Teleport benchmarks
All measured on this machine (Apple Silicon, macOS) on 2026-05-24 unless noted. Numbers are
labeled measured (taken here), reference (standard physics/coding values fed to the
calculator), or modeled (computed from a model, e.g. light-time). Cold vs. warm is called
out for every build/runtime figure. See LIMITATIONS.md for what these do and do not claim.
1. RF over the air (measured, real radios)
protocol69 single-symbol teleport (Pluto -> NESDR, 433.92 MHz)
| Metric | Value |
|---|---|
| TX | Pluto on-chip DDS tone at 66.094 kHz (= symbol 66), TX gain −20 dB |
| RX recovered tone offset | +66.281 kHz -> symbol 66 (FFT argmax, 65536-pt Hann, 20 kHz DC guard) |
| Tone SNR | 66.1 dB |
| TX-on vs TX-off at the tone bin | +49.07 dB |
| Live ambient capture energy (433.92 MHz, 0.2 s) | RMS 0.558, peak 1.414, DC 0.0069, energyDetected=true |
| protocol69 validation | HTTP 200 validated; Flow/UDP 200 ok; drift (expectedProjection 68) -> 400 fail-closed |
Full framed cache key over the air (8-FSK + RS + conv + Whip)
Frame: GNOSIS-RFKEY/1 prefixHash=66 prefixLen=7 boundaryDimension=12 stationCount=2 crc16=a4ce.
Modem: 2.4 MS/s, 8000 baud (300 samples/symbol), base 96 kHz, fec rs+conv.
| Pluto TX gain | RX result (NESDR -d 0) |
|---|---|
| −20 dB | recovered byte-exact, sync 0.998, rsErrors 0, CRC pass |
| −45 dB | ok, sync 0.744, rsErrors 0 |
| −50 dB | ok (rs+conv); FEC-off frame CRC-FAILS at the same power, sync 0.658 -> the coding gain on real silicon |
| −53 dB | fails (sync 0.41, body too corrupt) |
| −60 dB | below sync threshold (sync 0.229) |
2. FEC / modem A/B (measured, software, deterministic seeds)
Eb/N0 sweep, 8 seeds, payload = the cache-key frame, channel = AWGN at the target Eb/N0 (noise fixed near full-scale, signal scaled — the SDR-AGC-faithful model):
| Eb/N0 (dB) | raw (none) | conv only | rs+conv |
|---|---|---|---|
| 1–2 | 0 | 0 | 0 |
| 3 | 0 | 0 | 4/8 |
| 4 | 0 | 1/8 | 7/8 |
| 5 | 0/8 | 8/8 | 8/8 |
| 6 | 0 | 8/8 | 8/8 |
The FEC cliff is ~5 dB: raw 0/8, FEC 8/8 — the headline A/B. Whip fork/race/fold recovers the exact key with a fully-faded replica. RS(GF256) corrects exactly 10 byte-errors across 200 random trials; the K=7 Viterbi corrects scattered bit errors; both verified in unit tests.
3. Link-budget distance ladder (reference physics)
Reference config: X-band down 8.42 GHz, 20 W (43 dBm), 4 m spacecraft dish, 34 m ground dish, T_sys 30 K, 1000 bps, 8-FSK + RS+conv + 4-way Whip. Required Eb/N0 ≈ −4.2 dB after coding.
| Rung | Link margin (dB) | Closes | Round-trip the teleport avoids |
|---|---|---|---|
| ISS / LEO | 152.93 | yes | ~0 s |
| GEO | 114.32 | yes | ~0 s |
| Moon | 93.70 | yes | 2.6 s |
| JWST / L2 | 81.88 | yes | 10 s |
| Mars (opposition) | 50.66 | yes | 364 s |
| Mars (mean) | 38.36 | yes | 1,501 s |
| Mars (conjunction) | 33.34 | yes | 2,675 s |
| Jupiter (mean) | 27.58 | yes | 5,190 s |
| Saturn (mean) | 22.29 | yes | 9,540 s |
| Uranus (mean) | 16.21 | yes | 19,213 s |
| Neptune (mean) | 12.33 | yes | 30,021 s |
| Pluto (mean) | 9.98 | yes | 39,361 s |
| New Horizons | 6.91 | yes | 56,039 s |
| Voyager 1 | −2.42 | no (break rung) | 164,114 s |
| Proxima Centauri | −66.68 | no | ~2.68e8 s |
Properties (proved + checked): margin is monotone non-increasing in distance; teleport saving is monotone non-decreasing. So a rung that closes dominates every nearer rung. The break rung (Voyager 1 at this dish/power) is reported, not hidden — raise the dish/power and it moves outward. Mars-mean detail: FSPL 277.998 dB, EIRP 91.36 dBm, Rx power −119.69 dBm, N0 −183.83 dBm/Hz, received Eb/N0 34.13 dB, coding gain 7.5 dB, Whip gain 6.02 dB, margin 38.36 dB.
4. Teleport latency A/B (measured replay vs modeled round trip)
geodesicLength:0 local replay measured at ~0.19 ms (193 µs). Round trip is the modeled
light-time (never actually waited; the model records the true seconds).
| Rung | Round trip | Local replay | Speedup |
|---|---|---|---|
| Moon | 2.6 s | 0.19 ms | 13,267x |
| Mars (opposition) | 364 s | 0.19 ms | 1,884,465x |
| Mars (mean) | 1,501 s | 0.19 ms | 7,765,652x |
| Mars (conjunction) | 2,675 s | 0.19 ms | 13,840,118x |
| Jupiter (mean) | 5,190 s | 0.19 ms | 26,851,899x |
| Neptune (mean) | 30,021 s | 0.19 ms | 155,313,042x |
| Voyager 1 | 164,114 s | 0.19 ms | 849,044,631x |
Relay chain Earth -> Moon -> Mars -> Jupiter avoids a 6,693.9 s round trip. The teleport admits the over-the-air-recovered 66 (projection 69), and a forged key is rejected (fail-closed).
5. Skymesh (measured, two real NESDRs)
Two NESDRs (SN 94092559, 78922015) on the strong local FM carrier 102.49 MHz, 2.4 MS/s. Witness = gain-invariant top-3 of 64 spectral bands, agreement = top-3 overlap >= 2.
| Check | Result |
|---|---|
| Witness, dev0 / dev1 | 1935 / 1929 -> AGREE (dominant band 31 on both) |
| HIT (same sky) | both TELEPORTED, match=true, in lockstep |
| MISS (different sky, e.g. 98 MHz) | top-3 overlap 0 -> DENIED |
| TAMPER (genuine witness, forged broadcast 67) | projection 64 != 69 -> DENIED (fail-closed) |
| Band-count discrimination | 64 bands: HIT shares 3, MISS shares 0 (16 under-discriminates, 128 splits the carrier) |
| TS tests | 12/12 pass |
6. Build + test timings (cold vs warm)
| Target | Cold | Warm / incremental | Tests |
|---|---|---|---|
| libiio v0.25 (source, cmake -j4) | ~1–2 min from clean clone | — | tools usable from build tree |
teleport.rs lib + tests |
5.41 s | — | 3/3 |
| fat-station teleport route | 13.95 s (deps cached) | 0.18–0.24 s no-op | 32 lib tests pass |
| moonshine (teleport gate) | — | 11.61 s warm | 4/4 |
| monster-guard / swarm / resident | 8.44 / 9.34 / 9.43 s cold | — | 2/2 each |
| gnosis-uring (teleport gate) | 54.0 s cold debug; 2m07s cold release (LTO) | 0.96 s incremental, 0.25 s no-op | 146 lib + 5 integ + 4 flow-bench |
aeon-monitor --skymesh |
~11.9 s cold | 1.8–1.9 s incremental, 0.16 s no-op | 7/7 |
| rtlsdr-mock-sim vitest (full PHY + skymesh + bpsk) | ~7 s | — | 40+ pass, 2 hardware-skipped |
Lean SkymeshTeleport |
— | 1.0 s | builds, no sorries |
Lean MarsDistanceLadder |
— | 288 ms | builds, no sorries |
mesh /chain 7B (reference, R1<->R1 WiFi) |
— | 766 ms warm | argmax 12095 ("Paris") |
All RF/inference benchmarks: report the substrate path and the executed shape, not just wall-clock. These are component figures, not an end-to-end production-throughput claim.