SafeWall

A concrete threat model (radio tracking, telemetry domains) aimed at a real gap for journalists, with unusually honest limits. Pocket Tor routers are a known category and every differentiating feature is listed in the team's own docs as planned or untested, so a 6 rewards strong reasoning over delivered novelty.

Unambiguously security and resilience work: preventing radio-based exposure and traffic identification for civil-society users. The score sits below the top because the resilience half, operation in a degraded scenario, is described only as future acceptance criteria and never demonstrated.

The intended flow of one cable and a glance at a panel is right, but the LTE link measured about 1.4 kB/s with unresolved timeouts, the dashboard runs on demo data, and operation needs root Linux rigs and SSH over USB. Member B's 2.5 is defensible given the broken data path, but the evidence and the team's own honest documentation support the median view that the concept is right while nothing is field-usable today.

Coherent craft: a full brand system with scripted logo exports, a generated 8-slide deck, Blender CAD renders of the powerbank-like enclosure, and a clean CSP-hardened dashboard with offline preview. The score reflects real polish that rests on renders and simulated statuses rather than verified live views, which supports the median over Member C's 6.

The router base is real and tested (nftables isolation, DHCP/DNS, 28 test cases including a namespace suite), but the defining security features are absent from code: no torrc, no Tor redirect rules, no fail-closed kill-switch, no DNS filter, and the API returns 503. Member B's 1 is too harsh given the tested router base and Member A's 4 slightly overcredits unimplemented claims; Member C's claim that DNS filtering is present in the repo is contradicted by STATUS.md and the other two reviews, so the median of 3 stands.