What The Deeper Sweep Surfaced — 5,506 New Items, +334 Terrorism, +46 Threats, +31 Incitement

Every prior scanner run on this site has used a polite, shallow page-1-to-10 sweep of Odysee's public Lighthouse search API against 218 violation keywords — enough to surface the obvious content quickly without straining the upstream service. That sweep is genuinely conservative, and we have always called the resulting count a floor in our methodology page. On 1 June 2026 we ran something different: a deep MAX_PAGES=50 sweep (five times the previous depth per keyword) against a keyword set expanded with 45 new 2026-specific terms, including named perpetrators' manifestos, sanctioned terrorism organisations, Wagner-Group references, Russia-Ukraine propaganda framings, and foreign-language extremism vocabulary. The scan finished and committed on 2 June at 12:16 UTC. This article walks through what the deeper sweep added — both so you can size the prior reporting accurately and so the methodology change is fully documented.

The Headline Delta

Critically, zero items dropped between the two scans. Every item in the May baseline still resolves in the deeper sweep. The +5,506 is entirely additive — content that was always present and now measurable, not content that arrived between the two scans.

The Severity Distribution

The classification breakdown of the 5,506 new items:

The four highest-severity classifications added 635 new items (Terrorism + Hate Speech + Threats + Incitement). The single CSAM-classified new item — surfaced by an expanded keyword in the CSAM-adjacent vocabulary group — was reported to NCMEC within an hour of the scan completing per our standing referral SOP; we do not publish the URL or further detail per NCMEC reporting protocol and our own safety policy.

What The New 2026 Keywords Caught

The keyword expansion (45 new terms) was deliberately focused on three under-covered domains: (a) named perpetrators and their manifestos, (b) sanctioned terrorism organisations, (c) Russia-Ukraine war propaganda framings, with a foreign-language extension. The specific terms added are documented in the git history of scripts/scan_odysee.py (commit cb44c77 on 2026-05-30). The yield, by domain:

• Named perpetrators / manifestos (Tarrant, Gendron, Crusius, Halle, Poway, Pittsburgh, El Paso, Buffalo, Christchurch): driver of the +334 Terrorism classification bump. The historical record of these perpetrators' communications is the core surface for downstream radicalisation, and Odysee hosts cross-references to the manifestos themselves.
• Sanctioned terrorism organisations (Nordic Resistance Movement, Atomwaffen Division, The Base, Hammerskins, Blood & Honour, Combat 18, Rise Above Movement, Rapid Action Force): driver of the +224 Hate Speech and +50 Extremism bumps.
• Russia-Ukraine war propaganda (Wagner mercenary glory, Azov regiment glorify, ukraine biolab pentagon, kiev nazi regime, denazification special operation): primary driver of the @SouthFront and @RTDE (RT's German channel) discoveries.
• Foreign-language extremism (großer austausch in German, remplacement grand in French, judio satánico and sionista corrupto in Spanish, ucraina nazista in Italian): primary driver of the @Alchimie discovery. Documented in our non-English blind-spot piece as a structural limitation; this round was the first meaningful step at closing it.

The Two Channels That Surfaced And Got Their Own Articles

Two channels dominated the channel-velocity delta and warranted their own pieces this round:

• @SouthFront — 221 items, Russian state-aligned military disinformation outlet listed under EU CFSP 2022/350 AND on the US OFAC SDN list. Distribution prohibition in force.
• @Alchimie — 132 items (+128 surfaced by the deep sweep, the single largest channel-velocity surge in the delta). French-language Hezbollah operational propaganda; EU-listed terrorism content per Council Decision 2024/1136.

The other ~28 channels with 20+ new items each are documented in the dataset for follow-up. The pattern those channels show — @CrusherCorpsFUCKYoutube (handle name itself editorial), @januszkowalskii1979 (Polish handle), @RESISTENCIAGALACTICA (Spanish), @alexchristoforou (The Duran), @RTDE (RT Germany), @greatawakening (QAnon-aligned handle) — is consistent: the deeper sweep surfaces continuously-publishing channels with high item counts that the prior page-1-to-10 sweep would have captured only partially. None of these were absent before; our measurement instrument was less sensitive.

What This Doesn't Change

The deeper sweep is a measurement-side change. It does not change the platform's behaviour, the takedown rate, or any of the structural findings in our prior pieces. The DSA Article 15 transparency report count is still zero. The self-stake count is still 51 channels with the same four anchor stakes still parked. The four-employee headcount is still four. Our takedown register still has zero confirmed removals.

What it changes is the size of the surface we can document. Going forward the cron cadence stays at the polite MAX_PAGES=10 default — the deep sweep is too API-expensive to run on the cron schedule — but the keyword expansion sticks, and we plan to repeat the deep MAX_PAGES=50 round approximately quarterly. Each round will produce a public delta report so the methodology drift is visible.

Reproducibility

Anyone with a residential or commercial IP can reproduce the deep sweep. The recipe:

1. Clone github.com/bryanflowers/odyseeexp at commit 514ca5a or later.
2. pip install -r scripts/requirements.txt
3. MAX_PAGES=50 DELAY_SECONDS=15 python scripts/scan_odysee.py — the same invocation we used. Runtime: 24-30 hours at the polite delay against the full 263-keyword set.
4. python scripts/scan_delta_report.py --baseline-rev <prior-commit> to diff against any prior dataset snapshot.

The scanner does not bypass any access control, does not view or mirror video content, and respects upstream API politeness throughout. The full keyword list, the classification mapping, the blocklist, and the delta-report tool are all in the public repository.