← Return to the implementation

The experiment that determined the boundary

These are selected fields from retained physical-run logs, not a simulated packet trace. Station addresses, RF payloads, device identifiers and credentials are omitted. The UTC timestamps and numeric measurements are unchanged.

Stock Pluto really did stream samples

At 2026-09-05 00:13:48.170 UTC, the concurrent ESP/Pluto run recorded:

blocks                 609
samples                1,218,000,000
sample-seconds         60.900
wall-seconds           232.167
host_delivery_ratio    0.262
application_queue_drops 0

The ratio is sample-seconds delivered divided by elapsed host time. Zero application queue drops does not prove continuous acquisition upstream of that queue. This measurement does not locate each hardware gap. It does establish that the configured 20 MS/s did not arrive as a gap-free real-time stream at the application.

Optimized USB ACK, real iPhone

The final stock-Pluto attempt began at 2026-09-05 07:23:40.280 UTC with concurrent USB enabled, the M3 retry train disabled, 20 MS/s RX and cached late-ACK transmission enabled. Its event stream contains two verified M2 events, two verified M4 events, seven deauthentication events and no DHCP completion. Deauthentication counts are log events, not seven independent people or seven unique connections.

2026-09-05T07:24:29.923Z  event 190  tx_begin / ACK
queue_to_push_ms        0.012
synthesis_ms            0.000
radio_bus_wait_ms       0.000
waveform_cache_hit      true

2026-09-05T07:24:29.923Z  event 191  tx_complete / ACK
push_bytes              36784
elapsed_ms              2.383
rf_airtime_ms           0.460

The equal log timestamps result from publishing timing records after the push returns. They do not mean the push was instantaneous. Queue-to-push is a software interval; elapsed_ms is push-call duration; rf_airtime_ms is calculated from the sample count. None is a calibrated measurement of end-of-uplink to first-downlink RF sample. In particular, subtracting waveform airtime from push duration would not reconstruct the actual air-start delay.

The cached waveform eliminated synthesis for this ACK. It did not eliminate RX buffering, USB scheduling or the delay before calling TX. The contemporaneous password complaint is contradicted by verified M2/M4 cryptographic checks. The station failed to progress after key exchange, rather than supplying the wrong passphrase.

Final E310/Windows implementation

The accepted compact-fanout FPGA run on September 9 completed verified M2/M4, authenticated CCMP, DHCP, ARP and three complete HTTP transfers. ESP UART reported HTTP 200 and 491/491 body bytes on attempts 906, 907 and 908. These are client-side read results, not host-side TX intentions. The release contains a different public example page and passphrase configuration; it does not expose the lab key.

The independently rebuilt publication copy reproduces 2,774 LUTs, 5,185 FFs, 1,420 occupied slices, 14.5 BRAM and zero DSP, with +2.305 ns setup and +0.091 ns hold slack. Windows self-tests and the ARM C++ build also pass. Those build results verify reproducibility of the released source; the physical claim comes from the retained station run, not compilation.

The earlier operator-supplied iPhone page photograph documents the E310 implementation before the last compact-fanout revision. It is not a fresh iPhone test of that final revision. Neither client result constitutes a full spectrum-mask, coexistence or 802.11 certification campaign.

Continue with the portable interface contracts