host: Linux 7.0.0-1012-aws aarch64 | AWS Graviton3 c7g.2xlarge (8 vCPU, 15.3 GB RAM) | GitHub Actions self-hosted runner df-runner-trustdb-arm64-1 (instance i-0f50fc14e7d367a06, ca-central-1) | rustc: rustc 1.94.1 (e408947bf 2026-03-25) (aarch64-unknown-linux-gnu) | compiled 2026-09-12T01:37:34Z | commit: 673fe85c (CI merge commit of PR #3487 head e6fdd5ec onto main 960a20bd) | cmd: cargo test --release --features v2-mera --test cnsa_m1_bench -- --ignored --nocapture executed by .github/workflows/pqc-interop.yml, job "interop aarch64 + P10 evidence harness", run 34662891115, through scripts/pqc_evidence.sh; release compile 33 min 15 s on the runner; load average 9.39 9.53 8.32 immediately before the timed run (the harness's own compile had just finished; no other cargo or rustc process on the host), so this run is NOT quiesced. Workflow artifact pqc-evidence-aarch64-673fe85cbe168578e043021bd5fe57fb228a57cd; the harness's host.txt, cnsa_m1_bench.txt and KAT summary follow verbatim. ---- host.txt ---- utc: 2026-09-12T01:04:19Z host: Linux 7.0.0-1012-aws aarch64 os: linux arch: aarch64 cpu: nproc: 8 mem_gb: 15.3 rustc: rustc 1.94.1 (e408947bf 2026-03-25) cargo: cargo 1.94.1 (29ea6fb6a 2026-03-24) rustc_host: aarch64-unknown-linux-gnu commit: 673fe85cbe168578e043021bd5fe57fb228a57cd branch: uncommitted_files: 0 profile: release features: v2-mera bench_cmd: cargo test --release --features v2-mera --test cnsa_m1_bench -- --ignored --nocapture kat_cmd: cargo run --release --features v2-mera --bin run_acvp_demo compiled_at: 2026-09-12T01:37:34Z concurrent_cargo_rustc_before_run: 0 unknown load_average_before_run: 9.39 9.53 8.32 ---- cnsa_m1_bench.txt ---- host: Linux 7.0.0-1012-aws aarch64 | rustc: rustc 1.94.1 (e408947bf 2026-03-25) | 2026-09-12T01:37:34Z | commit: 673fe85c | cmd: cargo test --release --features v2-mera --test cnsa_m1_bench -- --ignored --nocapture nproc: 8 | cpu: ----- Finished `release` profile [optimized] target(s) in 0.63s Running tests/cnsa_m1_bench.rs (target/release/deps/cnsa_m1_bench-1ddc7b9b30aa521d) running 3 tests ==== CNSA Cat-5: four-way COST comparison (host-measured, RELEASE) ==== N=100000 x 115B | AES-256-GCM-SIV seal med 1045 ns (ct 136 B) | ML-DSA-87 sign 0.642 ms / verify 0.529 ms (every value is a COST -> LOWER IS BETTER) unencrypted AES-encrypted legacy-PQC-Cat5 8DB(M1) storage B/record 115 136 4763 136.07 write ns/record 0 1045 643349 1062 auth-vfy s / 100k records 0 0 52.9 0.0005 capability matrix: unencrypted encrypted:NO quantum-safe:NO per-datum-Cat5:NO tamper-evident:NO AES-encrypted encrypted:YES quantum-safe:NO(classical) per-datum-Cat5:NO tamper-evident:per-record-tag legacy-PQC-Cat5 encrypted:YES quantum-safe:YES per-datum-Cat5:YES tamper-evident:YES (at ~40x storage) 8DB (M1) encrypted:YES quantum-safe:YES per-datum-Cat5:YES tamper-evident:YES (+O(logN) braid) 8DB vs legacy-PQC-Cat5 (higher = our win): storage : 35x less /record write : 606x less /record auth-verify: 100000x less /set 8DB vs plain AES-encrypted (the honest 'cost of adding Cat-5'): storage : +0.073 B/record (+0.0539%) write : 1.02x of plain-encrypted write ==================================================================== test bench_four_way_unencrypted_encrypted_legacy_trustdb ... ok ==== CNSA 2.0 Cat-5 M1 micro-benchmark (host-measured) ==== provider: MlDsa87Provider (FIPS 204) | N = 100000 records of 115B unit sign (ns): med=260413 (min 258189 / max 268352) unit verify (ns): med=529679 (min 525882 / max 543371) ML-DSA-87 signature: 4627 bytes | boundary-sig wire: 7332 bytes STORAGE — signature overhead: row-level Cat-5 : 462700000 B (462.7 MB) [= N x 4627] MERA boundary : 7332 B (one signature) => 63107x less signature storage single-record bloat: 115B -> 4742B = 4023% overhead SIGN CPU @ N: row-level : 26.041 s [= N x med sign] boundary : 0.002174 s (BLAKE3 over N + 1 sign) => 11981x less sign CPU VERIFY CPU @ N (edge / audit): row-level : 52.968 s [= N x med verify] boundary : 0.000536 s (1 verify) => 98739x less verify CPU =========================================================== test bench_cat5_rowlevel_vs_mera_boundary ... ok ==== plaintext vs classical-encrypted vs PQC: storage / speed / search (RELEASE) ==== N=100000 x 115B records (storage/speed) | N_search=2000 terms (search demo) ONE-TIME key establishment (paid once per key/session, NOT per record): classical KDF (PBKDF2-HMAC-SHA-512, NIST SP 800-132): 662954438 ns, 0 B wire (only the 16-byte salt is public) PQC KEM encapsulate (ML-KEM-1024 + X25519 hybrid) : 151716 ns, 1616 B wire (ml_kem_ciphertext 1568 B + wrapped_key 48 B) PQC KEM decapsulate (recipient side) : 148068 ns => PQC key-establishment wire overhead amortized over N=100000 records: 0.012336% of total dataset size PER-RECORD storage + speed (AEAD held IDENTICAL across classical & PQC): plaintext classical-encrypted PQC (ours) storage B/record 115 131 131 write ns/record 0 1176 1175 read ns/record 0 1189 1184 => PQC adds 0.0000% storage and -0.0850% write latency vs classical (same AEAD; the only PQC cost is the one-time KEM above) SEARCH -- can we query the data in each state, and how fast: plaintext : 1303 ns (1 hits / 2000 terms, direct in-memory index) classical-encrypted : 1923007 ns (1 hits / 2000 terms) -- NO purpose-built encrypted index exists for this path; this is a full decrypt-and-scan (the only secure option -- a plaintext side-index would leak) PQC (ours) : 426857 ns (1 hits / 2000 terms) -- the search INDEX is itself sealed under the PQC-KEM-established key; one decrypt-the-index op, then native-speed search (search::sealed_content_index) => our sealed-index search is 5x faster than the classical decrypt-and-scan at N_search=2000 ================================================================================= test bench_plaintext_vs_classical_vs_pqc_storage_speed_and_search ... ok test result: ok. 3 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 36.49s EXIT=0 ---- run_acvp_demo.txt (summary line) ---- Local KAT result: 35 / 35 passed (0 failed). ---- SHA256SUMS.txt (as produced by the harness) ---- 4a51b1a25b3c8210ad5b40ac146d1ba29fa70e3d3b6460f2fe0f14d9e09fde02 host.txt e3c8b097e13febf3b3d42d7fa61c6347ab4d5276e5f6f63864dafe2c3bbbd18c build.txt e0d6ed650d064fb57be29c5ab69001381683b9a658a6f1fa01dbee7cb86db754 cnsa_m1_bench.txt 4a51b1a25b3c8210ad5b40ac146d1ba29fa70e3d3b6460f2fe0f14d9e09fde02 host.txt 9e6173270524747c219cd39cd8056b3fe4b8dc4a9d168a3edfbda6e37d634aa1 pqc_live_algorithm_swap.txt 081997e24417abb8d020d11eaa2eac4a0804fa218f4c8dc0c34137f8b211934f run_acvp_demo.txt