Relay configuration reconciliation — production action measurement Environment: {"commit":"a6d7695064443c8bb4e4200d481228bd5004e5eb","platform":"linux","arch":"x64","node":"v26.13.1","cpu":"AMD Ryzen 9 9950X 16-Core Processor","logicalCpus":8,"rounds":31,"warmups":5,"units":"milliseconds","timedAction":"production PublishClientConfig Run: exists + read + reconcile + write + log; seed write excluded"} Protocol: order base, head, head, base; fresh Node process per session; 5 warmups + 31 measured calls per input per session (62 measured per side). Each timed action performs real exists/read/reconcile/write/log operations. Synthetic seed write is outside timing. No mocks, test runner, cloud deployment, or UI. Every output checked byte-for-byte against expected seven entries; KEEP lines preserved in control. Action time: median [min, max] milliseconds, pooled over both sessions N kind base head speedup 1024 multiline 1.051 [ 0.778, 7.699] 0.542 [ 0.308, 5.616] 1.94x median improvement 0.509 ms; base full spread 6.921 ms; head full spread 5.308 ms 1024 single-line 0.569 [ 0.400, 0.885] 0.496 [ 0.408, 1.136] 1.15x 2048 multiline 2.490 [ 2.235, 6.665] 0.606 [ 0.475, 3.607] 4.11x median improvement 1.883 ms; base full spread 4.430 ms; head full spread 3.131 ms 2048 single-line 0.749 [ 0.651, 3.678] 0.761 [ 0.598, 1.350] 0.98x 4096 multiline 8.537 [ 7.834, 12.497] 1.085 [ 0.901, 1.986] 7.87x median improvement 7.452 ms; base full spread 4.663 ms; head full spread 1.085 ms 4096 single-line 1.363 [ 1.119, 3.482] 1.367 [ 1.141, 2.398] 1.00x 8192 multiline 30.825 [ 28.158, 40.869] 1.892 [ 1.582, 3.011] 16.29x median improvement 28.933 ms; base full spread 12.711 ms; head full spread 1.429 ms 8192 single-line 2.422 [ 2.113, 4.097] 2.513 [ 2.162, 3.668] 0.96x 16384 multiline 117.995 [107.351, 161.786] 3.955 [ 3.189, 6.360] 29.84x median improvement 114.041 ms; base full spread 54.435 ms; head full spread 3.171 ms 16384 single-line 5.912 [ 4.348, 9.375] 4.744 [ 4.396, 8.681] 1.25x Output SHA-256 (multiline, every size, both commits): 6abefcd17d11bba6e6c27e25c19185e2936f48575ab1956bf4503f1deeb492f0 All 1,440 production writes matched expected output across two commits and two sessions. Limit: 1,024 and 2,048 assignment differences are smaller than observed outlier spread. The clear timing proof is at 4,096, 8,192 and 16,384 assignments. Finite timing measurements support linear scaling for these inputs; they do not prove an asymptotic bound. Full relay cloud deployment and malformed/unclosed quote cases were not measured.