Repository navigation
Audit existing T27C + T27B + T27O stack before proposing new architecture #1644
Description
Purpose
Audit and document the capabilities that already exist in gHashTag/t27 before proposing any new canonical-specification, reference-implementation, compiler, or code-reuse architecture. This is an audit task, not a request to build a parallel system.
Required reading / surfaces
t27c: parser, type checker, HIR and current code-generation backends; inspect the compiler implementation and generated-code/test gates.t27b: native AArch64 backend, IR, interpreter, JIT/object output, current supported subset and host limitations.t27o(Ouroboros): Rust-to-T27 converter, conversion refusal rules, oracle vectors/tests, DRY reuse, seals and provenance; inspect epic #8998 and its existing golden criteria.- Existing reference specs, board-side twins, compiler ports, corpus sweeps, and conformance/oracle infrastructure.
Initial evidence already located
specs/tools/t27o/rust_convert.t27documents a deterministic Rust subset converter and its oracle pipeline.specs/tools/t27o/golden.t27defines evidence, equivalence on recorded vectors/tests, DRY, sealing, and provenance; explicitly says this is not a proof for all inputs.- Recent t27o ports convert real compiler functions and route the hand-written Rust wrappers to generated T27-derived Rust; the HIR cost-model port documents tests, corpus sweeps, and limitations of each test layer.
docs/now/2026-10-04-t27b-native-aarch64-backend.mddescribes a native AArch64 backend reusing t27c parsing/type-checking, with its own IR/interpreter/encoder/JIT/object writer and host constraints.- Trinity already has a live t27c WebAssembly stage explorer and says accurately that backends do not all follow a single linear pipeline.
Deliverables
- A concise architecture map of T27C, T27B and T27O and their actual boundaries.
- A table separating implemented behavior, measured behavior, documented limitations, and unverified assumptions.
- A list of the existing tests/oracles/conformance gates and any uncovered gaps.
- Recommendations that extend existing mechanisms only; identify duplicates and avoid parallel implementations.
- No new implementation until this audit identifies a concrete missing capability and its acceptance test.
Acceptance criteria
- Every claim links to an existing source file, issue, test, or reproducible measurement.
- Do not describe existing work as absent just because one search query misses it.
- Do not create a duplicate compiler/provider/reference architecture.
- Separate conversion evidence on sampled vectors from proof of equivalence for all inputs.
Activity
[-]Research principle: T27 canonical specifications, golden reference implementations, and measured physical limits[/-][+]Audit existing T27C + T27B + T27O stack before proposing new architecture[/+]dharmarooms commentedon Oct 16, 2026
T27 agent-domain audit — start at Z, then A (2026-10-16)
Repository audit confirms that the current alphabet is already defined in
docs/agents/AGENTS_ALPHABET.mdv3.0 (2026-04-07), with generated per-agent source-of-truth specs underspecs/agents/*.t27. Do not create another agent roster before reconciling this canon with the current architecture.Existing definitions confirmed
specs/agents/a.t27; domainArchitecture / ADR / SOUL; owns architecture/constitution decisions, ADR consistency and composability. It currently declares no exclusive skill/tool cards.specs/agents/z.t27; domainZero-Touch UX / Docs; skills includetrinity/blog-post,trinity/doctor,trinity/queen-hive-visuals,trinity/ux-wave; tools includemcp/tri-ssot,tri/now.specs/agents/c.t27; domainCompiler Core / AR Language; 19 tools are currently attributed to compiler/spec analysis and emission. Do not silently repurpose C as Security.Reserve / Security, with security policy / AAIF compliance as invariants, but the row has no concrete key-file artifact. This is a likely candidate to review for security ownership, not evidence that a complete security agent already exists.Initial domain decisions
t27/Cto avoid a domain collision.Architecture rule for the full audit
The 27 slots are a fixed capacity, not a requirement to invent 27 unrelated job titles. Each domain needs: a unique decision boundary, owned artifacts, entry/exit invariants, tools/skills with provenance, escalation rules to Queen, and evidence that the domain is distinct. Shared cross-cutting responsibilities (security, evaluation, memory, identity, orchestration) must not be duplicated across several agents without a clear owner.
Next audit steps