About | Krystal Trace

Supplier due diligence

What Krystal Trace is

Evidence preparation, not recovery

Krystal Trace is a blockchain transaction tracing engine, claimant evidence intake system, case timeline builder, and evidence-pack generator. It helps users organise TXIDs, wallet addresses, claimant records, confidence grades, possible exchange endpoints, and open questions before engaging legal or expert support.

What it is not

It is not a fund recovery guarantee, hacking tool, exchange account access tool, way to bypass legal process, wallet-access tool, substitute for solicitor advice, or substitute for a formal forensic expert report where litigation requires one.

LLM architecture

Three-tier processing with named sub-processors

Krystal Trace processes claimant evidence through:

(a) Ollama Inc. (US-hosted, transient processing — prompts not retained beyond request fulfilment, not used for AI model training) for bulk extraction, classification, and draft narrative work;

(b) Anthropic PBC (US-hosted, GDPR Data Processing Agreement available, prompts not used for AI model training) for final report drafting, solicitor-facing analysis, and high-stakes group-cohort pattern work;

(c) on-server local inference only — no third-party processor — when the matter's sovereign flag is set. Slower processing, but the evidence never leaves Krystal Unity's own server in [DC location].

Krystal Trace does not route private claimant evidence through DeepSeek's direct China-hosted API. DeepSeek's V4-Flash model is used via Ollama Inc.'s US infrastructure (option (a) above), which gives the model's reasoning capability without the data transiting Chinese jurisdiction. DeepSeek-direct is configured only as a fallback for non-private public-blockchain summary work.

Sub-processor register
Ollama Inc. - US-hosted transient processing, no training
Anthropic PBC - US-hosted GDPR DPA available, no training
DeepSeek-direct - fallback only for non-private public-chain summaries

Confidence

Grades are part of the evidence model

Every claim is graded. Heuristics never get presented as facts.

ADirectly evidencedBStrong inferenceCPlausible inferenceDWeak — reviewEUnknown

Directly evidenced on-chain, with a visible transaction path or source record.

Strong inference from repeated on-chain behaviour or corroborating evidence.

Plausible inference that should be treated as a working lead.

Weak signal requiring manual review before relying on it.

Unknown or not supportable from the evidence available.

See worked examples

Data posture

GDPR-aware by design

Data residency

Phase 1 uses US-hosted processors for selected LLM tasks unless a matter is flagged for sovereign mode. Sovereign-mode matters route through local on-server processing only.

Sub-processor disclosure

Sub-processors are named in onboarding packs and report covers so UK and EU law firms can run supplier due diligence without guesswork.

Built by

Krystal Unity - independent UK-based forensic tooling team building preliminary evidence workflows for claimants, lawyers, and experts.