Stats

Agent access, traffic and the evidence behind each request.

Agent access

Which work AI fetchers, search clients, crawlers and historical signed clients access, in which served format. Names and roles describe stored request evidence.

Window
28 completed days
Calendar
UTC reporting days
Controls in this report
Article, client and role
Select agent access
Declared role

Counts beside choices cover the other two selections. Article choices include registered evidence.

Loading the completed-day agent access report…

Agent access JSON · Current collection covers eligible served pages. Citation and answer-use observations are separate outcomes.

Page requests

Recorded page requests across the selected traffic categories. The filters in this report apply to its overview, rankings and evidence.

Window
Selected period
Calendar
Your timezone
Controls in this report
Traffic, period and page
Traffic

Categories describe request signals, not verified people. Agents includes on-demand fetchers, AI search, AI crawlers and historical or unknown-role signed clients. Other crawlers includes search, SEO and preview declarations. Automation includes legacy rules.

Period

Loading stats…

Page viewsRecorded page requests
Client identifiersRotating identifiers, not people

Page views over time

Local time · recorded page requests
View chart as a table

Pages & sources

Where recorded requests went, and the referrers they reported.

Top pages

Reported referrers

Frequent page transitions

Request signals

Traffic categories, software claims and network locations. None identifies a person or physical device.

Traffic mix

Five groups, one total

These five groups add up to the selected total. They describe request signals and declared roles, including agent access.

Device-type claims

Record coverage · some facts overlap

Claims, not hardware

Request headers describe software claims and mobile-experience preferences, not authenticated hardware. Non-mobile does not establish desktop; Android without Mobile does not establish tablet. Missing and conflicting evidence remain explicit.

Recognized records can span parser profiles. Earlier desktop includes default guesses. Differing indicators can reflect preferences; unsupported hints and earlier heuristics stay separate.

Top network countries

Named crawlers, tools and services

Name coverage · adds up to the selected total

Names and missing evidence

Names come from this site’s catalogue of crawlers, tools and services. Browser-family signals are reported separately and do not identify the operator.

A missing selected name can mean no catalogue match, several competing names, or earlier missing data. It does not identify a person or tell us whether a signer is known. The two coverage counts add up to the selected total.

Evidence behind the categories

Stored classifications, signatures and network checks, with missing and historical evidence kept explicit.

Request classifications

Signature evidence by classification

Scroll to compare all columns

What a signature establishes

Each included observation appears once. Classifications describe declared roles or request rules. A signature covering User-Agent protects the declaration; it does not establish the operator, human involvement or purpose. Select a kind to inspect its vendor and network checks.

The stored verifier profile shows which implementation produced each result. Current means its identifier matches today's verifier; it does not mean this report reverified the request. Historical profiles can have different checks. Missing profiles, inconsistent records and injected verdicts remain explicit. Each included event appears once across this table.

Names, categories & signatures

Scroll to compare all columns

Reading the combined evidence

A crawler category can be available while its software name is missing. An unsigned request can match a vendor's published network ranges elsewhere in this report. A selected name is a claim from the request, and a signature authenticates only its covered components. Neither establishes human reading or the initiating party.

Stored verified signing outcomes

Stored signing history

These counts follow the top-level stored verified flag, including earlier records without proof detail or with inconsistent detailed records. The classification table separates those cases and stored verifier profiles. Reports do not reverify historical requests. A reviewed display name does not establish a request’s role or purpose.

Unsuccessful verification

Unsuccessful does not mean impersonation

Consistent stored unsuccessful outcomes distinguish invalid input, unsupported features, unavailable keys and policy rejection. Inconsistent local records appear in the verification panel. Mixed candidate outcomes remain indeterminate. None establishes impersonation or intent.

Published network checks

Network attribution limits

Stored checks compare an edge-supplied address with documented vendor ranges. Membership supports network attribution; a nonmatch does not prove impersonation. Outcome counts require a structurally recognized record.

This report does not cross-check declared names against feed assignments; the offline analyzer does. Declaration counts come from the stored network record. Missing addresses, Worker relay addresses, unavailable feeds, failed revalidation and budget skips stay explicit. Historical outcomes are retained, without fresh address verification.

Several checked feeds can produce overlapping outcomes. Earlier records remain unknown. Configured methods and sources.

Why vendor checks were skipped or reused

Cache diagnostic coverage

Cache record coverage

Each included event appears once as recognized, unrecorded or inconsistent. Recognized records separate no applicable method, guards only and recorded cache access. Old records remain unknown. Known and unknown policy contexts can overlap when several sources apply.

Recorded access paths

Overlapping access paths

Counts are events with at least one matching source check. Paths can overlap and are not upstream-request totals. A cooldown stays unassessed; a shared fetch may fail. These facts do not identify a vendor, operator or reader.

Last accepted cache policy

Policy context, not current freshness

A cooldown can carry the last accepted policy context without a membership snapshot. Positive and blocked intervals refer to that validation, not freshness now. Sources and blockers overlap; do not add these counts. Failure without accepted context remains unknown. HTTP cache directives.

Declared vendor checks

Declared signing methods

Stored method assessments use their recorded declaration map. The current map covers ExaSearchBot, ExaUser, Google-Agent, ShapBot and Shap-User using documented signing origins; the two Parallel declarations name distinct origins. Inconsistent method records remain separate.

Google’s rollout is experimental; unsigned does not mean impersonation. Conflicting declared origins stay visible even when the selected proof matches one. A matching key does not bind an unsigned User-Agent, identify the initiating user or prevent replay. Counts overlap; declarations outside the recorded map remain unmapped. IP checks are separate; no DNS verification is performed here.

Signature verification

Coverage and unknowns

Consistent stored local results and Cloudflare assessments are separate. Missing and inconsistent records remain explicit. Coverage facts require typed detail consistent with the verifier’s output invariants; their counts overlap.

Native availability means at least one stored boolean is usable; each field has its own available, true and false counts. A missing signed-agent value is not false. Native booleans are assessed in a recognized envelope independently of the local result; an unrecognized envelope contributes neither field. Known-bot recognition is broader than signed-agent recognition, so local verification with native verified-bot false is not cryptographic disagreement. None establishes a human trigger.

Selection counts describe detailed locally verified events, not signer totals. Outcome-only signing history remains in the classification table above. Verification stops at the first valid paired candidate tagged Web Bot Auth; other candidates can remain unassessed. Historical absence and unrecognized assessments stay separate.

Published key directories

What the directory check means

The reviewed key-directory map includes first-party endpoints for Exa, Ahrefs, Google and Parallel. This assessment compares the locally verified key publisher with the declared client's reviewed endpoint; it does not establish a per-client signing promise or authenticate an unsigned User-Agent. Conflicting publishers, partial mappings and missing history stay separate. Historical assessments keep their recorded profile. Each included event appears once. Reviewed endpoints and sources.

Cloudflare categories by classification

Provider categories are separate evidence

Categories can be available without native verification booleans. They are Cloudflare’s classification, not our signature verdict or proof of a human trigger. Empty, unavailable and earlier unrecorded values stay distinct. Each included event appears once.

Dispatch

GK-001

A note when there is signal.

Engineering records, failed experiments, and the life around the work.

gkoreli.com

Find a thought.

Search the full text of published articles. Results link directly to source passages.