Final Reconciliation
Pass history
Pass 1
found the correct machine-centered architecture
Pass 2
added useful status, availability, portability, delivery, schema,
identity, and ECS work, but over-split the machine into packages
Pass 3
restored Pass 1 as the base and rejected the storage split,
empty umbrella, universal type drawer, and package-by-purity rule
Final v1
places runtime, PakVM, Bvisor, world, and ports inside SyncBat;
retains the useful Pass 2 laws under the recovered machine
Pass 2 is not a normative predecessor. Its retained ideas have been rewritten into current owners. Its rejected package map has no authority.
Final machine
MacBat declarations ─┐
├→ BatPak contracts, schemas, ProgramImages, WorldImages
BatQL source ────────┘ │
▼
SyncBat world
┌─────────────────┼─────────────────┐
▼ ▼ ▼
runtime PakVM Bvisor
│ │ │
└────────── world + port ───────────┘
│
▼
.fbat truth
NetBat transports declared world calls. TestPak proves the entire graph.
Locked corrections
.fbatremains BatPak’s journal format; no separate storage package exists.- PakVM names the interpreter;
.vpaknames the executable package format. - Bvisor and PakVM are internal SyncBat planes.
- The root concept-file spine replaces a universal type directory.
- HostBat is dissolved by ownership, not copied into SyncBat.
- The OS-specific Bvisor backend ladder is evidence, not product architecture.
- BatQL is an independently consumable compiler crate targeting BatPak-owned image contracts.
- TestPak owns repository proof; Muterprater owns mutation only.
What “final” means
Architecture, ownership, dependency direction, package boundaries, language boundaries, and semantic laws are frozen.
Exact opcode numbers, exact byte offsets, benchmark thresholds, and adapter-specific physical guarantees remain implementation constants. They may be selected only inside their named gate with goldens and conformance evidence. They cannot reopen the architecture.