Secure entry runtime report
Stage 1 emits <executable>.secure-runtime.json. This is a distinct UTF-8 JSON format, not the Stage 2 .secure-entry.json production receipt. Its fixed report_kind is sigil-secure-entry-runtime, schema_version is 1, assurance is provenance-unverified, and production_receipt is false. Unknown versions, missing fields and additional fields are rejected.
The report records compiler revision and executable digest, effective config identity/digest, fixed entry/hook identity, input closure marker digest and final image digest. It pins the x86_64 Linux musl secure-native-v1 profile, stock allocator and sigil-secure-entry-v2 native provider. Handoff and C spawn ABI versions remain 1. The launch primitive is clone3/pidfd with descriptor-only execveat.
The compiler revision is its verbatim embedded Git stamp: a lowercase hex abbreviation or full revision, optionally followed by -dirty. It is never expanded using the application's checkout, and a dirty stamp is never presented as clean. This self-reported stamp is not source attestation; the executable digest identifies the actual compiler bytes. The Stage 2 receipt retains its separate full-revision requirement.
Six inventories are inline JSON values with individual canonical SHA-256 digests: modules, constructors, runtime state, context schema, capability exclusions and closure tests. They preserve ordered startup observations and actual test privilege modes/results. The report is canonical JSON without a trailing newline. Digests use lowercase sha256: strings. It contains no timestamp or ambient host identifier.
When a final package declares static application providers, the module inventory includes native_providers: ordered module/library/initializer identities, exact public name/implementation/encoded-arity triples, observed contained C source hashes and include directories, and owned object/function code identities. These values enter the module digest and retained input closure marker. The top-level fixed native_provider and count still describe the secure-entry provider; they do not summarize application declarations. An observed source or object hash does not establish reviewed provenance.
The initial schema records both real-setuid and root-init qualification as not-run. Unprivileged and forced-AT_SECURE results cannot become privileged qualification. A public build does not invent successful closure tests: an empty test inventory means no per-image execution tests are attached. Consumers must run their required tests and review their results before enablement.
Structural validation and captured-byte binding only prove that the report is well formed and its hashes agree with the supplied image/inventories. They do not establish ELF policy, startup order, hardening, source/toolchain provenance, test completeness, security review or installation authority. The public build producer separately performs the supported runtime/image checks. Schema 1 cannot claim Stage 2 provenance or privileged qualification.
The application must explicitly accept this Stage 1 contract. A consumer that requires the Stage 2 production receipt must reject this report. Its trust in the compiler, SDK, build host and hook is explicit. Authentication, PAM/NSS, executable/identity authorization, TTY/environment/resource policy and final installation remain application responsibilities. Real privileged VM runs remain a separate qualification step.