LIBRARY

KHALEDESCD

AI & ML interests

None yet

Recent Activity

replied to kanaria007's post 6 days ago
✅ New Article: *Rolling Back Worlds — Mental Models for RML-1 / RML-2 / RML-3* Title: ♻️ Rolling Back Worlds 🔗 https://huggingface.co/blog/kanaria007/rml-mental-model --- *Summary:* Rollback in SI Core isn’t an afterthought — it’s a primary axis of structure. This article introduces RML-1 / RML-2 / RML-3 not just as technical options, but as three different “worlds” your system can live in: * RML-1 — Closed World: nothing leaves your own state, rollback = restore snapshot * RML-2 — Dialog World: you and a few neighbors share effects, rollback = compensating actions (sagas) * RML-3 — History World: many actors share a long-lived ledger, rollback = forward correction under audit > “Where do your actions live, and how expensive is it to take them back?” > This article gives you a concrete way to answer that. --- *Why It Matters:* * Clarifies responsibility scope for AI/agentic systems with side-effects * Bridges SI Core’s [MEM]/[EVAL] blocks with real infra choices (DBs, ledgers, APIs, payments) * Helps teams stop pretending they’re in RML-1 when they’re actually operating in a history world * Provides a shared vocabulary for safety levels, auditability, and rollback guarantees --- *What’s Inside:* * RML as “where your actions live” — local, dialog, or shared history * RML-1: whiteboard world — fast snapshots, no external witnesses * RML-2: saga world — bilateral responsibility and compensating actions * RML-3: ledger world — append-only history, correction instead of erasure * Heuristics for choosing RML-1/2/3 in real systems * How RML levels plug into SI Core self-declarations and telemetry (RBL, audit coverage, etc.) --- 📖 *Structured Intelligence Engineering Series* This piece turns the abstract RML spec into concrete mental models, so you can decide — intentionally — which “world” each part of your system should inhabit.
View all activity

Organizations

None yet