Primary Source Multiplier

The multiplier is triggered by verifiable first-party data generation. This includes internal telemetry, controlled A/B tests, proprietary sensor logs, and original survey results. The system validates provenance by checking for unique data signatures—statistical distributions, timestamps, and collection methodologies that cannot be replicated by scraping secondary sources. Content merely citing public datasets receives a baseline score, while content publishing the raw, previously unseen dataset receives the full multiplier weight.

The multiplier's magnitude scales with the granularity of the documented methodology. To qualify for maximum amplification, the source must detail:

• Sample size and selection criteria
• Statistical significance thresholds (p-values, confidence intervals)
• Instrumentation and collection protocols
• Edge cases and exclusion criteria This allows AI models to assess internal validity. A proprietary benchmark without methodology is treated as an anecdote; a benchmark with full reproducibility documentation is treated as a primary source.

The multiplier applies a time-decay function to reward the first publication of a novel data point. The originating source receives the full multiplier for a defined exclusivity window. As secondary sources cite and re-publish the data, the multiplier for those derivative works approaches zero. This mechanism explicitly rewards post-training knowledge—facts that did not exist before the model's cutoff date—and penalizes the aggregation echo chamber.

The multiplier distinguishes between observational data and interpretive analysis. Raw, unmanipulated datasets (e.g., a CSV of transaction logs) receive a higher multiplier than opinion pieces about that data. The system checks for executable verification paths:

• Can the result be reproduced with provided code?
• Is the data available for direct query?
• Are negative results and null findings disclosed? A published Jupyter notebook replicating the analysis acts as a strong multiplier catalyst, transforming a claim into a verifiable artifact.

The multiplier receives a non-consensus bonus when primary source data contradicts the prevailing statistical consensus in the training corpus. If a model's training data suggests a 2% industry churn rate, and a company publishes a verified primary source study showing a 7% churn rate, the Contrarian Viewpoint Index boosts the multiplier. This prevents AI models from drowning out novel empirical findings with the weight of outdated or generalized aggregate statistics.

A true primary source is defined by its position as a root node in the citation graph. The multiplier algorithm traverses the citation chain backward. If a document is cited by 100 others but cites no prior source for a specific claim, it is identified as the origin. This root-node status triggers the full multiplier. Documents that are hubs (citing many sources) but not roots receive a lower multiplier, as they function as aggregators rather than originators.