Living document — updated as phases land. The planned layout comes from PLANNING.md §39; crates marked (planned) do not exist yet.
opencodifier/
├── Cargo.toml # workspace: shared lints, deps, release profile
├── rust-toolchain.toml # local dev channel (stable)
├── rustfmt.toml # stable-only settings
├── deny.toml # cargo-deny: licenses/advisories/bans/sources
├── justfile # task runner; `just ci` = the pre-push contract
├── .gitforge.yml # GitForge CI — the pipeline of record
├── .github/workflows/ci.yml # GitHub mirror config (not a failure signal)
├── .aegis/baseline.json # aegis pattern-scan baseline (new findings = failures)
│
├── infrastructure/
│ └── docker/ci-rust.Dockerfile # runner-local CI image recipe (`just ci-image`)
│
├── scripts/ # single-purpose validation + release tooling
│ ├── e2e_validate.py # release-binary e2e over the real wire (PLAN 18d)
│ ├── coverage_floor.py # lcov DA-basis floor assertion (CI coverage lane)
│ ├── generate_attestation.py # verified per-artifact build attestation (schema opencodifier.attestation/1)
│ └── generate_release_notes.py# renders notes from attestations + docs/RELEASE_NOTES_TEMPLATE.md
│
├── dist/ # release staging (gitignored): matrix artifacts,
│ # attestations, rendered notes (`just release-matrix`)
│
├── crates/
│ ├── opencodifier-core/ # canonical decision IR (done)
│ ├── opencodifier-schema/ # native/OpenAI/Anthropic/Jev adapters (done)
│ ├── opencodifier-engine/ # DAG executor, rules, caches, narrowing, BM25,
│ │ # relational solver, ladder + calibration seams (done)
│ │ # facts.rs = the extraction grammar itemgen renders against
│ ├── opencodifier-itemgen/ # D37 relational/contrastive item generator:
│ │ # sample → render → solver-verify (binary `itemgen`;
│ │ # corpus emission only, never a serving dependency)
│ ├── opencodifier-runtime/ # InferenceBackend traits, mocks; `onnx` deferred by gate (done, D2)
│ ├── opencodifier-model/ # decision serving contract + embedding classifier + manifests (done)
│ │ src/decision.rs, embedding.rs, manifest.rs — serving contract
│ │ src/kai.rs + kai/ — kai contract + committed fixtures
│ │ src/julia.rs + julia/ — julia contract + committed fixtures
│ │ src/nli.rs + nli/ — NLI verbalization verifier contract + fixture (PLAN 24j, `nli` feature serves it)
│ │ src/render.rs — shared answer-key/question-text rendering (llamacpp + nli)
│ │ src/llamacpp.rs — llama.cpp loopback backend (D26, `llamacpp` feature)
│ ├── opencodifier-http/ # axum 0.8 /v1 API (done)
│ ├── opencodifier-cli/ # clap CLI binary `opencodifier` (done)
│ ├── opencodifier-mcp/ # rmcp 2.2 stdio server, six codify_* tools (done)
│ ├── opencodifier-wasm/ # wasm32 binding: WasmEngine over the zero-ML stack (done; D23)
│ └── opencodifier-ffi/ # C-ABI staticlib over the same wire module (done; D24)
│
├── fixtures/ # wire-format fixtures per adapter (done, byte-locked)
├── recipes/ # runnable decision graphs + captured responses (done)
├── skills/ # agent-facing usage skills: 8 §33 areas (done)
├── models/ # model artifacts (never committed; SHA-256 verified;
│ # created on demand by the models/CLI path)
├── benchmarks/ # per-release criterion baselines (first committed: D9, 2026-10-01)
│ ├── baselines/criterion/ # engine + http d9-baseline estimates (committed from B1 on)
│ ├── decision-model/ # Phase 13 model-pick benchmark: byte-locked suite, arms, results
│ │ results/REPORT.md # comparison page of record; docs/BENCHMARKS.md links here
│ └── validation/ # validation-campaign record (#104; VALIDATION.md summary)
│ └── runner/ # compute-host parity runners (own workspaces; ORT via
│ # runtime-onnx-dynamic + ORT_DYLIB_PATH): kai-parity-rs,
│ # julia-parity-rs, nli-parity-rs (PLAN 24j)
├── site/ # static marketing page (Cloudflare Pages direct-upload;
│ # Tailwind runs as a dev-time pipeline only —
│ # `npm run build:css` writes the committed tw.css)
│ ├── sdk/decisions-sdk.js # the Decisions SDK the playground games share:
│ │ # seeded rng, wire-contract builders, engine
│ │ # bridge (WASM), session boards, meters, shell kit
│ ├── *-core.js # one DOM-free, node-testable core per game
│ │ # (pong, life, pacman, racing, fighter, quiz,
│ │ # tower, antcolony, civ60, traffic, spacecraft,
│ │ # dilemma, rpsls, door, cube + cube-n, clock,
│ │ # pyraminx) — seeded, deterministic; the twisty
│ │ # cores carry proved solvers, not heuristics
│ ├── <game>.js # per-game rendering shells (canvas/DOM chrome;
│ │ # rendering never touches an rng stream)
│ ├── playground.html # merged playground: maze demo + game library
│ └── tests/ # node --test suites: board page, maze
│ # core/shell/deep-link, SDK games, cube cores,
│ # clock (27 files)
└── docs/ # PLANNING.md (founding), PLAN.md (live plan), …
# RELEASE_NOTES_TEMPLATE.md is the release-notes
# skeleton; `just release-notes` fills its
# mechanical slots from the dist/ attestations
cli ──► http ──┐
├──► engine ──► core ◄── schema
model ──► runtime ──► core
└────────► engine
itemgen ──► engine ──► core (itemgen also declares the direct core edge)
(mcp/wasm, when they land, take the same engine-facade edge as cli/http)
opencodifier-core must never depend on HTTP, MCP, CLI, Tokio, or any ML
runtime (PLANNING.md §39). Heavyweight dependencies are each confined to
exactly one crate — see docs/DECISIONS.md (feature-flag matrix).
- Modules:
snake_case.rs; one concern per module; module docs cite the PLANNING.md § they implement. - Tests: unit tests in
#[cfg(test)]modules (allow list:clippy::unwrap_used, clippy::expect_used, clippy::panic, clippy::float_cmp), integration tests intests/, property tests withproptest, benchmarks inbenches/(criterion,harness = false). - Errors: one error enum per crate,
thiserror-derived, with a stablecode()string (seecrates/opencodifier-core/src/error.rsfor the pattern).anyhowis banned from public APIs. - Every public item is rustdoc-documented;
missing_docs+ CI-D warningsmake that enforceable.