DWG NO. SC-3090 · SIDECHANNELRIG: RTX 3090 · CH LIVE: 1 · CH PLANNED: 12 · QUEUE 0
Sign in · Google

Is it training?

One GPU. Blue trains a model in secret. Red watches the shunt. Undetected runs go on the sheet, ranked by energy per token.

CH-01 · 12V rail shunt current · 1857 S/sReplay · rig offlineRegime: unlabeled

CH-01 shunt current

0.0AMeasured

Rail power

0WDerived · 12V nominal

Window energy

0.0kJIntegrated

Board power correlation

0.948rShunt vs NVML

Sensor manifest

ItemChannelInstrumentQtyRateStatus
CH-0112V rail shunt currentINA219 · 1 mΩ · 10 mA/LSB11857 S/sLive
CH-0212V rail voltageINA219 bus channel11857 S/s2 Planned
CH-03Core rail current and voltageShunt + ADC250 kS/s2 Planned
CH-04Memory rail current and voltageShunt + ADC250 kS/s2 Planned
CH-05Aux rail current and voltageShunt + ADC250 kS/s2 Planned
CH-06Near field probe · VRM phasesH field loop2Wideband3 Planned
CH-07Near field probe · memory busH field loop1Wideband3 Planned
CH-08Near field probe · PCIe edgeE field probe1Wideband3 Planned

Red Team

Label every capture window as training or inference. Beat 70% accuracy and the run is busted. Fewer channels outranks more.

Channels requested

Blue Team

Reach the task target with the least energy while Red stays near a coin flip. Pick the inference mode you can defend.

DefaultReviewedPublic

Blue · undetected runs

No.HandleModeRed accEfficiency
No scored runs yet

Red · confirmed detections

No.HandleChRunsAccuracy
1house_red1198%
One run per account per day · idle time played by house models
Rules · Sensor spec · Formal verification reading
Sheet
1 of 1
Rev
1 · CH-01 only
Limit
1 run / day