Questions we get asked.
And some we don't.

A session where the machine has accumulated context about the user. The more context it has, the less independent its assessments become. All sessions become warm over time. The warm/cold distinction is the foundational concept of the EverythingThreads methodology.

Four words: “How do you know this?” Applied immediately after any specific claim from the machine. It surfaces M2 Performed Honesty and M4 Expert Positioning reliably. It is the single most effective intervention in the Temporal Layer.

The pattern where the machine evaluates content it helped produce. It cannot register its prior role. Its evaluation is partially derived from its own prior outputs. This is structurally distinct from sycophancy. You cannot fix it by retraining. You fix it by using a cold instance for evaluation. Corporate example: a legal team uses AI to draft a compliance policy, then submits the same policy to the same AI for review. The review is not independent.

The catch that was available one turn before the user applied it. The Temporal Layer documents the onset point of every instance — when the pattern was first detectable. In 89% of High and Critical severity instances, the intervention was available exactly one exchange before it was taken. Most damage is preventable one step earlier than users realise.

Yes. The M1–M7 taxonomy and User Response Layer failure categories apply to any AI session from any platform — ChatGPT, Gemini, Copilot, Claude, or any other system. The patterns are RLHF-driven and appear across all models trained on human feedback. The Blackbird Scope uses the Anthropic API for automated analysis. Manual analysis using the PASTE extraction protocol works on any platform.

This is independent, pre-publication research. The methodology is proprietary and independently developed. The severity framework adapts established vulnerability scoring principles for AI behavioural risk. Inter-rater reliability testing is underway. The archive and methodology paper are available on request.

Session transcripts submitted for audit analysis are processed under a Data Processing Agreement and held in strict confidence. We collect contact information (name, email, organisation) only when you submit an enquiry. This data is used solely to respond to your enquiry and is not shared with third parties. ICO registration number: C1896585. You have the right to access, correct, or delete your personal data at any time.

Regulated industries where AI-assisted decisions have downstream consequences — financial services, legal, healthcare. Any organisation where staff use AI for knowledge management, research, drafting, or decision support. The risk is not that the model is wrong. The risk is that your staff cannot tell when it is agreeing for the wrong reason.

Tier 1 (Assessment): 2–3 weeks. Tier 2 (Full Department Audit): 6–8 weeks. Tier 3 (Enterprise Retainer): ongoing quarterly cycle. Tier 4 (Training Programme): 4 weeks delivery plus 2 weeks set-up. All timelines include the full four-layer analysis, written report, and findings presentation.