Wireless modes and optional fixed capture

raw_phy uses clean PHY rates. solo_80211 includes single-link MAC overhead without concurrent-link effects. full_wireless starts from the same Solo goodput matrix and applies active contention and hidden-terminal interference. The CLI and ncsim.models.wireless.configure_wireless use the same rate and RTS/CTS normalization. Explicit bandwidths are clean PHY rates in MB/s. Legacy none and csma_bianchi names remain accepted.

The default hidden-terminal treatment is effective_rate: active unsensed transmitters contribute receiver power to an SINR-based service-rate calculation. Shared endpoints conflict because they share a half-duplex radio. Zero service pauses transfers; if no future event can restore progress, the run reports blocked. This is distinct from a time limit or an unreachable route. The CLI returns nonzero for any status other than completed. A diagnostic positive outage floor is explicit and changes the model; it is not a remedy for outages.

Fixed-capture option

Select --interference full_wireless --hidden-terminal-model fixed_capture_overlap, or pass hidden_terminal_model="fixed_capture_overlap" to configure_wireless. It is disabled by default. The option holds each clean MCS fixed and uses an airtime-overlap and capture approximation for isolated hidden-link pairs. Pure sensed contention follows the default Bianchi calculation.

Supported operation is 802.11ax, 20 MHz, default MAC timings, and no RTS/CTS. Each receiver may have one active hidden interferer and neither member may simultaneously have sensed contenders. Unsupported active combinations raise an error instead of silently switching models. This is not a general multi-interferer or packet-level retry model. Saved packet comparisons cover equal 30 m links at MCS 5; they do not establish accuracy outside that setting.

Run python -m pytest tests/test_optional_hidden.py tests/test_public_interfaces.py for bounded regression checks. The study artifact contains the saved packet observations and its separate hidden comparison command. These checks do not launch ns-3 or a research campaign.

Other implementation changes

The event engine credits in-flight bytes before applying changed rates and serializes tasks on each compute node in ready order. minimum_hop routing minimizes hop count over usable positive-bandwidth links; shortest_path continues to minimize latency. These are correctness and routing changes, not effects of enabling the optional capture model.

The built-in conflict_aware_heft and uniform_discount_heft policies use modified communication estimates for static placement; all_on_fastest places all tasks on the fastest node. They supplement round_robin, manual, and the installed SAGA catalog and do not replace the default HEFT policy. Their placement estimates are not an interference-aware execution schedule.