docs: Wave N+2 β — military-technical benchmark vs MPU5/Rajant/Silvus (#22)

* docs: wave N+2 beta — military-technical benchmark vs MPU5/Rajant/Silvus

Deliverables:
- BENCHMARK_VS_MANET_2026-07-04.md (212 lines): 8-metric rubric, scored table
  for Tri-Net vs 6 competitors, 3 strategic conclusions
- _recon/BENCHMARK_RECON.md (303 lines): 40+ URL-cited source data,
  per-competitor tables, 8 independent field tests, 6 arXiv benchmarks,
  honest data-gaps section

Key findings:
- We win uniquely on M5 spec-openness, M6 audit-verifiability, M7 silicon-anchor
- MPU5 vendor claim (150 Mbps) vs field (2.5-9.3 Mbps Aerobavovna) — verifiability card
- Babel beat OLSR/BATMAN in independent testbed (9s repair) — validates E4 choice
- US Army documented MPU5 range degradation on SPOKE damage — validates E5 k=2 design

Guardrails preserved:
- 4 dies SKY26b marked submitted, not returned
- Tri-Net M2/M3 marked as gap where pre-silicon
- Score 0 for goTenna/TrellisWare = data not found, not product judgment

Anchor: phi^2 + phi^-2 = 3

* docs(benchmark): merge local 337-line format with cloud 212

Added 5 additional source URLs from local version:
- csrc.nist.gov CMVP list
- silvus.com SC4400 datasheet + 559-node demo
- doodlelabs.com Multi-Hop PDF
- persistentsystems.com/products/mpu5

Updated M3 self-heal with Sprint 2 done (176/0 tests, d640423).
Kept compact single-line rubric format from cloud.

Anchor: phi^2 + phi^-2 = 3

---------

Co-authored-by: Perplexity Computer <agent@perplexity.ai>

Author: gHashTag