Trust

Methodology validated at scale on public corpora; production accuracy pending Q3-Q4 2026 customer pilots. Honest framing, not marketing inflation.

IP & defensibility

ARIADA is built on a 9-module integrated platform. Module-by-module IP-portfolio status, prosecution roadmap, and conversion deadlines are disclosed for accredited-investor due diligence at /investors/legal. This Trust page focuses on validation evidence, security posture, and self-conformance.

Validation evidence

AI authorship attribution PoC v3.0 (March 2026). Methodology cross-referenced against 6,439,303 code samples spanning 64 distinct AI models (Llama, Qwen, DeepSeek, Gemini, Phi, GPT-4, Claude, IBM Granite, StarCoder, …) across 13 programming languages and 9 published datasets (AICD-Bench T1/T2/T3, DroidCollection, CodeMirage, GPTSniffer, DevGPT, SWE-bench, Stack Overflow Survey). All seven research claims CONFIRMED with STRONG-to-VERY-STRONG evidence at scale.

Three confirmed findings

  1. Multi-signal detection necessary. Single-feature classifiers (e.g. code length alone) are insufficient — Cliff's δ 0.07-0.28 (negligible-to-small effect size across 6M+ samples). The attribution module's multi-signal architecture is necessary, not optional.
  2. Per-model fingerprinting feasible. Distinct line/character/stdev signatures per LLM (e.g. o3-mini avg 169.9 lines vs llama3.3 avg 83.2 — a 2× difference). Enables per-tool quality scoring on customer codebases.
  3. AI adoption is exponential. DevGPT corpus shows 145% growth in 77 days (Spearman rs = 0.98, p < 0.001); power users dominate (Gini 0.68 — top 7% of authors produce 41% of AI commits).

Honest framing

The PoC validates the methodology at scale on public corpora. End-to-end ariada.ai classifier accuracy on production customer code is pending Q3-Q4 2026 customer pilots (PoC v4.0 in scoping). We do not claim "validated on N customer sites today."

Backlog optimizer + multi-domain scan research record

For investor due-diligence sources and the full IP-portfolio register, see /investors/legal.

Security & compliance footprint

Risks (honest framing)

Top risks from the umbrella PRD §14, abridged. Full risk register lives in the internal product spec.

# Risk Impact Mitigation
1Enterprise sales cycle (180-360d) > runwayCriticalLand-and-expand from blamer/clamper Pro; partner agency program; founder-led outbound to 50 named accounts
2Incumbent (Deque, Siteimprove) adds 1-2 ariada.ai capabilitiesHigh9-module integrated platform means single-capability copy non-fatal; arxiv-monitor weekly + competitor quarterly
4Pilot accuracy below 80% canonical scoring agreementHighPhase 1 limited to high-confidence rule mappings; iterative D-weight tuning; manual override
7LLM cost overrun on Tier-2/3 remediationMediumTier cascade + cache + similarity reuse; per-org budget; auto-downgrade; dual-vendor
8Founder bandwidth — 1-2 ppl can't ship 6 capabilities + salesCriticalPhase 1 narrowed to 6 capabilities, not all 9; build-prompts dispatched in fresh sessions
10IP-litigation posture (AudioEye 10-K aggression)CriticalFiled-IP portfolio covering nine platform modules (detail) + Evinced / AudioEye prior-art scans + threat-triage; PCT on time; design-around reserve

Cobbler's shoes

ariada.ai detects accessibility violations. ariada.ai itself must be impeccable. Every page on this site is checked with axe-core (zero violations target) and Lighthouse (a11y score 100). If you find an accessibility issue, email a11y@ariada.ai and we will fix it.

Request a security review See pricing