The research starts before the model.

Make every record explainable.

Test the schema. Preserve revisions. Distinguish missing values from zero. A practical reference for the interface between data delivery and investment research.

Run the contract lab Download working samples

fixture://fred-v1 / SYNTHETIC DATA / NO API KEY

observation  2024-01-01
first vintage  2024-01-02 → 100
revision       2024-02-02 → 101

selectAsOf("2024-01-31")
→ value: 100
→ availability: date precision
→ source record: retained

A working offline example, not a historical market record or proof of intraday availability.

Three adapters. One explicit contract.

Inspect the provider specifications

FRED / V1 observations shape

Keep the vintage.

Normalize the observation date and realtime date without inventing a precise publication timestamp. Preserve missing values and revisions.

Run the FRED fixture

WORLD BANK / V2 shape

Complete the pages.

Validate page counts and totals. Preserve annual period meaning and use retrieval time when row-level publication time is absent.

Read the pagination case

SEC / Company Facts shape

Retain the filing.

Keep units, filing dates and accession identifiers. The included adapter deliberately supports a narrow instant-fact example.

Inspect the sample kit

A pipeline with visible boundaries.

01 / RECEIVE

Record a response, request context and retrieval clock.

02 / VALIDATE

Reject partial pages, mixed units and conflicting identities.

03 / NORMALIZE

Retain dates, period semantics, provenance and nulls.

04 / REPLAY

Apply a declared cutoff with precision limits visible.

The integration reference.

All 20 technical guides

Data contracts

Specify fields, units, identity and compatibility before a feed reaches research.

Time & revisions

Separate event dates, publication times, retrieval times and revision visibility.

Reliable ingestion

Make pagination, retries, partial data and replay behavior explicit.

Provider interfaces

Read actual public specifications and identify the permissions needed to integrate.

Rights & delivery

Keep credentials, redistribution conditions and evidence attached to every integration.

Failure modes you can reproduce.

Worked fixture case

A January cutoff must not select a February revision

Walk through two synthetic FRED-shaped versions of the same observation and reconcile the value visible at a declared cutoff.

Trace the example

Worked fixture case

Two pages are one batch only after the contract reconciles

Validate a synthetic two-page World Bank-shaped response, preserve page metadata and distinguish retrieval time from publication time.

Trace the example

Worked fixture case

A missing observation must survive normalization as missing

Follow a synthetic missing-value token through parsing, export and analysis without converting it into an invented zero.

Trace the example

Worked fixture case

Retry once, validate the whole batch, then advance the checkpoint

Trace an illustrative 429 then 200 sequence with a two-second Retry-After and a maximum of three attempts.

Trace the example

Featured research provider / Promotional

Start with research evidence. Establish delivery separately.

Vector Ridge’s research and Darren O’Neill’s documented credentials can be inspected before discussing an integration requirement. The examples here are this publication’s own technical teaching material.

Inspect the evidence and inquiry path

Take the code with the assumptions.

Download the adapter, fixture responses, Node runner, independent Python reference and tests. No account, API key or external service is needed to run the bundled examples.

Open the reproducible sample kit