Top filterers
Transit ASes
ranked by filter_score × log(paths_through), so the
list elevates ASes with BOTH substantial path coverage AND a high
fraction of hops where our forged PTB went unhonoured. Filter score is
not_honored / (honored + not_honored) across all hops we
tested in that AS's ranges. Minimum 10 distinct paths and 5% filter score to
appear here.
*shadowed: not-honored verdicts observed downstream of an earlier break on the same path. Our PTB may never have reached those hops, so they are counted here for transparency but excluded from the filter score and from ranking — they are unattributable, not exonerating. "Seen in runs" counts how many of the last daily runs also ranked this AS: persistent presence is policy, a single appearance may be a fluke.
| AS | Holder | paths | honored | not_honored | shadowed* | filter % | seen in runs | countries (paths) |
|---|---|---|---|---|---|---|---|---|
| AS6939 | HURRICANE - Hurricane Electric LLC, US | 486 | 27 | 303 | 616 | 91% (n=330) | 7/7 | ita(303), aut(137), slo(29), srb(17) |
| AS1764 | NEXTLAYER-AS - Next Layer Telekommunikationsdienstleistungs- | 32 | 1 | 15 | 19 | 93% (n=16) | 7/7 | aut(32) |
| AS9002 | RETN-AS - RETN Limited, GB | 159 | 69 | 99 | 45 | 58% (n=168) | 7/7 | ita(107), aut(42), slo(8), srb(2) |
| AS174 | COGENT-174 - Cogent Communications, LLC, US | 67 | 22 | 57 | 220 | 72% (n=79) | 7/7 | ita(41), slo(20), aut(6) |
| AS41327 | FIBERTELECOM-AS - Fiber Telecom S.p.A., IT | 77 | 40 | 56 | 5 | 58% (n=96) | 7/7 | ita(61), aut(11), slo(5) |
| AS3356 | LEVEL3 - Level 3 Parent, LLC, US | 37 | 18 | 20 | 6 | 52% (n=38) | 7/7 | ita(19), slo(12), aut(6) |
| AS31543 | MYNET-AS - myNet GmbH, AT | 14 | 23 | 7 | 12 | 23% (n=30) | 4/7 | aut(14) |
| AS5405 | INTERDOTLINK - Inter.link GmbH, DE | 30 | 16 | 3 | 0 | 15% (n=19) | 6/7 | slo(13), aut(12), ita(5) |
| AS47147 | AS-ANX - Anexia Cloud Solutions GmbH, AT | 29 | 38 | 4 | 5 | 9% (n=42) | 6/7 | aut(29) |
| AS8412 | TMA - T-Mobile Austria GmbH, AT | 29 | 15 | 2 | 3 | 11% (n=17) | 7/7 | aut(29) |
This tests host-mode PTB acceptance (PTBs aimed at the hop itself). A router that honours PTBs to itself can still be filtering PTBs transiting through it, and conversely a router may ignore ICMPv6 addressed to its own control plane (a common and legitimate CoPP policy) while forwarding transit PTBs perfectly. A high filter score is therefore consistent with a PTB-filtering problem, but is not on its own proof of one; a low score is not proof of transparency either. Treat this as a list of networks worth investigating, not as a finding about any of them. Many hops drop ICMPv6 Echo to themselves or rate-limit it via CoPP; those are recorded as
no_response
and excluded from the score denominator. Anycast destinations are
skipped explicitly.
Local-vantage only -- every walk starts from the orchestrator host. This is the biggest blind spot here, and it is structural, not cosmetic: an AS can only be scored if the orchestrator's own path traverses it AND the destination still answers Echo, because the forged PTB is piggybacked on a working Echo exchange.
So the networks where traffic stops altogether are exactly the ones this table cannot see. Worked example: paths entering AS3212 (Telemach) toward AS28682 and AS60435 get no Echo reply at all from five of our eight vantages, while the three reaching those destinations by another route see them fully open -- a strong, reproducible, path-dependent signal. AS3212 scores nothing here, because the orchestrator reaches both destinations over a different path and never walks the failing one. Findings of that shape appear on the per-AS pages instead, under “reachability depends on the path”.
Per-vantage walks over SSH would close this gap and are the single most useful improvement to this page.
Source data: data/global/ptb_atlas.json ·
rolling history at data/global/ptb_atlas_history/
(last 30 days).
Built daily by scripts/28_global_ptb_atlas.py.