{M} Deferred

Deferred after Phase 19

Items discussed for memory but not shipped in P19 (or only partial). Use this before opening the next memory PR.

Related: {M}-Architecture · {M}-Phases · {M}-Papers

---

Not added (or partial)

Item	Status	Why not	Suggested owner area
Backfill arousal_score on legacy rows	Not done	NULL → rank bonus 0; full-text backfill deferred for Jetson migrate cost	Offline migration / eval
Dream / reflect / journal consume arousal	Explicit non-scope	P19 = write + rank + filter + lineage only; dream already uses valence/salience	Lifecycle
Monthly gate uses arousal	Not done	Paper I weights stay stable until design note revised	Research + config
Studio lineage UI (timeline)	API only	GET …/lineage enough for debug; full panel is front-end	Web UI
Hard filter on all recall entry points	Partial	End of search; some broad/recent helpers may skip	Backend audit
LLM-based arousal	Not done	Keep zero extra tokens; mirror valence later behind a flag	Extraction
Arousal −2 lexicon	Not done	Rank uses abs(a); −2 ≈ +2 for ordering; little value until sign is used	Heuristics
Tune soft-avoid vs hard-filter weights	Defaults only	Needs live chat logs	Evaluation
Experience spreading full parity	Partial	Knobs exist; not P19 scope	Experience store
Cross-user global entity graph	Out of scope	Multi-user is path-isolated	Product/security
Physical delete of superseded vectors	Not done	Breaks lineage / Studio; needs explicit purge policy	Data retention
Parametric in-weight memory	Out of scope	Aiko is explicit SQLite memory	Research
Filesystem-as-memory agent trees	Out of scope	Different research track	Research

---

Why P19 stayed small

flowchart TD
  A[Post-P19 goal] --> B[Add arousal score]
  A --> C[Filter unsolicited strong negative recall]
  A --> D[Expose lineage]
  B --> E{Requires new tokens?}
  E -->|No, lexicon| F[Ship]
  C --> G{Deletes memories?}
  G -->|No, recall-time only| F
  D --> H{Requires Studio UI?}
  H -->|No, API first| F
  F --> I[Flags off restores pre-P19 behavior]

Loading

1. Additive schema only (arousal_score).
2. Heuristic arousal — no new LLM on write beyond extract.
3. Filter is recall-time — does not delete negative memories.
4. Lineage is read-only — audit first, UI later.

Flags off → pre-P19 behavior.

---

Suggested next slices

journey
  title Next memory work slices
  section Stabilize Post-P19
    Smoke test: 5: Maintainer
    Audit recall helper coverage: 4: Backend
  section Improve data quality
    Optional arousal backfill: 3: Tools
    Tune negative recall weights: 4: Eval
  section Improve UX
    Studio lineage timeline: 4: Frontend
    Daily journal 5x3: 3: Product
  section Research alignment
    Paper I arousal revision: 4: Research

Loading

1. Smoke test + merge PR #97.
2. Optional offline arousal backfill.
3. Hard filter on remaining recall helpers.
4. Studio chain panel consuming the lineage API.
5. Reflect / daily journal 5×3: cheer, grateful, remember, learned, experiences.
6. Paper I revision: optional W_AROUSAL; dual-axis affect section.
7. Eval for negative-filter precision/recall on chat logs.

---

Priority matrix

quadrantChart
  title Deferred work priority
  x-axis Low effort --> High effort
  y-axis Low impact --> High impact
  quadrant-1 Big bets
  quadrant-2 Quick wins
  quadrant-3 Park
  quadrant-4 Schedule carefully
  "Recall helper audit": [0.25, 0.80]
  "Studio lineage UI": [0.55, 0.70]
  "Arousal backfill": [0.40, 0.55]
  "LLM arousal": [0.70, 0.45]
  "Global entity graph": [0.90, 0.30]
  "Parametric memory": [0.95, 0.20]

Loading

---

Decision flow for the next PR

flowchart TD
  A[New memory proposal] --> B{Does it change recall safety?}
  B -->|Yes| C[Add eval fixture + config flag]
  B -->|No| D{Does it change schema?}
  D -->|Yes| E[Prefer additive migration]
  D -->|No| F[Keep patch scoped]
  C --> G{Can Studio/API inspect it?}
  E --> G
  F --> G
  G -->|No| H[Add debug surface or defer UI]
  G -->|Yes| I[Update wiki + Paper checklist]

Loading

---

Back: Home · {M}-Phases