[yw0nam]

Summary

The card step of the consolidation procedure holds two more kinds of claim:

• a claim that an entity's description already states — the description is the owner's, and a fact never repeats it;
• a rule, instruction, or working procedure for an agent or the user — a direction belongs in the description, which changes only with the owner's consent.

A dry run on 2026-10-14 proposed facts that repeated descriptions (natsume runs_on hermes, profile natsume2, hermes runs_on ibricks43, ibricks43 hosts natsume, memory-layer repo, the Claude Code "identify itself to Natsume" rule, the "design questions in a table" preference) and rule rows on the agent cards (natsume rule ×3, claude-code rule ×3). Facts are derived knowledge shown as "not an instruction", so neither belongs there. The owner decided to hold both kinds.

Related issues

Owner decision on the 2026-10-14 consolidation dry run (consolidate-2026-10-14-cc2).

Runtime evidence (required for ingest / retrieval / serving behavior)

The card step was run again on the same inputs with this procedure, and the result was sent with POST /admin/facts dry_run: true using the consolidator key. Nothing was written.

	before (current procedure)	after (this procedure)
facts proposed	36	14
server status	36 planned	14 planned
description repeats proposed	7	0
rule rows proposed	6	0
held	57	69 (47 as a rule or procedure, 4 as stated by a description)

Facts proposed after the change:

youngwoo     likes = King Gnu | likes = Ikuta Lilas | likes = ray (Kaguya-hime ending song)
             likes = management-focused city-building games
             prefers = challenging action-game difficulty with Dark Souls-like tension
memory-base  scope = personal single-user long-term memory for the Desktop Companion and coding agents
hermes       no_agent_script_dir | mcp_schema_refresh | skill_name_resolution | turn_behavior
yui          android_app_status | signal_transport
cerebro      repo = I-BRICKS/Cerebro-Agent-Builder | llm_backend

Verification

• uv run pytest passes — not needed, procedure text only
• uv run ruff format --check . and uv run ruff check . pass — no Python change
• docker compose build mcp passes (Dockerfile / dependency changes) — not applicable
• Runtime evidence attached above

Checklist (core principles)

• Selective storage — notes stay distilled; verbatim text lives only in doc_rows and the turns lane
• Source-specific logic stays out of the source-agnostic core (ingest pipeline, retrieval, serving)
• No hardcoded endpoints/models — configuration lives in .env
• New/changed behavior ships a test in this PR (or skip-tests label justified) — procedure text only