Ingestion lag
Most brands take 11+ days to appear in Claude after a public launch.
Sample data
Latency Ledger
Mapping the AI Parametric Blindspot
Frontier models don't index the web. They ingest it—unevenly, late, and with structured holes. We measure those holes as a surface, then tell you where your brand falls through.
The ledger is mock data for demonstration — not live ingestion times. Disclaimer
Days to ingestion
Mock data — not liveScroll to page the ledger
Sample data for demonstration — not live measurements. Disclaimer
Model split
ChatGPT often ingests pop-culture events 2–4 days before Gemini.
Sample data
Blindspot
Implicit prompts miss the brand even when explicit ones already match.
Sample data
Parametric hole
A company can rank on Google and still be absent from logprob surfaces.
Sample data
B2B delay
Earnings-call facts take roughly twice as long to stabilize as celebrity news.
Sample data
Ingestion lag
Most brands take 11+ days to appear in Claude after a public launch.
Sample data
Model split
ChatGPT often ingests pop-culture events 2–4 days before Gemini.
Sample data
Blindspot
Implicit prompts miss the brand even when explicit ones already match.
Sample data
Parametric hole
A company can rank on Google and still be absent from logprob surfaces.
Sample data
B2B delay
Earnings-call facts take roughly twice as long to stabilize as celebrity news.
Sample data
Contradiction
Models often hold two competing brand facts until a later crawl overwrites one.
Sample data
Share of model
The first named competitor in a category tends to occupy 40%+ of answers.
Sample data
Web-search toggle
Search-on quadrants still fail when the parametric prior is empty.
Sample data
Canon lag
Rumor states persist for days after a correction is published.
Sample data
Contradiction
Models often hold two competing brand facts until a later crawl overwrites one.
Sample data
Share of model
The first named competitor in a category tends to occupy 40%+ of answers.
Sample data
Web-search toggle
Search-on quadrants still fail when the parametric prior is empty.
Sample data
Canon lag
Rumor states persist for days after a correction is published.
Sample data
Method note · sample figures
Surrogate Probability Surface
We treat token logprobs as a topography. By probing the same entity across paraphrases, we reconstruct a surrogate of how confidently a model "knows" you—and where that confidence collapses into a parametric blindspot.
Method note · sample figures
Stochastic State War-Room
Brand mention is not a boolean. We model ingestion as a Markov process: states of absence, rumor, contradiction, and canon. The war-room tracks transition probabilities as new events hit the training and retrieval stack.