SLOTRAIL-223

01 — the idea

One accurate feature on the block. Use it.

The block is 223 × 65 × 50 mm of polyurethane tooling board. Every dimension on it carries a 1–2 % tolerance — up to half a millimetre on the thickness. A rigid pocket that grips one block will not accept the next, and a set of shims falls out the moment you pick the block up.

But the block does arrive with one feature you can trust: a 6.00 × 6.00 mm groove down the centre of its underside — the tether-line channel — machined straight and parallel over the full 223 mm.

So the groove becomes the datum. A 6 mm square rail sits in it, and the block rests on that rail by the ceiling of the groove, inside the groove, where nothing else can touch it.

Flip the block 180° about the rail and the groove goes back over the same rail. Its height above the machine table now differs by exactly the rail's height — 6.00 mm you can measure with calipers — and not by the foam's thickness, which you cannot. That single fact is why the two halves line up.

02 — datums

Six degrees of freedom, three contacts

03 — print

Five parts, no supports (measured in §10)

Print settings

• Beds: 0.4 mm nozzle, 0.2 mm layers, 3 perimeters, 15 % gyroid. They only locate the rail; they are not a machine vice.
• Rails: 4 perimeters, 100 % infill. This is the part whose height sets the flip, so print both halves in the same job if you can — then measure the height h in three places and use the average.
• Every screw-head pocket is printed in. Ø7.5 × 4.3 mm deep over the two tower clamps and on the collar, and a slotted Ø10.5 × 6.3 mm deep pocket along each of the eight bed slots. That extra 0.3 mm of depth is what keeps every head inside its outer face: check them in the slicer's section view before printing, because they are the first thing a coarse layer height will close up.
• No supports anywhere. The rail socket's roof is at 49.4° — above the 45° threshold, so no slicer paints it — and the collar prints with its bore vertical. §10 has the per-part measurement.
• Better than printed: a 279 mm length of 6 × 6 mm ground key stock is straighter than anything you will print and it is what the machined version uses (§11) — in one piece, no lap. On this printed jig a single bar would not fit a 180 mm bed, so it stays in two halves; if you fit the bar anyway, set rail_h in jig.py to your measured bar.

Hardware

04 — assembly

Fifteen minutes, once

1. Join the bed. The two half-round notches at the joint face make one Ø8 mm hole. Put a drop of CA in each, press the dowels in, slide the halves together and clamp them flat while the glue sets. Check it on a flat table — if it rocks, sand the bottom until it doesn't. The two towers now sit 223 mm apart, exactly.
2. Thread the collar on. It is a closed clamp, so it has to go onto rail_a from the inner end before anything is glued. Slide it on and leave it loose.
3. Insert the rails. Push rail_a's outer end into the left tower's socket until it bottoms out, and rail_b into the right tower's socket the same way. They meet in the middle over a 20 mm half-lap.
4. Glue the lap. Sight along the rail or roll it on a flat table until the two halves are flush, then CA the lap. It is now one 277 mm bar held at both ends — straightness matters, length does not.
5. Tighten the rail clamps. One M4 over each socket, finger tight plus a quarter turn. The tower is not tapped: the screw cuts its own thread in a 3.3 mm pilot, so the first turn is stiff and the second is easy — that is normal, and it is why the hole is a pilot and not a 3.6 clearance hole. It presses the rail onto the socket floor, and that floor is the Z datum. Do not overtighten — it is printed plastic.
6. Check the fit. Slide the block's groove over the rail. It should drop on with a light push and no rattle. If it rattles, see Troubleshooting — one turn of masking tape on the rail fixes it and is allowed to stay.

05 — machine setup

Z0 is the rail, never the foam

Bolt the bed to the table through the eight counterbored slots. Once it is clamped it does not move again — the whole point is that both operations run on the same clamped fixture.

X: the right tower's inner face is the stop. Push the block against it.
Z: touch off on the rail's top face. That is Z0, in both operations.

Then measure two things with calipers and write them down:

06 — the two operations

Top, flip, underside

OP 1 — top half

Block in its shipping orientation, groove down, sitting on the rail. Machine the nose, cockpit, halo bosses and wing mounts — everything above zb ≈ 26.

The block's bottom face hangs 6 mm below the rail's top face, 62 mm above the bed plate, so nothing under the work gets in the way. The only obstacles are the two towers, standing 12 mm above the rail's top face.

OP 2 — underside

Loosen the collar, lift the block straight up off the rail, turn it 180° about its long axis, drop it back on the rail, push it against the stop, slide the collar to the end face and tighten.

The underside now faces up. Machine everything except the middle: the rail is still sitting in the groove.

Use a Ø3 mm ball nose for the fine detail and keep at least 14 mm of tool stickout (18 mm gives full clearance) so the collet nut clears the towers when you machine within 10 mm of either end.

07 — rules your car design has to follow

Four numbers

Inside those four numbers you can machine whatever the rules allow. The fixture has no opinion about your car.

08 — compliance

What the regulations ask for that a fixture cannot give you

The fixture is not a scrutineered part — the technical regulations say nothing about how you hold the block. But several regulated features live on the same faces this fixture and your CAM touch, so they are worth stating once, in order.

The order to do things in

1. Drill the halo holes with the official printed jig. Slot empty, block loose.
2. OP 1 — block on the rail, groove down, top half machined.
3. Flip — 180° about the rail, collar up against the end face.
4. OP 2 — underside machined, 6 mm channel untouched.
5. Axle holes — 11 mm above the bottom face, drilled from the machined sides.
6. Fit halo, tether guides, wheels, sticker.

09 — what has been checked

Numbers, not adjectives

Every figure below comes from the solid model in jig.py, which rebuilds and re-runs these checks on every run (report.json).

10 — can you print it?

Yes. Here is the arithmetic, not the opinion.

Every STL was re-opened and every triangle measured against the build plate. A face needs support when it points downward and lies flatter than 45°. Across all five parts that total is 95 mm² — and none of it is an overhang you would actually support. The three parts that carry the tolerances come out at zero.

The two places the slicer will still show something

• Right bed half, 40 mm² at Z 20. That is the 0.4 mm rim relief behind the stop face — one extrusion width. The layer above just runs 0.4 mm past the wall and lands on the next one; it is not a bridge and it needs nothing.
• Collar, 55 mm². 13 mm² is the ceiling of the 1.6 mm clamp slit, and 42 mm² is the round ceiling of the M4 head pocket, which is horizontal because the bore is printed vertically on purpose (a vertical bore is the only orientation with no bridge at all in the bore). The pocket is clearance for a bolt head. Its roof drops a layer thickness and the head still goes in.

What this section is not. It is a measurement of the geometry, not a photograph of a finished part: no physical print of the current model exists yet. What is proved is that no surface exceeds the support threshold, that every part fits a 180 mm bed with its lowest face flat on the plate, and that the thinnest feature in the set is the 1.6 mm slit — four extrusion widths. The rest is your printer's calibration.

11 — the machined version

SLOTRAIL-223-M. Same three numbers, cut on a 3-axis mill.

You cannot hand the printed parts to a machinist and expect them back as drawn. Three features are print-only: the V-shaped socket roof lives inside the tower where no end mill can reach it from above; the collar's 1.6 mm slit is a saw cut; and each bed half is a 92.3 mm block with a tunnel along X, which is a fourth-axis part. So the machined version is a different set of parts that arrives at the same datum chain — seat 74.000, mirror 77.000, stop 223.000 — so a car cut on either jig comes out the same.

What it is made of

Why a 3-axis machine can cut all of it

Every face on the seven machined parts is either vertical or horizontal — measured, not intended: 0 sloped faces across all seven solids, against 18–86 faces each. That is the definition of 2.5D work. Every pocket is a plunge from one side, every wall is a side cut, and the only thing that changes between setups is which face is up. No part needs a fourth axis, a fixture plate or a sine table.

The two details that make it work

1. The bar sits 1.0 mm proud of its pocket. Rail pocket 5.0 deep, bar 6.0 tall, so the bar's top stands 1.0 above the saddle's top face. The cap bears on the bar and never reaches the saddle face, so the clamp force is the screw's tension instead of a stack of tolerances fighting each other. It also means the cap cannot pinch the rail seat up or down: the seat is a machined surface, and the bar presses on it.
2. The bar's bottom corners are relieved 1.5 × 45°. A 6.2 mm pocket cut with a ⌀3 cutter leaves R1.5 fillets in the floor corners, and a square-cornered bar would sit on those fillets 0.75 mm above the seat — the Z datum would be fiction. Relieving the bar's two bottom edges costs one pass with a chamfer tool on bought stock and puts the bearing surface back on a true flat.

Tools

• ⌀3 mm 3-flute carbide end mill — the two rail pockets, 5.0 mm deep (1.7 × D). Nothing else in the set is that small.
• ⌀6 mm 3-flute — the table slots, the collar pocket, the locator shoulders, profiles.
• ⌀10 mm or a helical bore — the ⌀12 counterbores.
• Drills 3.3 / 4.2 / 5.0 mm and M4, M5, M6 taps; a 90° spot drill.
• 4 mm chamfer tool — the bar's 1.5 × 45° bottom relief.

Order of work

1. Machine the base plate both sides, then check that the shoulder floor to the underside is 12.000 all round. The saddles hang off that face.
2. Machine both saddles, top first: the pocket floor and the stop face come off the same setup, so they are square to each other by construction. Then flip and tap the M6s.
3. Bolt it together loose, drop the bar in, fit both caps. Do not tighten the saddles yet — they locate on the shoulders, so they can only go where the plate puts them.
4. Indicate the seat. Height gauge or a dial on the spindle: the pocket floor must read 74.000 from the base plate's underside at both saddles, and 223.000 from the left saddle's inner face to the right one's stop face. If they disagree by more than 0.05, the shoulder has a burr — clean it, do not shim it.
5. Fit the collar on the bar, slide it to the block's end, tighten. That is the only adjustment in the fixture, exactly as in the printed version.

What this section is not. There is no CAM file here and no machined part has been cut yet: what is proved is that the geometry can be cut by a 3-axis machine — every face axis-aligned, every pocket reachable from +Z in its own setup, two setups maximum per part, standard cutters, no undercuts. The toolpaths, feeds and workholding are still yours to set: clamp the base plate on parallels, hold the saddles in a vice on the 110 mm face, and leave 0.3 mm on the pocket floor until you have indicated the vice flat.

12 — the zero-hardware version

SLOTRAIL-223-ZH. Five printed parts, nothing to buy, nothing to lose.

Both versions above depend on hardware: eleven screws and eight T-nuts in the printed one, cap screws and a bought 6 × 6 mm ground bar in the machined one. This version does the same job with zero purchased parts and zero fasteners — and it keeps all three datum numbers: seat 74.000, mirror 77.000, stop 223.000. A car cut on any of the three jigs comes out the same.

What it takes out

How the block is held without a screw

Z — the same trick. The 6 × 6 groove nests over the rail. In OP 1 the groove ceiling rests on the rail's top face; in OP 2 the block is rolled 180° about the rail and hangs from its underside. Nothing new here, and that is the point.

X — a wedge you push with your thumb. The right tower is a fixed stop at x = 223. The pusher's back face runs on a 10° incline cut into the left tower, so pushing it down walks it towards the block. 10° is under the friction angle of PLA on PLA (19.3°), so it is self-locking: the block cannot push it back, and lifting it 11.52 mm releases the block. A 40 N push becomes 76 N of clamp force; the two prongs bear on 665 mm² of the block's end face, i.e. 0.114 MPa even at that worst-case push. You should not need it: the clamp only has to beat a couple of newtons of cutter load, and it brings 27 N of friction to do it. Push until it stops moving, no harder. The take-up is 2.0 mm over a 225.0 mm loading gap, so a block anywhere in the official ±0.2 mm on length loads freely and still gets clamped.

The two halves — a dovetail that carries tension. The pusher pushes the left half away from the right half; that load has to go somewhere. A plan-view dovetail at plate level takes it: 38 mm of engagement, a 11.89° undercut, 0.15 mm of clearance. It is checked the only way that proves anything — move the right half 1 mm and measure the interference. Separating them costs 86.7 mm³ of solid collision, so the joint cannot pull apart; lifting the right half costs 0 mm³, so it drops on in seconds. The dovetail carries load only. It never sets a datum — the wedge absorbs any slack.

OP 2 — the improvement, honestly. In the screwed version the flipped block hangs from a 6 × 6 mm printed bar spanning 236 mm, and beam theory says that bar deflects about 0.4 mm under a 50 g foam block (simply supported, E = 3 GPa, I = 108 mm⁴ — calculated, not measured). In OP 1 the block rests on the bar's high ends near the sockets, which is nominal; in OP 2 it hangs from the bar's underside and therefore finds the bar's lowest point at mid-span. The two operations would sit about 0.4 mm apart in Z. That is a real flaw in the screwed version, and it is the one thing this variant fixes: the block rests on plate ribs at Z = 30 mm, so in OP 2 the rail carries none of the block's weight and only ever touches the groove ceiling near its supported ends, where its height is nominal. The pocket floors between the ribs are the same 4 mm plate the model already has.

The parts

439 g of PLA, about 18–24 h, tallest part 100 mm, largest 188 mm — still inside a 180 mm bed. Total unsupported area across all five parts: 49.1 mm², and it is one 7.1 mm bridge over the slot the rail passes through, which FDM spans without support. Print the pusher as exported — rolling it upright was measured at 476 mm² of unsupported face and rejected.

Using it

1. Print the five parts. No supports, no raft.
2. Press the two rails together at the 20 mm lap.
3. Slide the rail into the right tower until it butts the blind end (0.4 mm of clearance), then lay it into the left tower's socket.
4. Drop the right bed's dovetail hooks over the left bed's tongue.
5. Drop the pusher into the left tower's guide channel from above.
6. Grip the bed across its 110 mm width in the machine vice. No bolts, no T-nuts, no clamps.
7. OP 1 — lift the pusher, lower the block onto the rail, push the pusher down by the block's rear face until it is firm.
8. OP 2 — lift the pusher, lift the block clear of the towers, roll it 180° about the rail, lower it back onto the rail and the ribs, push the pusher down again.

What this version cannot do

• It needs a vice. The screwed version bolts to a table and works on any machine. This one is held like any fixture, so you need a vice that opens past 110 mm — a 150 mm vice is plenty.
• It is heavier: 439 g against 298 g. The bed now carries the block instead of screwing itself down, and the ribs cost material.
• The clamp is set by hand, not by torque. A screw is repeatable in a way a thumb is not. Self-locking keeps it where you put it; it does not measure it. The X datum is the stop face in both versions, so this costs nothing in accuracy — only in feel.
• It is print-only. The dovetail's undercut cannot be cut with a 3-axis end mill and the pusher guide is an internal slope, so there is no machined zero-hardware variant. If you are milling the fixture, use SLOTRAIL-223-M in 11 and buy the bar.
• None of this has been printed or cut yet. Every number above is measured on the solid model — zero.py rebuilds and re-checks it on every run (zh/report.json). The first print is the first real test.

13 — the over-engineered version

SLOTRAIL-223-X. Ten printed parts, nine pieces of hardware, and the flip is a lever throw.

Every version above asks the same thing of you: pick the block up, turn it over, and put it back on the rail. That lift is the biggest error in the whole process — bigger than any tolerance in the fixture, because it is unmeasured and unrepeatable. This version never lets go of the rail. The rail is the axle. Turn the lever 180°, drop a pin, and you are in OP 2 with the rail still sitting in the same two bearings it started in. Same datum chain: seat 74.000, mirror 77.000, stop 223.000 — a car cut on any of the four jigs comes out the same.

What the extra parts buy

How it works

1. The rail is the axle. Its two ends pass through 46 mm hubs that sit in U-cradles cut into the stands. The cradle is the lower half of the bore, so the whole axle lifts out in one piece when you want to change the rail, and the caps only stop it lifting while you work.
2. The lever sets the angle. It clamps on the rail's outboard end with two M4 grubs, and carries two 5 mm holes in its boss, 180° apart. A 5 mm dowel drops through the cap's arm into whichever hole is on top. That is the whole index: no scale, no setting.
3. The clamp pushes along X. Two M6 grubs thread through the left stand and their plain tips sit 8 mm inside sockets in the jaw, so the jaw hangs on them and slides on them. Set the block against the stop, wind both in, done.
4. The stop is a face, not a bolt. The right stand's inner face is at 223.000 and carries a single full-height pad; the rail passes through it in a Ø9 round hole, because a square hole in a square bar jams at 45° when the bar turns.

What the checks say

What this version cannot do

• It does not machine the two side faces. The carrier indexes to 0° and 180° and that is all. At 90° the block would hang sideways off the rail with nothing but foam-on-rail corner contact holding it, and the 0.1 mm of clevis play in the groove turns into 0.95° of droop — a 0.83 mm error across the face. Four index positions would have been easy to draw and dishonest to claim, so there are two.
• OP 2 still hangs the block from the rail. That is what makes the sag number above matter: print the rail and the seam between your two faces is 0.61 mm out. Buy the 6 × 6 ground bar and it is 0.01 mm. This is the one version that expects you to spend money.
• Back the clamp off two turns before you flip. The pads bear on the end-face plane, which does not move when the block turns — so you can physically rotate with the clamp on, and the sweep check passes either way. But 24 N of pad pressure scrubbing across foam for 180° will score the end face, so the correct move is: two turns out, lever, pin, two turns in. The stop and the rail never move, which is the part that costs accuracy if you get it wrong; re-clamping a face that never left its stop does not.
• The clamp has no return spring. The grubs push the jaw out; you push it back with your thumb. A screw thread cannot pull what it is not attached to, and adding a spring would have added a part that wears.
• It is the heaviest and the slowest to make. 448 g against 439 g for ZH, ten parts to print instead of five, and two caps to unscrew every time you want the axle out. Over-engineering is a cost, and this is where you pay it.
• Print-only. The dovetail's undercut and the U-cradles cannot be cut with a 3-axis end mill, so there is no machined X variant. If you want to mill the fixture, that is 11 and a bought bar.
• Nothing here has been printed or cut. Every number is measured on the solid model; index.py rebuilds and re-checks all of it on every run (index/report.json). The first print is the first real test.

14 — which one do you build?

Four fixtures, one datum chain, four different bills

All four hold the block by the same 6.00 × 6.00 mm tether groove, put the bottom face at Z 74.000, the mirror plane at Z 77.000 and the X stop at 223.000. A car cut on any of them comes out the same, so choose on cost and on the one error you are most likely to make — not on accuracy.

Then the three exceptions, in the order they usually apply:

• Your printer is 180 mm square. Build 223 (03–06). It splits into parts that fit, and its eleven screws are M4 and M6 — hardware you already have. It is also the version whose assembly has been worked through longest — 04 has the figures.
• You have a mill and the fixture has to survive a season. Build −M (11). Metal does not creep, a tapped thread in 6061 does not strip on the fourth reassembly, and the bar is ground stock instead of a glued print.
• The flip is where your tolerance dies, and you have bought the bar. Build -X (13). It is the only version where OP 2 does not involve lifting the block at all. It is also 448 g, ten parts and the most to go wrong — read 19 before you commit.

15 — bills of materials, all four

What to put on the bench before you start

These are the complete lists. The printed parts have no hidden extras: if it is not on the list, the fixture does not need it. Quantities come from the same models that export the STLs, and every screw site is one the model probed for a head clearance of 0.3 mm.

A · SLOTRAIL-223 — printed, screwed

B · SLOTRAIL-223-M — machined

Finished sizes are what the model exported; stock is the smallest convenient oversize. The collar is a two-part split clamp — thread it onto the bar before the bar goes near the saddles, because its bore is closed.

C · SLOTRAIL-223-ZH — zero hardware

D · SLOTRAIL-223-X — rotary carrier

Printed rail sags 0.6 mm in op2; steel sags 0.01 mm.

Fasteners, once

16 — every printed part, in one place

Twenty STL files, twenty-one pieces, no raft anywhere

Three of the four fixtures are printed. This is the complete set: what each part measures, what it weighs at 15 % infill, and where the slicer will put support. Sizes are bounding boxes of the exported, print-oriented solids — the same files that are in each stl/ folder — so they are what your slicer will report, not what the design intended.

A · SLOTRAIL-223 — 5 parts, 298 g

B · SLOTRAIL-223-ZH — 5 parts, 439 g

C · SLOTRAIL-223-X — 10 unique, 11 pieces, 448 g

Settings that matter

Thin features, and why they are survivable

• rail 5.90 x 6.00 mm - printed lying down, 6 mm in Z (the datum face)
• pocket floor 4.00 mm - 10 lines at 0.4, printed as the part's base
• dovetail flanks are vertical walls - no support, no undercut in Z
• socket roof ridge at 49 deg - self supporting by design

17 — assembly, all four

The same eight steps, four different mechanisms

Start with the comparison: each column is a complete build, and the row you are on is the same decision in all four. Then the numbered sequence for the fixture you are actually building — 04 has the printed version's in full, with the exploded view, and the other three are below it on this page.

Build sequences, one per fixture

Eight steps is enough to compare, not enough to build from. Each fixture's own section carries the reasoning and the figures; below is the same order written for the bench — parts listed first, then one numbered sequence with the check that catches that step going wrong written into it. The printed version's full sequence with the exploded view is 04. If you skip the checks you are not building a jig, you are guessing at the datum with extra steps.

B · SLOTRAIL-223-M — machined, 6061

You need: base, 2 saddles, 2 clamp caps, collar body + cap, one 6 × 6 × 325 mm ground bar, 8 × M6 T-bolts + T-nuts (table), 4 × M6 × 25 (saddles), 4 × M5 × 12, 2 × M4 × 10. Figure: 11.

1. Stone every edge first. All seven parts. A burr on the base's top face is a Z error that no offset can fix, because the saddle sits on it and the bar sits in the saddle.
2. Bar and collar before anything else. Thread the collar body onto the bar and park it at the far end — its bore is closed, so it cannot go on later. Chamfer the bar's two bottom edges 1.5 × 45° now, while there is room to hold it.
3. Base onto the table, snug only. Eight M6 T-bolts through the base's eight slots. Leave them loose: the saddles are about to define where the base actually sits.
4. Saddles onto the base. Each drops onto its machined 0.5 mm locating shoulder, then four M6 × 25 come up from underneath through the base's counterbores into the saddles' tapped holes. Torque in pairs — one screw fully home tilts the saddle and the shoulder stops being a shoulder. Then tighten the eight table bolts.
5. The bar. Its pockets are open at the top, so lay the bar down into both saddles at once. The bar must finish 1.0 mm proud of the pocket floor; if it sits high or rocks, the pocket's R1.5 floor fillets are holding it — go back to step 2 and check the chamfer. Then the two clamp caps, two M5 each, snugged in pairs.
6. Close the collar on the block's end face with the cap and two M4 × 10. It must slide on the bar with no detectable rock.
7. Indicate it before you cut. Sweep a dial test indicator along the bar's top face over its full length and write the number down. That number, not the drawing, is the Z repeatability of the fixture — it is the measurement 19 asks for.

C · SLOTRAIL-223-ZH — no hardware at all

You need: 2 bed halves, pusher, 2 rail halves. No fasteners, no bar, no glue. Figure: 12.

1. Knock the elephant's foot off the dovetail. 240 grit on the tongue's two flanks, four or five strokes. A printed vertical wall is 0.1–0.2 mm fat at the bottom and the joint will not drop together until it is not.
2. Press the two rail halves together at the 20 mm lap. Dry fit first: they should go together with thumb pressure and come apart with a firm pull. A drop of CA is allowed; a hammer is not.
3. Drop the halves together. Right half's hooks over the left half's tongue, straight down — never slid in from the end. Check: lift either half 1 mm and they come apart freely; push them sideways and they do not.
4. The rail. Push it into the right tower's blind end until it butts — there is a designed 0.4 mm gap — then lay the left end down into the left tower's socket. If your rail feels loose in the socket, that is the 0.1 mm designed clearance and it is fine.
5. First fit. Slide the block's groove over the rail, push the end face against the stop, and drop the pusher into the tower's channel behind it. The pusher's two prongs should straddle the rail with 0.45 mm to spare on each side.
6. Load it. Press the wedge down with a thumb. Test the self-lock properly: push the block forward (the direction the cutter will push it) and let go. If it creeps back, the incline has dust on it — wipe both faces and rub pencil graphite on the ramp.
7. OP 2. Lift the wedge, lift the block 11.5 mm clear of the towers, roll it 180°, and lower it back onto the rail. It now rests on the seven ribs at Z 30, and the rail only locates. Wedge down again.

D · SLOTRAIL-223-X — rotary carrier

You need: 2 bases, 2 hubs, 2 caps, lever, jaw, detent pin, 2 rail halves (or one bought 6 × 6 × 325 bar), 4 × M4 × 12 caps, 2 × M4 × 8 grubs, 2 × M6 × 60 grubs, 5 mm dowel. Figure: 13.

1. Press a hub onto each rail end before the rail goes anywhere near a cradle. Both hubs, one end each, square to the rail. Roll the assembly on a flat table: a hub that shows runout is a wobbling flip, and no detent will fix it.
2. Join the deck. Same dovetail as ZH: hooks over tongue, straight down. Lift-test it — 1 mm of lift must free it.
3. Lay the rail into both U-cradles. The hubs sit in the cradles, not on the deck. Sight along the deck edge: the rail should be parallel to it within a millimetre over its length.
4. Cap the cradles — two M4 × 12 per cap. Snug one, then the other, then torque in pairs. The test is by hand, not by torque wrench: the hub must spin with a finger and have no visible play. Pinched caps are the most common −X build fault.
5. The lever onto the left hub with its two M4 × 8 grubs. Set it so the pin hole lines up with the cap's arm at both 0° and 180° — the arm is the stop, and it should be the arm that stops the lever, never the lever's boss.
6. The detent pin through the cap's arm into the lever's boss. It should drop 8 mm under its own weight. If it stops at the mouth, run a 5.0 drill through by hand and try again.
7. The jaw into the left deck's guide, then the two M6 grubs, a quarter turn at a time each. The jaw must slide without cocking — alternate the grubs and it will.
8. First fit. Block's groove over the rail, end face against the jaw's pad, grubs in until the pad bears. Snug, not tight: these pads push on balsa foam.
9. Flip drill. Two turns out on the grubs, lever through 180°, pin, two turns back in. Do it five times and watch the Z: put a dial indicator on the rail's top face across a flip. A number that repeats within 0.19 mm is the design claim in 19 being true on your bench — a number that does not is the fixture telling you which step is wrong.
10. Never touch again: the cradle alignment and the lever's clamp. Everything else in this list is meant to be undone and redone; those two are the fixture.

18 — the numbers, all four

What is identical, and what each version changes

This is the table to keep open at the machine. The first four rows are the datum chain and they are the same in all four fixtures — that is the whole point of building four of them. Where a row differs, the difference is a design decision, not a tolerance.

What you touch off on, per fixture

19 — the evidence, in one table

Twenty-two claims, each with the number that backs it

Every row is produced by a script that raises and exports nothing if the check fails, so a STL in these folders is a STL whose numbers passed. The last column says where to look if you want to argue with one.

What is not proven — and how to settle each one

20 — the files, and what to re-run

Every file in the folder, and what to re-run after you touch it

Nothing here is a one-off export: every STL is regenerated by a script that first re-proves the geometry. If you change a number, the checks re-run and a failure means nothing is exported — a broken fixture cannot silently reach your slicer.

Changing a parameter

21 — troubleshooting

Each symptom, one cause

Problems that only happen on the later versions