MAKE YOUR SERVICE THE ANSWER.Discuss your service 
SAG TECHNOLOGY FRONTIER
We design how search, answers and generative AI understand brands, and explain the architecture behind SAG in clear technical terms.
SAG / ARCHITECTURE NOTES
From discovery and evidence to data boundaries, reliable execution and delivery—see how each engineering choice works.
80 articles · 3 / 8
The principle for calculating an aggregate metric while preserving the numerator and denominator of each ratio. A simple average of ratios from engines with different answer counts gives the same weight to small and large samples. This can misrepresent the actual set of answers.
A data model that stores content from the same address as records tied to each collection time. If specifications or policies change while the address stays the same, the evidence behind an earlier report may differ from what is currently displayed. Storing only the latest content makes it difficult to reproduce past decisions.
Evaluation data for comparing the effects of analysis changes by fixing reviewed questions, evidence, and expected judgments. If you change a model or extractor and check only for polished answers, past errors may return. Search, generation, and metric errors must be measured separately.
Treat web materials as analytical inputs, separate from task instructions and tool permissions. Competitor pages and uploads may contain text intended to change a model’s behavior. The richer the supporting evidence becomes, the more important the authority boundary around external text is.
Learn how to design tenant boundaries, server-side authorization, and RLS together so that multiple organizations can use one service without mixing their data or permissions.
A pattern for distinguishing unchanged and changed material using content hashes and structural comparisons. The more frequently pages are collected, the more collection runs accumulate. Sending inputs with no actual changes to an expensive model every time increases costs and can weaken consistency in the results.
A comparison method that fixes the first competitor data entered and tracks differences against the company’s monthly data. If the comparison target and questions change every month, it is difficult to distinguish improvements in the company’s performance from changes to the baseline. A consistent baseline helps keep work on track.
This process extracts titles, body text, links, and metadata signals from different HTML documents into a common structure. On sites with long menus and footers, repeated sentences may be captured more often than product descriptions. If structure is not distinguished, sentence counts can be misread as indicating that there is sufficient evidence about the product.
A presentation structure that connects monthly metrics to annual trends and the evidence for individual months. Even if the long-term trend looks positive, it is difficult to decide what action to take without knowing what improvements and observations occurred in each month. The overall trend needs to be connected to detailed records.
Manage the overall rate by answer separately from citation shares by URL. When one answer cites multiple official pages, the URL rates add up to more than the answer rate. Adding them together as overall performance inflates the figures.