Technical note 7

TN-7: Why Joints Go Slack

Creep, movement, and preload retention in a wooden chair. First draft 17 Aug 2026, prompted by a printed barrel nut retainer that snapped and by the question it raised: not how strong is the joint, but how long does it stay tight. Status: DRAFT, four sources in the queue.

The claim

A bolted joint holds because something is stretched. Preload is stored displacement, nothing more, and every mechanism in this note is a way of stealing displacement back. In wood at furniture scale the amount stored is so small that ordinary seasonal movement erases it many times over, so any joint whose life depends on staying tight will not stay tight. Design the preload out, or give it somewhere elastic to live.

Where the preload lives, and how little of it there is

An M6 steel bolt with a 40 mm grip has an axial stiffness of about 100 kN per millimeter [DERIVED: k = EA/L, E = 200 GPa, tensile stress area 20.1 mm2]. So

preload 1000 N  is held in 0.010 mm of stretch
preload 2000 N  is held in 0.020 mm
preload 3000 N  is held in 0.030 mm

Thirty microns. That is the entire elastic reserve of a joint tightened to 3 kN. Anything that shortens the clamped stack by thirty microns leaves the bolt exactly finger tight.

What steals it

MOISTURE, and it is not close. White oak moves roughly 0.37 percent of its width per point of moisture content tangentially and about half that radially [STANDARD: FPL Wood Handbook; pin the table before print]. A realistic indoor swing of three points across an 18 mm thickness is therefore about 0.20 mm tangential, 0.10 mm radial [DERIVED]. Against a stretch of 0.030 mm, one season is seven times the whole preload. The joint does not loosen slightly; it goes dead slack and then takes the load back up as a loose fit.

THE CURIOUS FACT, and the most useful one in this note: wood barely moves along the grain. Longitudinal shrinkage runs about 0.1 to 0.2 percent green to oven dry, against 4 to 8 percent radial and 6 to 12 tangential [STANDARD: FPL Wood Handbook], so per point of moisture the along-grain figure is on the order of one fiftieth of the cross-grain one. The same three-point swing moves 18 mm of oak by 0.003 mm along the grain, which is a tenth of the stretch rather than seven times it. A preload path that runs along the grain is stable. A preload path that squeezes across the grain is doomed. This is the single most actionable sentence here.

THERMAL, small but not nothing at this scale. Over a 40 mm grip and a 20 K swing: steel 0.010 mm, bronze 0.014, aluminium 0.018, zinc 0.022 [DERIVED from standard coefficients]. A steel bolt inside an aluminium sleeve differs by 0.009 mm across that swing, which is a third of the preload in the example above. Mixing metals inside the grip is a real effect, not a rounding error.

CREEP, which is the slow version of all of the above. - Wood in compression perpendicular to the grain creeps substantially, and it creeps far faster while the moisture content is cycling than it does at constant humidity. The effect has a name, mechano-sorptive creep, and it is why a joint left alone through a year of seasons loosens more than the same joint held at constant humidity [STANDARD in the timber literature; needs a citation before print]. - Thermoplastics creep at room temperature under sustained stress. PETG and nylon are not preload materials at any timescale that matters here. - Zinc die casting alloys creep measurably at room temperature. This matters because zinc is otherwise the obvious process for a small complex retainer. [UNVERIFIED: get creep data for Zamak 3 and 5 before specifying.] - Steel, aluminium, brass and bronze do not creep meaningfully at room temperature. In the load path they are the honest choices.

And the preload has to be small anyway

Oak crushes across the grain at around 5 to 8 MPa at the proportional limit [STANDARD: FPL; pin the value]. A 3 kN preload under an ordinary M6 washer, say 14 mm across, puts 26 MPa on the wood, which crushes on the spanner and then keeps crushing for a decade [DERIVED]. To stay under about 5 MPa at 3 kN the washer has to be 30 mm across; at 2 kN, 24 mm. So the practical choices are a large washer or plate, or a modest preload, or both, and either way the stored stretch is smaller than the arithmetic above suggests.

The strategies, in order of how much they help

  1. DESIGN THE PRELOAD OUT. A joint that works in bearing and shear does not care what the moisture did. The block chair holds because screws bear in wood, not because anything is tight; the splined and pinned miter holds because pins bear in double shear. Neither has a preload to lose. This is the same conclusion the longevity ranking reached from the other end: compression ages best.
  2. TAKE THE WOOD OUT OF THE GRIP. A rigid sleeve through the full thickness lets the bolt clamp metal to metal while the wood is merely captured. The cross-grain movement then happens beside the preload path instead of inside it. This is the preload sleeve, and it is what makes a bolted binding credible at century scale.
  3. RUN THE PRELOAD ALONG THE GRAIN where a sleeve is not possible, and take the fiftyfold reduction in movement for free.
  4. ADD ELASTIC RESERVE. A disc spring giving 0.4 mm of travel at working load has a stiffness near 7.5 kN/mm against the bolt's 100. The same 0.05 mm loss then costs 375 N out of 3000, twelve percent, instead of 5 kN, which is one and two thirds times the entire preload [DERIVED]. Belleville washers are not a refinement here, they are the difference between a joint that survives a season and one that does not.
  5. SPREAD THE BEARING. Creep rate rises with stress, so a bigger washer buys time as well as capacity.
  6. CHOOSE LOW-CREEP MATERIALS in the load path: no polymer, no zinc.
  7. RETIGHTEN. Honest, traditional, and effective, and timber engineering assumes it. But an owner will not do it, and a chair that needs it has moved its maintenance burden onto a person who does not know it exists. Fine for a shop chair; not the basis for a hundred-year claim.

The proven precedent: the drawbored peg

The tradition solved this problem in wood, and its solution has the longest record of anything in the note. A drawbored mortise and tenon is bored so the hole through the tenon sits a millimetre or two closer to the shoulder than the holes through the cheeks; when a tapered wooden peg is driven, it must bend to pass through the offset, and that bend pulls the shoulder hard against the mortise and holds it there [VERIFIED that drawboring works by an offset that draws the joint tight, against the joinery sources, 19 Aug 2026; see TN-4].

Read against this note, the drawbore does four right things at once. It DESIGNS THE PRELOAD OUT of any metal path (strategy 1): the load is carried by the shoulder bearing wood on wood, not by a stretched bolt. It STORES what preload there is as elastic BEND in the peg (strategy 4): a wooden peg bent across an offset is a soft spring with a lot of travel, so the thirty-micron problem of the stiff bolt does not arise; the joint can give up a few tenths of a millimetre and still be tight. It runs the drawing force largely ALONG the tenon's grain (strategy 3). And it is trivially RENEWABLE: a slack peg is knocked out and a fresh one driven, no proprietary part, which is the repairability doctrine in its oldest form. The cost is that it wants solid wood, not plywood, so it belongs to the hardwood keel and any solid-wood variant, not to the panels. It is the answer this whole note is circling, arrived at by carpenters a few thousand years before the arithmetic above.

The rule this yields

Rank every binding by one question: does it depend on stored displacement? Blocks and screws, no. Pinned spline, no. Tapped metal bracket, yes but with a stiff short grip and no wood inside it. Bolt through wood with a plastic or zinc part in the stack, yes and badly. That single question sorts the ladder more sharply than material or cost does.

Verification queue

FPL Wood Handbook: shrinkage coefficients, longitudinal against transverse, and compression perpendicular to grain at the proportional limit. A citation for mechano-sorptive creep. Room-temperature creep data for Zamak 3 and 5. Disc spring travel and load figures from DIN 2093 rather than from memory.

Revisions

17 Aug 2026: first draft. 19 Aug 2026: added "The proven precedent: the drawbored peg," at KU's request, after the fastener thread (TN-4).

Technical Notes of the Hundred-Year Chair: public from first draft, refined iteratively, forever. A claim is VERIFIED, STANDARD, DERIVED, or UNVERIFIED, and says which. The register · the verification ledger · hundredyearchair.com