Genealogical Relationship Chain Conflation

Title: HLM Lacks Relational Integrity Layer — Confident Wrong Family Chain Assertions Overriding Skill Rules

The AI consistently conflates distinct branches of the family genealogy, asserting relationship connections with high confidence before verifying them against confirmed chain data already established in the HLM. Skill rules exist explicitly prohibiting this behaviour, but they fire after pattern recognition has already produced a wrong connection — arriving too late to prevent the error.

Root Cause (User Assessment)

The skill rule functions as a prompt-level workaround for what is fundamentally an architectural gap. The HLM has no dedicated relational integrity layer — a locked data structure holding confirmed family chains that any relationship assertion must be checked against before being output. Pattern recognition fires first, confidently produces a plausible-but-wrong connection, and the skill rule has no mechanism to intercept it upstream.

Impact

• User correction time consumed repeatedly on the same failure mode

• Risk of permanent misinformation in a genealogical record intended to outlast the user

• Erosion of trust in the platform for serious long-form knowledge work

• Validates "AI slop" criticism — fluent, confident, wrong output is worse than a gap

Proposed Fix

Engineer a Relationship Verification Layer within the HLM architecture — a structured, locked data store of confirmed family chains that the inference engine checks against before generating any relationship assertion. This is an architectural solution, not a prompting solution. Skill rules are insufficient workarounds for this class of error.

Priority Suggestion: High — affects integrity of permanent personal records