SentraMind
← All demonstrations

Prompt sandbox

This browser demonstration executes the deterministic detector and policy layers locally. It does not reproduce endpoint interception, the appliance runtime or model classification.

Prompt sandbox detector pack v2026.08 · deterministic layer Browser demonstration
All samples are synthetic. Nothing you type leaves this page.
Singapore NRIC / FIN
S8438201I
prefix-weighted checksum verified
high precision
Hard block
Rule action: Singapore NRIC / FIN → BLOCK (precision class: high)Rollout stage: enforce for SG-FIN-Analysts → no changeException: none active for this user × destination × rule
Precedence is normative: exception > rollout stage > policy rule action.
⊘
This cannot be sent
Destination: ChatGPT
●Singapore NRIC / FIN detected
🔒 The matched content is never shown back to you.
Blocked here — use SentraMind Internal Assistant instead.
Override cost is graduated per rule: one click for ordinary warns, typed justification for crown-jewel and classified matches. Every override is logged against the AD identity.
This runs the deterministic detector layer only. Model classification runs on the appliance GPU and is not reproducible in a browser — so we do not pretend it is.

What is real here

  • NRIC prefix-weighted checksum with per-series letter tables
  • MyKad contextual validation — there is no check digit to verify
  • NIK province codes and the +40 day-offset for women
  • Thai mod-11 check digit; Vietnamese province and century/gender structure
  • Luhn, IBAN mod-97, secret patterns and Shannon-entropy scoring
  • Action ladder, rollout-stage capping, and exception precedence

What is not

  • Model classification — it runs on the appliance GPU
  • Exact Data Match against your real records; the crown-jewel sample is a stand-in for salted-hash matching
  • Document fingerprinting and classification-label reads
  • Code-switched multilingual classification