Ternary Fabric Strategic Convergence Note
This note records the current architectural decision for how T81 relates to
the separate t81dev/ternary-fabric repository.
Current Decision
Do not merge ternary-fabric into T81 now.
Treat ternary-fabric as a parallel strategic track, not as:
- a T81 submodule
- a subtree import
- a direct code merge target
- part of T81’s current stable build-against surface
Why
The two repositories are thematically related, but they currently center
different architectural truths:
- T81 today:
deterministic, policy-gated runtime;
CanonFS/Axion provenance and admission;
bounded AI OS-object family;
RFC-00D1 interchange hardening
ternary-fabric today:
LD_PRELOAD/interposer acceleration;
TFMBS device ABI;
MLIR and compiler lowering;
FPGA/RTL/hardware bring-up;
performance/offload-first execution story
A full merge would blur T81’s current runtime-first adoption wedge and add
subsystem breadth in exactly the areas T81 is trying to keep narrow.
Allowed Near-Term Convergence
Allow only concept transfer and planning alignment:
- borrow clear contract-writing patterns after rewriting them into T81’s
runtime-first framing
- treat
ternary-fabric as one future input into an experimental accelerator
or backend note
- keep MLIR/GGUF/T3_K overlap in mind without merging compiler plugins,
interposers, or hardware flows
Explicitly Out Of Scope
Do not import or vendor:
- LD_PRELOAD interception
- fabric residency heuristics
- predictive scheduler/orchestrator code
- FPGA or ASIC bring-up flows
- TFMBS public ABI headers
- MLIR dialect/plugin stacks from
ternary-fabric
Future Integration Seam
If convergence is revisited later, the only sanctioned seam is:
- an experimental accelerator/backend boundary behind T81’s runtime, outside
current stable public APIs
That seam must preserve:
- T81 as the governed execution substrate
- CanonFS as the artifact/provenance system
- Axion as the pre-side-effect policy gate
Any future hardware or fabric acceleration should therefore be treated as an
execution-backend candidate, not as a replacement for T81’s current project
identity.
Revisit Conditions
Do not revisit deeper convergence until all of the following are true:
- RFC-00D1 is no longer an active hardening lane
- the bounded AI OS-object family remains stable without posture corrections
- T81’s runtime/backends story is clear enough to add one experimental backend
seam without confusing the public story
- there is a concrete T81 consumer for acceleration, not just thematic overlap
Public Framing Rule
If ternary-fabric is mentioned alongside T81, describe it as:
- a related ternary acceleration effort
- not part of T81’s current stable build-against surface