Feed live·rebuilt ·every 6 hours·28/29 active sources OK·1 needs review: sblam·1 dormant: feodo_tracker (acknowledged)
xfeedspublic threat intelligence
Evidence surface

How the feed gets smaller before it gets useful.

Aggregate analysis of the pipeline, its sources, and the deliberately conservative choices that keep a blocklist from becoming an outage.

Run health

Every refresh leaves a trace.

Feed size and change per run40 recorded runs
Focus a run for exact counts.

sblam (failed); 1 dormant source acknowledged (visible in the source table). A stalled source degrades a run rather than stopping publication; its status stays visible in the source table. A source whose evidence has been frozen past the expiry ceiling is marked expired and contributes nothing at all until a maintainer reviews it.

Refresh history is recorded per publication run. The chart shows the primary run; the non-commercial tier is rebuilt on the same cadence but counted separately.
Publication method

The funnel is the product.

The feed publishes corroborated evidence, not every address a source observed. Each subtraction is a safety or licensing rule applied before an address can become a firewall action.

229,509
observed
Observed across public sources
165,221
licence-restricted evidence only
Licence permits scoring, not redistribution
6,394
allowlisted
Known cloud, CDN, crawler, and resolver space
978
tagged only (Tor exits)
Visible for context; never automatically blocked
44,480
withheld as uncorroborated
Did not reach the publication confidence threshold
12,436
published
8,828 high + 3,608 medium
Corroboration across independent classes12,436 published records

Each published record is counted by the independent classes that backed it, not by the number of source files that might repeat the same evidence.

1,8361 independent class
8,7132 independent classes
1,5933 independent classes
2594 independent classes
295 independent classes
46 independent classes
27 independent classes

1,836 records (15% of published records; 21% of the high tier) were admitted on a single high-precision source. Why some records are promoted without a second classSpamhaus DROP hijacked netblocks and active abuse.ch command-and-control servers are high-precision findings that do not need a second opinion. Everything else requires agreement across independent classes.

3,096 indicators had benign-scanner evidence and were capped from high to medium rather than deleted, preserving the consumer's policy choice. GreyNoise is reported only in this aggregate; it is never named against an individual record.

Before publication controls, 186,208 observed candidates had evidence from exactly one independent class.

The non-commercial tier has its own licensed publication counts, but this aggregate evidence funnel is intentionally shown only for the primary corpus so restricted-source observations are never relabelled.
IPv4 distribution

Where observations land in IPv4 space.

One cell per IPv4 /8, log-scaled208 /8 blocks observed
0
16
32
48
64
80
96
112
128
144
160
176
192
208
224
240241242243244245246247248249250251252253254255
0.0.0.0/8 at upper left → 255.0.0.0/8 at lower rightobserved multicast / reserved

Two /9 buckets are folded into each /8. Counts are logarithmically shaded so sparse activity remains visible without revealing individual addresses.

The /8 grid divides the 32-bit IPv4 space, so it has no IPv6 equivalent. IPv6 is examined by prefix rather than by address block in the IPv6 coverage section.
The /8 observation grid is a primary-corpus aggregate. Tier selection never rewrites or invents a download path.
Network patterns

Networks that recur across sources.

9,982 distinct ASNs were seen. Small-ASN results remain aggregated to avoid turning this section into an address disclosure channel.

Persistence, not provider size91 days recorded

Ranked by days seen, then address-days. Per millionAddress-days per million announced addresses normalise for network size, so this is not merely a list of the largest hosting providers. uses announced address space; a missing denominator is shown as an em dash.

ASNNetworkDays seenAddress-daysPer millionAnnounced
AS4837CHINA169-Backbone - CHINA UNICOM China169 Backbone30206,5253,609.257,221,376
AS396982GOOGLE-CLOUD-PLATFORM - Google LLC30142,2136,458.622,019,072
AS14061DIGITALOCEAN-ASN - DigitalOcean, LLC30141,63944,952.73,150,845
AS4134CHINANET-BACKBONE - No.31,Jin-rong Street30125,3911,266.199,036,416
AS9541CYBERNET-AP - Cyber Internet Services (Pvt) Ltd.30102,376572,111.9178,944
AS25369BANDWIDTH-AS Hydra Communications Ltd3099,486791,481.0125,696
AS45102ALIBABA-CN-NET - Alibaba (US) Technology Co., Ltd.3082,52914,811.85,571,840
AS212238CDNEXT Datacamp Limited3082,03876,922.91,066,496
AS8075MICROSOFT-CORP-MSN-AS-BLOCK - Microsoft Corporation3077,190955.980,749,056
AS203020HostRoyale HostRoyale Technologies Pvt Ltd3063,92364,655.4988,672

Only the sources above publish dated history, so days before this project began running are covered by those feeds alone and are thinner than recent days. Coverage evens out as our own run history accumulates. Dated upstream history: bruteforceblocker, ipthreat_30d. Its earlier left edge is thinner than recent project-run history, so it is not a trend line.

Current-run volume (context only)

Raw address count is retained as context, but it favours the largest providers; use persistence above for the comparable ranking.

ASNNetworkAddressesSources
AS55933CLOUDIE-AS-AP - Cloudie Limited32,60615
AS203346AS-232WEB Proper Support LLP27,6512
AS4134CHINANET-BACKBONE - No.31,Jin-rong Street27,14724
AS152194CTGSERVERLIMITED-AS-AP - CTG Server Limited20,8009
AS7018ATT-INTERNET4 - AT&T Enterprises, LLC20,38623
AS25369BANDWIDTH-AS Hydra Communications Ltd20,23918
AS4686BEKKOAME - BEKKOAME INTERNET INC.17,4081
AS4837CHINA169-Backbone - CHINA UNICOM China169 Backbone15,61823
AS396982GOOGLE-CLOUD-PLATFORM - Google LLC15,29923
AS14061DIGITALOCEAN-ASN - DigitalOcean, LLC13,89224
AS18068ACROSS - Dream Wave Shizuoka Co. Ltd.13,3121
AS3257GTT-BACKBONE GTT Communications Inc.10,71516
AS8075MICROSOFT-CORP-MSN-AS-BLOCK - Microsoft Corporation10,21822
AS51396PFCLOUD Pfcloud UG (haftungsbeschrankt)8,96420
AS701UUNET - Verizon Business7,72518
AS45839SHINJIRU-MY-AS-AP - Shinjiru Technology Sdn Bhd7,69411
AS4538ERX-CERNET-BKB - China Education and Research Network Center7,21210
AS398324CENSYS-ARIN-01 - Censys, Inc.7,16814
AS137409GSLNETWORKS-AS-AP - GSL Networks Pty LTD6,78715
AS210558services-1337-gmbh 1337 Services GmbH6,75216
AS263821AS263821 - Soluciones Favorables6,6905
AS213412ONYPHE ONYPHE SAS6,52011
AS9541CYBERNET-AP - Cyber Internet Services (Pvt) Ltd.6,40514
AS2516KDDI - KDDI CORPORATION6,19617
AS9009M247 M247 Europe SRL6,19523
Network enrichment maps IPv4 announcements to autonomous systems. The IPv6 records here are too few to attribute to networks without narrowing the aggregate further than disclosure allows.
Network enrichment is held to the primary aggregate in this surface; non-commercial selection leaves only paths that exist in the feeds.
IPv6 coverage

Prefixes need different judgement than addresses.

Wide on purpose. General IPv6 practice treats a /64 as one actor and a /32 as an entire ISP that should almost never be blocked outright. These wider entries are Spamhaus DROP netblocks leased or stolen outright by criminal operations, published for firewall and backbone use, where the whole allocation is the finding. The widest entry in this run is /29; review the widest entries before deploying them.
Prefix widthEntries
/2918
/3227
/366
/4833
Which sources report IPv68 sources

Observations, not published records. A source may report IPv6 and still contribute nothing to the feed if its records are never corroborated by an independent redistributable source.

SourceIndependence classIPv6 observationsPublication
stopforumspam_listedstopforumspam4,298scoring only
dataplane_sshclientdataplane482scoring only
blocklist_deblocklist_de332republished
turris_greylistturris96scoring only
spamhaus_drop_v6spamhaus90republished
abuseipdb_blacklistabuseipdb42scoring only
reportedipreportedip37republished
dataplane_telnetlogindataplane4scoring only
Only IPv6 records are analysed here. The IPv4 view of the same corpus is the /8 observation grid and the network patterns section.
IPv6 structural analysis is derived from the primary corpus and is not relabelled for the non-commercial tier.
Disclosure control

The corpus is visible only in aggregate.

What entered scoring

No individual address appears in this section. 229,509 addresses were observed from 28 sources before publication controls. There is no top-offending-addresses list.

Why restricted sources appear here

A count is a derived fact, not an extract. A scoring-only source may be named against a count, but never against an address; GreyNoise and other restricted-source identities are not published per record.

Tier view

The non-commercial feed is licensed to retain additional share-alike material. Its filenames remain stable under noncommercial/.

Source health

Independence beats source count.

Independence classSources that share upstream data contribute at most one vote, preventing copied lists from manufacturing false confidence. and scoring onlyThese sources can influence a confidence score, but their licences prohibit republishing their addresses. are explained beside the table, never inside its scrolling container.

SourceIndependence classRecordsStatusPublication
abuseipdb_blacklistabuseipdb10,000okscoring only
binary_defensebinary_defense320okyes
blocklist_deblocklist_de25,408okyes
blocklist_de_strongipsblocklist_de286okyes
bruteforceblockerbruteforceblocker602okyes
carpathiancarpathian651okyes
cins_armycins15,000okyes
dataplane_proto41dataplane51,667okscoring only
dataplane_smtpgreetdataplane9,683okscoring only
dataplane_sshclientdataplane25,920okscoring only
dataplane_sshpwauthdataplane13,905okscoring only
dataplane_telnetlogindataplane40,845okscoring only
dataplane_vncrfbdataplane4,592okscoring only
dshield_attacksdshield9,994okscoring only
dshield_blockdshield20okscoring only
et_compromisedbruteforceblocker590okyes
feodo_trackerabusech0expirednot contributing
greensnowgreensnow1,349okscoring only
ipsum_levelsMETA_aggregate32,292okyes
ipthreatipthreat3,558okyes
isc_research_scannersisc_labels812okscoring only
jacobrakai_honeypotjacobrakai441okyes
reportedipreportedip10,636okyes
sblamsblam0failedscoring only
spamhaus_drop_v4spamhaus1,747okyes
spamhaus_drop_v6spamhaus90okyes
stopforumspam_listedstopforumspam53,866okscoring only
threatfoxabusech1,984okscoring only
tor_exitstor1,296okyes
turris_greylistturris9,413okscoring only
Highest class overlap8 highest pairs

Jaccard similarity measures shared observations over the combined observations of a pair. These low overlaps support treating the classes as independent, while the scorer still counts at most one vote per class.

Class AClass BJaccardShared observations
abuseipdbblocklist_de28.1%12,022
abuseipdbturris24.9%10,415
cinsdshield24.4%7,142
abuseipdbdshield23.0%7,693
abuseipdbcins19.9%6,990
dshieldreportedip19.0%4,768
cinsturris16.8%6,818
dataplaneturris16.4%21,624
What each class contributes17 classes

Overlap between independent sensors is the agreement the score is built on. These columns show how much of the published feed each class stands behind, how many records would be withheld or lose high confidence without it, and where most of its observations are also seen. Voting-only classes can raise confidence but never admit a record on their own.

ClassRoleObservedOnly this classSupports publishedWithheld without itLose high without itMost contained in
dataplanevoting only124,84493,3768,4900120turris 17%
abuseipdbvoting only23,4121,5267,6370448dataplane 61%
blocklist_deadmitting31,44716,2187,0675,5194,883abuseipdb 38%
turrisvoting only28,7904,3666,132027dataplane 75%
reportedipadmitting12,0482,3676,0374,4101,858dataplane 59%
dshieldvoting only17,7594,8474,543023dataplane 48%
cinsadmitting18,6935,1003,9533,2171,170dataplane 45%
ipthreatadmitting5,7031,4623,8042,9702,865abuseipdb 68%
greensnowvoting only4,3995002,379012abuseipdb 77%
spamhausadmitting1,8361,8341,8361,8361,836dshield 0%
binary_defenseadmitting1,927451,623828334dataplane 82%
bruteforceblockeradmitting673109400241237dataplane 74%
jacobrakaiadmitting47850367151132abuseipdb 71%
carpathianadmitting6733521739067abuseipdb 33%
stopforumspamvoting only53,90852,43617301blocklist_de 2%
sblamvoting only28504600stopforumspam 96%
abusechvoting only1,6541,6201300abuseipdb 1%
The non-commercial tier uses its own manifest and preserves the same stable feed filenames. Source-level aggregate health is presented for the primary publication run.
Terms and provenance

Licensing is a publication control, not a footnote.

Primary tier

12,436 addresses suitable for commercial and non-commercial use. Restricted sources may contribute only to scoring and are never republished.

Non-commercial tier

19,804 addresses under CC BY-NC-SA 4.0. It can retain share-alike material that the primary tier must withhold.