Feature audit — September 7, 2026

This is a historical audit of the source revision below, not the current feature inventory. Its Phase 1 follow-ups were subsequently resolved by the September 8 functional completion; the September 11 extension report records the later optimizations, infrastructure comparisons and Phase 2 handoff. The dated findings and validation results below retain their original meaning.

The audit reviewed development at d5a7692, the Phase 1 milestone and every open issue/PR, with the closed foundation tickets used to check delivered scope. The repository's integration branch is development; there is no dev branch. Corrections are on chore/feature-audit-2026-09-07 in PR #69.

Phase 1 foundations are implemented and have executable regression coverage. The complete standalone release-to-invocation product remains unfinished. Passing the checks below does not close the Phase 1 gate or the outstanding correctness issues found by this audit.

Feature inventory

Feature group	Current implementation and validation boundary	Remaining owner
Build and contracts	Locked Rust workspace, generated WIT/Protobuf bindings, seven JSON Schemas, descriptor checks, cross-language SDK compilation, deterministic test utilities.	#2 and #36 closed; SDK parity follow-up #65.
Declarative manifests	Bounded JSON decoding, exact numeric handling, duplicate-key/depth/collection limits, canonical encoding and stateless Phase 1 validation in latent-manifest.	#3 closed.
Release storage	DirectoryArtifactRepository: exclusive ownership, component digest verification, bounded indexes/pages/recovery, local publication, retry and restart. No guest preparation during catalog operations.	#4 closed; durability/integrity follow-ups #66 and #68.
Deployment and routing	DirectoryDeploymentRepository: atomic persisted deployment state, immutable generations, tenant/namespace checks, deterministic revisions/weighting, pinned snapshots and recovery.	#5 closed; management repository prerequisite #67.
Budgets and cancellation	Effective monotonic deadlines, budget intersection, concurrent consumption/reservations, terminal reconciliation and cancellation registries in core/node/executor.	#6 closed; full runtime integration remains #9–#11.
Admission and scheduling	Phase 0 fixed generic cell pool with bounded FIFO acquisition, cancellation, affine leases, return/quarantine and race tests. General admission and fair scheduling are interfaces/pending work.	#7 and #8.
Wasmtime and capabilities	Real Phase 0 echo/containment backend, fresh stores, fixed engine/cache topology, fuel/deadline/trap/memory containment, context/log imports. Generic dispatch and hardened clock/context/log are pending.	#9 and #10.
Activation orchestration	Phase 0 runner and new budget/cancellation primitives have unit/integration tests. Production orchestration, retained status and cleanup composition are pending.	#11.
Invocation and management RPC	Generated messages, clients/server wrappers and conversion helpers exist. Merged code does not provide the standalone service implementations/listeners.	#12, #37, and #14.
Telemetry and operations	Phase 0 measurement/collector infrastructure exists. Shared Phase 1 telemetry, inventory, health, draining and operational lifecycle are pending.	#13 and #14.
Operator CLI and SDKs	latent and latent-control are placeholders. SDKs define compiling interfaces; they do not implement transports. latentd phase0-spike remains the executable demonstration.	#15 and SDK contract follow-up #65.
Conformance and evidence	Maintained Rust/Python regression tests and real-component fixtures accompany retained Phase 0 evidence. The full Phase 1 CLI/node conformance gate is pending.	#16.
Later phases	OCI/signing, general policy/capability providers, ingress/triggers, blobs, state/effects, commits, workflows, cluster routing and remote invocation remain architectural seams/research.	Roadmap Phases 2 onward; Angular hosting remains Phase 3 #44.

The roadmap, API map and validation baseline now distinguish these implementations from planned contracts.

Minor corrections

• Normalize a submitted capsule manifest before comparing immutable publication retries. A valid reversed import order previously published once but failed an identical retry, including recovery from an uncertain directory sync. A tiny regression covers ordinary duplicates, sync-failure retry and reopen.
• Seek directly to the artifact-list cursor instead of repeatedly traversing every earlier digest. Existing entry/byte-boundary pagination tests remain.
• Move the durable 100,000-release CI step behind manual workflow_dispatch input run_catalog_scale, default false. Ordinary catalog/supervisor checks remain automatic.
• Make copied collector-runner test fixtures model native kernel identity consistently on WSL/container hosts. Real collectors retain their native-host restrictions; fake workload tools fail on unexpected execution, and tests exercise WSL/container rejection explicitly.
• Refresh stale feature descriptions, schema count, crate summaries and test documentation. Remove the validator warning that treated any implemented binary as a scaffold violation, retaining rejection of implementation placeholder tokens. Use the existing area:contracts label in the architecture issue template.

GitHub findings and disposition

Finding	Disposition
SDK calls cannot provide the ID needed to cancel/inspect their own unfinished invocation; C cancellation lacks transport-error separation.	#65, Phase 1, priority P1, size M.
New artifact roots do not synchronize the parent chain that makes acknowledged releases reachable after a crash.	#66, Phase 1, priority P1, size M. Source-level durability finding; no power-loss experiment performed.
Deployment RPC generation preconditions/receipts and tenant pagination cannot be implemented atomically through the current repository port.	#67, Phase 1, priority P1, size L; extracted repository prerequisite of #37.
A valid-JSON one-byte mutation changes persisted release metadata without changing its digest or causing reopen/fetch rejection.	#68, Phase 1, priority P1, size M. Confirmed with one tiny production-API artifact in 0.04 seconds.
Weighted routing integration lacks an explicit key-source decision.	Clarified #11: define derivation/default/retry semantics and small deterministic resolution tests. A new wire field is unnecessary unless caller-controlled affinity is intended.
Budget-wrapper normal-return finalization is insufficient for dropped/panicking futures.	Recorded against the existing cleanup scope of #11; no duplicate implementation ticket.
Epic #1 marks open management ticket #37 complete.	Corrected its checkbox and the completed Phase 0 handoff; added #65–#68 as completion dependencies. #16 and #37 reference their new prerequisites.
PR #52 implements an older alternative for already-completed #5.	Closed as superseded by merged #64; its branch is retained.
PRs #53 and #54 remain open.	Retained for #13/#12. At review, #53 conflicts with development; #54 is mergeable but blocked. Neither is counted as delivered functionality.

New issues use the repository's existing area/size/priority labels and the Phase 1 — Single-Node Stateless Fabric milestone. General SDK transports, signed supply chains and other later-phase features were not added as Phase 1 requirements.

Validation evidence

Validation used the existing pinned Linux toolchain container on Docker/WSL, with a copied checkout and an isolated target directory. Windows compilation was also attempted: catalog directory synchronization fails with access denied, and the full shell/archive suite requires Linux tools. Those results are not a Windows support claim; Linux is the documented persistence/reference platform.

Check	Result
Formatting and whitespace	cargo fmt --all --check and git diff --check pass.
Ordinary Linux workspace tests	214 passed; no failures. Command below excludes the 100,000-entry index unit and large metadata memory probe in addition to the default ignored tests.
Workspace Clippy	Passes with existing warnings; this audit does not claim a warning-free workspace.
Exact SDK toolchains and surfaces	tools/validate_sdks.sh passes for Go, TypeScript, Java, .NET and Zig/C; Rust SDK is included in workspace tests.
Repository/contracts Python suite	All 186 tests pass; repository and foundation validators and Phase 1 descriptor validation pass.
Collector fixture correction	60 focused Python tests pass, including every previously failing Linux/WSL fixture.
Real Component Model contracts	tools/validate_contracts.sh passes: WIT/Buf/generated bindings, byte-identical echo rebuilds, three echo/backend checks and four real containment checks.
Executable outcome/recovery matrix	The explicitly selected latentd Phase 0 end-to-end test passes with the generated fixtures, covering success, failures and same-runtime recovery.
Workflow syntax	Actionlint 1.7.12 accepts the updated CI workflow.

The lightweight workspace command was:

cargo test --workspace --all-targets --all-features --locked -- \
  --skip one_hundred_thousand_index_adoptions_are_bounded \
  --skip large_release_metadata_has_a_bounded_compilation_working_set

The 100,000-release publication/reopen acceptance probe, 100,000-activation soaks, native calibration/profiling and full Phase 0 authorization gate were not run. Reading and validating retained archives does not generate replacement benchmark evidence. The existing successful development CI run 34051729068 is historical context, not evidence that this audit reran its heavy job.