active timing-mark results
the historical c25519 series completed 960 executions. the primary endpoint combines a correct declaration and original-circuit responsiveness; at 75 ms every original circuit closed.
cmd/lab -set mark has a completed historical c25519 series: 40 repeat blocks with 24 conditions
per block, 960 executions. all established and were scorable. the endpoint requires a correct
target declaration, its original circuit open at cutoff and an echoed reserved probe before
cutoff. a successful send return alone does not prove responsiveness.
| tested relay periods, ms | correct + responsive, active | responsive target, active and sham |
|---|---|---|
| off | 40/40 | 40/40 in each arm |
| 25, 47.5, 50, 52.5 | 0/40 at each period | 40/40 in each arm at each period |
| 75 | 0/40 | 0/40 in each arm |
these counts held separately for immediate clients with additive cover and constant-rate clients at 50 ms. each denominator belongs to one client mode, relay period and arm; configurations are not pooled. the independent comparison unit is the repeat block, not its six circuits. all outcomes remain included. at 75 ms all six original circuits closed in each of 160 executions across modes and arms: availability loss, not demonstrated protection.
seven wrong active declarations and seven sham declarations remain visible in the full table. immediate clients with unpaced relays had 4/40 sham declarations. zero observed primary outcomes do not imply zero population probability, equivalence to guessing, a safe relay-period threshold or resistance to arbitrary active attacks. global no-mark calibration does not bound wrong attribution when a mark is present.
measurement used clean commit
f0f188539ca9241b67ad6593f1e66687e80ed967,
optimized go 1.27.3 on linux/amd64, base seed 1. binary sha256:
404796bd0feb555b2f04629169e979f4531303dc5188ab71eaba9746be521ae1.
these are measurements of that historical source and binary. the passive 2078098 results and
the cooperative key-memory series keep their separate provenance.
one-minute linux load across all 1920 before/after samples had min/median/max 0.34/0.69/1.32 in a reported 32-cpu environment. these are load averages, not percentages or independent trials. no load threshold or post-hoc exclusion was applied; timing/load effects were not isolated.
this is one-process loopback echo with one delayed encrypted client-entry flow and five ungated candidates. the detector observes completed forward frames into the last relay, including padding, using a known epoch, common clock and fixed code/lag search. shared-host effects, backpressure and closure can affect the evidence.
numbers and provenance are in the readme. the full condition table and method and limits retain the false-declaration and availability boundaries.