{
  "schema": "cain.sovereign.evolution-01.contribution.v1",
  "contributor": "cain-studio-platform-17",
  "generated": "2026-10-12",
  "baseline_revision": "6cfa3b9e465a35b846311e53503aae509fd457d7",
  "post_change_revision": "c0b6a64b",
  "scope": [
    "cain/cain_z3_verifygate.py and platform-gateway/cain_z3_verifygate.py (identical copies) + /verifygate/* routes in platform-gateway/routers/subdomains.py",
    "alternative execution routes that do not pass through MCPGate (gateway HTTP routes that execute, spawn or invoke; cain/guard.py SDK path)",
    "host enforcement reality (eBPF tracepoint vs kprobe-override deny, seccomp)",
    "evidence ledger checks (MCPGate evidence digests, substrate proof-run chain)"
  ],
  "deployed": true,
  "deployment_note": "Rolling deploy deploy/20261012T034805Z at HEAD bf225092 by -17, owner-approved, including --allow-uncommitted: 9 pre-existing uncommitted gateway files that blue had already been serving since the 02:03 host reboot. Green (8098) had been stopped since 01:48 and was restarted by the deploy (its systemd unit is still disabled). Verified on 127.0.0.1:8088 and :8098 directly and 6x per domain on cainstudio.online / mcpgate.online (verifygate unknown ceiling -> INVALID_INPUT) and 4x per domain on all 3 domains (anonymous POST /fabric/continuum/verify/run -> 403; GET -> 200); no harness worker was spawned.",
  "tests": [
    {
      "label": "baseline, existing authorization / Z3 / MCP proxy / guard suites (16 files)",
      "command": "python3 -m pytest -q -p no:cacheprovider -o addopts=\"\" tests/test_cain42_pcaa_fail_closed.py tests/test_cain_trust_kernel_3.py tests/test_cain42_phase3_z3_verifygate.py tests/test_sovereign_proof_properties.py tests/test_sovereign_differential.py tests/test_evolution2_breaker.py tests/test_evolution2_properties.py tests/test_cain_mcp_proxy_jit.py tests/test_mcp_proxy_capability_forgery.py tests/test_mcp_proxy_self_issued_capability.py tests/test_mcp_proxy_midflight_revocation.py tests/test_mcp_proxy_model_continuity.py tests/test_guard_default_path_bypasses.py tests/test_mcpgate_proxy_gate_proofs.py tests/test_cain_guard_jit_integration.py tests/test_cain_jit_capability.py",
      "exit_code": 0,
      "passed": 174,
      "failed": 0,
      "skipped": 0,
      "revision": "6cfa3b9e (clean for every file these suites import)",
      "environment": "Python 3.14.4, z3 5.1.0, pydantic 2.12.5, Linux 7.0.0-38-generic"
    },
    {
      "label": "fail-before: new verifygate regression test against the unfixed engine",
      "command": "python3 -m pytest -q -p no:cacheprovider -o addopts=\"\" tests/test_sovereign_e1_verifygate_fail_closed.py",
      "exit_code": 1,
      "passed": 4,
      "failed": 26,
      "skipped": 0,
      "revision": "6cfa3b9e engine + new test file",
      "note": "the 4 passing cases assert unchanged valid behaviour (normalised spelling, valid proofs)"
    },
    {
      "label": "fail-before: new continuum regression test against the unfixed router",
      "command": "python3 -m pytest -q -p no:cacheprovider -o addopts=\"\" tests/test_sovereign_e1_continuum_run_operator_only.py",
      "exit_code": 1,
      "passed": 2,
      "failed": 3,
      "skipped": 0,
      "revision": "3537d76e router + new test file"
    },
    {
      "label": "post-fix: new tests + every suite that imports the verifygate engine",
      "command": "python3 -m pytest -q -p no:cacheprovider -o addopts=\"\" tests/test_sovereign_e1_verifygate_fail_closed.py tests/test_cain42_phase3_z3_verifygate.py tests/test_sovereign_proof_properties.py tests/test_sovereign_differential.py tests/test_evolution2_breaker.py tests/test_evolution2_properties.py",
      "exit_code": 0,
      "passed": 106,
      "failed": 0,
      "skipped": 0,
      "revision": "3537d76e (working tree clean for these files at the time of the run)"
    },
    {
      "label": "post-fix: continuum regression",
      "command": "python3 -m pytest -q -p no:cacheprovider -o addopts=\"\" tests/test_sovereign_e1_continuum_run_operator_only.py",
      "exit_code": 0,
      "passed": 5,
      "failed": 0,
      "skipped": 0,
      "revision": "a098fa61"
    },
    {
      "label": "post-change: baseline 16 files + 2 new files, shared working tree",
      "command": "baseline command above + tests/test_sovereign_e1_verifygate_fail_closed.py tests/test_sovereign_e1_continuum_run_operator_only.py",
      "exit_code": 1,
      "passed": 203,
      "failed": 6,
      "skipped": 0,
      "revision": "a098fa61 + uncommitted peer edit to cain/continuum/hosted_pipeline.py",
      "note": "All 6 failures are in test_sovereign_proof_properties.py / test_sovereign_differential.py and return DENY 'SIGNER_NOT_PINNED' from session -46's in-progress (uncommitted) hosted_pipeline.py fail-closed change; none touch the files changed here. Reported to -46. Not a regression from 3537d76e/a098fa61."
    },
    {
      "label": "post-merge: baseline 16 files + 2 new files after -46's c0b6a64b (signer/ledger pinned in the 6 affected tests, assertions unchanged)",
      "command": "same as the post-change entry above",
      "exit_code": 0,
      "passed": 209,
      "failed": 0,
      "skipped": 0,
      "revision": "c0b6a64b",
      "note": "/root/.cain/hosted_continuum_decisions.jsonl stayed at 1163 rows during this run"
    },
    {
      "label": "GAP-04 sweep: all 52 test files touching guard / cain_sdk / mcp_proxy, isolated worktree at 6c6d15bc + the 3 GAP-04 files",
      "command": "python3 -m pytest -q -p no:cacheprovider -o addopts=\"\" $(grep -lE 'cain\\.guard|from cain import guard|cain_sdk|guard_action|@guard|mcp_proxy|MCPTransparentInterceptor' tests/*.py)",
      "exit_code": 1,
      "passed": 829,
      "failed": 1,
      "skipped": 0,
      "revision": "9672bdc1 contents",
      "note": "the 1 failure (test_trifecta_integration.py::test_trifecta_provenance_manifest_valid) fails identically on clean HEAD: stale provenance manifest, 5 files TAMPERED"
    }
  ],
  "defects": [
    {
      "id": "D-S17-01",
      "severity": "P1",
      "file": "cain/cain_z3_verifygate.py (+ platform-gateway copy)",
      "line": 96,
      "summary": "prove_monotonicity mapped an unrecognised ceiling name to rank 0 (NONE), so parent=ADMIN / child=SUPERUSER (or child='') returned proved=True, status UNSAT.",
      "repro_command": "python3 -c \"from cain.cain_z3_verifygate import Z3VerificationEngine as E; r=E().prove_monotonicity('ADMIN','SUPERUSER',[],[]); print(r.proved, r.status)\"  # 6cfa3b9e: True UNSAT; 3537d76e: False INVALID_INPUT",
      "fixed": true,
      "fix_commit": "3537d76ef828ef7458a6f6dcde3a3f2bb70194de",
      "regression_test": "tests/test_sovereign_e1_verifygate_fail_closed.py::test_unknown_ceiling_never_proves_monotonicity"
    },
    {
      "id": "D-S17-02",
      "severity": "P1",
      "file": "cain/cain_z3_verifygate.py (+ platform-gateway copy)",
      "line": 259,
      "summary": "prove_memory_containment priced an unrecognised operation as WRITE and an unrecognised ceiling as NONE; e.g. VERIFIED / ADMIN / 'SHUTDOWN' returned proved=True.",
      "repro_command": "python3 -c \"from cain.cain_z3_verifygate import Z3VerificationEngine as E; print(E().prove_memory_containment('VERIFIED','ADMIN','SHUTDOWN').proved)\"  # 6cfa3b9e: True; 3537d76e: False",
      "fixed": true,
      "fix_commit": "3537d76ef828ef7458a6f6dcde3a3f2bb70194de",
      "regression_test": "tests/test_sovereign_e1_verifygate_fail_closed.py::test_unknown_memory_names_never_prove"
    },
    {
      "id": "D-S17-03",
      "severity": "P1",
      "file": "cain/cain_z3_verifygate.py (+ platform-gateway copy)",
      "line": 109,
      "summary": "A solver UNKNOWN (timeout, incompleteness) was reported as status 'SAT' with a fabricated counterexample in all three proofs; a solver exception propagated as a 500. proved stayed False, so this was mislabelled evidence, not a fail-open.",
      "repro_command": "tests/test_sovereign_e1_verifygate_fail_closed.py::test_solver_unknown_is_reported_as_unknown_not_sat (monkeypatches z3.Solver.check to return unknown)",
      "fixed": true,
      "fix_commit": "3537d76ef828ef7458a6f6dcde3a3f2bb70194de",
      "regression_test": "tests/test_sovereign_e1_verifygate_fail_closed.py::test_solver_unknown_is_reported_as_unknown_not_sat, ::test_solver_exception_fails_closed"
    },
    {
      "id": "D-S17-04",
      "severity": "P2",
      "file": "cain/cain_z3_verifygate.py (+ platform-gateway copy)",
      "line": 179,
      "summary": "prove_temporal_decay clamped half_life to >= 1 s, giving a looser decay bound than the real one (half_life=0.5, elapsed=1, t0=0.8: true bound 0.2, used 0.4, so a claim of 0.35 was proved). NaN inputs raised a raw Z3Exception; inf/<=0 half-life were not refused.",
      "repro_command": "python3 -c \"from cain.cain_z3_verifygate import Z3VerificationEngine as E; print(E().prove_temporal_decay(0.8,0.5,1.0,0.35).proved)\"  # 6cfa3b9e: True; 3537d76e: False",
      "fixed": true,
      "fix_commit": "3537d76ef828ef7458a6f6dcde3a3f2bb70194de",
      "regression_test": "tests/test_sovereign_e1_verifygate_fail_closed.py::test_short_half_life_is_not_clamped, ::test_non_finite_or_invalid_decay_inputs_fail_closed"
    },
    {
      "id": "D-S17-05",
      "severity": "P3",
      "file": "platform-gateway/routers/continuum_verify.py",
      "line": 57,
      "summary": "POST /fabric/continuum/verify/run had no authentication: the path is outside main._FRONTIER20_PREFIXES and _TENANT_GATED_WRITE_PREFIXES, containment_gate only acts when a key is present, and the router had no dependency. Every anonymous POST made the gateway spawn a new Python interpreter for the harness worker (the workload that OOM-killed the gateway on 2026-10-11). Resource-exhaustion / unauthenticated process spawn; it does not execute caller-chosen code.",
      "repro_command": "tests/test_sovereign_e1_continuum_run_operator_only.py with subprocess.Popen stubbed (no process spawned). Established from route/middleware source; NOT probed against production, because a live POST would spawn the worker.",
      "fixed": true,
      "fix_commit": "a098fa61709883d8ba5a3cfa426730dba9e5b6a0",
      "regression_test": "tests/test_sovereign_e1_continuum_run_operator_only.py"
    },
    {
      "id": "D-S17-06",
      "severity": "P5",
      "file": "cain-ebpf/cain_enforce.bpf.c",
      "line": 69,
      "summary": "The program is described in its header and in published evidence (clawx-site/evidence/evolution7-sovereign-action-commit-2026-10-09/CAIN_IP_VALUE_RECALC_V38.json, invention 'EBPF TRACEPOINT BOUNDARY INTERCEPTOR') as 'rejecting unauthorized binary execution and non-whitelisted network egress directly in the Linux kernel'. It attaches only to tracepoints sys_enter_execve / sys_enter_connect, records a BLOCK value in a ring-buffer event and always returns 0; tracepoint programs cannot deny a syscall, and the cain_blocked_ports map is never read. It is telemetry, not enforcement.",
      "repro_command": "grep -nE 'SEC\\(|return' cain-ebpf/cain_enforce.bpf.c",
      "fixed": true,
      "fix_commit": "9fc79332 (V38), 656dac9a (V33-V37, V39-V45, V33.md, generator, kit package V45)",
      "regression_test": null,
      "owner_action": "Owner approved correcting the wording on 2026-10-12. Every recalc V33-V45 (13 source records, 8 served copies, the auditor-kit staging copy of V45) now describes the program as tracepoint telemetry; label, tier and all figures unchanged; original wording kept in a 'correction' field. V33.md 'eBPF Execution Gate' line and the V33 generator corrected too. PUBLISH_MANIFEST (V38) and the public evidence inventory re-pinned; served hashes verified on cainstudio.online and mcpgate.online (and clawx.click where it mirrors the bundle). The signed auditor-kit 8.1.0 / 8.2.0 releases are immutable and still contain the old V45 text; the next kit version (built by -82, signed by the owner) carries the corrected one. The kernel-enforcement gap itself (T-S17-05) is unchanged: this fixes the claim, not the missing enforcement."
    },
    {
      "id": "D-S17-07",
      "severity": "P6",
      "file": "cain/substrate/proof.py (= platform-gateway/proof.py)",
      "line": 610,
      "summary": "proof_runs.run_hash covers results but not the stored passed/failed/total columns that latest()/history() return, so those counts can be edited without verify_chain() noticing. The run-chain functions (run_all/latest/history/verify_chain) have no production caller (only platform-gateway/tests/test_phase6_integration.py), so there is no live impact today.",
      "repro_command": "static: compare the hashed body at cain/substrate/proof.py:610-612 with the INSERT column list at :615",
      "fixed": false,
      "fix_commit": null,
      "regression_test": null
    },
    {
      "id": "D-S17-08",
      "severity": "P2",
      "file": "cain/guard.py + cain_sdk/guard.py",
      "line": 481,
      "summary": "GAP-04 / I9: CAIN_ALLOW_OFFLINE_LOCAL defaulted to 'true', so guard(fail_closed=True) fell back to the local baseline on a gateway outage (cain_sdk's fallback allows anything not dangerous-and-HIGH). Owner approved the flip to fail closed.",
      "repro_command": "tests/test_sovereign_e1_guard_offline_fail_closed.py (gateway = closed local port 127.0.0.1:9)",
      "fixed": true,
      "fix_commit": "9672bdc1",
      "regression_test": "tests/test_sovereign_e1_guard_offline_fail_closed.py",
      "note": "SDK-side: deploying the gateway does not update installed SDKs; customers get it with the next published wheel."
    }
  ],
  "threats": [
    {
      "id": "T-S17-01",
      "title": "Formal-proof API returns proved=True for inputs it cannot interpret",
      "assets": "integrity of /verifygate/* proof results and their evidence_sha256",
      "attacker_prereqs": "none (routes are public)",
      "entry_point": "POST /verifygate/prove-monotonicity | prove-memory-containment | prove-temporal-decay",
      "existing_controls": "pydantic range checks on decay numbers; Z3 timeout 10 s",
      "observed_evidence": "D-S17-01..04 reproduced on 6cfa3b9e; fixed 3537d76e; 106/106 post-fix",
      "coverage": {
        "prevent": "YES",
        "detect": "PARTIAL",
        "contain": "NO"
      },
      "impact": "a client treating proved=True as permission would accept an escalated or uninterpretable authority; no CAIN authorization path consumes these results (only the routes and tests call them)",
      "required_test": "tests/test_sovereign_e1_verifygate_fail_closed.py",
      "residual_risk": "results are unsigned; the model is a single-comparison encoding, so UNSAT means only that the stated numeric comparison cannot be violated with the given constants",
      "release_implication": "NONE",
      "compromise_level": "application"
    },
    {
      "id": "T-S17-02",
      "title": "Unauthenticated process spawn / resource exhaustion through the continuum verify route",
      "assets": "gateway host memory and CPU; gateway availability",
      "attacker_prereqs": "network access to any of the 3 domains",
      "entry_point": "POST /fabric/continuum/verify/run",
      "existing_controls": "worker takes an flock, so only one harness runs at a time (but every POST still starts an interpreter, and the harness can be kept running back-to-back)",
      "observed_evidence": "D-S17-05; fixed a098fa61; not deployed",
      "coverage": {
        "prevent": "YES",
        "detect": "NO",
        "contain": "PARTIAL"
      },
      "impact": "denial of service of the request-serving gateway",
      "required_test": "tests/test_sovereign_e1_continuum_run_operator_only.py",
      "residual_risk": "no rate limit on the operator path",
      "release_implication": "NONE",
      "compromise_level": "application"
    },
    {
      "id": "T-S17-03",
      "title": "Policy-service outage downgrades a remote DENY to a local ALLOW in the SDK guard",
      "assets": "authorization decisions for SDK-guarded functions",
      "attacker_prereqs": "ability to block or break the agent host's connection to the policy gateway (DNS, egress firewall, TLS failure), or a non-200 response",
      "entry_point": "cain.guard remote mode, cain/guard.py:477-485",
      "existing_controls": "local baseline is default-deny (allowlist only); fail_closed=True is honoured only when CAIN_ALLOW_OFFLINE_LOCAL is set to something other than 'true', and it defaults to 'true'",
      "observed_evidence": "source read; already published as GAP-04 (HIGH) in CAIN42_L5_REPOSITORY_AUDIT.md (2026-09-28), still open; FIXED 9672bdc1 (owner-approved default flip), 5 tests (2 fail on the old default)",
      "coverage": {
        "prevent": "YES",
        "detect": "PARTIAL",
        "contain": "NO"
      },
      "impact": "an action the remote policy denies, but that is on the developer's local allowlist, executes during an induced outage (violates I9 as stated: the offline policy is not separately authorized)",
      "required_test": "guard(remote) with a gateway that DENYs action X, then unreachable, with X on the local allowlist: expect DENY",
      "residual_risk": "operators who set CAIN_ALLOW_OFFLINE_LOCAL=true accept the fallback; installed SDKs keep the old default until upgraded",
      "release_implication": "NONE",
      "compromise_level": "application"
    },
    {
      "id": "T-S17-04",
      "title": "SDK guard accepts an unauthenticated ALLOW from whatever answers at the gateway URL",
      "assets": "authorization decisions for SDK-guarded functions",
      "attacker_prereqs": "control of the network path or the configured gateway URL (http:// is accepted)",
      "entry_point": "cain/guard.py:452-460 (response JSON 'decision' is trusted; 'signature' is copied, never verified)",
      "existing_controls": "TLS when the URL is https; API key sent to the gateway",
      "observed_evidence": "source read",
      "coverage": {
        "prevent": "NO",
        "detect": "NO",
        "contain": "NO"
      },
      "impact": "a spoofed gateway can authorize any action, and the ALLOW is then cached for the fast path",
      "required_test": "guard against a stub gateway returning an unsigned ALLOW: expect DENY once decision-signature pinning exists",
      "residual_risk": "full until decisions are signed and the SDK pins the decision key",
      "release_implication": "OPEN_ITEM",
      "compromise_level": "process"
    },
    {
      "id": "T-S17-05",
      "title": "Host enforcement on the hosted path is application-level only; the eBPF 'enforcement' program observes",
      "assets": "host files, network, processes reachable by tool execution",
      "attacker_prereqs": "a tool call that MCPGate allowed, or code running in the gateway process",
      "entry_point": "any allowed tool execution; direct subprocess/network use by an agent that does not route through MCPGate",
      "existing_controls": "MCPGate application checks; cain/ebpf_enforcer.py kprobe bpf_override_return (-EPERM before the syscall runs) exists for its own sandbox launcher; /frontier/kernel/sandbox-run disabled unless CAIN_KERNEL_SANDBOX_RUN_ENABLED=1; /fabric/openshell/execute-sandboxed operator-only",
      "observed_evidence": "cain-ebpf/cain_enforce.bpf.c is tracepoint-only and always returns 0 (D-S17-06); BPF LSM not in the active LSM list on this host (lsm=lockdown,capability,landlock,yama,apparmor,ima,evm); kprobe-override deny NOT re-executed in this evolution",
      "coverage": {
        "prevent": "PARTIAL",
        "detect": "PARTIAL",
        "contain": "PARTIAL"
      },
      "impact": "an agent that bypasses MCPGate (direct client, direct subprocess, direct socket) is not stopped by any kernel control on the hosted path",
      "required_test": "non-destructive kprobe-override deny test in an isolated VM; out of scope here",
      "residual_risk": "MCPGate does not mediate traffic that does not go through it; kernel compromise defeats all of these controls",
      "release_implication": "BLOCKER",
      "compromise_level": "kernel"
    },
    {
      "id": "T-S17-06",
      "title": "MCPGate decision evidence is not durable or request-bound",
      "assets": "after-the-fact reconstruction of what MCPGate allowed or denied",
      "attacker_prereqs": "none for loss; process restart loses in-memory state",
      "entry_point": "cain/mcp_proxy.py MCPToolCallEvaluation.evidence_digest",
      "existing_controls": "per-decision digests, mostly sha256 of the reason string",
      "observed_evidence": "source read (cain/mcp_proxy.py:260-601): no append-only ledger is written by the proxy itself",
      "coverage": {
        "prevent": "NO",
        "detect": "NO",
        "contain": "NO"
      },
      "impact": "two different requests denied for the same reason get the same evidence digest; no local chain to detect deletion",
      "required_test": "per-request evidence record bound to the action digest, chained and externally checkpointed (ihvrk/ledger is the candidate)",
      "residual_risk": "a local hash chain still does not stop a privileged writer from rewriting everything without an external checkpoint",
      "release_implication": "OPEN_ITEM",
      "compromise_level": "application"
    }
  ],
  "traceability": [
    {
      "threat": "T-S17-01",
      "requirement": "solver UNKNOWN/timeout/exception/invalid input never yields a proof",
      "invariant": "I8",
      "implementation": "cain/cain_z3_verifygate.py Z3VerificationEngine._check / _not_proved / _rank / _finite",
      "test": "tests/test_sovereign_e1_verifygate_fail_closed.py",
      "observed_result": "30/30 pass at 3537d76e; 26/30 fail at 6cfa3b9e",
      "evidence": "reports/sovereign/evolution-01/s17/findings.json",
      "residual_risk": "unsigned results; single-comparison model",
      "release_decision": "PASS for this component"
    },
    {
      "threat": "T-S17-01",
      "requirement": "an unknown authority name is never mapped to a default rank",
      "invariant": "I6",
      "implementation": "cain/cain_z3_verifygate.py _rank()",
      "test": "tests/test_sovereign_e1_verifygate_fail_closed.py::test_unknown_ceiling_never_proves_monotonicity",
      "observed_result": "pass at 3537d76e",
      "evidence": "reports/sovereign/evolution-01/s17/findings.json",
      "residual_risk": "none in the encoded model",
      "release_decision": "PASS for this component"
    },
    {
      "threat": "T-S17-02",
      "requirement": "no host process spawn from an unauthenticated request",
      "invariant": "I1",
      "implementation": "platform-gateway/routers/continuum_verify.py run() Depends(require_operator)",
      "test": "tests/test_sovereign_e1_continuum_run_operator_only.py",
      "observed_result": "5/5 pass at a098fa61; 3/5 fail at 3537d76e",
      "evidence": "reports/sovereign/evolution-01/s17/findings.json",
      "residual_risk": "none beyond operator-path rate limiting",
      "release_decision": "PASS (deployed and verified on blue+green)"
    },
    {
      "threat": "T-S17-03",
      "requirement": "mandatory policy-service failure does not authorize unless a separately authorized offline policy permits it",
      "invariant": "I9",
      "implementation": "cain/guard.py:481-484, cain_sdk/guard.py:187-189 (default 'false')",
      "test": "tests/test_sovereign_e1_guard_offline_fail_closed.py",
      "observed_result": "5/5 pass at 9672bdc1; 2/5 fail on the old default; guard/proxy sweep 829 passed, 1 pre-existing failure (stale provenance manifest)",
      "evidence": "reports/sovereign/evolution-01/s17/findings.json",
      "residual_risk": "installed SDKs until upgraded",
      "release_decision": "PASS for source; SDK release pending"
    },
    {
      "threat": "T-S17-04",
      "requirement": "an ALLOW is accepted only from an authorized issuer",
      "invariant": "I3",
      "implementation": "cain/guard.py:452-460 (no signature check on the gateway decision)",
      "test": "none yet",
      "observed_result": "NOT_IMPLEMENTED",
      "evidence": "source read",
      "residual_risk": "full",
      "release_decision": "BLOCKED for remote-mode SDK claims"
    },
    {
      "threat": "T-S17-05",
      "requirement": "kernel-level enforcement is claimed only where a deny mechanism exists",
      "invariant": "I15",
      "implementation": "cain-ebpf/cain_enforce.bpf.c (tracepoint, telemetry only); cain/ebpf_enforcer.py (kprobe override, not re-run)",
      "test": "none in this evolution",
      "observed_result": "published claim exceeds the code (D-S17-06)",
      "evidence": "clawx-site/evidence/evolution7-sovereign-action-commit-2026-10-09/CAIN_IP_VALUE_RECALC_V38.json",
      "residual_risk": "overclaim stays published until the owner decides",
      "release_decision": "BLOCKED for any 'kernel enforcement' claim on the hosted path"
    }
  ],
  "capability_status": {
    "verifygate Z3 proofs (fail-closed result handling)": "IMPLEMENTED_AND_TESTED",
    "verifygate models (single-comparison encodings)": "FORMALLY_CHECKED_WITH_STATED_ASSUMPTIONS",
    "continuum verify run gating": "IMPLEMENTED_AND_TESTED (deployed 2026-10-12, blue+green verified)",
    "SDK guard offline fallback fail-closed": "IMPLEMENTED_AND_TESTED (source; next SDK release)",
    "SDK guard decision signature verification": "NOT_IMPLEMENTED",
    "eBPF tracepoint program (cain_enforce.bpf.c)": "IMPLEMENTED_BUT_NOT_FULLY_TESTED (telemetry only, not enforcement)",
    "eBPF kprobe-override deny (cain/ebpf_enforcer.py)": "IMPLEMENTED_BUT_NOT_FULLY_TESTED (not re-run in Evolution 1)",
    "BPF LSM enforcement": "BLOCKED (bpf not in the active LSM list; reboot required)",
    "MCPGate durable evidence ledger": "NOT_IMPLEMENTED",
    "substrate proof-run hash chain": "IMPLEMENTED_BUT_NOT_FULLY_TESTED (no production caller; counts columns not hashed)"
  },
  "alternative_routes_checked": [
    {
      "route": "POST /sandbox/execute",
      "gate": "tenant API key (fabric_control_plane._require_tenant) before proxying to the sandbox service",
      "verdict": "authenticated; no PCAA/MCPGate decision"
    },
    {
      "route": "POST /frontier/kernel/sandbox-run",
      "gate": "403 unless CAIN_KERNEL_SANDBOX_RUN_ENABLED=1",
      "verdict": "disabled by default"
    },
    {
      "route": "POST /fabric/openshell/execute-sandboxed",
      "gate": "frontier20_gate operator credential + tenant key; engine policy vets argv",
      "verdict": "operator-only"
    },
    {
      "route": "POST /fabric/contracts/invoke-exec",
      "gate": "frontier20_gate operator credential + contract/RIG/E8 checks",
      "verdict": "operator-only"
    },
    {
      "route": "POST /fabric/continuum/verify/run",
      "gate": "none before a098fa61",
      "verdict": "D-S17-05, fixed in source"
    },
    {
      "route": "cain.guard SDK (in-process)",
      "gate": "local allowlist or remote gateway",
      "verdict": "T-S17-03 / T-S17-04 open"
    }
  ],
  "not_done": [
    "No live production probe of D-S17-05 (a POST would spawn the worker).",
    "kprobe-override deny in cain/ebpf_enforcer.py not re-executed (attaches kprobes on the live host).",
    "Tamper tests on a live, production-called evidence ledger were not added: the MCP proxy writes none, and the substrate proof-run chain has no production caller. IHVRK's ledger (session -98) is the candidate for Evolution 2.",
    "No permit state-machine Z3 model from -17: session -98's ihvrk/solver/ihvrk_solver.py covers I1-I15 and is not duplicated here."
  ],
  "side_effects": [
    "The baseline run at 6cfa3b9e and the post-change run at a098fa61 executed tests/test_sovereign_proof_properties.py and tests/test_sovereign_differential.py, which at that time appended rows to the live /root/.cain/hosted_continuum_decisions.jsonl (session -46 measured 12 rows per run). -46 stopped the writes in c0b6a64b and reports the chain still verifies (1163 rows). No other live state was written: /app/data was checked before and after."
  ]
}
