v0.435.0 — THE IN-MEDIUM SAMPLE SHIP

B277 (PAPER_2272): the pair cap's SECOND branch meets real data.

- PAPER_2264 (B269) split the NS enstrophy cap by the B112 context
  rule: vacuum-coupled flows cap at 17/20, in-medium flows at 197/200
  - a flagged falsifiable prediction. B276 graded the vacuum-side
  statistic on isotropic8192; B277 puts the in-medium branch in front
  of wall-bounded turbulence.
- JHTDB channel flow (Re_tau ~ 1000), t = 1.0, fd4lag4 gradients,
  official REST service, publicly sanctioned testing token, 2 x 1,000
  deterministic points (seeds 28/29; bit-reproducible). The walls
  (|y| > 0.9) were EXCLUDED - disclosed in the paper, the module, the
  provenance file, and the registry.
- THE IN-MEDIUM BRANCH HOLDS: sampled 1182-form ratios 0.039143 /
  0.031072, sub-batch max 0.073405 - thirteen times under 197/200 and
  eleven times under 17/20. Verdict issued by the tier-3 harness, now
  carrying the `cap` argument its B269 docstring promised (first real
  use). Positive-stretching fraction 0.736/0.764 - the DNS skewness
  reproduced in shear flow.
- Rule 7, the headline: at these margins the sampled statistic CANNOT
  DISCRIMINATE the two branches - both pass by an order of magnitude.
  This is a CONSISTENCY PASS for the in-medium branch, NOT the pair-cap
  discrimination; that lives in the far tail with the extreme-event
  kill test and stays OPEN. ns_proof_set() data front 2 says exactly
  that; four fronts and three flags unchanged in count.
- Daniel ledger unchanged: JHTDB full token / Kerr letter = the kill
  test + the pair-cap discrimination; THz bench; GFZ deep-sonic; U_i
  spec; SEG-Y; quartz bench; tester answers.
- Housekeeping, guarded: the staleness sweep's own finds are fixed AND
  pinned as SHIP GUARD v10 - README whitepapers badge measured live,
  README census sentence carries the live figures, WHITEPAPER_INDEX
  title column repaired on 2262-2272 (PAPER_2261's title had been
  copy-propagated); SHIP GUARD v5 label regex case-insensitive.

Gate 6,102/0. Dispatch 2,323 keys. Registry 7,112 rows. Acceptance
109/109 (simulator v1.90.0 untouched).
