The orientation problem
A watch on a wrist spends its day in continuously changing orientations: dial up on a desk, crown down with the arm hanging, every angle in between as you type, drive, gesture, and sleep. Each orientation changes the mechanical conditions at the most sensitive point in the watch. In horizontal positions, the balance spins on the polished tip of its staff pivot — minimal contact, minimal friction. In vertical positions, the pivot lies over and runs on its cylindrical side — more contact area, more friction, visibly lower balance amplitude (typically 20 to 40 degrees less). Add the effect of gravity on any residual imbalance in the balance itself and on the hairspring's off-centre breathing, and the result is unavoidable: a mechanical movement runs at slightly different rates in different orientations. This positional variation is one of the largest honest sources of rate error in daily wear — and the clearest single measure of how carefully a movement was made.
What the six positions reveal
A thorough test measures rate in six positions — dial up, dial down, crown up, crown down, crown left, crown right — and the spread tells the story. A well-adjusted movement in good health might hold the spread to 5–10 seconds per day; an unadjusted movement can spread 20–40. COSC certification bounds the variation across its five tested positions as one of its seven criteria; the great American railroad grades advertised "adjusted to five positions" as a headline feature a century ago, because the work was expensive and the buyers understood what it bought.
The positional spread is the single most useful number about how a watch will behave on a wrist, because wrist wear is all positions: the rate you live with is a weighted average across them. A movement that tests +2 dial-up but −12 crown-down will disappoint an owner who only ever saw the dealer's flat timing slip. This is also where a knowledgeable owner can cheat constructively: because the positional rates differ predictably, resting the watch overnight in a chosen position — crown up if it runs fast, dial up if slow — nudges the 24-hour average. Watchmakers have quietly prescribed this for decades; it is the only free regulation in horology.
Where the differences come from, and how they are fought
Three mechanisms dominate. Pivot friction, as above — attacked with finer pivots, better polish, cap jewels, and healthy lubrication; this is why low amplitude from old oil makes positional spread worse. Balance poise: if the balance's mass is not perfectly even around its rim, gravity torques it differently in each vertical position; watchmakers remove invisible amounts of material (or adjust screws) to "poise" it, statically on rubies and then dynamically on the timing machine. Hairspring geometry: a flat spiral breathes asymmetrically and shifts the system's centre of gravity as it expands and contracts; precise centring, terminal curves, and the Breguet overcoil exist to tame exactly this. The index regulator's curb pins add a fourth, gratuitous source — their contact with the spring varies with orientation — which is why free-sprung balances (Rolex, Patek Philippe, Lange, and most chronometrically serious calibres) eliminate the pins entirely and adjust rate through inertia masses on the rim instead. Free-sprung movements characteristically post tighter positional spreads, and that, not snobbery, is the argument for them.
Ask for (or make) rate measurements in at least four positions: dial up, dial down, crown up, crown down. Healthy pattern: horizontal positions nearly identical, vertical positions a few seconds apart and a few seconds different from horizontal, amplitude dropping no more than ~40 degrees from horizontal to vertical. Warning patterns: dial-up and dial-down differing sharply (bent pivot or cap-jewel issue), one wild vertical outlier (poise error or hairspring touching), or spread that grows dramatically at half-wind (isochronism and torque trouble). Five minutes of data tells you more than any adjective in a listing.
The tourbillon: the solution that doesn't apply
The tourbillon was invented for exactly this problem — in a pocket watch, which rode for hours in one fixed vertical orientation, accumulating one direction's error all day. Rotating the entire escapement once a minute averages the vertical-position errors toward zero, and in its native context it genuinely worked. A wristwatch, however, performs the averaging by itself: ordinary wear shuffles the watch through orientations all day long. The tourbillon on a wrist solves a problem the wrist has already solved — the technical basis for the well-supported conclusion that tourbillons do not improve wristwatch accuracy in normal use. They remain magnificent craft demonstrations, and the tourbillon article treats them on those honest terms.
Accuracy in one position is easy to achieve and easy to misunderstand. Consistency across positions is hard to achieve and impossible to fake — and it is the consistency, not the headline rate, that a wrist actually experiences, because a wrist is never in one position.