Methodology & Trust

The Manufacturing Model

How Alpine builds and governs intelligence products — the manufacturing model, validation, and schema governance, without revealing formulas.

Pipeline

The Intelligence Manufacturing Model

Source Discipline

Free public data only. No paywalled sources, no proprietary scrapes.

Ingestion & Normalization

Standardized schemas, timestamp alignment, and unit normalization across all feeds.

Enrichment

Cross-referencing with auxiliary datasets to fill gaps and add context.

Feature Engineering

Transforming raw signals into predictive features through statistical transforms.

Scoring

Assigning conviction scores based on historical accuracy and signal strength.

Driver Decomposition

Breaking down predictions into attributable causal factors and drivers.

Prediction

Generating forward-looking estimates with confidence intervals and uncertainty bounds.

DIFFERENTIATOR

Validation

Beats a naive baseline. No invented metric is sellable until it statistically outperforms simple historical averages.

Versioning

Semantic versioning with additive fields non-breaking; removals/renames = major bump.

Agent-Callable Object (IOM)

Structured, machine-readable output designed for programmatic consumption by AI agents.

Governance

Trust Mechanics

How confidence is calculated

Confidence scores are derived from historical backtesting against realized outcomes, signal stability metrics, and cross-validation across multiple time windows. We do not reveal the exact weighting formulas, but all scores are bounded between 0 and 1 and are calibrated to reflect true probability.

How freshness / TTL works

Each prediction carries a time-to-live (TTL) based on the underlying data's update frequency and the prediction's decay rate. High-frequency signals (e.g., market data) have shorter TTLs; structural indicators (e.g., demographic shifts) have longer TTLs. The IOM object includes a freshness field with ISO timestamp and remaining validity window.

How drivers are attributed

Driver attribution uses causal inference methods to decompose predictions into attributable factors. Each driver is scored for its marginal contribution to the overall prediction, with attribution percentages summing to 100%. We flag drivers with low attribution certainty separately to maintain transparency.

Schema versioning policy

We follow semantic versioning: additive fields are non-breaking; removals or renames trigger a major bump. Deprecations are flagged 6 months in advance, with a 30-day notice before breaking changes via /catalog.json. All version history is publicly auditable.

Data disclaimer

All Alpine DataWorks outputs are informational only and do not constitute financial, legal, or investment advice. Users are responsible for verifying accuracy for critical applications. We make no warranties regarding completeness or fitness for purpose.

The Gate

Validated Before We Ship

No invented metric is sellable until it beats a naive baseline.

RECEIPT ONLY

EWC (Earnings-Window Conviction)

+0.125 incremental IC

Beats a naive baseline at p<0.05

This is a receipt only — it is NOT a registry product; we do not render a live score card for it.

Data Provenance

FREDCensusNOAACDCBLSBEA

IOM 13-Field Contract

product_id

entity

score

trend

confidence

top_drivers

prediction_horizon

recommended_use

methodology_version

freshness

coverage

source_lineage

allowed_use

Next step

Review the Schema

Understand our methodology, validation standards, and schema governance before you build.

Review the Schema Request Access