On this page
Continuity and Evolution๏
Reader Question๏
How does Melder preserve a runtime world and retain deliberate change history?
Short Answer๏
Crystallizer records pure-data structural twins and checkpoint windows so a fresh
process can rebuild a world through normal runtime verbs. MutationResearch is a
separate process-owned record for version lanes, composition history, source/impact
views, and candidate rehearsal. Persistence reconstructs; research organizes evolution.
Continuity Path๏
Structural units emit their current record at defined lifecycle points.
A checkpoint seals a profile window into a detached persistence artifact.
Local cache or user-provided external handlers store that artifact.
A fresh process reloads records, verifies/preflights them, and rebuilds through public configuration, bind, conjure, link, and cluster verbs.
The restore report names built units, translated identities, and shortfalls.
Evolution Path๏
MutationResearch organizes content-addressed spell versions and grouped compositions into lanes with append-only journal history. Read operations can expose recorded source, structural/part diffs, drift, and impact. Candidate preview remains separate from actual binding or promotion.
Why This Design Is Strong๏
Records are value data rather than serialized live Python instances.
Fresh runtime identities avoid pretending old process objects survived restart.
Restore uses the same public construction semantics as normal runtime setup.
Research history and persistence custody remain separate responsibilities.
Tradeoffs๏
Not every live value or callable can be reconstructed from data alone. Restore therefore requires code participation and reports shortfalls rather than hiding them. Fresh identities require translation maps for structural records, while content-derived spell identities remain stable where the source contract permits.
Where to Go Next๏
Source entry points: