Repository navigation
SYS REQ 029
The system shall use local Ollama assessments only as bounded, evidence-bound advisory results that cannot bypass deterministic validation or operator authorization.
An installed model, a schema-valid answer, or a completed inference request is not proof that the approved model reviewed all relevant evidence or that the underlying build succeeded. Epic #229 isolates this trust boundary from model selection, LBA capture tooling, VM lifecycle, and native-build orchestration.
An assessment is eligible for use only after the caller-approved model identity, input authorization, evidence coverage, output validity, and applicable execution limits have been checked. Failed, cancelled, incomplete, ungrounded, and identity-mismatched results are distinguishable from accepted assessments. No result grants tool authority, changes an approval, overrides a deterministic failure, or establishes native-build proof. Advisory does not mean optional: a feature requiring model review, including REQ-108, cannot claim validity when review is absent or invalid. A bounded sample supports conclusions only about that sample. Private evidence and retained results follow the approved disclosure and retention policy; a shared daemon and model store remain intact on failure. Software allocation covers identity, aggregation, budgets, frame attribution, privacy, lifecycle, and resumption; each allocation records its actual implementation, test basis, supported platform and remaining live-proof limits.
Cross-adapter offline boundary coverage associated with TEST-118 is recorded through TEST-111, TEST-112, TEST-113, TEST-114, TEST-115, TEST-116, and TEST-117. The original integrated epic passed 2,142 Python and 91 source-config .NET tests, with seven applicable checks passing for every PR #237 through #243. Follow-up REQ-118 / TEST-119 and REQ-119 / TEST-120 integrated through PRs #245/#247; merge fbd69b0ae6480ae97b725c325b4ad05fe5189ba1 passed 2,166 Python and 108 .NET tests and all seven PR #247 gates. REQ-120 / TEST-121 integrated through PR #249, merge 69db2c8616f976cdab207a78c715430204697dcb; the current full suites pass 2,171 Python and 108 .NET tests and all seven PR #249 gates passed, including Windows and Ubuntu. A separately authorized CPU-only public synthetic transport smoke passed identity, budget, completion and postflight checks without private inputs; neither metadata-only live L1 nor L2 yielded a successful assessment. L2 preserved a private inference-deadline refusal with zero retries and unknown server completion, not a final receipt. The offline planner reuses canonical cost/hash calculations, but performs no inference and makes no latency or model metadata compatibility claim. These are component/contract regression and bounded transport/diagnostic observations, not end-to-end live desktop/capture, hostile-host/model attestation, native LabVIEW build, stable release, or universal tokenizer/resource-control qualification. POSIX checkpoint custody is implemented; unsupported platforms fail closed. Refusal custody is POSIX for the Python coordinator and supported Linux/Windows private paths for source-config; recording a refusal never grants retry or VM authority.
System-Requirements-Specification; epic #229
High
TEST-111, TEST-112, TEST-113, TEST-114, TEST-115, TEST-116, TEST-117, TEST-118, TEST-119, TEST-120, TEST-121
tools/source-config/ProjectBuddy.SourceConfig.Tests/GenerationTests.cs and ConfigurationTests.cs; scripts/tests/test_assisted_benchmark_contract.py, test_assisted_benchmark_capture.py, test_assisted_benchmark_review.py, and test_assisted_benchmark_run.py; .github/workflows/ci-assurance.yml and the documentation-link gate. Each allocated software requirement identifies concrete methods and reproduction commands. Epic #229 holds the batch authority, integration links and bounded-smoke evidence; no historical run ID alone is treated as permanent proof.
REQ-109, REQ-110, REQ-111, REQ-112, REQ-113, REQ-114, REQ-115, REQ-118, REQ-119, REQ-120
Implemented components and offline cross-adapter contract coverage verified; bounded public transport smoke passed, broader live/system/release qualification not claimed