CPW-FED quasi-Yagi v2, 915 MHz, single copper layer on FR4 1.6 mm — NO BAY.
Rear-fed through the SPLIT reflector on the family tab; the feed is printed
except ONE series NP0 0402 (the matching network in one part).
.gtl top copper (rear tie + rails + trace with the C gap + split reflector
     halves + strips + driven arms = one piece; 2 floating directors)
.gko board outline: the family flare, widened to +-81 mm over the driven
     element (l_drv 166 tips at +-83); 6 routed windows + the CPS slit
.drl Excellon: SMA pin 1.5 + 4 posts 1.7 (5.08 mm square; official KiCad
     footprint SMA_Amphenol_132134_Vertical) + 2x zip-tie 3.0
.gts top soldermask (NEGATIVE: openings ONLY at the 0402 lands and the
     SMA pin/post rings; the whole antenna stays under mask)
.gbs bottom soldermask (no openings - bottom fully covered)
.gto top silk: SMA drill-pattern square + brackets at the 0402 site

HOW THE FEED WORKS
The reflector is SPLIT into two halves — they are the CPW ground planes
(the published rear-fed topology; cf. Baek 2016 for the stub-less UHF
variant). A 4 mm trace with 0.5 mm gaps runs from the SMA pad at
x=-23.5 through the reflector channel on a RAISED AXIS (y -0.5..+3.5: the
trace bottom edge sits at the transition height, so the lower gap runs in
ONE straight line from the tie into the 2 mm CPS slot — no jog; the split
line rides the axis, upper half from +4.0, lower to -1.0); at x=0 the
trace merges into the UPPER half (the bond) and the lower gap continues
as the CPS slot to the driven element. The halves are solid-tied 2 mm
behind the pad circle (tie face -25.5). On this axis the pad sees
~26+45j (a single series C saturates at ~-11 dB — the lambda/4 that
would fix it alone does not fit a 17 mm tab; that was v1's 58 mm bay),
so the match is the two-part L-NETWORK: series 4.7 pF in the straight-run
gap + shunt 0 pF across the upper CPW gap. No bay, board 44 mm shorter.

ASSEMBLY
1. Solder the SERIES 4.7 pF NP0/C0G 0402 (C1) across the trace gap at
   (-20..-19.5, 0) - BEFORE the connector. The lands are MASK-DEFINED:
   copper is the continuous 2.0 mm neck with a 0.5 mm gap, and the mask
   exposes two 0.65 x 0.7 mm lands (1005-metric pattern, 0.25 mm
   termination overlap per side; silk brackets mark the site).
   Technique: tin ONE land with a small bead (fine tip, 300-330 C,
   flux), slide the part in with tweezers while reflowing, check it
   sits flat and centred across the gap, then solder the second end.
   Iron only, no hot air needed. INSPECTION: there is no continuity
   test (it is a series capacitor) - inspect for solder bridging
   across the gap instead, and reflow apart if bridged.
2. Mount an Amphenol 132134 (4-post 6.35 mm-sq flange SMA jack) on the
   BOTTOM face at (-23.5, +0) — the family x-spot, raised +1.5 mm
   onto the feed axis: pin up through the board, soldered to its round
   3.0 mm pad on top; all 4 posts soldered to the ground halves.
   (Footprint-style copper: round pad, 2.0 mm neck between the posts,
   ONE 45-deg taper, then a straight 4 mm run — binding gap 0.34 mm
   neck-to-post mask web, above JLCPCB's ~0.2 dam floor.)
3. Clamp the stand on the rear tab (zip-tie holes at (-23.5, +-16) take a
   backup tie). Plug a straight SMA cable, route it STRAIGHT DOWN off the
   jack and tie it to the stand.
4. The cable is part of the design (the counterpoise): FDTD shows
   lambda/4, lambda/2 and 3lambda/4 tied-down cables behave identically
   (within 0.4 dB), so any tidy length works. Do NOT dress the cable
   along the board or forward past the reflector.

ORDERING (JLCPCB, per their published capabilities): order as 2-LAYER
with the provided EMPTY bottom copper (true 1-layer is NPTH-only there
and NPTH pads need 0.45 mm rings; 2-layer plates the holes, same price).
The MASK LAYERS ARE IN THIS ZIP (.gts/.gbs) - openings only at the
solder spots, everything else stays masked (any color; HASL fine; all
mask webs >= 0.5, above their ~0.2 dam floor). Regular +-0.2 mm outline
precision suffices - all copper keeps >= 0.2 mm to routed edges.

OUTDOOR HARDENING (from the environmental/structural research round;
sources on the gallery card):
* Coat BOTH faces AND every routed edge with 50-75 um SILICONE (SR)
  conformal coat after soldering - soldermask alone breathes with
  humidity (er +9%, loss x1.8 at 95% RH) and the skeletal windows
  multiply exposed glass-weave edges that wick moisture. Expect a
  <10 MHz downshift (verified margin absorbs it). Avoid polyurethane
  on the elements (10-20x lossier).
* Pot ONLY the SMA/rear-tab zone, NEUTRAL-CURE silicone (acetoxy RTV
  corrodes copper). Self-fusing tape over the mated SMA joint.
* Mount COPPER FACE DOWN: the FR4 then separates rain/snow from the
  metal and roughly halves the wet detuning. All wet effects shift
  the band DOWN - the design's +70 MHz high-side margin is the icing
  defense (matches the published design-5%-high rule for iced Yagis).
* SMA torque 0.34-0.56 N-m (gold-plated BRASS - not the 0.8-1.1
  stainless spec); counter-hold the body; first cable tie within
  100 mm; >=64 mm radius drip loop; ties every <=0.9 m down the stand.
* For mast service (ETSI 300 019 class 4.1, 50 m/s): keep the clamp
  on the solid rear tab, and epoxy-bond an FR4 spine along the boom
  axis to lift the first bending mode above the vortex-shedding band.
* Consistency: FDTD corner-swept er 4.0..4.6 - S11@915 never worse
  than -12.5 dB, so JLCPCB laminate spread cannot detune it. Order
  high-precision (+-0.1 mm) routing (the standard +-0.2 lands on the
  balun slit), log the laminate vendor per batch, and acceptance-test
  each unit clamped on the real stand: S11 <= -12 dB over 902-928.

FDTD-verified (openEMS PML_8): 8.40 dBi, F/B 13.4 dB, S11@915 -15.4 dB
with the cable modelled (lambda/4 cable identical to 0.1 dB) — see
cpw_yagi.json on the gallery card for the full campaign log (the failed
topologies, the bay proof, the fold negative
result, and the research-driven series-C endgame).
