This index tracks RFC status and intended disposition.
draft: active authoringproposed: under formal reviewaccepted: design accepted, pending/ongoing integrationintegrated: merged into normative specssuperseded: replaced by a newer RFCrejected: closed without adoptionAuxiliary note:
| RFC | Title | Status | Notes |
|---|---|---|---|
| RFC-0000 | T81 Base-81 Ternary Computing Stack | accepted | Umbrella RFC: all component RFCs (0001–0033, 00B0–00B7, DPE-0001–0009) accepted; all 23/23 AC met |
| RFC-0001 | Architecture Principles | accepted | All §1–§3 principles satisfied: deterministic execution (RFC-0002), canonical data (tensor/float/fraction), ternary-native ISA, safe recursion (RFC-0003 RCS), no-UB policy, privileged boundary mediation; 49/49 axion+ethics+tier tests pass |
| RFC-0002 | Deterministic Execution Contract | accepted | Cross-layer invariants proven: 27-program conformance suite, EvidenceCollector, full stack; §11 fulfilled |
| RFC-0003 | Axion Safety Model | accepted | All 9 criteria met: AXREAD/AXSET/AXVERIFY mediation, fail-closed policy parse, deterministic audit log, tier-supervision invariant, instruction ceiling, CanonFS observability, AI-model hooks, interrupt rate-limit; 49/49 axion+ethics+tier+policy tests pass |
| RFC-0004 | Canonical Tensor Semantics | accepted | §2.1–2.5 met: immutable shape tuples, 1-based handle pool, TVECADD/TMATMUL/elementwise with shape guards, IR lowering, Axion shape-metadata hooks; 30+ tensor tests pass |
| RFC-0005 | TISC v0.4 Extensions | accepted | All 7 criteria met: structural opcodes (46–49), VLoad/VStore/VAdd/VFma, ChkShape, ReadIsaVersion, Axion trace; 10/10 tests pass |
| RFC-0006 | Deterministic GC | accepted | All 6 criteria met: mark_and_sweep, compact_heap, GcSafepoint opcode, byte-threshold (3^12), Axion events, policy veto; 8/8 tests pass |
| RFC-0007 | T81Lang Standard Library | accepted | All 7 criteria met: arith/tensor/option/result/axsafe_io modules in kBuiltinTable; MakeOptionSome/None/MakeResultOk/Err opcodes in TISC; tier+purity rules enforced by SA; ISA-version freeze replaces per-module versioning |
| RFC-0008 | Formal Verification Harness | superseded | Superseded by RFC-0027 (spec-as-executable is the concrete realization) |
| RFC-0009 | Axion Policy Language (APL) | superseded | Superseded by RFC-0022; s-expression policy syntax, PolicyEngine::evaluate(), and guard/loop predicates all implemented; forward path is RFC-0022 |
| RFC-0010 | TISC Float/Fraction Ops | accepted | Caveat resolved: FDIV is IEEE 754 exactly-rounded; transcendentals route through t81_soft_math (RFC-0030); no host libm dependency |
| RFC-0011 | T81Lang Grammar Modernization | accepted | All 7 AC met: annotations, generics, expanded types, var/break/continue, record/enum, @module namespacing, legacy artifact superseded; realized via RFC-0003/0004/0007/0015/0029 (2026-03-16) |
| RFC-0012 | Ternary Tensor Quantization | superseded | Superseded by RFC-0026 (QMATMUL covers quantized matmul path) |
| RFC-0013 | Ternary Matmul | superseded | Superseded by RFC-0026 (QMATMUL) |
| RFC-0014 | Neural Primitives | superseded | Superseded by RFC-0026 (ATTN + EMBED cover the neural primitive surface) |
| RFC-0015 | Agentic Constructs | accepted | First-class agent/behavior declarations, AGENT_INVOKE opcode, Axion audit; 9/9 AC met (2026-03-15) |
| RFC-0016 | SIMD Limb | superseded | Superseded by RFC-0017; t81::simd namespace retained for internal helpers only |
| RFC-0017 | Introduce T81 Native | accepted | All 6 criteria met: t81::T81 struct, negation (AVX2+scalar), addition (carry-map+prefix scan), conversion, master header, fallback guard; 43 assertions pass |
| RFC-0018 | T81 Native SIMD Arithmetic | accepted | All 6 criteria met: AVX2 carry-map addition, scalar fallback, multiplication, subtraction, benchmarks, AddEntry table; proved by t81_native_property_test + t81_simd_add_helpers_test |
| RFC-0019 | Axion Match Logging | accepted | Runtime/spec/CLI depend on canonical match metadata and guard-audit strings |
| RFC-0020 | Axion Segment Trace | accepted | Runtime/spec/CLI depend on canonical segment-trace strings |
| RFC-0021 | Tier4 Cognition | accepted | All 10 criteria met: Tier4Loop observe/reflect/refine, SelfModel (81-entry ring buffer), RecursiveImprovementBounds, TierAwarePlanner, promotion heuristics, VM tier limits, Axion CheckTier enforcement; 4 test suites pass |
| RFC-0022 | Axion Policy Language Evolution | accepted | CLI/compiler/runtime policy surface is active; supersedes RFC-0009 in practice |
| RFC-0023 | T81Lang Print Canonical Runtime | accepted | All 5 criteria met: Opcode::Print + State::printed_output; IR lowering via binary_emitter; Int/Bool/Float/Fraction/Symbol rendering; determinism verified by vm_print_test + e2e_print_runtime_test |
| RFC-0024 | C++23 Wording Alignment | accepted | All 3 criteria met: no normative changes, consistent C++23 wording, CI unaffected; 344/344 tests pass |
| RFC-0025 | Policy-Gated Tensor Loading via CanonFS | accepted | TLOADHASH + allowed-tensor-hashes are active; operational hardening continues |
| RFC-0026 | AI-Native Inference Opcodes | accepted | Phase-1 opcode surface implemented; WLOAD promotion review resolved; RFC-0030 float-domain policy complete |
| RFC-0027 | Spec-as-Executable Conformance Model | accepted | Conformance suite is wired into CMake/CTest; optional annotation follow-ons remain |
| RFC-0028 | Deterministic Trace JIT | accepted | All 6 criteria met: §2 trace_hash, §3 flat register file, §4 CanonFS JIT cache (JitTraceCache; 19/19), §5 AxionBoundary OSR, §6 repro oracle (15/15) |
| RFC-0029 | T81Lang Feature Registry Drift Prevention | accepted | All 6 criteria met: kBuiltinTable as single source, tier-gate enforcement in SA, distributed builtins blocked in DCP, effect-surface blocking in @pure/Tier≤1, VM DecodeFault safeguard, spec/t81lang_features.md feature registry |
| RFC-0030 | Deterministic Math Subsystem | accepted | t81_soft_math integer-backed implementation covers all transcendentals; FDiv is IEEE 754 exactly-rounded; RFC-0010 caveat resolved |
| RFC-0031 | Deterministic AI Execution Contract | accepted | All 5 RFC-0032 promotion phases complete; conformance programs authored; phase_status advanced to spec_conformant |
| RFC-0032 | AI Subsystem Promotion Pathway | accepted | All 5 phases complete: ternary codec, doc, Axion hooks, T81VmBackend, EvidenceCollector + AI CLI; conformance suite closed |
| RFC-0033 | Dual TUI Frontends | accepted | All 4 phases complete: FTXUI infra, t81 studio (7 views, palette, REPL), t81 agent (14 slash cmds, session save/load), CI snapshot test (11/11 assertions) |
| RFC-0034 | T81-Native AI Inference | accepted | In-repo implementation closure is complete: opcode/runtime path, Axion policy surface, @ternary_inference lowering, conformance programs, CanonFS-backed .t81w execution evidence, and native benchmark evidence are all in place; a first native VM fast-path set (TExp, TQUANT, TACT, TERNACCUM, TSiLU, TSoftmax, TRMSNorm, TRoPE, TWEMBED, TWMATMUL, TATTN) for balanced-trit weights is now implemented, with matched VM native-vs-binary wins recorded at roughly 62x-67x for the unary set, 6259.43x for TQUANT, 6834.34x for TACT, 80.57x for TERNACCUM, 11.99x for TRMSNorm, 4.07x for TRoPE, 769.63x for TWEMBED, a scale-dependent TWMATMUL crossover from 0.84x at 64 to 11.84x at 256 and 570.42x at 4096, and TATTN at 110970.48x on the current 256-trit VM comparison; remaining work is cross-platform evidence refresh plus broader optimization of the post-decode result-representation path |
| RFC-0036 | T81Lang Foreign Function Interface Grammar | accepted | Language/frontend contract for the governed FFI surface defined by RFC-00B8 |
| RFC-0037 | T81Lang TNN Standard Library | accepted | TNN library surface for ternary-native AI/runtime work; depends on RFC-0034 |
| RFC-0038 | Ternary Lattice Cryptography Primitives | accepted | POLYMUL / POLYMOD lattice-crypto substrate is implemented and is the accepted parent for RFC-0039 |
| RFC-0039 | NTRU-KEM: Ternary Key Encapsulation Mechanism | accepted | Accepted narrow KEM construction built on the lattice-crypto direction; revisit if RFC-0038 changes materially |
| RFC-0040 | SWAR Formalization | accepted | In-repo implementation closure is complete: stable SWAR API, explicit TISC opcodes (0xD5-0xD7), Setun bridge mnemonics, VM/JIT/CanonFS coverage, Axion trace/policy coverage, performance evidence, and experimental/packed_trit_vector.hpp deprecation wording all in place (2026-03-22); remaining stable-promotion item: refreshed x86_64 cross-architecture evidence (pending CI runner) |
| RFC-0041 | SIMD Formalization | accepted | In-repo implementation closure complete: AVX2/NEON kernels, t81/simd/simd.hpp facade, migration guide, ARM64 evidence (2026-03-18 + 2026-03-22 snapshots), formal deprecation wording (#pragma message guard on direct experimental include), and RFC-0041 evidence note all in place; remaining stable-promotion item: x86_64 evidence refresh (pending CI runner) |
| RFC-0042 | Deterministic Backend Equivalence Contract | accepted | Defines scalar as the canonical oracle and makes scalar ↔ SWAR ↔ SIMD ↔ future backend substitution a governed equivalence surface rather than an optimization-only convention |
| RFC-0043 | Deterministic Conformance and Validation Framework | accepted | Defines how determinism claims are proven: corpus classes, replay artifacts, backend matrix, breach classification, and CI enforcement model |
| RFC-0044 | Stable Packed Trit Vector Interface | accepted | Stabilizes the shared packed-trit substrate under SWAR and SIMD so backend governance no longer depends on experimental representation assumptions |
| RFC-0045 | Deterministic Memory Model | accepted | Defines canonical memory visibility, segment semantics, handle identity, aliasing constraints, and the relationship between immediate VM mutation and deferred epoch commit |
| RFC-0046 | Deterministic Scheduling and Execution Ordering | accepted | Defines the constitutional scheduling rule across program order, dependency order, canonical commit order, and future multi-core/distributed execution |
| RFC-0047 | Deterministic JIT and Lowering Rules | accepted | Constrains Trace-JIT and future lowering-based acceleration so optimization remains interpreter-equivalent, policy-boundary-safe, and subordinate to backend/memory/scheduling governance |
| RFC-0048 | Deterministic Surface Definition and Governance Boundaries | accepted | Defines the constitutional classes for DCP, governed non-DCP, experimental, and out-of-scope surfaces, plus the promotion/demotion rules that control deterministic claims |
| RFC-0049 | Canonical Ternary Arithmetic Semantics | accepted | Defines the mathematical oracle for addition, subtraction, multiplication, comparison, carry, normalization, and overflow so every governed backend shares one arithmetic contract |
| RFC-0050 | Vectorized Ternary Operations for TISC | accepted | Defines when vector execution is an ISA/VM semantic surface versus an internal backend optimization, including lane ordering, width, trace, and fallback rules |
| RFC-0051 | Deterministic Heterogeneous Acceleration | accepted | Constrains GPU and other accelerator backends through explicit kernel-class, transfer, reduction, fallback, and promotion rules so heterogeneous execution cannot silently expand the deterministic surface |
| RFC-0052 | Canonical Dataflow and State-Driven Execution | accepted | Defines canonical dependency graph, state identity, readiness, propagation, and CanonFS-linked execution rules above DPE |
| RFC-0053 | Distributed Deterministic Execution Protocol | accepted | Defines cross-node state identity, global commit order, conflict resolution, replay artifacts, and failure behavior for future distributed execution |
| RFC-0054 | CanonFS Object Identity and Persistence Contract | accepted | Reconciles CanonFS object identity, persistent-driver guarantees, capability semantics, parity semantics, and user-visible reference naming |
| RFC-0055 | Native Ternary Hardware Target and Interop Contract | accepted | Separates native ternary hardware targets from accelerator-only targets; defines TISC semantic authority, hardware interop obligations, and promotion rules for future external ternary platforms |
| RFC-0056 | Google Axion-T81 Integration Contract | accepted | Contract specifies Axion as a Verified Lowering Target (RFC-0055): TISC-to-ARM64 lowering, Titanium offload governance (RFC-0048 boundary classification), cloud monitoring separation (§5.2), and multi-instance/multi-region determinism via RFC-0053 epoch protocol; all 4 open questions resolved (2026-03-22); Phase 1–3 implementation is post-acceptance work |
| RFC-00B0 | Axion HAL Specification | accepted | First non-hosted promotion path and HAL contract for the Axion OS stack |
| RFC-00B1 | Ternary MMU | accepted | TVA layout, radix page table, and MMU fault model for Axion |
| RFC-00B2 | Device Driver Architecture | accepted | Phase 4 storage/display/network boundary and guest-device model |
| RFC-00B3 | Axion Governance Kernel Architecture | accepted | All §8 acceptance criteria met; two open questions resolved (capability model, device registry) |
| RFC-00B4 | Userland Service Contract | accepted | Stable service/runtime request and result boundary before syscall and ABI widening |
| RFC-00B5 | Governed Event Interrupt Model | accepted | Slices 26–28 complete: WaitForDevice, interrupt policy gate, unhandled IRQ governance (UnhandledInterruptDropped); 3214/3214 assertions; all open questions resolved or formally deferred (2026-03-22): nesting resolved by design, audit surface stabilized (6 events), priority classes and tooling deferred to future RFCs |
| RFC-00B6 | Minimal Syscall and Capability Boundary | accepted | All §8 criteria met; open question resolved (Slice 22): ClaimDevice/ReleaseDevice/QueryDevice wired as KernelCallKind entries with ownership enforcement; proved by [AC-22d-01..08] |
| RFC-00B7 | Pager Service ABI | accepted | PagerService capability + RequestPageMapping / WaitForPagerHandoff / ResumePageFaultedThread fully implemented and tested |
| RFC-00B8 | Governed Foreign Function Interface | accepted | RFC-0036 language surface and VM bridge implemented end to end: integer/string/double/bytes/mixed/StringVectorHandle/IntVectorHandle call shapes proved; required_capabilities wired to SyscallContext::trace_reasons; canonical ffi_call/ffi_quarantine/ffi_policy_deny/ffi_capability/ffi_error audit events in Axion event registry; sandbox boundary resolved (Q6: policy-mediated gate is normative per T81 determinism requirements); 405/405 pass (2026-03-22); post-acceptance: additional ecosystem bindings and stable promotion evidence |
| RFC-00B9 | TernaryOS User Environment Standard | accepted | t81-init, session model, t81sh, service registry, TTY contract, and Axion gates (BootService/SessionCreate/ServiceSpawn/ShellExec) are implemented with AC-1..15 passing; T81_ENABLE_TERNARYOS is now default-ON as of 2026-03-22 |
| RFC-00BA | llama.cpp GGUF Ingestion Bridge | accepted | The narrow llama.cpp-backed GGUF bridge is now accepted in-repo: build-gated metadata enumeration, float export, bridge-backed weights import --format gguf conversion, real TinyLlama import evidence, and source-hash/bridge-revision reporting are all implemented while runtime execution remains independent from llama.cpp |
| RFC-00BB | Native Model Architecture Compatibility | accepted | Compatibility-state/profile gating active for all 7 profiles; real TinyLlama GGUF -> .t81w -> VM TWEMBED proof for llama-dense-v1; synthetic GGUF -> .t81w -> VM TWEMBED execution evidence added for gemma, mistral, phi3, qwen2 (2026-03-22, gguf_import_bridge_test.cpp); post-acceptance: real GGUF execution for non-llama families once reference models are available |
| RFC-00BC | TernaryOS EL0 Userland Bring-Up | accepted | Foundational bring-up program for the freestanding EL0 lane; parent roadmap for the later 00BD..00CE chain |
| RFC-00BD | KernelCall ABI Ordinal Freeze | accepted | Freezes the freestanding KernelCall ordinal map; this is the contract that later 00C* RFCs activate rather than redefining |
| RFC-00BE | Freestanding Cooperative Scheduler | accepted | Establishes the baseline EL0 scheduler/state model for the slice6 lane |
| RFC-00BF | Freestanding KernelCall Observability | accepted | Adds the observability ring used by later CanonFS and scheduler proofs |
| RFC-00C0 | CanonFS Executable Identity | accepted | Freezes T81X executable identity and the first CanonFS-hosted EL0 load proof |
| RFC-00C1 | CanonFS Per-Binary Call Sequence Manifest | accepted | Defines T81M manifest verification keyed by code_hash |
| RFC-00C2 | Hardware-Interrupt-Driven WaitForDevice Wake | accepted | Activates IRQ-driven wake for the freestanding scheduler |
| RFC-00C3 | Axion Async Context Switch Audit Trail | accepted | Adds governance/audit semantics for async IRQ wake and context-switch paths |
| RFC-00C4 | Per-Device Wake Filtering | accepted | Adds device_id discrimination to WaitForDevice wake behavior |
| RFC-00C5 | Concurrent Device Wait | accepted | Proves multiple EL0 threads can park and wake under the same scheduler lane |
| RFC-00C6 | Per-Thread TTBR0 Address-Space Isolation | accepted | Introduces per-thread L3 isolation for the freestanding EL0 lane |
| RFC-00C7 | EL0 Fault Containment | accepted | Defines contained fault handling and retained per-thread fault evidence in the scheduler |
| RFC-00C8 | Concurrent Fault Isolation | accepted | Phase 19 validates the previously untested fs_sched_fault_handler() context-switch branch: faulting tid=7 is contained while healthy tid=6 continues through WaitForDevice(30) and exits cleanly |
| RFC-00C9 | EL0 Fault Evidence Query | accepted | Freezes the minimal post-mortem read path over retained fault_ec and fault_far, strengthening Phase 18/19 proof without widening the EL0 ABI |
| RFC-00CA | EL0 Fault Summary Query | accepted | Activates reserved KernelCall ordinal 21 in the freestanding bridge and proves an EL0 sibling can query retained fault summary state immediately after fault-handler handoff |
| RFC-00CB | EL0 Fault Detail Query | accepted | Activates reserved KernelCall ordinal 15 in the freestanding bridge and proves an EL0 sibling can query retained subject_tid / fault_ec / fault_far detail immediately after fault-handler handoff |
| RFC-00CC | EL0 Fault Acknowledgement and Drain | accepted | Activates reserved KernelCall ordinal 16 in the freestanding bridge and proves an EL0 sibling can drain retained fault state deterministically after querying it |
| RFC-00CD | Supervisor Fault Recovery Status | accepted | Activates reserved KernelCall ordinal 23 in the freestanding bridge and proves supervisor recovery remains pending even after thread-level fault drain |
| RFC-00CE | Supervisor Fault-Group Acknowledgement | accepted | Activates reserved KernelCall ordinal 17 in the freestanding bridge and proves supervisor recovery moves from pending to acknowledged after thread-level drain |
| RFC-00CF | Slice6 CanonFS Operator Actions | draft | Narrow operator-shell surface over existing CanonFS/runtime machinery; treat follow-on shell command work as implementation unless it freezes a new stable contract |
| RFC-00D0 | Base-81-Aware TCP/IP Stack | draft | Initial architectural direction for standards-compatible TCP/IP with base-81 identity, representation, and observability layers |
| RFC-00D1 | CanonFS Foreign File System Interchange | draft | Defines CanonFS import/export as a governed interchange boundary with provenance, normalization, and explicit round-trip limits; RFC remains draft overall, but a narrow seed contract is now stable enough to build against: canonfs import / canonfs export, the v1 import/export/provenance/manifest JSON schemas, host-file / host-directory, and the current policy-profile names. Deferred questions still include symlink posture, archive/bundle export, text-output contract status, and schema-catalog promotion. |
| RFC-00D2 | DAIOS Target Architecture and Sequencing | draft | Informational planning RFC that defines the target DAIOS layers and sequencing rule while explicitly preserving the current runtime-first posture: architecture ahead, implementation behind proof. It does not promote new APIs or authorize broad subsystem expansion. |
| RFC-00D5 | Checked Floating-Point Conversion | draft | Proposed numerical contract: checked binary64 bridges, finite representation, exact/rounded/error outcomes and migration decisions. No production or ISA changes; maintainer approval pending. |
These series are the right planning unit for backlog work. Do not treat every increment inside them as grounds for a new RFC.
RFC-00BC..00CE is one coherent program:
bring-up, ordinal freeze, scheduler/observability, CanonFS identity,
IRQ wake/audit, isolation, fault lifecycle, and supervisor recovery.RFC-DPE-0001..0009 is a closed execution series for task/epoch semantics.RFC-0034, RFC-0037, RFC-0038, RFC-0039, RFC-00BA, and RFC-00BB
together define the model-ingestion and compatibility lane.| RFC | Title | Status | Notes |
|---|---|---|---|
| RFC-DPE-0001 | Deterministic Parallel Execution — Vision and Motivation | accepted | Informational; vocabulary and principles for the DPE series |
| RFC-DPE-0002 | TISC Task Graph Primitives | accepted | All §5 criteria met; DpeTaskRunner + [DPE-02-05] (single-task epoch ≡ direct TISC execution) proved in Slice 14 via task_runner_test |
| RFC-DPE-0003 | Epoch Execution and Canonical Commit | accepted | [DPE-03-01..04] met (Slice 15); kernel wiring + [AC-22s] (Slice 16); [DPE-03-06] met via KernelEpochPolicyGate + EpochAbortedPolicyFault audit event (Slice 17); [DPE-03-05] met: DeltaRecord::word_tags + FloatHandle canonical serialization + intern_float() + t81_dpe_float_reduction_test 21/21; all 6/6 criteria met; [DPE-07-01..06] met (Slice 21): SubmitEpoch KernelCallKind ABI boundary fully wired through axion_kernel_call() |
| RFC-DPE-0004 | DAG-Ordered Multi-Task Epoch Execution | accepted | [DPE-04-01..04] met (Slice 18): topological_sort_epoch() + DpeTaskInputSnapshot + predecessor delta propagation; proved by t81_dpe_epoch_dag_test |
| RFC-DPE-0005 | Level-Parallel Epoch Execution | accepted | [DPE-05-01..04] met (Slice 19): topological_levels_epoch() + std::thread level dispatch + noexcept fallback; proved by t81_dpe_epoch_parallel_test |
| RFC-DPE-0006 | Bounded Thread Pool for Epoch Execution | accepted | [DPE-06-01..04] met (Slice 20): DpeThreadPool (N workers, queue, idle-wait, shutdown) + optional pool param in axion_kernel_submit_epoch(); proved by t81_dpe_thread_pool_test |
| RFC-DPE-0007 | Epoch Execution Timeout | accepted | [DPE-08-01..06] met (Slice 23): optional timeout_ms param on axion_kernel_submit_epoch(); per-level steady_clock check; Aborted_Timeout → RetryLater/EpochTimedOut at syscall boundary; proved by t81_ternaryos_epoch_timeout_test |
| RFC-DPE-0008 | Epoch Audit Events | accepted | [DPE-09-01..06] met (Slice 24): EpochSubmitted/EpochCommitted/EpochAborted wired into axion_kernel_submit_epoch(); epoch_audit_submissions/commits/aborts counters + last_epoch_audit_kind/sequence retained state; proved by t81_ternaryos_epoch_audit_test |
| RFC-DPE-0009 | Epoch History Ring | accepted | [DPE-10-01..06] met (Slice 25): EpochHistoryRecord + kEpochHistoryCapacity=8 ring in EpochRuntimeState; populated on commit; evicts oldest on overflow; exposed as vector in KernelRuntimeStatusView; proved by t81_ternaryos_epoch_history_test |
| RFC | Title | Status | Notes |
|---|---|---|---|
| RFC-00A0 | AI Experiment Sandbox | superseded | Superseded by RFC-0032; 3-stage promotion lifecycle executed to completion |
| RFC-00A1 | Deterministic Evidence Protocol | superseded | Superseded by RFC-0032 §5; evidence-schema-v1 (FNV-1a, key=value, no timing) replaces SHA-256/JSON/wall-clock design |
| RFC-00A2 | AI Benchmark Specification | accepted | 6/6 AC met: VM throughput benchmarks, CanonHash81 determinism validation (score=1.0), Google Benchmark JSON output, t81 internal benchmark CLI; BM_DeterminismValidation.cpp added (2026-03-16) |
| RFC-00A3 | Model Artifact Provenance | accepted | Implemented as TLOADHASH + allowed_tensor_hashes + model_load Axion event (RFC-0025) |
| RFC-00A4 | Ternary Quantization Codec | accepted | quantize_threshold() + pack_ternary_to_base81() promoted per RFC-0032 Phase 1; float metrics removed |
| RFC-00A5 | LLM Backend Adapter | superseded | Superseded by RFC-0032 Phase 4; T81VmBackend (deterministic VM-only) replaces llama.cpp/ONNX vision |
| RFC-00A6 | Axion Policy Hooks | accepted | AIHookEngine + PolicyEngine live per RFC-0032 Phase 3; event registry in spec/supplemental/axion-event-registry.md |
| RFC-00A7 | UX Integration | superseded | Superseded by RFC-0033 (t81 agent/t81 studio) + existing CLI; t81 ai command hierarchy not implemented and not on roadmap |
| RFC-00A8 | AI Native VM Opcodes | superseded | Superseded by RFC-0026; ATTN/QMATMUL/EMBED/WLOAD live in TISC ISA |
t81::T81); RFC-0016 is formally superseded. RFC-0018 documents the add/mul arithmetic implementation.RFC-00BC..00CE should be planned as a single
architectural series. New RFCs in this area require a new stable contract,
not just another phase, command, or proof step.spec/rfcs/template.md.