Identity inference,
collapsed to chance.

4.7%re-id · shield on

exposed shielded

Live scorecard

Re-ID · off

100%

attacker unhindered

Re-ID · on

4.7%

chance 6.25%

Throughput

97.6%

of baseline link

Emission

1.000×

not jamming

Collapse curve

re-ID vs mixing

Givens passes →op: 96 · re-ID 4.7%

Throughput optimum

vs feedback bits

Feedback resolution (bits) →5-bit · 97.6%

Sensing activity

solicitations / s

threshold 5.0 Hz — shield auto-engages above0.0 Hz

Shield controls

live model

Givens passes96min robust 48

Feedback resolution5bits · 802.11 {5,7,9}

Candidate identities16chance 6.25%

Link SNR20dB

Energy in

= energy out · 1.000×

out

Attacker vs. protector

synthetic · L0

Passive re-ID — shield off—

Passive re-ID — shield on—

Link throughput retained—

Deployment presets · adaptive shield

Prefer the terminal? The same instrument ships as veil — a dependency-free TUI & scriptable harness inside the crate (cargo run -p wifi-densepose-privshield --bin veil). Live-steer the shield with on/off · passes · bits · preset · optimize, or run veil doctor in CI.

Compliant waveform controls only — never jamming. The shield rotates its own beamforming feedback with keyed Givens rotations (energy-preserving), so a sniffer can't average out a stable identity while the associated receiver, holding the key, decodes normally. All figures are SYNTHETIC / evidence-level L0 from the reference model — not measured on hardware.