Functional Design that Grabs- The Picatinny Rail

It hangs off the two angled sides. The flat top surface everyone assumes is the reference is, in the original 1995 standard, clearance. NATO changed that in 2009, and the change is a big part of why some quick-detach mounts are fussy on one rifle and fine on the next.

What a Picatinny rail actually is

Strip away the tacticool and a Picatinny rail is a dovetail with slots cut across it. Look at it end-on: a flat top, a 45° chamfer running down and out from each top corner, a short vertical land at the widest point, then a 45° undercut running back in to a narrower neck, then the base. The slots across the top are the recoil grooves. That's it, that's the rail.

The standard behind it is MIL-STD-1913 (AR), Dimensioning of Accessory Mounting Rail for Small Arms Weapons, dated 3 February 1995. And no, 1913 isn't a year, it's a document number. Its neighbours in the same block are MIL-STD-1911, a 1993 document about hand-emplaced ordnance, and MIL-STD-1916, a 1996 document on product acceptance. If 1913 were a year, 1911 would be a 1911 document.

It's eleven pages, two of them drawings, and it's public domain, so you can read it yourself (link in Sources). Its own stated purpose is dry: "uniform accessory mounting rails and requirements that are interchangeable among the different units of the Defense Department." Interchangeability is the word, and everything that follows is a tug of war between that and precision. The document never calls the rail "Picatinny"; the name only appears in the return address on the foreword, U.S. Army Armament Research, Development and Engineering Center, Picatinny Arsenal, NJ 07806. The Army credits a Picatinny mechanical engineering technician, Gary Houtsma, with having "developed the dimensional requirements and created MIL-STD-1913", and gave him a medal for it nineteen years later.

The three numbers most people know are in there: rail width 0.835" (21.2 mm), slot width 0.206" (5.23 mm), slot spacing 0.394" (10.01 mm).

How a 1913 mount really locates

Here's the part Spuhr got right. Open the profile drawing (Figure 1) and count the datums. There's one. It's called C, and it hangs off a width, .748 ±.002 inches, not off the top surface. A datum on a width is a centre plane: the mount is referenced to the middle of the rail, found from its two sides. The top face gets a flatness call-out (.005) and nothing else. No datum, no height reference.

The clearest walk-through I found is a 2016 write-up by Mitch Thomas, a mechanical designer who reads the drawing the way an inspector would. He reads the datum as four points of contact on the four angled faces, two per side, pushed inward until all four touch: "When touching, the points must be .748 +/- .002 inch apart." And then the line that made me sit up: "mathematically the angles of the surfaces are irrelevant, as long as the surfaces move inward or outward to maintain the .748 +/-.002 dimension." The .748 dimension lines on the drawing run to the top corners of the dovetail, so you can also read it simply as the width across the top of the angled sides. Either way the reference is the sides. The 1995 drawing puts no specific tolerance on the 45° itself; it just says 2 X 45° and leaves it to whatever general tolerance the drawing inherits.

Now look at what the top flat gets. The distance from the top face down to the underside contacts is .164 −.020. That's a 0.020" band, half a millimetre. The datum width is held to ±.002, a 0.004" band. Five to one. Spuhr's line was "the tolerance to the top flat surface was deliberately made very large." The standard itself doesn't say why (its explanatory section is one page of generalities), so "deliberately" is Spuhr's word, not the Army's. But the number is real and it's right there on the page.

What that buys you: a pair of 45° jaws closing on the rail finds the same centre no matter how wide the rail is. The width band is about a tenth of a millimetre, so the mount lands in nearly the same place on any rail that meets the drawing, which is what Spuhr means by attaching "on a huge variety of rails with extremely tight control". The mount never needs to touch the top.

What it costs you depends on the jaws. Most real mounts have one fixed jaw and one that moves, so a change in rail width shifts the whole mount sideways a hair, and a jaw that rides on a single 45° face also rides up or down by the same amount. If the two sides don't match (the angles aren't controlled, remember), one side settles lower than the other and the mount rolls. Spuhr: "The mount will always sit slightly canted depending on the exact width of the rail. For most practical purposes, this doesn't matter at all... but it drives people with OCD absolutely crazy." I couldn't find anyone who's actually measured how much. A jaw shaped as a full V that captures the whole ridge wouldn't care about width at all, and Spuhr didn't say which kind they meant.

What NATO changed in 2009

Ten nations, working under NATO's Research and Technology Organization, spent the 2000s turning the American rail into a NATO one, with Aimpoint, Beretta, Colt, FN Herstal and Heckler & Koch at the table. The result, STANAG 4694 "NATO Accessory Rail", was approved on 8 May 2009 and published in March 2011.

The best primary source I found is the slide deck the working group's chairman, Per Arvidsson, presented at an NDIA symposium in May 2009. His four official differences from MIL-STD-1913, verbatim: "Metric drawing. Added some new necessary measurements and tolerances. Adjustment of some measurements. Reduction of straightness tolerances with approx 50%." I went looking for the slide that says "datum" and there isn't one. The change is on the next slide, under Recommendations:

"On a typical Mil-Std-1913 rail the grabber is clamping the rail on the v-angles."
"Our tests have shown that this does not provide good repeatability."
"We recommend instead that the top surface is used as a reference and alignment of the grabbers."
"Our tests have shown that this provides excellent repeatability."

Two things worth noticing. NATO's reason was repeatability, from their own tests; the word "cant" doesn't appear anywhere in the deck. And it's a recommendation attached to a drawing, which is why both mounting philosophies still coexist on the same rails today.

The deck has a two-panel graphic of the contact surfaces. The 1913 panel shows four contacts, all on the angled faces. The NATO panel shows three: the top face, plus the two underside bevels. The upper chamfers touch nothing. Thomas again: "the preferred mounting arrangement doesn't involve the topside bevels at all; they're still present mainly for backward compatibility."

Now put the NATO metric drawing next to Figure 1 of MIL-STD-1913 and convert. 0.835" is 21.2 mm, 0.748" is 19.0 mm, 0.164" is 4.17 mm, 0.108" is 2.74 mm, and the datum just got renamed from C to B. Two numbers moved a little, and they're the "adjustment of some measurements" from the slide: the neck went from 0.617" (15.67 mm) to 15.6 max, and the minimum height from 0.367" (9.32 mm) to 9.4 min. Otherwise it's the Picatinny rail in millimetres. What NATO added: a tolerance on the 45° angle for the first time (45° ±0°20'), a flatness call-out on the angled faces (0.06 mm per 100 mm), a corner radius (R1.5 max), and the top-to-undercut height tightened from a 0.51 mm band to 0.25 mm, which is half.

One more oddity. Before 4694 there was STANAG 2324, NATO's first attempt to simply adopt MIL-STD-1913 as written. Wikipedia's NATO Accessory Rail article calls it "Draft STANAG 2324"; the Picatinny rail article calls it a NATO standard outright; I couldn't find any NATO record of it being ratified. If the draft reading is right, the thing everyone called "the NATO rail" had no ratified NATO standard behind it for its first fourteen years.

The tolerance stack, with the numbers

Spuhr's argument against the three-surface approach is a stack argument. If your mount references the top flat and pulls on the two undercuts, the fit depends on the width (0.13 mm band) and on the height from the top face to the undercut contacts (0.25 mm band). The 45° geometry converts vertical to horizontal one-to-one: a jaw that has to reach 0.25 mm further down also has to travel 0.25 mm further in. Both jaws. So the width band plus the height band, counted once per jaw: 0.13 + 0.25 + 0.25 = 0.63 mm. Spuhr says "up to ~0.6 mm (0.024")". I ran it three times before I believed the arithmetic closed that neatly, and I haven't found anyone else who publishes the figure, so I take it as Spuhr's own sum. It's a worst case, it's only the rail's half of the stack (the mount has its own), and nobody has published the cam travel it's being compared against.

Why that's fine on a screw clamp and bad on a lever: a screw just turns a little further. A quick-detach lever has a fixed throw. Its cam can only take up so much, so most lever mounts ship with some way to soak up rail variation. Spuhr's own manual for their QDP mounts says the tension "is adjusted from the factory, but due to variances between different rails the tension may require re-adjustment prior to use." American Defense sells the same idea as a feature ("accommodate both in-spec and out-of-spec rail systems"), adjusted a flat or two at a time on an eight-sided nut until you feel resistance about two-thirds of the way closed. LaRue moved the adjustment to a click nut so it can be done without a wrench. Bobro went the other way and uses a spring so the lever self-adjusts. Four companies, four answers to the same 0.6 mm.

My own napkin math, for what it's worth: run a three-surface mount on an old inch-spec 1913 rail, where the same height carries a 0.51 mm band, and the stack roughly doubles.

So who's right, Spuhr or NATO?

Spuhr's post and the primary documents disagree in three places.

• Did NATO add the 0.25 mm, or halve it? Spuhr writes "you now also add 0.25 mm tolerance to the top flat" and closes with "NATO really missed an opportunity by not tightening up that loose 0.25 mm top-flat tolerance." But the 1995 drawing already carried that dimension at .164 −.020, which is 0.51 mm. The NATO drawing shows 4.17 −0.25. NATO cut it in half, and added the angle tolerance and the flatness control that never existed before. Whether 0.25 mm is still too loose for a QD lever is a fair opinion. "Didn't tighten it" isn't what the drawings say.
• Which surfaces? Spuhr describes NATO's scheme as "the two 45-degree sides plus the top flat", and the red drawing in the post marks all four angled faces. NATO's own graphic contacts the top plus the two underside bevels only. It's a fair simplification, but it isn't what NATO drew.
• Why did NATO do it? Spuhr: so "the mount will always sit perfectly straight and level." NATO: "repeatability", twice, and never cant.

Where nobody disagrees: the original scheme references the angled sides, the NATO scheme promotes the top face, three references stack more than two, and a lever mount feels every extra tenth. Spuhr has the architecture right and the history a little off. To be fair to them, it was an Instagram caption.

Two different things get called "error" in this argument, and it's worth separating them, because Spuhr and NATO are each talking about a different one. A mount that seats 0.0012" higher at the front than it did last time has tilted the line of sight by one minute of angle - Thomas's number, for a 4-inch mount base - and the group moves about an inch at 100 yards. That's repeatability, NATO's complaint, and 0.0012" is a third of the thickness of a sheet of printer paper. A mount that rolls a degree does almost nothing to a zeroed rifle at 100 yards; the sideways error is the bullet drop times the sine of the roll angle, so it only bites when you're dialing a lot of elevation. That's cant, Spuhr's complaint. Same unit, wildly different quantities.

What Spuhr does about it

Their answer is to not choose. Fixed (screw) mounts follow the NATO three-surface recommendation for straightness. QD mounts stick with the original two-surface approach "to keep tolerance stack as low as possible and ensure compatibility with as many rails as possible." The third photo in the post is one of their inspection fixtures, a rail-shaped steel gauge with a dial indicator riding on it, used to check that a QD mount sits straight across rail variation. Which is how a maker actually knows.

Their QDP manual is worth reading even if you never buy one, because it's a closed-loop procedure rather than a torque number, and it has two surprises. Step one: "Wipe the rail clean and apply a light coat of oil to it (CLP is fine)." Oil. On a friction clamp. Then seat the mount and push it forward, close the levers, and if one won't close, open it and back the T20 adjustment screw off two turns, close, snug it up, repeat. The pass/fail is step eight: "Grab hold of the mount and pull it to the rear. If the mount moves on the rail repeat from step 6." Ring screws go to "25 in/lb (2.9 Nm) or to the scope manufacturer's specification, if lower", and the scope gets levelled with a 10° wedge indexed off the scope's own flat bottom rather than by eye.

The slots, and which way to push

The standard defines the recoil groove in one sentence: "a groove on the accessory mounting rail that prevents forward and aft movement of an accessory." Figure 2 gives it .206 +.008 wide, .118 +.008 deep, on a .394 pitch. Read the notes under the figure, though: "Center to center dimension applies between adjacent grooves." The pitch is only controlled between neighbours. Nothing in the standard says where slot twelve ends up relative to slot one. That's why AR upper receivers carry those little T-numbers etched beside the slots - BCM sells them as "laser etched T-markings for addressing the remount of optics", which is the polite way of saying put it back in the same slot, because the slots aren't all in the same place on the next receiver.

The push-forward rule comes straight from the physics. On firing, the rifle jumps back into your shoulder. The scope and mount, by inertia, want to stay where they were, which relative to the rail is forward. So the recoil lug ends up against the muzzle-side wall of its slot whether you like it or not. Start it there and nothing moves. Start it against the rear wall and the first few shots walk it forward, with your zero along for the ride. Spuhr's manual makes "push it forward" step three of eight; Tract Optics says the same for rings: push them toward the muzzle before torquing. If you run two separate rings, push both, or one lug carries the whole load.

This is also where Weaver comes in. William Ralph Weaver started W.R. Weaver Co. in 1930, and his tip-off mount became the de facto rail for two generations of hunting rifles. There is no Weaver standard document. Nobody ever wrote one. The numbers everybody quotes (0.180" slots, spacing that varies) are what people measured off Weaver's products. A Weaver-spec accessory fits a Picatinny rail; go the other way and a full-width Picatinny lug won't drop into a Weaver slot. The failure mode people miss is the one that works: a 0.180" lug in a 0.206" to 0.214" slot has up to 0.034" (0.86 mm) of fore-aft slop, nothing binds, nothing warns you, and the optic walks to the front wall a shot at a time. That's my arithmetic from the two slot widths, not a published number.

Pro tips, most of them from people who make the stuff

• Ring-cap torque and clamp torque are different numbers on the same mount. Fix It Sticks' compiled chart (which openly says they "do not guarantee that they are correct") lists Nightforce at 25 in-lb on the rings and 68 on the crossbolts, Badger at 15 to 18 and 65. Put the clamp number on the ring caps and you crush a scope tube. Warne notes ring torque has been drifting down as tubes got thinner, some now 15 in-lb.
• A scope leveller levels the scope to the mount, not to the rail. Arisaka's wedge tool pushes an 11° ramp up against the turret housing - clever, and it faithfully reproduces whatever cant the mount itself has. Check the mount before you trust the level.
• Quick field check for "is this even a Picatinny rail": Håkan Spuhr's own three points, from a 2016 rant about brand-name rails that miss spec. Cross slots at least 5.23 mm, slot bottoms square, 10 mm centre to centre. "If there not is that, it's not a Picatinny, and not a NATO rail either!" For a real answer, EGW sells a certified go/no-go gauge with steps at nominal and ±0.002", and a Scandinavian shop that machines rails suggests ±0.05 mm as the line between made properly and not.
• Anodizing eats the tolerance. Type III hardcoat grows about 0.001" per surface for a 0.002" coating, half in, half out. Across a rail that's 0.002", which by my arithmetic is 40% of the entire width band. It's also a decent guess at why the width tolerance is minus-only and the slot tolerance is plus-only: machine to the low side, anodize back to nominal. That reading is mine; the standard doesn't explain itself.
• A lot of aftermarket rail is extruded, not milled. Two aluminium extruders advertise MIL-STD-1913 profile stock, and the shop then cuts slots, length and holes. So the 45° faces come from a die that wears, not a cutter you can compensate.
• Return to zero, measured: Recoil magazine ran one scope through eight QD mounts on one rifle, remounting between strings, and got an average shift of 0.07 MOA. Bobro 0.016, Burris 0.223, a fourteen-fold spread inside "they all return to zero." Their own caveat: "We would hardly call our exercise unassailably scientific." And that was one rifle; cross-rifle is exactly the case Spuhr's stack argument says will be worse.
• What cant costs, roughly: one trade magazine's figure is about 5 inches sideways at 1000 yards for 1° of cant. The better mental model is the one above, drop times the sine of the angle, so a flat-shooting load forgives cant and a heavy-drop one doesn't.
• The standard doesn't specify screw holes, rail length, or how many slots. Section 5.3 leaves length to the application; the figure notes leave slot count to the application. Every one of those is the gun maker's call.

How to read the drawing

If you've never read a GD&T drawing, the Picatinny profile is a good first one because it's small and strange. A triangle flag with a letter (C on the American drawing, B on the NATO one) marks the datum, the thing everything else is measured from. Here it hangs off a width, which means the datum is a centre plane between two features, not a surface, and that one choice is the whole story above. The ⌖ symbol in a box is position tolerance: how far the feature's centre may wander from the datum. The Ⓜ after it means "at maximum material condition", the size at which the part has the most metal (biggest for the rail head, smallest for a slot), and it comes with a gift called bonus tolerance: as the feature departs from that size, the allowed wander grows by the same amount on top of the stated figure. Thomas works this through for the rail base and lands on a "virtual condition" of .627", the widest space the base can ever occupy, and therefore the narrowest a grabber can be cut. A boxed number like .108 is a basic dimension, exact by definition, controlled by a tolerance elsewhere. "min." means the rail can be as tall as you like; the grabber just needs that much clearance. The parallelogram ▱ is flatness, and on the NATO drawing 0.06/100 means 0.06 mm over any 100 mm.

Those symbols exist as ordinary characters, by the way: position is U+2316 (⌖), the circled M is U+24C2 (Ⓜ), the flatness parallelogram is U+25B1 (▱). I went down that hole in the Unicode post a while back.

Where the rail came from

Weaver, 1930, as above. The push to standardize Weaver-type rails for the M16 family is usually credited to the A.R.M.S. company and Richard Swan in the early 1980s, and Wikipedia says Picatinny Arsenal asked for Swan's help (the same article spells him Swan in one paragraph and Swanson in the next, so I'm not going to pretend the record is tidy). The standard itself came out of the Armament Research, Development and Engineering Center at Picatinny in February 1995, with Houtsma's name on the Army's credit.

The name is older than the Army. A peak "then known as 'Piccatinny'" above Clifford Pond already had a colonial forge on it around 1749. The War Department showed up on 6 September 1880, called the place the Dover Powder Depot, and renamed it Picatinny Powder Depot four days later. The word is Lenape, and what it means depends on who you ask: "rugged cliff by water", "water by the hills", others. One local history puts it plainly - researchers disagree about the derivation and agree only that it's Native American.

What came after, and what didn't

M-LOK (Magpul, 2014) and KeyMod (VLTOR/Noveske, 2012) are the systems that replaced rail on the sides of handguards. Neither builds on 1913; both are negative-space slots cut into the handguard itself. KeyMod was published open, public domain. M-LOK is licensed free of charge but through an approval process, which is free-but-gated. When SOCOM had Naval Surface Warfare Center Crane A/B test them in 2017, the numbers were brutal: point-of-aim shift after remounting was 0.2 to 14.6 MOA for KeyMod versus 0.0 to 6.6 for M-LOK, and M-LOK carried 215% more load before failure. Repeatability separated them, the same thing NATO moved the datum for. And notice what the industry kept on top of every one of those handguards: a rail. Lights and grips moved to slots, but the optics stayed on the dovetail, because that's the only part with a datum tight enough for a sight.

NATO's next rail standard, STANAG 4740 "NATO Powered Accessory Rail" (2015), keeps the grabber sides of a normal NATO rail and hollows out the top surface for two lines of electrical contacts. 4694 promoted the top surface to primary reference in 2009, and 4740 cut a trench in it six years later.

The furthest departure is the Zeiss rail, on the scope rather than the gun since 1990: a stepless dovetail machined into the underside of the tube, no slots, no index positions at all, the opposite philosophy to a 10 mm pitch. Its cousin, the Schmidt & Bender convex rail, deliberately lets you tilt the reticle up to 1°, a degree of freedom the Picatinny architecture can't offer. Small one for the pile: the physical STANAG 4694 sample NATO circulated on a plaque was built by B&T.

Where this sits in the error budget

None of this is the main reason a zero moves. Receiver fit, a flat-top that flexes, the scope's own erector, ring alignment, the ammunition: any of those can swamp a rail-to-mount interface that's holding a tenth of a millimetre. Thomas says as much in an aside I nearly skipped: rail straightness tolerances are usually such that "zero will be lost just by moving the mount forward on the rail." The interface argument matters at the level where people are chasing that last fraction of a minute, or swapping one optic between rifles and expecting it to come back. Everyone else gets to enjoy it as the answer to a question they didn't know they had.

If you've measured your own rails, or have a QD mount that behaves on one gun and not another, I'd like to hear about it in the comments. Bonus points if you have ever put calipers on a "Picatinny" rail and found Weaver.

Glossary

• Picatinny rail — the slotted dovetail rail defined by MIL-STD-1913; named after Picatinny Arsenal, New Jersey.
• MIL-STD-1913 — the 1995 US military standard that dimensions the rail profile and recoil groove. 1913 is a document number, not a year.
• STANAG — NATO Standardization Agreement, a document member nations agree to follow.
• NATO Accessory Rail (NAR) — the metric rail defined by STANAG 4694 (approved 2009, published 2011), backward compatible with Picatinny.
• Datum — the theoretical reference (a plane, an axis, or a centre plane) that other dimensions are measured from. The physical surfaces that establish it are the datum features.
• GD&T — Geometric Dimensioning and Tolerancing, the symbolic language on engineering drawings for controlling shape, position and orientation.
• Tolerance — the allowed variation on a dimension. Written as ±, or as a one-sided limit like −0.13.
• Tolerance stack — the total variation a fit has to absorb when several toleranced dimensions add up.
• Position tolerance (⌖) — how far a feature's centre may deviate from its true position relative to a datum.
• MMC (Ⓜ) — maximum material condition, the size at which a part has the most material: the largest permitted size for an external feature like the rail head, the smallest for an internal one like a slot. Used with position tolerance to grant bonus tolerance as the part departs from it.
• Virtual condition — the worst-case boundary a feature can occupy once size and position tolerance are combined.
• Basic dimension — a boxed, theoretically exact value on a drawing; its variation is controlled by a geometric tolerance elsewhere.
• Flatness (▱) — how far a surface may deviate from a perfect plane; 0.06/100 means 0.06 mm over any 100 mm.
• Dovetail — the wedge-shaped cross-section of the rail head; the standard's own word for it.
• Recoil groove — the cross slot on the rail; the standard says it "prevents forward and aft movement of an accessory". The lug on the mount that sits in it is the recoil lug.
• Grabber — NATO's word for the clamp on a mount that grips the rail.
• QD mount — quick-detach mount, clamped with a cam lever instead of screws; fixed throw, needs a pre-set tension.
• Crossbolt — the bolt that pulls a screw-type mount's clamp across the rail; its torque is a different number from the ring-cap screws.
• Cant — roll of the scope or mount relative to the rifle; makes elevation adjustments push shots sideways.
• Return to zero (RTZ) — how close a remounted optic's point of impact lands to where it was before removal.
• MOA — minute of angle, 1/60 of a degree; about 1.047 inches at 100 yards, 10.47 at 1000.
• T-marks — numbered marks etched beside the slots on a flat-top receiver so an optic can go back in the same slot.
• Weaver rail — the earlier commercial rail the Picatinny grew out of; narrower 0.180" slots, no published standard.
• Type III anodizing (hardcoat) — a thick, hard anodic coating on aluminium; roughly half of it grows outward and adds to the part's dimensions.
• CLP — cleaner, lubricant, preservative; the general-purpose gun oil.
• in-lb / N·m — inch-pounds and newton-metres, the two torque units in this post. 25 in-lb is about 2.8 N·m (Spuhr's manual rounds it to 2.9).
• M-LOK / KeyMod — slot-based handguard attachment systems from Magpul (2014) and VLTOR/Noveske (2012); neither derives from MIL-STD-1913.
• NDIA — National Defense Industrial Association, whose symposia host briefings like the 2009 NATO rail deck.

Sources

• Soldier Systems Daily — "Picatinny – How It Really Works!" (16 Feb 2026) — the repost of Spuhr's Instagram text, the two marked-up drawings and the fixture photo that started this.
• Spuhr on Instagram — the original post — age-gated, needs a login to view.
• MIL-STD-1913 (AR), 3 February 1995 (PDF) — the standard itself, public domain; Figure 1 profile, Figure 2 recoil groove. Official index entry at DLA ASSIST.
• EverySpec — MIL-STD-1800 to 1999 index — neighbours 1911 (1993) and 1916 (1996), which is how you know 1913 is a document number.
• Per G. Arvidsson, NATO Army Armaments Group — NDIA Infantry Small Arms Symposium briefing, May 2009 (PDF) — the four official differences, the recommendation slide, and the rail/grabber interface graphic.
• STANAG 4694 metric reference drawing (Wikimedia Commons) — 21.2 −0.13, 19 ±0.05, 4.17 −0.25, 45° ±0°20', flatness 0.06/100.
• NATO Accessory Rail presentation sample (Wikimedia Commons) — the B&T plaque.
• Thomas Mechanical Design — "Picatinny and NATO Rails - GD&T" (2016) — the datum-C analysis, the halved tolerance, the 1 MOA = 0.0012" figure, the .627" virtual condition, the straightness aside.
• Wikipedia — NATO Accessory Rail — dates, participants, "Draft STANAG 2324", STANAG 4740.
• Wikipedia — Picatinny rail — A.R.M.S./Swan history, Picatinny Arsenal asking for Swan's help, slot dimensions, STANAG 2324 called a NATO standard.
• Wikipedia — Weaver rail mount — W.R. Weaver Co. 1930, 0.180" slots, inconsistent spacing.
• Weaver — Picatinny vs Weaver rails — the one-way fit rule from the maker's side.
• US Army — Picatinny engineer recognized for developing venerable Picatinny rail (2014) — Gary Houtsma and the Order of Saint Maurice.
• Picatinny Arsenal — official history — Dover Powder Depot, 1880, the renames.
• NJ Skylands — History of Picatinny Arsenal — the 1749 forge and the disputed etymology.
• Wikipedia — Picatinny Arsenal — the Lenape glosses.
• Spuhr — ISMS QDP mount manual (2023, PDF) — oil the rail, push forward, lever adjustment loop, 25 in-lb (2.9 N·m) ring screws, 10° wedge.
• The Firearm Blog — Violations of the Picatinny rail MIL-STD-1913 (2016) — Håkan Spuhr's three-point field check.
• GlobalSpec — STANAG 4694 listing — March 2011 publication.
• American Defense Mfg — tech support and Accu-Shot — installing and adjusting the ADM lever — the eight-flat nut, two-thirds-closed resistance, in-spec and out-of-spec rails.
• The Firearm Blog — LaRue Click Adjust Nut (2018) — wrench-free lever adjustment.
• GunCreed — Bobro Engineering's self-adjusting lever — the spring approach.
• Recoil — Guide to quick-detach scope mounts — the eight-mount return-to-zero test, 0.07 MOA average.
• Arisaka Defense — Optic Leveler guide — the 11° wedge, levelling the scope to the mount.
• Shooting Sports Retailer — Why rifle and scope cant matter — the 5 inches at 1000 yards figure.
• Tract Optics — How to properly mount a rifle scope — push the mount forward before torquing.
• Fix It Sticks — compiled torque specs (PDF) — ring vs crossbolt numbers, with their own disclaimer.
• Warne — Torque and scope mounts — ring torque drifting down to 15 in-lb.
• EGW — Certified Picatinny rail gauge — the ±0.002" go/no-go steps.
• Picatinny.se — Picatinny rail dimensions — the ±0.05 mm shop threshold, undefined screw holes.
• Anoplate — The impact of anodize on dimensions and Precision Coating — Type III hardcoat — half in, half out growth.
• Alcobra Metals — Picatinny rail extrusion and Minalex — Picatinny rails — extruded profile stock.
• Bravo Company — M4 upper with laser T-markings — "for addressing the remount of optics".
• Soldier Systems Daily — USSOCOM KeyMod vs M-LOK test at NSWC Crane (2017) — 0.2 to 14.6 vs 0.0 to 6.6 MOA, 215% load.
• Wikipedia — M-LOK, Magpul — M-LOK, Wikipedia — KeyMod — the two slot systems and their licensing (Magpul's licence PDF, cited by Wikipedia, was offline when I checked).
• T-Worx — Intelligent Rail — the powered rail behind STANAG 4740.
• Wikipedia — Zeiss rail — the stepless scope-side rail and the S&B convex rail.
• NSSF — Minute of angle — 1 MOA is 10.47" at 1000 yards.