Owned agent arrival trend, counted without turning fetches into customers

Name: Owned Agent Arrival Trend
Creator: US Tech Automations

Across 107,594 dated request rows observed from 2026-07-10 through 2026-08-14, the reconciled scan counted 11,691 agent-signature fetches and excluded 33,422 known self-probe rows. Two independent reads agreed on the same pinned source prefix before this page admitted any total.

This is an instrument for seeing which owned evidence and offer surfaces received signature-bearing requests. It is not a people counter and it is not a sales dashboard. The tables publish classifier-family labels and generalized surface names only; raw User-Agent strings, IP material, and request paths stay in the private source log.

11,691classified fetch rows in measured coverage

36calendar dates in the observed log window

0malformed or undated rows withheld

21,760,205source bytes pinned for both scans

Source: US Tech Automations first-party permits access log, derived 2026-08-14T03:21:13.419989+00:00; complete UTC observations through 2026-08-13. Report age at render: 2.83 hours. Source-prefix SHA-256: 08e98115b0496c11c9ea76cd0105bc9ccaa1edc963d6e49c6cd2bb9e21b08b2b.

Which classified agent families did we observe?

The recent column covers 2026-08-07 through 2026-08-13 and includes complete UTC dates only. A directional label is withheld until one family has observations on at least seven complete dates and the log contains two complete seven-day windows. That prevents one burst or a partial current day from being presented as a trend.

Classifier family	Observed fetches	Recent complete dates	Active complete dates	Directional reading
Anthropic crawler signature	5,394	4,574	9	MORE OBSERVED FETCHES
Meta external-agent signature	1,821	717	25	FEWER OBSERVED FETCHES
Perplexity crawler signature	974	794	17	MORE OBSERVED FETCHES
OpenAI user-initiated fetch signature	901	866	11	MORE OBSERVED FETCHES
GoogleOther crawler signature	808	704	14	MORE OBSERVED FETCHES
Amazon crawler signature	654	66	13	FEWER OBSERVED FETCHES
ByteDance crawler signature	460	36	35	NO MATERIAL COUNT CHANGE
OpenAI training crawler signature	334	283	15	MORE OBSERVED FETCHES
OpenAI search crawler signature	324	244	33	MORE OBSERVED FETCHES
DuckAssist crawler signature	9	9	4	WITHHELD — fewer than seven active complete dates
Common Crawl signature	6	0	3	WITHHELD — fewer than seven active complete dates
You.com crawler signature	6	5	2	WITHHELD — fewer than seven active complete dates
Anthropic AI crawler signature	0	0	0	WITHHELD — fewer than seven active complete dates
Anthropic web fetch signature	0	0	0	WITHHELD — fewer than seven active complete dates
Cohere crawler signature	0	0	0	WITHHELD — fewer than seven active complete dates
Google AI-use control signature	0	0	0	WITHHELD — fewer than seven active complete dates
Meta Facebook crawler signature	0	0	0	WITHHELD — fewer than seven active complete dates
Mistral crawler signature	0	0	0	WITHHELD — fewer than seven active complete dates
Perplexity user-initiated fetch signature	0	0	0	WITHHELD — fewer than seven active complete dates
Timpi crawler signature	0	0	0	WITHHELD — fewer than seven active complete dates

A classifier family records a token presented by the requesting client. Tokens are self-asserted and can be copied. The label therefore says signature, not verified operator identity.

Which owned surface families received those fetches?

Surface totals use stable, coarse families so a changing offer slug or evidence leaf does not create a misleading new series. The public projection groups all non-public service routes together and never emits a concrete request path. The surface totals reconcile to the same classified-fetch total as the family table.

Generalized owned surface	Observed fetches	Recent complete dates
Rankings outbound handoff	2,903	2,068
Product request forms	2,170	1,469
Earthquake evidence family	1,176	1,154
Grid queue evidence family	1,144	589
Rankings family	1,140	713
Crawler policy evidence family	890	890
Other public evidence families	818	623
Offer pages	431	362
Permit cost evidence family	358	50
API and service routes	232	104
Permit lookup tool	87	13
Service root	82	19
SEC filing evidence family	69	69
Recall evidence family	55	54
Offer checkout handoff	42	33
Agent behavior evidence family	25	25
Public evidence index	15	9
Agent registry evidence family	14	14
Streamgage evidence family	11	11
TTB permit evidence family	9	9
Crawler configuration review	6	6
Pricing archive evidence family	5	5
Promise archive evidence family	5	5
Model evidence audit	4	4

0 bytes appeared after the pinned cutoff during the second scan. They are deliberately absent from this result and will be measured by the next scheduled report; an append during a run must not change half of the totals.

What this observation cannot tell you

Fetch ≠ identity ≠ referral ≠ demand. A token match does not prove which organization operated the client. A request does not prove that an answer engine cited us, that a person was referred, that a buyer read the page, that a lead was created, or that a sale occurred. Our own known probes are excluded, but an unknown automated probe may remain. Zero means no matching row inside the stated first-party coverage, not that no client fetched us elsewhere or before logging began.

Use this table to prioritize observable surface coverage and to spot when this owned observation series has gone quiet. Use the separate attribution and revenue ledgers for referrals, contact, accepted work, and payment. This page never upgrades one evidence type into another.

From observation to an existing, bounded action

The evidence page itself is not a new product and states no new price. Site owners who need to review what their own public crawler configuration expresses can use the existing AI Crawler Configuration Review. Readers who want comparison evidence first can inspect the existing crawler-policy archive. Neither destination promises that a crawler will obey a draft, fetch a page, cite it, index it, or train on it.

Review the existing service Inspect the crawler-policy evidence

Browse public evidence · US Tech Automations

Mechanical method and failure behavior

Open the private append-only log and pin its byte length from the first descriptor.
Read exactly that prefix through two independent file opens. A rewrite, short read, replacement, or digest disagreement makes the result UNKNOWN.
Parse the first scan into dated rows and independently recount the second scan. Valid rows, invalid rows, self probes, classified fetches, family aggregates, and surface aggregates must all reconcile.
Fill every classifier family on every observed date, including explicit zeroes, and exclude the still-open current UTC day from trend windows.
Write one atomic derived report. The public reader rejects a missing, malformed, stale, future-dated, vocabulary-skewed, or unreconciled report with HTTP 503/UNKNOWN instead of rendering a plausible zero.

No language model participates in collection, classification, aggregation, reconciliation, page rendering, or sitemap admission. The source is an owned request log; this page does not make claims about traffic outside that source.