Source Recency Weight

A mathematical formula that systematically reduces a source's authority score as time passes from its publication date. Exponential decay is the most common implementation, where relevance halves at a fixed interval. Linear decay and logarithmic decay offer alternative curves for different knowledge domains. The decay rate is domain-specific: medical literature may decay rapidly (6-12 month half-life), while mathematical proofs remain stable for decades.

Different fields require distinct recency thresholds. Fast-moving domains like AI research, cybersecurity, and clinical medicine demand sources published within months. Slow-moving domains like philosophy, mathematics, and historical analysis tolerate older references. A well-tuned system applies variable decay curves based on the topic classification of the query, not a one-size-fits-all window.

Recency weighting must distinguish between original publication date and last substantive update date. A 2015 paper with a 2024 revision containing new data should receive a higher weight than an unrevised 2015 paper. Systems should parse schema.org dateModified and datePublished metadata to apply the correct timestamp. Minor typo fixes should not reset the decay clock—only substantive content changes qualify.

Not all valuable sources decay. Evergreen content—foundational research, seminal papers, and canonical references—should be exempt from aggressive recency penalties. A system can identify evergreen sources by analyzing persistent citation velocity: if a 10-year-old paper continues to receive citations at a steady rate, its recency penalty should be dampened or removed entirely. This prevents the algorithm from favoring shallow, recent content over deep, enduring expertise.

Source Recency Weight exists in tension with Source Credibility Score. A highly authoritative source (e.g., a peer-reviewed journal) may retain value longer than a low-authority source (e.g., a blog post). The system must calibrate the recency-authority balance per query type. For breaking news, recency dominates. For legal precedent, authority dominates. This is often implemented as a weighted ensemble where the recency multiplier and authority score are combined with domain-specific coefficients.

Recency weighting systems must flag temporal anomalies—sources that cite outdated information as current or present stale data without disclosure. This includes detecting when a source references a superseded study, uses deprecated APIs, or reports statistics from a prior decade as if they are current. Temporal contradiction detection compares the source's timestamp against the timestamps of its own references to identify anachronistic evidence chains.

The mathematical formula governing how a source's authority diminishes over time. Common implementations include:

• Exponential decay: authority = base_score × e^(-λt)
• Half-life models: score halves every N months
• Step functions: discrete drops at threshold ages

The decay rate (λ) is domain-specific—medical research may use a 2-year half-life, while historical citations may use decades.

The process of identifying when a cited source's content has been updated or altered post-citation, potentially invalidating the original evidence. This directly interacts with recency weighting: a source that was recent at citation time may have since been superseded by a correction or retraction. Drift detection systems monitor version histories and content hashes to trigger re-evaluation of the citation's validity.

An algorithmic weighting rule giving higher authority to direct, first-hand accounts or original research over secondary interpretations. When combined with recency weight, a recent primary source (e.g., a newly published clinical trial) receives the highest possible composite score, while an old tertiary source (e.g., a decade-old textbook summary) is heavily penalized by both signals.

A cryptographic fingerprint of a source document's content at the exact moment of citation. This immutable hash enables verification that the referenced material has not been altered, even if the live source is updated. Provenance hashing complements recency weighting by preserving the evidentiary value of a source as it existed at citation time, preventing temporal paradoxes where updates retroactively invalidate valid citations.

A hierarchical categorization system ranking sources by editorial rigor and authority type:

• Tier 1: Peer-reviewed journals, official government datasets
• Tier 2: Established industry publications, institutional reports
• Tier 3: News media, corporate blogs
• Tier 4: Social media, self-published content

Recency weight interacts with tier classification—a recent Tier 4 source may still be outranked by an older Tier 1 source, depending on the application's freshness requirements.