{
  "contract": "this hash is committed to git BEFORE the benchmark runs; a matching observed outcome then proves the prediction was pre-registered, not fit post-hoc.",
  "name": "bpf_kernel_load_bar",
  "note": "S06 candidate 6. Sealed before any load attempt. The committed artifact records kernel_loaded false; this candidate tests whether that gap is closable on this host.",
  "prediction": {
    "bar": {
      "FAIL_publishable_as": "'the Linux verifier rejects our compiled monitor, and here is its log' -- a rejection with the kernel's own reasoning is a result about our compiler's output and is published as one, not as an absence.",
      "PASS": "the real object is ACCEPTED by the in-kernel verifier (a program fd is created) AND the malformed control is REJECTED, with both verifier logs captured verbatim into the artifact.",
      "no_field_writing": "kernel_loaded and in_kernel_verifier_run are written by the loader from the observed outcome only. If the loader cannot run, the fields stay false and the artifact stays red with a note."
    },
    "both_outcomes_publishable": true,
    "candidate": "S06-6 the eBPF object meets the Linux kernel's own verifier",
    "committed_state_being_tested": {
      "in_kernel_verifier_run": false,
      "kernel_loaded": false
    },
    "design_constraint": "the loader is written as a REUSABLE producer that takes any BPF object path, not a monitorgen-welded script, so a sibling lane can point it at its own object.",
    "kill_criterion": "Docker Desktop's Linux VM denies the load (no CAP_BPF/CAP_SYS_ADMIN effect, or bpftool unavailable in every image tried): record NOT_RUN with the denial verbatim, no retry budget beyond two images.",
    "negative_control_run_first": "a deliberately malformed BPF object -- the real object with its instruction stream truncated mid-instruction -- must be REJECTED by the kernel verifier. If the kernel accepts the malformed object, the harness is not reaching a real verifier and no result about the real object is reported.",
    "why_this_matters": "this estate's own 2026-08 audit named 'code compiled from our proof passes the Linux kernel's own verifier' as its single best unspent hour, blocked for one reason: the host is a Mac. A privileged Linux container on this host dissolves that blocker."
  },
  "schema": "prereg-seal-v1",
  "sealed_at_commit": "c9cf3c818eef97eec8e0afb1bc3bbc131e4fd2e6",
  "sha256": "f9a13ba976b65806423f802b59373f3b9a9f890472e60317475258411f29e35a"
}
