Methodology and Data Sources

Methodology principles

1. Prefer official sources.
2. Keep source provenance per record.
3. Label incomplete or unverified data.
4. Separate raw, normalized, and published data.
5. Do not publish final claims when the primary data layer is missing.

See also

• Classifier coverage matrix
• Risk score methodology

Verification workflow

Data is first pulled into the raw layer, then cleaned in the normalized layer, and only then published in the final layer. During that process we compare PDFs, official portals, government lookup pages, and selected fallback sources. If sources conflict, the default priority goes to the latest official source that can be rechecked. If the conflict cannot be resolved, the page must be labeled as uncertain or withheld from publication.

Confidence tiers

• official: data comes directly from an official source and can be traced back
• fallback: data comes from a secondary source used to clarify or complete the structure
• brief: there is a plausible signal, but not enough for a final claim
• needs-verification: there is an early indication, but manual confirmation is still required
• pending-import: the target dataset has not been loaded into the publish pipeline yet

Review cadence

Pages are not reviewed in one bulk batch. Each page carries its own review date and that date should change when metadata, source references, or cross-links are updated. Seasonal datasets such as minimum wages are reviewed more aggressively near the annual update cycle. Regulatory datasets such as KBLI, KLU, OSS, and foreign ownership are reviewed when there is evidence of rule changes or editorial corrections.

How official files are turned into published pages

The pipeline does not start from copywriting. It starts from raw files pulled into the raw layer, then cleaned into normalized JSON, and only then published into the site. In this review pass we checked again that the definitional material from the BPS KBLI PDF and the DJBC BTKI FAQ is strong enough to support hub and trust pages, while still keeping partial data layers clearly constrained.

How official files and fallback files are treated differently

When an official file provides a clear definition, structure, or effective date, we use it to strengthen hub pages, methodology notes, source pages, and explanatory blocks. When a fallback source only helps orientation, it is not allowed to replace the main claim that should come from an official source. That is why KLU and wage coverage still carry some fallback dependency, while the explanatory layer for KBLI and BTKI is now anchored more directly in official material stored locally.

What we still do not publish even if the file was pulled locally

Some local files remain outside the publish layer on purpose. Access-denied pages, third-party search markup, or discovery scripts without a solid regulatory basis may still be useful for internal exploration, but they are not strong enough to upgrade a YMYL page into a fully verified claim. Keeping that boundary explicit is part of the site’s anti-spam and trust discipline.

This portal is informational. Confirm the final obligation and competent authority before filing, licensing, payroll, tax, or investment decisions. Read methodology.