Eve library

Kuan Yin — Privacy Implementation

Guide

Revision 1

On this page

Technical doc 32.16.1.13. Engine in @kuanyin/dharma-analytics/privacy-protection.ts; user-facing controls in user-autonomy.ts.

Privacy is engineered in, not bolted on. The core principle: analyse the minimum, retain less, and keep the user in control.

Data minimisation at the source#

  • Typing dynamics — analyzeTypingOnDevice reduces raw keystrokes to aggregate features on-device; raw timings are discarded and never leave the client.
  • Content analysis — analyzeEphemeralContent works over a SHA-256 fingerprint with no retention of the original text.
  • Aggregates — aggregatePatternsWithKAnonymity enforces a k-anonymity floor; laplaceNoisyCount adds calibrated differential-privacy noise before any pattern count is exposed.
  • enforceDataMinimization strips fields not justified by the stated purpose.

manageConsent({ subjectId, purpose, action, now, ttlMs? }) is a state machine with expiry; getConsentStatus resolves the effective state (granted | withdrawn | expired | never_set). Predictive features and learning are opt-in and honour withdrawal immediately (user-autonomy.ts).

Data subject rights (GDPR)#

  • Art. 20 export — exportUserData(subjectId, categories, now) returns a bundle with a record count and integrity checksum.
  • Art. 17 erasure — processRightToDeletion(records) deletes everything except legal-hold records, reporting what was retained and why (fullyErased: false when a hold applies).
  • Retention — applyRetentionPolicy flags records past their window for deletion.

Pseudonymisation & audit#

  • pseudonymizeIdentifiers / hmacPseudonym replace identifiers with HMAC-SHA-256 pseudonyms; anonymizeDataset generalises quasi-identifiers.
  • Every data access is appended to a tamper-evident hash chain (auditDataAccess); verifyAuditChain() detects any break.

Encryption & breach response#

assessEncryptionAtRest / assessEncryptionInTransit validate NIST-aligned policy; the module includes a GDPR 72-hour breach-response planner.

User controls#

user-autonomy.ts exposes opt-out for predictive features, friction-skip preservation (with a no-shadow-penalty guarantee), appeal rights, human-review requests, notification/data-sharing/learning/recommendation/visibility controls, account portability, and complete account deletion. These map to the user-facing "Privacy controls guide" (../user/07-privacy-controls.md).