Verification Methodology

Every knowledge unit on knowledgelib.io carries a confidence score, source provenance, and freshness indicator. This page explains how we assign and maintain these.

Confidence Score (0.0 - 1.0)

The confidence score reflects our assessment of how reliable and complete the knowledge unit is. It is NOT a probability that the content is correct — it's a composite measure of source quality, cross-verification, and completeness.

Score	Meaning	Requirements
0.90 - 1.00	High confidence	3+ independent sources, including at least 1 primary/authoritative source. All key claims cross-verified. No significant conflicting data found.
0.75 - 0.89	Good confidence	2+ independent sources. Most claims cross-verified. Minor gaps or minor disagreements between sources noted in caveats.
0.60 - 0.74	Moderate confidence	Sources available but limited cross-verification. Some claims rely on single sources. Domain may be rapidly changing.
Below 0.60	Low confidence	Unit is published as provisional. Limited sources or significant uncertainty. Clearly flagged for agents and users.

Source Reliability Ratings

Each source cited in a knowledge unit carries a reliability rating:

Rating	Definition	Examples
authoritative	Primary authority on the subject. Official, legally binding, or canonical.	Government regulations (FERC orders), official standards (IEEE), manufacturer specs
high	Rigorous methodology, peer review, or established editorial standards.	LBNL research, RTINGS.com lab measurements, academic papers, Wirecutter
moderate_high	Professional, generally reliable, but with potential biases or limited methodology transparency.	Industry reports (ACP, IEA), established review sites (Tom's Guide, SoundGuys)
moderate	Useful data but methodology unclear or potential commercial bias.	Manufacturer white papers, sponsored reviews, industry press releases
low	Used only for context or when no better source exists. Clearly flagged.	User forums, unverified community data, social media

Freshness & Update Cycle

Each knowledge unit declares a freshness field indicating how often it's re-verified against primary sources:

monthly — rapidly changing domains (pricing, availability, market data)
quarterly — domains with regular update cycles (product launches, regulatory filings, industry reports)
yearly — stable domains (established regulations, fundamental frameworks, historical data)

The last_verified date shows when the unit was last checked. If a unit's last_verified date exceeds its freshness cycle, it is flagged as potentially stale. We do not hide stale units — we mark them transparently.

Verification Process

Source identification — identify all primary sources for the topic. Prefer government/institutional sources, peer-reviewed research, and established testing organizations.
Claim extraction — extract specific, verifiable claims from each source. Every number, date, and factual statement must trace to a source.
Cross-verification — check claims across multiple independent sources. Note disagreements in caveats.
Completeness check — ensure the unit answers the canonical question fully without requiring external sources. Check against common follow-up questions.
Confidence scoring — assign confidence based on the criteria above.
Peer review — domain expert reviews the unit before publication.
Ongoing monitoring — track primary sources for updates. Re-verify on the declared freshness schedule.

What We Don't Do

We don't inflate confidence scores. A 0.75 means exactly what the table says.
We don't hide uncertainty. Caveats and limitations are explicit in every unit.
We don't publish units on topics where reliable sources don't exist. We'd rather have no unit than a bad one.
We don't accept payment to influence content. Knowledge units are editorially independent.

Version History

Every knowledge unit maintains a version number. Changes are tracked:

Major version (1.0 → 2.0): Significant structural changes, new data sources, major corrections
Minor version (1.0 → 1.1): Price updates, freshness re-verification, minor corrections