Waterfall's Overt User Simulator caches previously loaded overt-website responses and replays them to generate cover traffic, overcoming Slitheen's 40% downstream throughput ceiling (caused by restricting covert replacement to leaf HTTP objects only). Because downstream-only decoy routers intercept all downstream TLS records — not just leaf content — Waterfall achieves higher covert capacity while perfectly mimicking overt browsing patterns against traffic analysis.

Through Internet-scale BGP simulation against China, downstream-only decoy routing (Waterfall) with a single decoy AS provides equivalent resistance to routing attacks as a traditional upstream decoy system (e.g., Telex) with 53 decoy ASes. This efficiency gain arises because censoring ISPs can selectively re-route upstream traffic per destination but must re-route all or none of downstream traffic through each provider AS, making downstream-only routing far more expensive to evade.

§4.1.1 defense

Evaluation of the top 10,000 Alexa websites finds that 3,916 (39%) support HTTPS, of which 1,976 (50%) perform HTTP 3XX redirects that echo the requested path in the Location header and 812 (20%) replay the URL in HTTP 404 error responses — both usable as upstream covert channels readable by downstream-only decoy routers without intercepting upstream traffic.

§7.1.1 defense

Table 2 shows that with 50 decoy ASes, the most powerful practical routing attack on downstream-only systems (rewiring-I) impacts 93% of China's routes (22.4% unreachable, 70% re-routed), compared to only 18.2% total impact from RAD on traditional upstream designs. Table 3 shows that even for Syria, the rewiring-II attack with just 1 downstream-only decoy AS already impacts 81% of routes versus 1.5% for RAD on upstream systems.

§4.2.1, Table 2, Table 3 evaluation

BGP simulation shows that a censor's source-block attack against 100 downstream-only decoy ASes disconnects 23% of Chinese Internet destinations, versus only 8% when applying the standard RAD attack against 100 upstream decoy ASes — imposing nearly 3× more unreachability collateral damage on the censor for the same decoy count.

§4.1.1, Figure 3 evaluation

With infinite training time, Laserbeak achieves 93.5%, 95.9%, and 95.9% accuracy against ephemeral padding, FRONT, and Interspace respectively, compared to 96.5% undefended — confirming that padding-only defenses provide no meaningful protection against a sufficiently trained deep-learning WF adversary. Only ephemeral blocking defenses retain measurable protection, reducing Laserbeak to 71.8% accuracy under infinite training versus 96.5% undefended.

2026-pulls-ephemeral-network-layer-fingerprinting §5.3, Table 2 ml-classifierwebsite-fingerprinttraffic-shape

A plug-and-play Boundary Preserving Aggregation Module (overlapping window partitioning with joint packet- and burst-level features, W=20ms, stride=10ms) consistently improves existing WF baselines without architectural modification: applied to DF, AUC rises from 0.780 to 0.901 and P@5 from 0.315 to 0.545; applied to ARES'25, P@5 rises from 0.869 to 0.900 in the open-world 5-tab setting. The module's consistent gains across all three tested baselines confirm that fixed non-overlapping window segmentation is a structural vulnerability in prior WF pipelines.

2026-yuan-demux-boundary-aware-multi-scale §V-G, Figure 5 website-fingerprintml-classifiertraffic-shape

DEMUX achieves a P@5 of 0.943 and MAP@5 of 0.961 in the closed-world 5-tab multi-tab website fingerprinting setting, outperforming the strongest prior baseline (ARES'25) by 9.2 and 6.2 percentage points respectively. ARES'25's P@K degrades from 0.900 at 2-tab to 0.851 at 5-tab (a drop of 4.9 pp), while DEMUX improves from 0.926 to 0.943 over the same range, expanding the absolute margin from 2.6 to over 9 points.

2026-yuan-demux-boundary-aware-multi-scale §V-B, Table II website-fingerprintml-classifiertraffic-shape

Censorship classifiers and traffic analysis attacks consistently exploit the initial seconds of a proxy connection, where packet-size, inter-arrival-time, and burst features are maximally discriminative. Cited work demonstrates that website fingerprinting classifiers trained solely on the first few seconds of Tor traffic achieve high accuracy, and real-world GFW detection of fully-encrypted protocols also targets early-connection bytes.

2025-pereira-extended §1 traffic-shapewebsite-fingerprintml-classifierdpifully-encrypted-detect

The framework's GAN-based schedule generator trains on short session windows (e.g., the first 10 seconds) of real browsing traffic from the Tranco Top 1000 sites, learning joint distributions of packet sizes, inter-arrival times, and burst patterns to produce realistic synthetic schedules. This repurposes GAN architectures previously used for traffic analysis (e.g., GANDaLF) as a defense-side cover-traffic generator.

2025-pereira-extended §2.2 traffic-shapewebsite-fingerprintml-classifier

Local onion association—periodically downloading the full set of onion associations from a CT-log-based API and performing each lookup locally—produces a traffic pattern from the guard's perspective that is indistinguishable from generic onion service access, eliminating both the OLF fingerprint and the DNS-based Website Oracle attack vector. This approach requires no per-connection clearnet exit circuit and imposes negligible overhead given the current ~1,500 stable O-L site count.

2025-syverson-onion-location-measurements-fingerprinting §2.2.4, §6 traffic-shapewebsite-fingerprint