Target researchers describe a monitoring loop for semantic search failures
The MOSS preprint links model-based relevance alerts to human investigation. Its search and recommendation evaluations have different scopes and limitations.
-
-
A new QA study tests why supported answers can still miss the question
Joint Constraint Aggregation compares answers against the same requirements. The preprint tests candidate selection, with limits on retrieval and model generalization.
-
Claude adds customer attribution and enforcement events for API platforms
User profiles identify the customer behind a Claude API request. Beta webhooks report enforcement changes, with access and reconciliation limits.
-
Cloudflare starts checking whether verified bots respect site preferences
Cloudflare is assessing robots.txt adherence and responses to rate limits. Internal use has begun; customer-selectable trust controls are planned.
-
Claude adds OpenTelemetry export for tracing research failures
The beta exposes request and tool traces. Inspect content mode, truncation and original sources before assigning the cause of an answer failure.
-
A Gemini study tests whether brand-mention measurements hold for 20 days
The preprint reports useful mention-history forecasts in one setup. It does not establish lasting web citations, traffic or AI-search rankings.
-
Google updates Indexing API quota guidance: eligibility comes first
The October documentation refresh clarifies the current allowance. Check supported page types before counting notifications or claiming indexing success.
-
Exa 3.0 adds Lite search: check source support before switching
Exa’s new search option changes price and retrieval choices. Compare source support, excerpt completeness and total workflow cost.
-
Humanity’s Last Choice proposes a GEO evaluation standard, without new experiments
The paper separates source visibility, task quality and valid edits. Its first version proposes a method rather than reporting optimization results.
