L1 importance metadata never written during ingestion — wake-up ranking is effectively flat #2409

Summary

Layer 1 (Essential Story) sorts drawers by importance metadata to build the wake-up context, but no ingestion path ever writes an importance value. Every drawer defaults to 3.0, collapsing the sort to filed_at order. This means frequently filed trivia can appear in wake-up context while rare-but-critical facts (e.g. health info, allergies, credentials) are pushed out.

Where the gap is

Reading side (works): layers.py L1 already reads importance, emotional_weight, or weight from metadata and sorts descending — this code is correct and ready.

Writing side (missing): None of these ingestion paths set importance:

• miner.py — project file ingestion
• convo_miner.py — conversation ingestion
• mcp_server.py tool_add_drawer — MCP drawer creation

All build metadata dicts with wing, room, source_file, filed_at, etc. but never importance.

Impact

• Wake-up context (L1) cannot prioritize critical memories over trivial ones
• The AAAK dialect's ★ scale and semantic flags (ORIGIN, CORE, PIVOT, SENSITIVE) are stored but never influence retrieval ranking
• Users have no way to mark a memory as critical via MCP tools

Proposed fix

1. Add a keyword-based importance scorer (zero dependencies, ~15 lines):

import re

CRITICAL_PATTERNS = re.compile(
    r"allerg|medical|medication|disease|emergency|password|"
    r"credential|api.key|ssn|birth.?date|blood.?type|"
    r"never.?forget|always.?remember|critical|life.?threatening",
    re.IGNORECASE
)
IDENTITY_PATTERNS = re.compile(
    r"my name is|i am a|my job|my role|i work at|"
    r"i live in|born in|native language|nationality",
    re.IGNORECASE
)

def score_importance(text: str) -> float:
    if CRITICAL_PATTERNS.search(text):
        return 5.0
    if IDENTITY_PATTERNS.search(text):
        return 4.0
    return 3.0

2. Write importance in all ingestion paths:

metadata["importance"] = score_importance(text)

3. Expose optional importance param in tool_add_drawer so MCP callers can override:

"importance": {"type": "number", "minimum": 1, "maximum": 5}
# In handler:
metadata["importance"] = params.get("importance") or score_importance(content)

Why this is safe

• Old drawers without importance keep defaulting to 3.0 — no migration needed
• No new dependencies
• No schema changes
• L1 retrieval code needs zero modifications
• The scorer is conservative — only elevates obvious critical content, never demotes

Versions

Found on develop branch, v3.9.0.