memory — the tree, report search, and session recall

Home branch in Memora: root/skill/memory. Roots: root/canon (owner decisions), root/workspace, root/project/<name>, root/skill/<name>.

Choosing a part

Task	Part
Walk, branch, verdict or search _memory/memora.db and _memory/reports/	this page, inline: the memory CLI in _scripts/ (walk, branch, verdict, search, render)
Recover a lost conversation or a past answer through the host's session store	parts/sessions/
Turn an unstructured brain dump or inbox batch into nodes without losing items	parts/capture/

Rules

• Walk first. Before a non-trivial task, walk the matching root and read the promoted champion and every pruned dead-end with its failure reason.
• Write at entry, during the work and at exit. A node carries its provenance, parent_id, a supersedes edge when it replaces something, and the period it holds for (valid_from / valid_until) kept apart from when it was recorded.
• Search reports by fragment. search reads _memory/reports/ by section instead of pulling whole files into context.
• Render only on request. The graph file is generated, never committed, and regenerated whenever it is shown.
• A hypothesis is labelled as one. If a node was not measured or decided by the owner, it names its source or says it is unverified.
• One writer. A node created by another session is not this session's to commit; shared paths are claimed under _memory/parity/.

Related

planning (task nodes and the task list), grill-me (recording decision batches), dev (what to record as a lesson).