Trust, Compliance & Security
The boundary is the product. LyraMind decides and orchestrates, it never executes, custodies, advises, or predicts. Here is exactly what that means, how we handle your data, and how our analysis works.
Regulatory posture
LyraMind is not a registered investment adviser, broker-dealer, or custodian. The service provides educational analysis of public and provided information, legibility, clarity, machine-readability, and decision-readiness. It does not provide personalized investment advice, recommendations to buy/sell/hold, or predictions of security prices. Buy/sell/hold questions are redirected to a data-readiness verdict, never answered as advice.
- No advice, no prediction. Every response is educational and source-cited; the compliance status is on the response.
- Never a claim about a named model. We analyze the clarity and machine-readability of public text via transparent analysis. We do not claim to game, optimize inside, or predict how GPT / Claude / Gemini / Bloomberg will respond, and never label a synthetic persona as a real institution's AI.
- Never executes or custodies. An execution-readiness verdict is a control/gate, not a recommendation. LyraMind moves no funds and holds no assets.
How our analysis works (methodology)
For AI Readiness, LyraMind scores your disclosure text with transparent, deterministic legibility heuristics, hedging density, quantification, negation clarity, structural clarity, evidence coverage, and machine-extractability, and, optionally, a panel of synthetic analyst archetypes plus (when you enable it) live frontier models. When live models are enabled, the spread of their comprehension is the empirical misread-risk signal, measured, not asserted. You always see the exact spans and the suggested fix. Your disclosure decisions remain yours and your counsel's; LyraMind provides editorial/legibility analysis, not legal or disclosure advice.
For market intelligence, reads are composed from a deterministic Market Brain (regime, participation, mean-reversion, confluence) and a multi-reasoner verification panel whose confidence falls when reasoners disagree, with every read carrying its confidence, sources, and data mode.
Data handling & security
- Analyzed in-process by default. Your draft text is not transmitted to any third-party model unless you explicitly enable the live-model panel.
- Never used to train models, yours or anyone's.
- Ephemeral by default; zero-retention mode available. Draft text is processed in-memory and not persisted unless you enable saved history.
- Encryption in transit (TLS 1.2+) and at rest for anything persisted.
- Privacy by construction, the question flywheel stores no principal; the Trust Ledger keeps a one-way actor hash, never raw identity or key.
- Auditability, every verdict is recorded to a tamper-evident, hash-chained Trust Ledger you can verify.
The Trust Layer, on every answer
Every response carries: confidence · confidence_label · sources[] · evidence[] · reasoning · data_mode · freshness · generated_at · compliance{status, checks, grounded}. It's what a regulated buyer can embed and audit, not a claim you have to take on faith.