We need one written, reviewable capability response for the recorded system configuration: macOS 27.0.1, build 26A434; Darwin 27.0.0; xnu-13432.1.9~1/RELEASE_ARM64_T8103; arm64. These are retained host observations, not a new claim of kernel/source equivalence.
The target is an application-owned non-platform Conda CPython process using its original subprocess.run/Popen calls with stdin/input omitted. We must keep System Integrity Protection enabled, preserve the original interpreter and global Python settings, and preserve executable, argv, environment, Path objects, stdin choices, timeouts, return/exception behavior and cleanup. We cannot replace the launch with a debugger-owned PTY, DEVNULL, wrapper, or changed timeout.
A single harmless-target capability attempt has already been consumed. Authentication succeeded; the target's process credentials were uid/euid 501, gid/egid 20, with process group vector [20]. In that exact attempt DTrace reported at line 11: "probe description syscall:::entry does not match any probes. System Integrity Protection is on". No trace records or begin/end receipts were obtained; the target gate stayed closed and both recorded children were reaped during bounded cleanup. The diagnostic script path is intentionally omitted here; the original local raw text is preserved. We are not attributing every possible native-observation limitation solely to SIP, and we will not rerun that attempt.
The single requested deliverable is CAPABILITY-RESPONSE.json (or one equivalent signed/versioned document), answering whether a named, supported mechanism on this exact configuration provides EITHER:
A. A synchronous record of each original child's descriptor-0 underlying object at the accepted fork/exec boundary, including type, object/device identity and descriptor/access flags; OR
B. A complete no-transient-change proof from an independently accepted descriptor-0 identity anchor over the entire original interval, covering every relevant thread and child from birth without a registration race.
Please include these sections in the ONE response:
- disposition: SUPPORTED / UNSUPPORTED / NOT_ESTABLISHED; route A or B; responding organization/product, supported API/tool and exact version; links to public contract or a reviewable vendor coverage statement applicable to this OS/build/architecture.
- permissions: exact privileges, supported signing/entitlement eligibility and installation requirements; explicit statement that SIP and global Python stay unchanged. Do not assume administrative privilege alone gives kernel visibility.
- boundary_and_scope: how parent/child stable identities, fork/exec phase, all relevant threads and descendants, birth-registration ordering, PID reuse and cross-CPU ordering are handled; when capture starts and stops relative to the original anchor/call.
- coverage_and_loss: the exact guaranteed events/object fields; proof or documented guarantee of coverage; missing-event/drop/buffer-overflow/sequence-restart/error detection; the independent acceptance rule must reject any unknown interval or loss. For route B, enumerate applicable descriptor writers including direct/noncancel/guarded syscalls, open/close/dup/fcntl, received descriptors/fileports, spawn/exec file actions and close-on-exec, plus concurrency/cancellation/signal/startup paths. A short illustrative syscall list is insufficient.
- raw_contract: a minimal raw-record schema or already existing reference example, with observed fields distinguished from derived fields; stable process identity, phase, timestamps/sequence domain, descriptor/object identity, raw returns/errno, and coverage/error status. A source-version match or pathname alone is not an object-identity witness.
- preservation: how original call arguments, stdin/input omission, results/exceptions, original timeouts, original PID/wait/reap/raw streams and cleanup remain intact, and any observer effects. We cannot add waits, kills, reads or retries to fill missing original evidence.
- prospective_validation: a finite, independently reviewable validation plan only, with success criteria and false-acceptance controls for missing records, descriptor replacement/ABA, unobserved early children and stream loss. No launch, sudo command, new probe iteration or installation is authorized by this request.
If no supported mechanism satisfies either route, please say UNSUPPORTED and name the missing public capability; if applicability is unknown, say NOT_ESTABLISHED. Do not substitute a generic follow-fork setting, a late process/fd snapshot, an EndpointSecurity descriptor subset/type list, successful compilation, source version numbers, or a recommendation to disable SIP/change stdin.
A positive written response permits only independent design/source/coverage review. It does not prove our actual descriptor-0 chain and grants no project verification credit. Only after pre-review passes could a separately bound, new finite observation request be considered. No historical missing data will be reconstructed or old completed validations repeated.
This is one specific supported-API/capability question, not a request for Apple to develop our collector or disclose private implementation. No repository sources, account secrets, credentials, private paths or runtime databases are attached. No existing observation tool or build materials are requested from our end user.