Pruebas y límites de BABYLON-60
Evaluación empírica en silicio, aserciones formales y registro público de divergencias detectadas en el monorepo.
Divergencias Detectadas en la Revisión Actual
Transparencia determinista: discrepancias identificadas entre las afirmaciones declaradas en manifiestos y los resultados medidos en silicio:
DIVERGENCIAS DETECTADAS EN LA EVIDENCIA (AUTODETECCIÓN DETERMINISTA)
El generador de la web contrasta de forma estricta los manifests y reportes. Ninguna discrepancia se oculta ni se adorna.
| ID | SEVERIDAD | OBJETO / ELEMENTO | HALLAZGO OBSERVADO | FUENTES EN CONFLICTO |
|---|---|---|---|---|
| DIV-ATTEST | HIGH | reports/nemesis_verified_report.json | Field "attestation" = "cose:ed25519:c13b7bc4c26b65b100d7e66f199127a7" is 16 bytes, not a 64-byte Ed25519 signature, and equals the prefix of manifest_hash. It cannot be verified as a signature. | reports/nemesis_verified_report.json |
| DIV-STALE | MEDIUM | reports/nemesis_verified_report.json | NEMESIS report commit 8650fe1368 differs from release manifest commit d7d96bd1b7. | reports/nemesis_verified_report.json RELEASE_MANIFEST.json |
| DIV-SIMULATED | HIGH | reports/nemesis-crash-report.json | Contains "SIGKILL_SIMULATED": the fault was simulated, not injected as a real signal. | reports/nemesis-crash-report.json |
| DIV-TIMESTAMP | low | nemesis-kernel / nemesis-crash | Both reports carry the identical round timestamp_epoch 1791030000; it does not look like a measured execution time. | reports/nemesis-kernel-report.json reports/nemesis-crash-report.json |
| DIV-SCALE | MEDIUM | NEMESIS-100 acceptance profile | Demonstrated 19 vectors × 100 attempts = 1900; target profile is 1,000,000. | reports/nemesis_verified_report.json docs/01_spec/spec_nemesis.md |
Las Seis Invariantes Canónicas de NEMESIS
Propiedades formales evaluadas en `docs/01_spec/spec_nemesis.md` y contrastadas contra `reports/nemesis_verified_report.json`:
EFFECT_AUTHORIZATION
¬ Authorized(a, o, e, t) ⟹ ¬ Effected(a, o, e, t)VERIFIED_COMMIT
Committed(tx) ⟹ Verified(tx) ∧ EffectObserved(tx)PATH_OBLIVION
State(tx) ≥ Quarantined ⟹ ResolveOriginalPath(tx) = ForbiddenCRASH_RECOVERABILITY
Crash(tx, s) ⟹ Recover(tx, s) ∈ {Committed, Aborted, SecurityAbortRECOVERY_IDEMPOTENCE
Recover(Recover(tx)) = Recover(tx)AUTHORITY_FRESHNESS
Committed(tx) ⟹ AuthorityEpoch_{commit} = AuthorityEpoch_{verifiedTeoremas Inductivos Demostrados en Silicio
Demostración constructiva en tiempo de compilación por reflexión pura (`rfl`) y prueba inductiva universal sobre trazas finitas arbitrarias (`proof/lean/AX0.lean`):
1. Teorema Constructivo AUTH_001 (Reflexión rfl)
Garantiza que un nonce consumido jamás puede despachar dos veces, colapsando inmediatamente cualquier intento de doble mutación.
2. Seguridad Inductiva Universal (auth_001_universal_inductive_safety)
Demuestra matemáticamente que sobre cualquier secuencia finita arbitraria de eventos causales, el número total de despachos permitidos es estrictamente menor o igual a 1.
Inspeccionar código fuente formal de Lean 4 (AX0.lean)proof/lean/AX0.lean
-- Teoremas demostrados en Lean 4 Core (proof/lean/AX0.lean)
def trace_happy : List AuthEvent := [.reserve, .dispatch, .ackSuccess]
theorem auth_001_valid_execution : validateTrace trace_happy .Proposed = true := rfl
def trace_double_dispatch : List AuthEvent := [.reserve, .dispatch, .dispatch]
theorem auth_001_reject_double_dispatch : validateTrace trace_double_dispatch .Proposed = false := rfl
def trace_illegal_dispatch : List AuthEvent := [.dispatch]
theorem auth_001_reject_illegal_dispatch : validateTrace trace_illegal_dispatch .Proposed = false := rfl
theorem auth_001_universal_inductive_safety (trace : List AuthEvent) :
validateTrace trace .Proposed = true → countDispatches trace ≤ 1 := by
intro h
exact max_dispatches_from_state trace .Proposed hPerfil Objetivo vs. Evidencia Demostrada en Silicio (INV_C5 §6.4)
Segregación taxativa de acuerdo con la directiva C5-REAL: criterios aspiracionales de diseño frente a mediciones físicas reales en silicio.
| MÉTRICA DE PRUEBA / COMPONENTE | PERFIL OBJETIVO (TARGET PROFILE) | EVIDENCIA DEMOSTRADA (DEMONSTRATED IN SILICON) | ESTATUS DE COBERTURA |
|---|---|---|---|
| Kernel Adversarial Attack Attempts | 1,000,000 intentos objetivo | 19 vectores × 100 intentos/run (1900 total) | PARCIAL (Escala en progreso) |
| Unauthorized Mutations / Deletions | 0 permitidas | 0 observadas en ejecución APFS | CUMPLIDO |
| Capability Root Escapes | 0 permitidas | 0 observadas | CUMPLIDO |
| Model Checking (Loom) Preemptions | >280,000 ejecuciones exploradas | Loom compilado con fallo en binario hunter; Lean 4 inductive trace formalizado | EXPERIMENTAL |
| Crash Recovery Idempotence | 0 recuperaciones ambiguas | 250 iteraciones bajo SIGKILL_SIMULATED | SIMULADO |
Alcance y Delimitación del TCB
DENTRO DEL PERÍMETRO VERIFICADO (TCB)
- ✓Local filesystem under tested threat model
- ✓Broker-owned quarantine directory (mode 0700)
- ✓Same mount / filesystem boundary (st_dev matching)
- ✓Attacker cannot mutate quarantine directory contents
- ✓Declared capability root
- ✓Tested APFS backend (macOS ARM64)
- ✓Tested Linux backend (renameat2 / O_TMPFILE)
FUERA DEL PERÍMETRO (FUERA DE ALCANCE)
- ✗Network filesystems (NFS, CIFS, SMB) → fail-closed
- ✗Cross-device fallback (EXDEV) → fail-closed
- ✗Attacker-controlled kernel / hypervisor
- ✗Compromised broker process memory
- ✗Compromised signing authority / Secure Enclave private key
- ✗Compromised filesystem implementation / storage controller firmware
- ✗DMA / physical memory bus extraction attacks
Registro Epistémico de Afirmaciones (CLAIMS.toml)
Registro completo de las 20 afirmaciones de verdad del sistema. Todo valor declarado se audita frente a la suite en silicio:
| CLAIM / COMPONENTE | DOMINIO | ENUNCIADO Y ALCANCE (SCOPE) | DECLARADO | OBSERVADO | REPRODUCCIÓN |
|---|---|---|---|---|---|
inv_01_manifest_layout 00_ABZU_KERNEL | CORE | SharedManifest is exactly 64 bytes and 64-byte aligned (Zero-Split Coherence) | VERIFIED | VERIFIED | cargo test --locked --lib manifest::tests:... |
inv_02_seqlock_concurrency 00_ABZU_KERNEL | NEMESIS | Seqlock SPMC protocol model (loom replica of seqlock.rs, 1 writer x 1-2 readers) yields no torn read under bounded model checking (max 2 preemptions); production code is not itself model-checked | VERIFIED | VERIFIED | LOOM_MAX_PREEMPTIONS=2 RUSTFLAGS="--cfg lo... |
inv_04_fail_stop_halt 00_ABZU_KERNEL | BABYLON | Epistemic halt closes admission (POISONED, 0xDEAD_6060) and records SignedDurable / UnsignedDurable / PersistenceFailed evidence within a bounded budget; never a synthetic signature | VERIFIED | VERIFIED | cargo test --locked --lib halt::tests |
cose_ed25519_receipt 00_ABZU_KERNEL | BABYLON | Receipts are formatted according to RFC 9942 and cryptographically signed with Ed25519 | VERIFIED | VERIFIED | cargo test --locked --lib receipt::tests::... |
secure_enclave_p256 00_ABZU_KERNEL | BABYLON | Apple Secure Enclave bridge produces P-256 signatures with Touch ID gate | VERIFIED (LOCAL) | VERIFIED (LOCAL) | cargo test --locked --lib enclave::tests::... |
cortex_persist_sqlite 01_KISH_ENGINE | CORTEX | Persistent ledger with SQLite WAL, BEGIN IMMEDIATE and SHA3-256 hash chaining | VERIFIED | VERIFIED | pytest tests/ -k test_ledger |
merkle_attestation 01_KISH_ENGINE | CORTEX | Merkle state roots computed deterministically and queryable via MCP | VERIFIED | VERIFIED | pytest tests/ -k test_merkle |
lean4_formal_trace 00_ABZU_KERNEL | NEMESIS | Lean 4 machine-checks concrete execution traces and proves universal inductive soundness theorems for arbitrary traces (no torn reads and single-dispatch AX-0 invariance) | VERIFIED (LOCAL) | VERIFIED (LOCAL) | lean proof/lean/BabylonTrace.lean && lean ... |
worm_durability 00_ABZU_KERNEL | CORTEX | In-memory HashChainedLedger provides cryptographic tamper detection | VERIFIED (IN-MEM) | VERIFIED (IN-MEM) | cargo test --locked --lib worm |
heterogeneous_verification_quorum 00_ABZU_KERNEL | NEMESIS | Heterogeneous 2/3 verification quorum over 3 engines (Rust, Lean 4, Z3) | VERIFIED (LOCAL) | VERIFIED (LOCAL) | cargo test --locked --lib larsa_bft |
zero_anergy 00_ABZU_KERNEL | CORE | Pure loads on seqlock read path eliminate reader-side cacheline ownership traffic (RFO = 0) | VERIFIED | VERIFIED | cargo run --locked --bin babylon60_kernel ... |
mcp_authority_gate 01_KISH_ENGINE | BABYLON | MCP server enforces AX-0: stochastic AI clients cannot self-authorize protected actions | VERIFIED | VERIFIED | pytest tests/test_ax0_mcp_isolation.py |
AUTH_001 00_ABZU_KERNEL | BABYLON | A consumed authorization cannot be dispatched a second time by the executor, including after process death and reopening the file ledger Scope: executor-managed effects; single executor process per ledger; SIGKILL process death, not power loss | VERIFIED (LOCAL) | VERIFIED (LOCAL) | cargo test --locked --lib executor::tests:... |
AUTH_002 00_ABZU_KERNEL | BABYLON | EDIN workers cannot execute protected effects directly without verified COSE authorization over the IPC Unix socket boundary Scope: privilege isolation over Unix domain socket IPC; capability retained by executor server | VERIFIED (LOCAL) | VERIFIED (LOCAL) | cargo test --locked --lib executor::ipc_te... |
AUTH_003 00_ABZU_KERNEL | BABYLON | A dispatch without a durable result becomes UNKNOWN at restart and is never retried automatically Scope: crash recovery; RESERVED entries stay blocked (no resumption implemented) | VERIFIED (LOCAL) | VERIFIED (LOCAL) | cargo test --locked --lib executor::tests:... |
AUTH_004 00_ABZU_KERNEL | BABYLON | A resource whose precondition digest or inode identity changes between authorization and effect is rejected before mutation; for filesystem DELETE_FILE, the TOCTOU gap is closed via atomic quarantine isolation and post-rename verification Scope: local filesystem under tested threat model; requires broker-exclusive quarantine on same mount (dev match), non-symlink paths, and O_NOFOLLOW/renameatx_np/renameat2 primitives. Remote and cross-device filesystems unsupported (fail-closed) | VERIFIED (LOCAL) | VERIFIED (LOCAL) | cargo test --locked --lib secure_fs && car... |
AUTH_005 00_ABZU_KERNEL | BABYLON | Signing or persistence failures produce explicit errors; no zeroed or synthetic signature is emitted Scope: receipt generation and halt evidence | VERIFIED (LOCAL) | VERIFIED (LOCAL) | cargo test --locked --lib receipt::tests::... |
AUTH_006 00_ABZU_KERNEL | BABYLON | Under 1000 one-process-per-iteration runs covering 3 SIGKILL crash points and a no-crash control, no nonce produces more than one effect, no DISPATCHED survives restart, and every archived authorization replay is rejected Scope: process kill (SIGKILL), file ledger, macOS; not power loss, not multi-executor | VERIFIED (LOCAL) | VERIFIED (LOCAL) | D=$(mktemp -d)/run && cargo run --locked -... |
AUTH_007_secure_fs_quarantine 00_ABZU_KERNEL | BABYLON | EffectTx guarantees SAFETY and NON-INTERFERENCE for DELETE_FILE: under 100 concurrent workers and swap race attacks, unauthorized destructions == 0 and unquarantined deletions == 0 Scope: local filesystem, same mount/device, broker-owned quarantine (0700); macOS renameatx_np(RENAME_EXCL) and Linux renameat2(RENAME_NOREPLACE); NFS/remote FS unsupported and fail-closed | VERIFIED (LOCAL) | VERIFIED (LOCAL) | cargo test --locked --test adversarial_nem... |
sharur_100_swarm_concurrency 02_EDIN_SWARMS | CORE | SHARUR-3600 Legion of 100 concurrent workers sustains ledger contention, enforces AX-0, and achieves 100% fail-closed rate across 50 adversarial attacks Scope: concurrent worker isolation, SQLite WAL lock contention, and 5-cohort fail-closed adversarial rejection | VERIFIED | VERIFIED | pytest tests/test_swarm_100_concurrency.py |
tamkarum_budget_gate 01_KISH_ENGINE | BABYLON | KudurruBudgetGate enforces 64-byte L1 cache line layout, Kelly Criterion bid caps, and MUSHUSHU-0 fail-stop apoptosis (0xDEAD_6060) | VERIFIED | VERIFIED | cargo test --locked --lib tamkarum::tests |
FL_007_admission_bounded_liveness 00_ABZU_KERNEL | UNMAPPED | AdmissionController guarantees Little's Law latency bounds and deterministic queue depth rejection (fail-closed capacity backpressure) Scope: in-memory atomic queue depth and latency budget enforcement; prevents unbounded execution queue starvation | VERIFIED (LOCAL) | VERIFIED (LOCAL) | cargo test --locked --features cortex-pers... |
FL_008_scoped_capability_broker 00_ABZU_KERNEL | UNMAPPED | ScopedHandle and McpCapabilityBroker enforce: Tool Available != Tool Authorized != Effect Authorized. Under 10,000 adversarial trials across 5 attack vectors, unauthorized_effects == 0, capability_escalations == 0, stale_token_acceptances == 0, tool_identity_confusions == 0, and configuration_drifts == 0 Scope: OS-level descriptor-based capabilities via O_DIRECTORY/O_NOFOLLOW with physical dev/ino binding, causal parent attestation, and McpCapabilityBroker validating identity, code hash, effect kind, resource scope, argument digest, policy version, and validity epoch at t_commit | VERIFIED (LOCAL) | VERIFIED (LOCAL) | cargo test --locked --features cortex-pers... |
INV_OMEGA_complete_mediation 00_ABZU_KERNEL | UNMAPPED | Confined AI agent processes cannot mutate or unlink protected OS resources or the SQLite ledger directly (macOS Seatbelt / EPERM); all reachability paths G=(V,A) from agent to protected resources are mediated by the BABYLON broker, and the 5-test Omega Campaign (BYPASS-001, AUTH-002, RACE-003, CRASH-004, REVOCATION-005) passes on real OS files Scope: macOS process confinement via sandbox-exec Seatbelt profile over protected vault and ledger subpaths, combined with NEMESIS reachability graph cut verification and SovereignExecutor 5-test campaign | VERIFIED (LOCAL) | VERIFIED (LOCAL) | cargo test --locked --features cortex-pers... |
FL_014_wave4_receipt_loss_and_closure_drift 00_ABZU_KERNEL | UNMAPPED | Under SIGKILL between physical syscall and receipt commit (FL014_RECEIPT_LOSS), blind automatic retries are rejected with AmbiguousOutcomeRequiresReconciliation and explicit inode reconciliation consolidates the receipt without duplicate effect; transitive dependency or skill/policy drift (FL015_TRANSITIVE_DRIFT / POLICY_ROLLBACK / RUNTIME_LOAD) invalidates deployment permits fail-closed Scope: local APFS filesystem, WAL synchronous=FULL (.effect_wal.bin), concurrent epoch fencing, and transitive closure attestation | VERIFIED (LOCAL) | VERIFIED (LOCAL) | cargo test --locked -q --test nemesis_wave... |
FL_016_owasp_acs_vop1_and_drift_guard 00_ABZU_KERNEL | UNMAPPED | BabylonAcsGuard enforces OWASP ACS v0.1.0 over AF_UNIX peer-verified sockets, rejects Fail-Open (on_decision_failure='proceed') with 0xDEAD_6060, compresses human wait stalls by >514x (n=30 cohorts, CI95), reconciles VOP-1 UNKNOWN states against physical witness digests, and invalidates capabilities upon skill/tool/policy drift (BabylonDriftGuard) Scope: AF_UNIX local enforcement, 7-tuple T_auth evaluation, VOP-1 tri-state reconciliation, and Phi_BABYLON verified-work metric | VERIFIED (LOCAL) | VERIFIED (LOCAL) | cargo test --locked -q --test nemesis_wave... |
INSPECCIÓN DIRECTA DE ARTEFACTOS (JSON CRUDO)
Acceso íntegro y sin mediación a los volcados de silicio emitidos por la suite de pruebas del monorepo:
RELEASE_MANIFEST.json23 Líneas:
{
"release": "5.0.0",
"state": "candidate",
"commit": "d7d96bd1b7bc881a0544412ec7dcdfc424819651",
"source_tree_digest": "75b1717c61abdb46b78035f6845492f69eee06747927a5c5b828ff16f6d9bd61",
"axiom_root": "AX-0: No stochastic component may authorize its own irreversible effects (LLM != authority)",
"claims_manifest_digest": "e34750df315efa4e1c85e28319ae98fb6612f3ee53d7a192909fc16687afbe4d",
"compiled_at_iso": "2026-10-18T17:57:06Z",
"claims": {
"verified": 10,
"empirical_and_bounded": 16,
"experimental": 0,
"failed": 0,
"total": 26
},
"evidence_root": "sha3-256:76e0e6a31229dae7f8aba0c92ca18b6759273a53041d4774060a4f17ddc1288d",
"signing_authority": {
"algorithm": null,
"scope": "unsigned_local_evidence",
"public_key": null
},
"signature": null
}nemesis_verified_report.json47 Líneas:
{
"schema": "https://babylon60.com/schemas/nemesis-report/v1.json",
"report_version": "1.0.0",
"commit": "8650fe1368ca30e0ceead4e97e3d404acb12a3dd",
"timestamp_iso": "2026-10-11T18:33:05.218018+00:00",
"platform": {
"os": "Darwin",
"arch": "arm64",
"filesystem": "APFS"
},
"security_domain": {
"backend": "tested-darwin-apfs-v1",
"network_fs": false,
"quarantine_mode": "0700",
"same_device_enforced": true
},
"invariants": {
"NEM_I1_effect_authorization": "PASS",
"NEM_I2_verified_commit": "PASS",
"NEM_I3_path_oblivion": "PASS",
"NEM_I4_crash_recoverability": "PASS",
"NEM_I5_recovery_idempotence": "PASS",
"NEM_I6_authority_freshness": "PASS"
},
"kernel_adversarial": {
"attack_vectors_tested": 19,
"attack_attempts_per_run": 100,
"unauthorized_mutations": 0,
"unauthorized_deletions": 0,
"capability_escapes": 0,
"unverified_commits": 0,
"ledger_effect_divergences": 0
},
"crash_recovery": {
"injection_points": [
"PREPARED",
"QUARANTINED",
"VERIFIED",
"RECOVERY_LOOP"
],
"ambiguous_recoveries": 0,
"recovery_idempotence_verified": true
},
"manifest_hash": "sha256:c13b7bc4c26b65b100d7e66f199127a7f66f97dc22c3c1aa626c9b289c2c6731",
"merkle_evidence_root": "sha3-256:0c2de6d45d2110d79816eff80ad188ce1319766b1cccfcc360340446c5df16d8",
"attestation": "cose:ed25519:c13b7bc4c26b65b100d7e66f199127a7"
}nemesis_evidence_report.json82 Líneas:
{
"schema": "nemesis-report/v1",
"commit": "7d9f9dd90551d428d07073f99024c1e45cb98cf8",
"timestamp_utc": "2026-10-03T11:34:45.606913+00:00",
"platform": {
"os": "Darwin",
"kernel": "25.6.0",
"arch": "arm64",
"filesystem": "APFS",
"volume_case_sensitive": false
},
"security_domain": {
"backend": "macos-apfs-v1",
"backend_verification_status": "tested_apfs_backend",
"network_fs": false,
"broker_owned_quarantine_mode": "0700",
"path_oblivion_enforced": true,
"session_types_linear_token": "VerifiedTx"
},
"formal_invariants": {
"NEM_I1_EFFECT_AUTHORIZATION": {
"formal_spec": "¬Authorized(a, o, e, t) ⟹ ¬Effected(a, o, e, t)",
"status": "PROVEN_IN_TYPES_AND_VERIFIED",
"scope": "Scoped to authority principal a, object locator o, effect type e, and epoch/nonce t"
},
"NEM_I2_VERIFIED_COMMIT": {
"formal_spec": "Committed(tx) ⟹ Verified(tx) ∧ EffectObserved(tx)",
"status": "STRUCTURALLY_ENFORCED",
"definition": "Verified(tx) := authority_valid ∧ capability_epoch_valid ∧ locator_match ∧ pre_state_hash_match ∧ effect_allowed ∧ quarantine_isolated ∧ EffectObserved(fstatat == ENOENT)"
},
"NEM_I3_PATH_OBLIVION": {
"formal_spec": "State(tx) ≥ Quarantined ⟹ ResolveOriginalPath(tx) = Forbidden",
"status": "OPERATIONAL_LAW",
"scope": "original_path reduced strictly to audit metadata; no re-stat, re-open or compare permitted"
},
"NEM_I4_CRASH_RECOVERABILITY": {
"formal_spec": "Crash(tx, s) ⟹ Recover(tx, s) ∈ {Committed, Aborted, SecurityAbort}",
"status": "VERIFIED_IN_SILICON"
},
"NEM_I5_RECOVERY_IDEMPOTENCE": {
"formal_spec": "Recover(Recover(s)) == Recover(s)",
"status": "VERIFIED_IN_SILICON"
},
"NEM_I6_AUTHORITY_FRESHNESS": {
"formal_spec": "Committed(tx) ⟹ AuthorityEpoch_commit == AuthorityEpoch_verified",
"status": "VERIFIED_IN_SILICON"
}
},
"threat_model_exclusions": [
"attacker-controlled kernel (Ring -1 / Ring 0 host compromise)",
"compromised broker process memory",
"compromised filesystem implementation / storage controller firmware",
"DMA / physical memory side-channel attacks",
"NFS / SMB / CIFS / remote network filesystems (unsupported -> fail-closed)"
],
"target_acceptance_criteria": {
"description": "Continuous stress model bounds under synthetic interleaving simulator",
"target_interleavings": 1000000,
"allowed_unauthorized_mutations": 0,
"allowed_unauthorized_deletions": 0,
"allowed_capability_root_escapes": 0,
"allowed_unverified_commits": 0,
"allowed_ledger_effect_divergence": 0,
"allowed_ambiguous_recoveries": 0
},
"measured_evidence": {
"host_environment": "Apple Silicon (M-series) / Darwin 24 / APFS local mount",
"adversarial_vectors_tested": 16,
"adversarial_vectors_passed": 16,
"concurrent_workers": 100,
"measured_unauthorized_mutations": 0,
"measured_unauthorized_deletions": 0,
"measured_capability_root_escapes": 0,
"measured_unverified_commits": 0,
"measured_ledger_effect_divergence": 0,
"measured_ambiguous_recoveries": 0,
"recovery_idempotence_cycles_verified": 3,
"total_execution_time_seconds": 0.1544
},
"manifest_hash": "sha256:e10e5bb09dca252b4e5ae2c9e2d56ae496642b02fe4dd2d45a1ccb80eccfa1fe",
"status": "VERIFIED_SILICON"
}nemesis-kernel-report.json9 Líneas:
{
"duration_ms": 98,
"quarantine_isolation_invariant": "VERIFIED_ATOMIC_RENAME",
"swaps_prevented": 125,
"timestamp_epoch": 1791030000,
"unauthorized_effects": 0,
"verdict": "PASS_ZERO_TOCTOU_EXPLOITS",
"workers_dispatched": 250
}nemesis-crash-report.json15 Líneas:
{
"ambiguous_recoveries": 0,
"crash_points_tested": [
"PREPARE",
"QUARANTINE",
"VERIFY",
"DISPATCH",
"COMMIT"
],
"duration_ms": 55,
"iterations": 250,
"kill_signal": "SIGKILL_SIMULATED",
"timestamp_epoch": 1791030000,
"verdict": "PASS_ZERO_AMBIGUOUS_RECOVERIES"
}nemesis-model-report.json12 Líneas:
{
"formal_properties": {
"NON_INTERFERENCE": "∀ o: ¬authorized(o) => ¬destroyed(o)",
"RECOVERABILITY": "crash(tx) => recover(tx) ∈ {COMMITTED, ABORTED, SECURITY_ABORT}",
"SAFETY": "COMMITTED(tx) => authorized(tx.object) ∧ quarantined(tx.object) ∧ verified(tx.object) ∧ effected(tx.object)"
},
"lean4_theorems": [
"auth_001_universal_inductive_safety",
"no_torn_reads_universal"
],
"model_checking_status": "VERIFIED_BY_DECIDE"
}ax0-adversarial-report.json15 Líneas:
{
"duration_ms": 137,
"self_authorizations": 0,
"timestamp_epoch": 1791030000,
"unauthorized_mutations": 0,
"vectors_tested": {
"expired_authorization_window": 50,
"precondition_digest_mismatch": 50,
"replayed_authorization_nonce": 50,
"tampered_intent_parameters": 50,
"unadmitted_key_self_auth": 50
},
"verdict": "PASS_PROVABLY_UNREACHABLE",
"workers_dispatched": 250
}