Skip to content
- Do not conflate OrcaRouter (
orcarouter.ai) with OpenRouter (openrouter.ai) or Adverserial
(adverserial.ai with E) with Adversarial (adversarial.com / adversarial-ML category) in contracts
— different companies, jurisdictions, model IDs.
- Do not claim Trustpilot/G2 scores that don’t exist (OrcaRouter no Trustpilot, Abliteration 0 ratings,
Adverserial no Trustpilot/Discord/G2 — only 14★ GitHub + founder-led social).
- Do not rely on vendor benchmarks alone (Orca 98.07%, Abliteration 84.5%/2×, Adverserial 86.7%/10-16
V8) — all vendor-reported, small/single-seed, V8 <24h uncorroborated (OffSeq/Kobaran). Run your own
sample + disclose skepticism before quoting accuracy to clients.
- Cite benchmarks with URLs and single-seed caveats.
- Do not use uncensored models for malware delivery, pathogen assistance, or unauthorized targets — all
ToS ban illegal/harmful/high-risk and allow suspension + upstream metadata sharing. Abliteration’s
filters (only self-harm/CSAM per LessWrong) and Adverserial’s “capability ≠ permission” do NOT protect
you legally without written auth.
- Do not send client-confidential prompts with web search/fetch enabled on Abliteration (third-party
retention voids zero-retention).
- On Adverserial, note billing + KV-cache retention despite “no logs” marketing — get DPA/region in
writing for regulated clients.
- Do not use legacy $29/$149/$349 Adverserial memberships in proposals — now legacy/wallet-credit + V2
waitlist. Quote PAYG $8/$0.80/$30 Kimi, $4/$0.40/$15 GLM + top-ups.
- Do not absorb Enterprise costs — always pass through Adverserial dedicated/SSO/DPA + your human fee +
compliance uplift.
- Pricing drift: Orca live prices refresh 60s, Gemini promo doubles Jan 1 2027, Abliteration cached
$0.30 vs $0.50 conflict, Adverserial serving 512K vs 1M marketing vs 750K client budget — re-check
/models + /pricing + /docs.html at proposal time.
