Valence + Intensity — Metadata mới cho knowledge entries từ Hippocampus

[magicpro97]

Mô tả

sk's mistake vs pattern là binary category. OpenWolf's Hippocampus PR #35 thêm valence (reward/neutral/penalty/trauma) + intensity (0.0-1.0) cho mỗi event.

Học hỏi từ OpenWolf Hippocampus (PR #35)

• WolfEvent: context + action + outcome triplet
• Valence: reward (0.3-1.0), neutral (0.2-0.4), penalty (0.4-0.9), trauma (0.6-1.0)
• Trauma events: NEVER evicted, NEVER decay, dedicated trauma_index
• Decay: 5%/week for non-trauma, ~5 months to reach 0
• Recall scoring: location 55% + recency 20% + valence 15% + intensity 10%

Đề xuất

• Thêm intensity REAL DEFAULT 0.5 column vào knowledge_entries
• High-intensity mistakes (user repeated "NEVER do this") → intensity = 0.9
• briefing.py sort by intensity DESC → high-priority warnings first
• Optional: trauma_immune BOOLEAN DEFAULT FALSE — entries không bao giờ bị pruned
• Migration: add column via migrate.py MIGRATIONS list

Ưu tiên

Trung

References

• OpenWolf Hippocampus PR #35
• yhyu13/openwolf event-store.ts

Acceptance Criteria

• DB schema thêm valence và intensity columns
• Migration backward-compatible
• learn.py hỗ trợ --valence và --intensity flags
• briefing.py sort theo intensity khi relevant