============================================================================== SYNTHETIC - classical software simulation on generated numbers. Not a medical device, not control software, not sensor fusion, not field evidence. ============================================================================== spark-membrane 0.2.0 | python3 -m spark_membrane demo --seed 42 protocol spark-membrane-demo-v2 sha256 69efc39ab63b4b11... locked hash: MATCH PLACEHOLDERS: every threshold and cap is UNCALIBRATED (upstream defaults or author choices; docs/PROTOCOL.md). READ FIRST - results from the pinned repos that do not flatter OES32: [public_dataset_negative] On the locked NASA SMAP/MSL protocol, OES32 did not meet its pre-stated success criterion: it beat only EWMA on SMAP, and maxabs and CUSUM scored a higher pooled F1 (not significantly). [synthetic_negative] On the oes-telemetry-bench SYNTHETIC held-out set, max-abs matched OES32's detections (6/7 each) with fewer false-alarm episodes per normal hour (11.54 vs 23.08); the rates come from about 5 minutes of synthetic normal time and are highly uncertain. [synthetic_negative] The multi-quantum-oes preregistered SYNTHETIC stress evaluation found no demonstrated advantage: the block score did not beat a simple max-abs baseline. [unrun] oes32-hls FPGA synthesis is UNRUN; its latency figure is a design target, not a measurement. [unrun] The qldpc_decoder_cpp HLS kernel has no belief-propagation updates yet; its HLS synthesis and hardware-in-the-loop runs are UNRUN. FRAME BUS: 40 native-32 frames, cadence 1 s, all audited; stream sha256 a074efbf7bcd91cf... EWMA/CUSUM calibration: separate 64-frame event-free stream (calibration-seed42.jsonl, never audited), sigma of frame means 0.00107; restart after alarm: yes. AUDIT (T=0): ACCEPT iff protocol_hash AND residual AND sidecar_A AND sidecar_C0 AND sidecar_C1 AND sidecar_Cfold AND weighted. Advisory, never gating: maxabs, ewma, cusum. Residual and weighted are separate checks, never blended. t event verdict failing gating checks (check@index value>threshold) | engines 0-19 none ACCEPT - | all quiet 20 single_channel_spike LATCH residual@13 0.1996>0.08; sidecar_A@13 0.1996>0.08; sidecar_C1@13<->29 0.1986>0.08; sidecar_Cfold@13<->14 0.1931>0.08 engines: DISAGREE fired=residual,sidecar,ewma,cusum quiet=weighted,maxabs residual R=0.1996 vs tol 0.08 at index 13 -> LATCH | weighted score=0.1045 vs 0.5 -> quiet (separate formulas, never blended) 21-23 none ACCEPT - | all quiet 24 spike_above_explorer_cap LATCH residual@5 0.4010>0.08; sidecar_A@5 0.4010>0.08; sidecar_C1@5<->21 0.3987>0.08; sidecar_Cfold@4<->5 0.4078>0.08 engines: DISAGREE fired=residual,sidecar,ewma,cusum quiet=weighted,maxabs residual R=0.4010 vs tol 0.08 at index 5 -> LATCH | weighted score=0.2089 vs 0.5 -> quiet (separate formulas, never blended) 25-26 none ACCEPT - | all quiet 27 two_channel_fault LATCH residual@27 0.1515>0.08; sidecar_A@27 0.1515>0.08; sidecar_C0@2<->18 0.1557>0.08; sidecar_C1@11<->27 0.1491>0.08; sidecar_Cfold@26<->27 0.1543>0.08 engines: DISAGREE fired=residual,sidecar quiet=weighted,maxabs,ewma,cusum residual R=0.1515 vs tol 0.08 at index 27 -> LATCH | weighted score=0.0841 vs 0.5 -> quiet (separate formulas, never blended) 28-29 none ACCEPT - | all quiet 30-33 slow_common_drift ACCEPT - | DISAGREE fired=ewma,cusum quiet=residual,sidecar,weighted,maxabs 34-35 none ACCEPT - | all quiet 36 global_common_shift LATCH residual@31 0.6072>0.08; sidecar_A@31 0.6072>0.08; sidecar_C1@15<->31 1.2092>0.08; weighted@31 0.6024>=0.5 engines: all fired residual R=0.6072 vs tol 0.08 at index 31 -> LATCH | weighted score=0.6024 vs 0.5 -> ALARM (separate formulas, never blended) 37-39 none ACCEPT - | all quiet Gate ACCEPTED while advisory engines (cusum, ewma) fired at t=30-33: the gating checks do not see these events. Advisory engines never change the verdict; this is reported, not hidden. EXPLORER (new code; SYNTHETIC; one index, |delta| <= protocol cap; reference and thresholds read-only): t=20 REPAIRED_IN_SIM one-index delta -0.183669 at index 13 passes every gating check (|delta| <= cap 0.25); original frame kept shield gate: EXPLORER WRITE -> REFUSED (only DECODEUR or GARDIEN may write; EXPLORER proposals stay in simulation) t=24 LATCH_HELD no single-index delta within cap 0.25 passed the audit in 16 seeded proposals (cap was binding) t=27 LATCH_HELD no single-index delta within cap 0.25 passed the audit in 16 seeded proposals t=36 LATCH_HELD no single-index delta within cap 0.25 passed the audit in 16 seeded proposals (cap was binding) FAIL-CLOSED: protocol tolerance edited from 0.08 to 0.5 after the lock -> sha256 0d1b61891f580816... -> LATCH [failed: protocol_hash; not evaluated: 6 checks]; explorer LATCH_HELD (explorer disabled: protocol hash mismatch (fail-closed)) 512-CHANNEL BUS: 16 blocks x 32 = 512 channels, resampling: none -> LATCH block 11: LATCH residual@ch7 (global 359), sidecar_A@ch7 (global 359), sidecar_C1@ch7 (global 359), sidecar_Cfold@ch6 (global 358); explorer REPAIRED_IN_SIM other 15 blocks: ACCEPT SUMMARY: 36 ACCEPT, 4 LATCH, 7 frames with engine disagreement, 1 REPAIRED_IN_SIM, 3 LATCH_HELD (explorer evidence class SYNTHETIC). EVIDENCE: trail 98 records (measurement-trail v1), head 2d2998ff451d73c1...; claims gate: 9 ledger claims admitted, 26/26 probe overclaims refused, 6/6 honest denials admitted. ============================================================================== SYNTHETIC - classical software simulation on generated numbers. Not a medical device, not control software, not sensor fusion, not field evidence. ==============================================================================