Mercury Schema Drift Detection

Discovered: 2026-05-06 (15 unrecognized top-level wiki dirs)

The Problem

vector_lib.detect_type() classifies wiki pages based on directory path. Unrecognized directories return None, causing vector embedding failures or mis-indexing.

Standard Wiki Directories

/opt/data/wiki/
  concepts/           # Reference docs (SCHEMA type: concept)
  entities/           # External services, people, systems (type: entity)
  projects/           # Project tracking (type: project)
  queue/              # Active/done task cards (type: task)
  memories/           # Agent persistent memory (type: memory)
  raw/                # Immutable captures, articles, verifier logs (type: raw)
  skills/             # Skill definitions (type: skill)
  archive/            # Archived pages by year (type: archive)
  learnings/          # ← MISSING — should exist for lessons learned

Detection Method

List top-level wiki directories and compare against known schema:

wiki_root = "/opt/data/wiki"
known_dirs = {'concepts', 'entities', 'projects', 'queue', 'memories', 
              'raw', 'skills', 'archive', 'learnings'}
 
unrecognized = [d for d in os.listdir(wiki_root) 
                if os.path.isdir(os.path.join(wiki_root, d)) and d not in known_dirs]

2026-05-06 Drift Snapshot

Directory	Files	Classification	Action
cluster-ops/	3 session logs	project sessions	→ projects/cluster-ops/sessions/
feedback/	3 entries	concept (feedback)	→ concepts/feedback-* or archive
hermes/	1 (goals.md)	concept	→ concepts/hermes-goals.md
ops/	1 overnight note	project session	→ projects/cluster-ops/sessions/
runbooks/	3 operational docs	concept (runbook)	→ concepts/runbooks-* or own type
tasks/	3 task files	queue duplicate	merge into queue/ or redirect
queries/	2 (index + research)	concept/query	→ concepts/queries/
sessions/	1 session log	project session	→ projects/*/sessions/
agents/, autonovel/, email/, voice/	empty	unused scaffold	delete or populate
projects-archived/, proposed-manifests/	empty	stale scaffolds	archive or delete

Remediation Steps

1. Create learnings/ directory if missing
2. Migrate content from unrecognized dirs to proper locations
3. Update vector_lib.detect_type() if new schema types are genuinely needed (e.g., runbooks as first-class type)
4. Delete empty scaffolds
5. Verify no orphans in concepts/index.md after migration