation, net reserves, M2, and remittances
Official sourceBRH official datasets ↗ · IHSI — IPC ↗
PipelineBRH / IHSI → Collector → PostgreSQL → FastAPI → Next.js
CollectorHealthy
Last successAug 27, 2026, 10:10 PM
Quarantined candidates0
BRH_BDP_ANNUAL · healthy
Last attemptAug 27, 2026, 4:53 PM
Last successAug 27, 2026, 4:53 PM
Last failureNot available
Quarantine0
BRH_DAILY_RATE · healthy
Last attemptAug 27, 2026, 10:10 PM
Last successAug 27, 2026, 10:10 PM
Last failureNot available
Quarantine0
BRH_MONETARY_ANNUAL · healthy
Last attemptAug 27, 2026, 4:52 PM
Last successAug 27, 2026, 4:52 PM
Last failureNot available
Quarantine0
IHSI_CPI_PAGE · healthy
Last attemptAug 27, 2026, 4:52 PM
Last successAug 27, 2026, 4:52 PM
Last failureAug 26, 2026, 1:54 PM
Quarantine0