Every spine–core link of the four released SUs (SU2, SU6, SU12, SU13), joined to its physical patch path and colored by whether the spine receives any light. The pattern repeats identically in all seven CNR frames: a replicated build-convention error, not localized damage.
If one frame were damaged, one row would differ. None does: rails 1–3 fail identically everywhere, rail 4 works everywhere. The defect is in how rails 1–3 trunks were terminated, fleet-wide.
Each board is one frame: columns are cassettes, rows are ports 1–6. Green cells are the only rail-2 positions that receive light — and they are the same cells in every frame: cassettes DEF-* / JKL-*, ports 5–6. Identical fixed points in seven independent frames is the signature of a deterministic ordering/polarity reversal — the survivors are the positions where the flipped mapping happens to coincide with the correct one.
Each duplex link is a fiber pair: transmit one way, receive the other. If a trunk cassette is built to the wrong polarity convention (Type-A where Type-B belongs), the pairs cross — light still flows toward the cores, but every return lands on the wrong fiber and dies. The link never trains, so it is invisible to topology tools; only light-level readings expose it.
Next action: light-meter one dead cassette (e.g. CNR-105 ABC-1) against its rail-4 neighbor to document the flip, then apply the same correction to all seven frames. Frame map: SU2 → CNR-105/108 · SU6 → CNR-117/211 · SU12 → CNR-229/232 · SU13 → CNR-138. Un-released SUs likely share the defect — correct their frames before release.
Live light/dark state of all 1,152 SU7 spine–core links by rail (CG1–4), polled at both ends. Each cell is one core-side transceiver; a red outline means both halves are dark — swap that optic. Embedded here for cross-reference alongside the polarity findings above. Open full page ↗