feat(airframe): Rev T1 — the tilt spar becomes a structural member of the wing (both halves)

[Stab-Rabbit-coding]

Summary

This branch carries both halves of Rev T1, in the order they happened.

0. The wing became the structure. The tilt spar stops being a rotating drive
   shaft and becomes a fixed 20 × 16.3 mm CF tube bonded into the wing over its
   full span — the wing's primary bending member. Everything below is a
   consequence of that. (8 commits, §0.)
1. The fuselage half of the Rev T1 wing-root joint is built. The wing side has been done since 2026-08-29; the fuselage still carried WING_SPAR_Y = +38.15 and WING_SPAR_BORE_D = 8.3, so tools/wing_root_deconflict.py failed on three findings that were all that one fact.
2. The nacelle tilt actuator is now a multi-turn device, per owner direction, with the fuselage drive stage that goes with it and a control spec for the loop it closes.
3. A weight audit of the cargo section and wing roots, asked for as a lightening exercise — which found the real problem is that the BOM understates printed mass by +521.6 g, 13.3 % of AUW.

---

0 — The wing half (Rev T1 / T1b)

The four 10 AWG ESC feeds had no viable path to the nacelles: they do not fit the
wing's Ø7 conduits, and those conduits sit 17.65 mm forward of the tilt axis, so
every transition dragged the harness through a ~45 mm arc.

A bearing transmits shear, not moment, so the Rev R2 rotating spar was
structurally incapable of reacting the wing-root moment — which is why both
previous root designs (enlarged tenon FOS 0.49, two-rod couple FOS 4.14) existed
only to work around it. A fixed bonded spar carries it directly, and the couple
arm goes from 48 mm of chordwise rod spacing to the full 86.7 mm span.

Spar station / bore	45.15 → 28.0 mm / Ø8.3 → Ø20.4
THICKNESS_SCALE	1.00 → 1.46 — the root OML moves for the first time
THICKNESS_SCALE_TIP	1.56 → 2.20
Root path	TENON_LOAD_PATH = "spar_carrythrough"; tie rods retired
Spar bending	33.28 MPa, FOS 9.0 (cantilever bound — skin credited with nothing)

Bore inventory, all constant-mm so taper cannot erode the webs: 8.0 Ø3.2 nav ·
28.0 Ø20.4 spar (now carrying the 4 × 10 AWG on the tilt axis) · 44.5 Ø6.5
AK7455 · 53.6 Ø4.4 drive shaft. The 2 × Ø7 double-D is retired — it could not
carry two 10 AWG conductors at any realistic wire OD, and its 27.5 mm station now
falls inside the spar bore.

Three owner directives that invalidated built geometry

1. The spar must not penetrate the thrust duct. The stub was 32 mm and
   reached 15 mm inside the duct wall. The spar runs along local X at Y = 0,
   so it clears only if it terminates at |X| ≥ 26 → max 15.7, built 15.0.
   It had been budgeted outward from the bearing stack and never checked against
   the duct — the same class of error as the Rev R2 through-duct spar this
   revision exists to remove. The nacelle's bearing pair, ring gear and ring
   magnet must now all fit inside 15 mm.
2. The centre of the cargo bay must stay clear. That killed the 55 mm socket:
   only 18.67 mm exists before the bay, where it gives FOS 0.51 on moment,
   and capacity goes as 1/L². Splitting the joint by load type instead — socket
   for shear, bonded 80 × 60 flange for moment — lands FOS 29.2 against the
   4.02 the long socket gave. The constraint produced a better joint, and the
   bay is untouched.
3. The shaft turns more than one revolution. This inverts the tip stage to a
   reduction and dissolves an "impossible ring gear" finding that was only
   true under the step-up reading a limited-rotation servo forces. Built 14T/50T
   module 0.8, i = 3.571, C = 25.6 → station 53.6. It voids the 180°/270°
   question and makes the encoder load-bearing for control, not telemetry.

The airfoil is no longer an S1223 — and is now described as such

Re-derived from the tabulated UIUC coordinates (REF-CAD-006) at 1/20000 chord,
not quoted. The built sections are designated S1223/t17.7 (root) and
S1223/t26.7 (tip): S1223's camber line with the thickness envelope scaled
×1.46 / ×2.20.

The previous header's chordwise locations were wrong — "22.6 %" and "39.4 %"
against a true 19.8 % and 49.0 %, which match the published S1223
characterisation exactly and were never traceable to the coordinate table the
file actually builds from.

Stated per claim rather than as a blanket disclaimer: camber (8.67 % at 49.0 %)
is exact; zero-lift angle and lift-curve slope partially survive at the
root only, on thin-airfoil grounds — that theory depends on the camber line
alone, which is why preserving it matters, but 26.71 % t/c is far outside
"thin"; CL_max, L/D and the 7.6 N cruise-lift figure do not survive.
docs/flight_envelope.md now carries a banner saying every aerodynamic result in
it rests on an input that no longer describes the aircraft. LE radius goes as
t_scale² — the tip's nose is 12 % of its own chord.

Errors found and corrected rather than propagated

• wing_tip_spar_through_bore() carried rotate([0,90,0]) and was cutting a
  chordwise hole. Harmless at Ø8.3 — the module's real job was simply never
  done — but a 20 mm gash at Ø20.4.
• Plan 004 KTD4's shaft station implied a 15 mm gear centre distance, which
  returns a ring gear smaller than the spar it encircles.
• wing_airfoil_integrity.py validated only the t_scale = 1.0 table and never
  read either thickness scale — it would have passed identically at 2.20, so it
  was not covering the risk it was cited for. Now validates the sections
  wing_solid() actually lofts.
• 14 CFR Part 23.303, cited twice, was removed by the 2017 Amdt. 23-64
  restructure → §23.2230 (new REF-FAA-004, verified verbatim from GovInfo
  CFR XML; eCFR bot-blocks automated retrieval).
• The FOS 4.0 joint target was described as bare judgment. It has a basis:
  §23.2265 requires a special factor for parts with manufacturing-process
  variability — exactly FDM-printed CF-PETG. The requirement is now cited; the
  numeric value stays the project's own choice.
• Also recorded to prevent a future fabrication: +3.8 / −1.52 appear nowhere in
  current Part 23 and must not be cited to it.

New: docs/WING_ATTACH_INTERFACE.md — the single spec for both joints, with a
WA-R1…R15 requirement register; tools/spar_bundle_fit.py; REF-MATH-001
(Melissen 1997) for the exact circle-packing ratios, because the n = 4 case
decided a load-bearing geometry change and the wrong constant (11.0 mm) had
already propagated through two independent documents by being carried in prose
instead of computed.

Unverified, and the FOS numbers should not be read as cleared: the 10 AWG
wire OD is assumed at 5.5 mm (bom_revS.csv records none, and the whole
Ø20.4 bore chain derives from it), and no ASTM D3039/D695 certificate exists for
any CF stock — the 300 MPa spar allowable is a conservative stand-in.

Aerodynamic content produced with the aeronautical-engineering skill, which
carries a standing ENGINEERING REVIEW REQUIRED notice: these are computed
geometric properties and an explicit statement of what is not known, not a
performance substantiation.

---

1 — Fuselage wing-root joint (WA-R1/R1b/R2/R4/R6)

airframe/blender-scripts/merge_cargo_interior.py now matches the Rev T1 wing:

• Socket at chord station 28.0 → hull Y +21.00, Z +66.85 (unscaled camber midline), bore Ø8.3 → Ø20.4, boss Ø27.7 → Ø30.1. Shear only, FOS 16.
• F688ZZ root bearing deleted. Not a tidy-up: a bearing here would let the fixed spar spin under the tilt pinion's gear reaction.
• Both wing-root tie rods retired — the wing already gates them off, and the forward one now intersects the Ø20.4 spar bore outright.
• Mortise 30.8 → 12.8 for the 12 mm locating tenon; fit lands at +0.40 mm/side in both axes.
• Conduits re-cut: nav Ø4.2 @ Y +1.0, AK7455 Ø7.5 @ Y +37.5, drive shaft Ø4.4 @ Y +46.6. The two Ø7 EDF ports are gone — the four 10 AWG feeds now enter through the spar's own bore.

The root flange turned out not to be printable hull material, and measurement is what decided it. Ray-casting the envelope on a 13 × 17 grid (221/221 hits) shows the sidewall skin moving 34.3 mm (port) / 37.0 mm (stbd) of hull X across the 80 × 60 footprint. Filling to a plane gives a 34 mm slab; bounding with a plane leaves that plane tangent to the skin (measured: a 0.46 mm non-manifold edge with 4 incident faces, plus a sliver body). Subtracting a translated envelope from the full 900k-face shell was worse still. Cut locally, in its own generator, it is clean.

New: airframe/stls/fuselage/generate_wing_root_flange.py → 24.3 / 24.4 cm³, both watertight single bodies.

2 — Multi-turn tilt actuator (WA-R15)

DS3225 body + LibreServo_v4, rotation-limit pin removed, run continuous-rotation and closed on the AK7455 — the architecture the cargo winch already uses. Same body, so the Rev S1d pad footprint, bolt pattern and 60 g mass all carry over.

Three things were forced, not chosen:

• Coaxial drive does not fit — the body overlaps the Ø30.1 socket boss by 13.5 mm in Y at either output-shaft orientation, and relieving the boss would cut into the socket bore. Hence a fuselage spur pair, module 0.8, 38T/38T, C = 30.40, 1:1; both axes run along hull X so no right-angle stage is needed.
• C is set by the landing-gear bay tops (Z +82.39 + the 3.0 mm budget ⇒ C ≥ 29.80). 36T and 37T fall under budget; 38T clears at 3.60 mm.
• The root tenon forces the gear plane axially. Its forward face is 4.90 mm from the shaft axis, so clearing it in Y needs a tip radius under 3.90 mm — no gear that can transmit at C = 30.40 qualifies. That costs an 18 mm actuator standoff and 18 mm/side of cargo-bay roof reach.

The stage is 1:1 deliberately: a step-up would trade the 48× torque surplus for slew rate but pull the actuator back to 260°, re-opening the 180-vs-270 question Rev T1 exists to close.

Control spec — docs/TILT_DRIVE_CONTROL_SPEC.md (new)

Cascade: LibreServo_v4 closes the actuator on its own absolute encoder; Pilot closes the nacelle on the AK7455. Sensors, sign conventions, spans, fail states, and the bench programme that would produce a plant model.

No gains, and the document says why — the plant is not identified and dead time is unmeasured. Two findings worth naming:

• The train is not self-locking (two spur stages, total 3.571) and has no holding provision. Gravity is nulled by the pivot-at-CG, but the aero moment is unquantified, and an unquantified moment on a non-self-locking train is an unbounded rate. TILT-CTL-01 / WA-R16 — flight-release blocker.
• Differential tilt needs a trip that is architecturally separate from both position loops. TILT-CTL-02.

3 — Weight audit — docs/MASS_AUDIT_CARGO_WING_ROOT.md (new)

All 23 PRINT-* rows whose description names an existing STL were measured. The BOM understates by +521.6 g.

Row	BOM	Measured	Δ
PRINT-CARGO-SECT	165 g	372.2 g	+207
PRINT-BATT-TRAY	22 g	140.2 g	+118
PRINT-HEAD-SHELL	83 g	177.6 g	+95
PRINT-CARGO-CRADLE	18 g	80.6 g	+63
PRINT-MIDDLE-CANONICAL	135 g	190.9 g	+56

The cargo-section row is not a Rev T1c regression — the pre-T1c shell already measured 316 g against that same 165 g row. Three rows corrected here; the rest measured and reported for their owning branches rather than edited across a governance boundary.

Foam and voids: cargo-section enclosed cavity 3.183 L, less the 2.135 L cargo bay (which the BOM itself says is not foamed) = 1.048 L foamable → 33.6 g, an upper bound before EPS void formers displace any of it. FOAM-PU-2LB carried 900 g — the A+B kit mass. Its own note says "~0.7 L mixed fill", which at 2 lb/ft³ is 22.4 g. Corrected; procurement quantity unchanged.

Correction to our own Rev T1c figure

Rev T1c is +102.8 g (+2.63 % AUW), not the +77.0 g first recorded. The root flange was costed as RHO_SOLID × 40 % infill. That is the wrong model: at 5 mm thick with 4 perimeters at 0.6 mm the plate is 4.8 mm of wall and 0.2 mm of infill — essentially solid. RHO_PRINT is a bulk figure and must not be re-derived that way.

4 — Weight-reduction plan — docs/plans/2026-08-30-001-... (new)

Eight ranked units. The largest lever is not structural: BATT-6S-4000 is 750 g — 19 % of AUW — and the 2800 mAh pack is 225 g lighter, more than every printed-part optimisation combined. Sized here; the call is the owner's.

#	Target	Recoverable
W1	Battery down-select	−225 g
W2	BOM ⇄ mesh reconciliation — gates everything else	0 g
W3	Battery-tray floor pocket	−23.1 g
W4	Actuator standoffs → hollow	−32.9 g
W5	Cargo cradle → windowed	~−25 g
W8	Wing root	none — CLOSED

• The tray's 8.5 mm floor is not a strength figure — the SCAD says it is set by RAIL_D = 6.5 so the slot does not breach the cavity. At 3.0 mm it is FOS 55 at limit and FOS 25 at 9 g × 1.5 against 77 MPa flexural, deflecting 0.045 mm under a 750 g pack.
• The actuator standoff's governing mode is mesh opening, not strength — 2.5 × 10⁻⁵ mm deflection against a 0.05 mm centre-distance budget.
• The cradle is not over-thick — already 2.5 mm walls; it is a fully-closed 110 × 80 × 60 box for a 250 g payload. Medium confidence, because the auto-latch works by flex-tab compliance and that snap force is uncharacterised.
• W6 could invert the ranking: PRINT-NACELLE-PORT/STBD name an STL that no longer exists; the two candidates measure 278 g and 658 g against a 132 g row.

Tool defects found while the geometry was being fixed

The gate had been red "by design" for one known cause, which meant its other checks were unexercised. Four real defects surfaced:

• tenon_params() matched only "two_rod", so the Rev T1 default path fell through to the enlarged 30 mm tenon and reported an 8.60 mm/side mortise foul against a correctly-sized mortise.
• The spar corridor is stepped (Ø20.4 outboard, Ø16.3 inboard) and was swept at one diameter, reading the intended annulus as "wall not cut".
• Every route was probed 12 mm past the wall bracket — right for a rigid shaft, wrong for a wire that turns.
• cargo_bay_envelope.py derived spar OD with the retired 0.15 mm/side rotating clearance and called a fixed bonded tube a "rotating tilt-spar".
• merge_cargo_interior.py finalize_watertight() dropped sliver bodies before closing slits, so a fresh 2-face zero-area body reached the published STL.

Verification

tools/wing_root_deconflict.py               CLEAR
tools/wing_internal_clearance.py            PASS
tools/wing_airfoil_integrity.py             PASS
tools/wing_spar_carrythrough.py             PASS
tools/spar_bundle_fit.py                    PASS
tools/cargo_bay_envelope.py                 PASS
tools/landing_gear_wing_clearance.py --proud PASS
tools/validate_stls.py                      PASS
tools/precommit_index.py --check            PASS
generate_wing_root_flange.py                PASS

Cargo shell regenerated: watertight, 1 body, 0 boundary and 0 non-manifold edges, CI-valid.

Known blockers carried out of this PR

ID	Item
WA-R16 / TILT-CTL-01	Tilt train is not self-locking; no holding provision. Flight release.
WA-R18	Rev T1c is +102.8 g; hover T/W was ~1.19 against a 1.2 minimum. Flight release.
MA-1	BOM understates printed mass by +521.6 g. Blocks any weight statement.
WA-R17	WA-R3 split-collar clamp is still not a part. Blocks wing removal/refit.
WA-R15a	Cargo-bay roof band needs re-measuring after the 18 mm standoff.
TILT-CTL-02..06	Plant model, slew rate, differential-tilt trip, LibreServo_v4 encoder part.

Nacelle side (WA-R7…R12) is unchanged and still specification-only.

---

Analysis and implementation by Claude (Claude Opus 5, Anthropic) under the author's direction, per AGENTS.md §3 "Attribution and Licensing".

🤖 Generated with Claude Code