AI capability control room
Visible intelligence, explicit guardrails
Every AI label maps to a working capability, provider status, deterministic fallback and a rule describing what the model is not allowed to decide.
Capability register
What is genuinely AI enabled
AI enhanced object understanding
Suggest object type, period, region, materials, visible marks and research queries from auction imagery.
Local deterministic engineGuardrail: Hypotheses are capped, source-labelled and never authenticate an object.
Fallback: Deterministic category, mark, exact-object and source research remain available.
AI enhanced signature and mark interpretation
Classify isolated marks, propose cautious transcription alternatives and generate search queries.
Local deterministic engineGuardrail: The model must return unreadable when uncertain and cannot create a verified identity.
Fallback: Local mark detection, cleanup, descriptor matching and source trust scoring still run.
AI assisted evidence synthesis
Summarize agreement, contradictions, missing evidence and next research actions from retrieved records.
Local deterministic engineGuardrail: Only supplied evidence may be used. Every claim retains its source reference.
Fallback: Rule-based certainty, contradiction and readiness engines remain authoritative.
AI driven query expansion
Generate multilingual and alias-aware search queries for marks, objects, provenance and prior sales.
Local deterministic engineGuardrail: Queries are bounded, sanitized and treated as discovery instructions only.
Fallback: Structured product playbooks provide deterministic queries when AI is unavailable.
AI assisted forensic explanation
Translate numeric signature-forensics outputs into a clear expert-review brief.
Local deterministic engineGuardrail: The model cannot change the forensic verdict or call a signature genuine or forged.
Fallback: The native forensic result and its raw metrics remain visible.
AI assisted human review
Prepare a concise evidence brief for the human-in-the-loop queue.
Local deterministic engineGuardrail: Final decisions and training eligibility remain human controlled.
Fallback: The review queue works without AI.
Telemetry
Recent model activity
| Time | Event | Task | Model | Duration | Tokens |
|---|---|---|---|---|---|
| No AI activity has been recorded yet. | |||||