A Wi-Fi test bench
you run yourself.

The lab rig, drawn as it sits on the bench

What you need to buy and cable up to run MangoWifi against a real access point. Stage 1 is the bench you build now; Stage 2 adds the conducted-RF gear the roaming and range tests need.

Stage 1 · build now

One AP, one Wi-Fi client, three wired boxes

Stage 1. Everything wired lives on the right; the Agent's only link is Wi-Fi. Because the Console pairs to the Agent's Wi-Fi address, the control channel, the latency probe and the Agent's iperf3 load all cross the AP under test — that one air hop is the thing every Wi-Fi 8 runner is measuring.

Stage 2 · roaming & range

Radios in shield boxes, the air replaced by coax

Seamless-roaming and throughput-at-range tests need signal strength you can control. You can't get that repeatably by walking a laptop down a corridor, so the radios move into RF shield boxes and a programmable attenuator sets the path loss.

Stage 2. Each radio sits in its own shield box with its antennas replaced by coax, so the only RF path is the cable. The Agent's signal is split and fed through a two-channel programmable attenuator: the Console ramps channel 1 up while channel 2 comes down, the Agent's link to AP-A fades as AP-B's grows, and the roam happens at a known, repeatable point. Drawn as one antenna chain — a 2×2 radio needs two splitters and four attenuator channels.

One binary, two roles

The same executable runs as Console or Agent. Which one it becomes is read from its config file at startup — there is no CLI flag — and the config location comes from MANGOWIFI_DATA_DIR.

Inside the console

Captured from the application itself, with the real catalog loaded. No rig is attached, so the app runs its preview path — which is why the radios are chipped MOCK and the checks read as not yet run. Click any shot to read it full size.

1 / 10

Four sections, one catalog

Every test is one struct in a single catalog, tagged with the section it belongs to and the clause it traces back to.

• Baseline

  Live

  Twenty-four pre-flight checks across environment, radio, station, RF, AP and traffic. Is the kernel new enough, is the radio bound, does the AP appear in a scan, does iperf3 exist.

• Criteria

  Live

  Per-metric rows in the shape commercial suites use — basic, TCP, UDP, RFC 2544, range, capacity, roaming, application and backwards compatibility.

• Wi-Fi 8

  3 of 18 live

  Draft 802.11bn D2.0 (July 2026) across MAPC, Enhanced MLO, Seamless Roaming, Reliability and Power. The three live rows are all Reliability KPIs.

• WFA

  Catalogued

  Wi-Fi Alliance certification-aligned rows — Wi-Fi 6/6E/7, Vantage, Voice-Enterprise, WPA3. Each carries a coverage rating; the runners are not written.

Three Wi-Fi 8 KPIs that run today

All three are UHR Reliability rows. Each scores against one percentile and reports the rest, and each states whether the concurrent load actually ran — so an unloaded pass is never mistaken for a UHR result.

• p95 latency floor

  WFA UHR KPI — p95 ≤ 5 ms (AI/XR target)

  wifi8_rel_p95_latency

  200 packets per second for 30 seconds, with udp_high_load fired once on the Agent alongside it. Passes at p95 ≤ 5 ms, warns to 10 ms; loss passes at 0.1% and warns to 1%.

• 30-minute endurance with drift

  WFA UHR KPI — p95 ≤ 5 ms held across the window

  wifi8_rel_stress_endurance

  The same thresholds over 30 minutes, with the load cycled continuously rather than fired once — a single iperf3 pass is 15–30 seconds and would leave most of the window idle. Split into six 5-minute buckets so drift is visible as a trend.

• p99.9 tail

  UHR AR/XR aspirational — p99.9 ≤ 10 ms

  wifi8_rel_p999_latency

  30 minutes at 200 pps, scored on p99.9: passes at 10 ms, warns to 20 ms. The project is explicit that 360,000 samples put roughly 360 in the top 0.1% — directional, not certification-grade, which would want about a million.

The Wi-Fi 8 catalog in full

Draft 802.11bn D2.0, dated July 2026. Eighteen rows across five categories, each carrying the clause it tests against. Three execute today; the rest are inventory with a spec reference and no runner, which is what makes the gap auditable.

Three rigs, two of them real

The same build runs on one laptop for development and on a two-box bench for real measurement. The third arrangement is described because it is designed, not because it exists.

• Works today

  A · Dev loopback

  One machine, no access point. Console and Agent on the same box, measuring sub-millisecond loopback — which verifies the wiring, not the radio.

• The target

  B · Real lab

  Two boxes and the AP under test, plus a LAN receiver on its wired side. The probe crosses the air both ways; throughput crosses air then copper.

• Needs hardware

  C · Not yet built

  A programmable attenuator for roaming, a multi-AP cluster for MAPC, Wi-Fi 8 station silicon for Enhanced MLO, and optionally an RF chamber.

Download

No account and nothing phoning home. Every release publishes checksums beside it.

What is shippable today

Taken from the project's own scope statement rather than softened for a product page.

• Ready for a lab bench

  • Wi-Fi 6/7 Baseline and Criteria
  • Wi-Fi 8 Reliability: p95 latency under load
  • Wi-Fi 8 Reliability: 30-minute endurance with drift analysis
  • Wi-Fi 8 Reliability: p99.9 tail latency

• Needs hardware first

  • MAPC rows — a multi-AP cluster
  • Enhanced MLO rows — real Wi-Fi 8 station silicon
  • Seamless Roaming rows — a programmable attenuator

• Not in the box

  • Certification-grade PHY measurement — chamber gear and a vector signal analyser
  • A co-located echo responder on the LAN receiver, which would let Agent-side tests capture p95 too